CN115982323A - Big data analysis method and artificial intelligence system applied to cloud online service - Google Patents
Big data analysis method and artificial intelligence system applied to cloud online service Download PDFInfo
- Publication number
- CN115982323A CN115982323A CN202310140645.4A CN202310140645A CN115982323A CN 115982323 A CN115982323 A CN 115982323A CN 202310140645 A CN202310140645 A CN 202310140645A CN 115982323 A CN115982323 A CN 115982323A
- Authority
- CN
- China
- Prior art keywords
- session
- conversation
- emotion
- intention
- template
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the application provides a big data analysis method and an artificial intelligence system applied to cloud online service, by mining a conversation emotion text and a conversation intention text in online conversation behavior data, compared with the online conversation behavior characteristics of the whole coarse granularity adopted in the prior art, a conversation emotion knowledge vector and a conversation intention knowledge vector can more accurately evaluate the characteristic difference between texts, so that the reliability of online conversation behavior data search is improved, and compared with the prior art, a characteristic vector is extracted for each template conversation behavior independently, by acquiring template conversation mining characteristics including the conversation intention knowledge vector and the conversation emotion knowledge vector, the characteristic vector of online conversation behavior data with larger data volume can be more accurately extracted, the effect of online conversation behavior data search is further improved, and further the follow-up recommendation of related conversation behavior users is facilitated.
Description
Technical Field
The application relates to the technical field of AI (artificial intelligence), in particular to a big data analysis method and an artificial intelligence system applied to cloud online service.
Background
With the continuous application of internet information technology, various internet service providers can provide various requirements for online users by arranging cloud online services (such as cloud online e-commerce services, cloud online medical services, cloud online e-internet services, cloud online logistics services, cloud online campus services, cloud online industrial services and the like). In the process of using the cloud online services, a user usually has some intention or doubts about initiating corresponding online conversation scene activities, and in the related technology, the online conversation scene activities are usually subjected to large data analysis and mining of online conversation behavior characteristics with overall coarse granularity so as to facilitate subsequent information push.
Disclosure of Invention
In order to overcome at least the above shortcomings in the prior art, the present application aims to provide a big data analysis method and an artificial intelligence system applied to cloud online service.
In a first aspect, the present application provides a big data analysis method applied to a cloud online service, which is applied to an artificial intelligence system, and the method includes:
according to an online session scene designated in cloud online service, performing session emotion analysis on a plurality of template session behaviors contained in template online session behavior data respectively, and determining one or more session emotion texts and corresponding session emotion knowledge vectors contained in the plurality of template session behaviors;
according to an online conversation scene designated in the cloud online service, respectively carrying out conversation intention analysis on the template conversation behaviors, and determining one or more conversation intention texts and corresponding conversation intention knowledge vectors covered by the template conversation behaviors;
determining one or more template session mining characteristics corresponding to the template online session behavior data according to the determined session emotion texts and the determined session intention texts, wherein each template session mining characteristic comprises a plurality of session intention knowledge vectors and a plurality of session emotion knowledge vectors of an online session user;
and determining target online conversation behavior data from the candidate online conversation behavior data according to the determined template conversation mining characteristics and target conversation mining characteristics corresponding to the candidate online conversation behavior data in the cloud online behavior big data of the cloud online service, and pushing a first user corresponding to the template online conversation behavior data to a second user corresponding to the corresponding determined target online conversation behavior data.
In one possible implementation of the first aspect, the method further comprises:
acquiring a user interaction window established by the second user and the pushed first user, and acquiring user interaction text data in the user interaction window;
mining an intentional knowledge point of progress on the user interactive text data to generate an intentional knowledge point of progress;
and performing content pushing on the user group where the first user and the second user are located based on the advanced intention knowledge point.
In a possible implementation manner of the first aspect, the step of performing advanced wisdom point mining on the user interaction text data to generate an advanced wisdom point includes:
analyzing an interactive semantic keyword directed graph from the user interactive text data, wherein the interactive semantic keyword directed graph comprises W groups of interactive semantic keywords with semantic relation, and W is an integer not less than 1;
acquiring an attention keyword directed graph by using the interactive semantic keyword directed graph, wherein the attention keyword directed graph comprises W groups of attention keywords with semantic relation;
utilizing the interactive semantic keyword directed graph and utilizing a first topic feature analysis unit in a neural network model to mine a first topic feature vector sequence, wherein the first topic feature vector sequence comprises W first topic feature vectors;
mining a second topic feature vector sequence by using a second topic feature analysis unit in the neural network model by using the attention keyword directed graph, wherein the second topic feature vector sequence comprises W second topic feature vectors;
determining intention knowledge point prediction information corresponding to the interactive semantic keyword digraph by using the first theme feature vector sequence and the second theme feature vector sequence and an intention knowledge point prediction unit in the neural network model;
and determining the advanced intention knowledge point of the interactive semantic keyword directed graph by using the intention knowledge point prediction information.
For example, in a possible implementation manner of the first aspect, the determining, by using the first topic feature vector sequence and the second topic feature vector sequence, intentional knowledge point prediction information corresponding to the interactive semantic keyword directed graph by using an intentional knowledge point prediction unit in the neural network model includes:
acquiring W first interactive knowledge point representations by using the first topic feature vector sequence and a first interactive knowledge point classification unit in the neural network model, wherein each first interactive knowledge point representation corresponds to one first topic feature vector;
acquiring W second interactive knowledge point representations by using the second topic feature vector sequence and a second interactive knowledge point classification unit in the neural network model, wherein each second interactive knowledge point representation corresponds to one second topic feature vector;
combining the W first interactive knowledge point representations and the W second interactive knowledge point representations to obtain W target interactive knowledge point representations, wherein each target interactive knowledge point representation comprises a first interactive knowledge point representation and a second interactive knowledge point representation;
utilizing the W target interaction knowledge point representations, and determining intention knowledge point prediction information corresponding to the interactive semantic keyword digraph by using the intention knowledge point prediction unit in the neural network model.
For example, in a possible implementation manner of the first aspect, the obtaining W first interactive knowledge point characterizations by using a first interactive knowledge point classification unit in the neural network model includes:
regarding each group of first topic feature vectors in the first topic feature vector sequence, acquiring first staged coding information by utilizing staged coding units in the first interactive knowledge point classification unit, wherein the first interactive knowledge point classification unit belongs to the neural network model; for each set of first topic feature vectors in the first sequence of topic feature vectors: acquiring first integral coding information by utilizing an integral coding unit in the first interactive knowledge point classification unit; acquiring first option information by utilizing the first staged coding information and the first overall coding information and utilizing a characteristic item selection branch based on expected cross entropy in the first interactive knowledge point classification unit;
acquiring a first interactive knowledge point representation by utilizing the first option information and the first topic feature vector and utilizing a first integral coding unit in the first interactive knowledge point classification unit;
acquiring W second interactive knowledge point representations by using the second topic feature vector sequence and using a second interactive knowledge point classification unit in the neural network model, wherein the acquiring comprises the following steps:
for each set of second topic feature vectors in the second sequence of topic feature vectors: acquiring second phased encoding information by utilizing a phased encoding unit in the second interactive knowledge point classification unit, wherein the second interactive knowledge point classification unit belongs to the neural network model; acquiring second integral coding information by utilizing an integral coding unit in the second interactive knowledge point classification unit; acquiring second option information by utilizing the second staged coding information and the second overall coding information and utilizing a characteristic item selection branch based on expected cross entropy in the second interactive knowledge point classification unit;
and acquiring a second interactive knowledge point representation by utilizing a second integral coding unit in the second interactive knowledge point classification unit by utilizing the second option information and the second theme feature vector.
In a second aspect, an embodiment of the present application further provides an artificial intelligence system, where the artificial intelligence system includes a processor and a machine-readable storage medium, where a computer program is stored in the machine-readable storage medium, and the computer program is loaded and executed in combination with the processor to implement the big data analysis method applied to the cloud online service in the first aspect.
By adopting the technical scheme of any aspect, the conversation emotion text and the conversation intention text in the online conversation behavior data are mined, compared with the online conversation behavior characteristics with overall coarse granularity adopted in the prior art, the conversation emotion knowledge vector and the conversation intention knowledge vector can more accurately evaluate the characteristic difference between the texts, so that the reliability of online conversation behavior data search is improved, compared with the prior art, one characteristic vector is extracted for each template meeting behavior independently, and by obtaining the template conversation mining characteristics including the conversation intention knowledge vector and the conversation emotion knowledge vector, the characteristic vector of online conversation behavior data with larger data volume can be more accurately extracted, the effect of online conversation behavior data search is further improved, and further the follow-up recommendation of related conversation behavior users is facilitated.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and it will be apparent to those skilled in the art that other related drawings can be obtained by combining these drawings without inventive efforts.
Fig. 1 is a schematic flow chart of a big data analysis method applied to a cloud online service according to an embodiment of the present application;
fig. 2 is a schematic block diagram of a structure of an artificial intelligence system for implementing the above big data analysis method applied to the cloud online service according to the embodiment of the present disclosure.
Detailed Description
The following description is presented to enable one of ordinary skill in the art to make and use the present application and is provided in the context of a particular application and its requirements. It will be apparent to those skilled in the art that various modifications can be made to the disclosed embodiments and that the general principles defined in this application may be applied to other embodiments and applications without departing from the spirit and scope of the application. Thus, the present application is not limited to the described embodiments, but should be accorded the widest scope consistent with the claims.
Step S101, according to an online conversation scene designated in cloud online service, conversation emotion analysis is respectively carried out on a plurality of template conversation behaviors covered in template online conversation behavior data, and one or more conversation emotion texts covered by the plurality of template conversation behaviors and corresponding conversation emotion knowledge vectors are determined.
In this embodiment, the online session scenario may refer to a service scenario applied in the online session, such as an e-commerce query scenario, a medical query scenario, and the like. The online session behavior data may refer to behavior data generated by any user during the process of initiating an online session in the specified online session scenario, for example, dialog behavior data generated during the process of initiating an online session with an online customer service in the cloud online service of the e-commerce service platform in the e-commerce query scenario. The conversation emotion text can refer to text content with emotion expressions (such as positive and negative) in a plurality of template conversation behaviors covered in the template online conversation behavior data, and can be analyzed and determined through a natural semantic processing model, and the conversation emotion knowledge vector can be used as a text keyword vector for representing the conversation emotion text.
Step S102, according to an online conversation scene designated in the cloud online service, conversation intention analysis is respectively carried out on the template conversation behaviors, and one or more conversation intention texts and corresponding conversation intention knowledge vectors covered by the template conversation behaviors are determined.
Similarly, the conversation intention text may refer to text content in which a conversation intention (e.g., a core principal point desired to search for a certain commodity category) exists among a plurality of template conversation activities included in the template online conversation activity data, and may be determined by analysis of a machine learning model, and the conversation intention knowledge vector may be used as a text keyword vector representing the conversation intention text.
Step S103, determining one or more template session mining characteristics corresponding to the template online session behavior data according to the determined session emotion texts and the session intention texts, wherein each template session mining characteristic comprises a plurality of session intention knowledge vectors and a plurality of session emotion knowledge vectors of an online session user.
And step S104, determining target online conversation behavior data from the candidate online conversation behavior data according to the determined each template conversation mining characteristic and the target conversation mining characteristic corresponding to each candidate online conversation behavior data in the cloud online behavior big data of the cloud online service, and pushing a first user corresponding to each template online conversation behavior data to a second user corresponding to the corresponding determined target online conversation behavior data.
The determined template session mining characteristics can be used for tasks such as repeated analysis of online session behavior data and correlation search of the online session behavior data. The target online conversation behavior data with the corresponding target conversation mining characteristics matched with the template online conversation behavior data is online conversation behavior data meeting the related business requirements.
Therefore, by mining the conversation emotion text and the conversation intention text in the online conversation behavior data, compared with the online conversation behavior characteristics with overall coarse granularity adopted in the prior art, the conversation emotion knowledge vector and the conversation intention knowledge vector can more accurately evaluate the characteristic difference between the texts, so that the reliability of online conversation behavior data search is improved.
In some exemplary design ideas, each conversation emotion text can be loaded into a target conversation emotion coding model meeting a model convergence condition to determine a corresponding conversation emotion knowledge vector, and illustratively, one or more conversation emotion texts covered by a plurality of template conversation behaviors are loaded into the target conversation emotion coding model meeting the model convergence condition to determine the corresponding conversation emotion knowledge vector. The updating process of the model weight parameters of the target session emotion coding model is shown in the following.
Taking the template conversation behavior as the online conversation behavior 1 as an example, for example, the online conversation behavior 1 includes a conversation emotion text a, a conversation emotion text B, and a conversation emotion text C, the conversation emotion text a is loaded into a target conversation emotion encoding model satisfying the model convergence condition to obtain a conversation emotion knowledge vector a, the conversation emotion text B is loaded into a target conversation emotion encoding model satisfying the model convergence condition to obtain a conversation emotion knowledge vector B, and the conversation emotion text C is loaded into a target conversation emotion encoding model satisfying the model convergence condition to obtain a conversation emotion knowledge vector C.
In some exemplary design ideas, according to an online conversation scene specified in the cloud online service, conversation intention analysis is performed on a plurality of template conversation behaviors, and a process of determining one or more conversation intention texts covered by the plurality of template conversation behaviors is similar to step S101. According to an online conversation scene designated in the cloud online service, conversation intention analysis is respectively carried out on the template conversation behaviors, and in an implementation process of determining one or more conversation intention texts covered by the template conversation behaviors, a conversation intention analysis model trained in the related technology can be adopted for analysis.
In some exemplary design ideas, the respective session intention texts may be loaded into the target session intention coding model satisfying the model convergence condition, and a corresponding session intention knowledge vector is determined, and, for example, one or more session intention texts covered by the plurality of template session behaviors may be loaded into the target session intention coding model satisfying the model convergence condition, and a corresponding session intention knowledge vector is determined. The process of updating the model weight parameters of the target session intention coding model is similar to the process of updating the model weight parameters of the target session emotion coding model, which is specifically referred to below.
Still taking the template conversation behavior as the online conversation behavior 1 as an example, for example, the online conversation behavior 1 includes a conversation intention text a, a conversation intention text B, and a conversation intention text C, the conversation intention text a is loaded into a target conversation intention coding model satisfying a model convergence condition to obtain a conversation intention knowledge vector a, the conversation intention text B is loaded into a target conversation intention coding model satisfying a model convergence condition to obtain a conversation intention knowledge vector B, and the conversation intention text C is loaded into a target conversation intention coding model satisfying a model convergence condition to obtain a conversation intention knowledge vector C.
In some exemplary design concepts, when step S103 is executed, the following embodiments may be implemented:
and step S1031, determining conversation emotion intention relation information between each conversation emotion text and each conversation intention text according to each determined conversation emotion text and each determined conversation intention text.
In some exemplary design concepts, for each template conversation activity of the plurality of template conversation activities, the following steps are performed:
taking the template conversation behavior k as an example, the template conversation behavior k is any one of a plurality of template recording conversation behaviors, and semantic association degrees between one or more conversation intention texts covered in the template conversation behavior k and one or more conversation emotion texts covered in the template conversation behavior k are calculated; and logically connecting the conversation emotion texts with the conversation intention texts, wherein the corresponding semantic association degrees of the conversation emotion texts are greater than or equal to the preset semantic association degrees, in the one or more conversation intention texts and the one or more conversation emotion texts.
Still taking the online conversation behavior 1 as an example, if the preset semantic association degree is 60%, calculating the semantic association degrees between the conversation intention text a and the conversation emotion text a, between the conversation emotion text B and between the conversation intention text a and the conversation emotion text C, and calculating the semantic association degrees between the conversation intention text a and the conversation emotion text a, between the conversation emotion text B and between the conversation intention text a and the conversation emotion text C as 100%, 0 and 0, respectively, logically linking the conversation intention text a with the conversation emotion text a, and correspondingly logically linking the conversation intention text B with the conversation emotion text B.
Therefore, the embodiment logically connects the session intention text and the session emotion text through the semantic association degree, can ensure the association precision between the session intention text and the session emotion text, improves the precision of session mining characteristics, and further improves the online session behavior data search effect.
Step S1032, one or more template conversation mining characteristics corresponding to the template online conversation behavior data are determined according to the conversation emotion intention contact information and the first correlation parameters between the conversation intention knowledge vectors corresponding to the conversation intention texts.
In some exemplary design concepts, when step S1032 is executed, the following steps may be adopted:
step S10321, calculating a first correlation parameter between the session intention knowledge vectors corresponding to the respective session intention texts, and determining the conversation intention linkage information of the two correlated conversation behaviors corresponding to the template online conversation behavior data according to the calculated first correlation parameters.
Illustratively, when step S10321 is executed, the following steps are performed in order for a plurality of template conversation activities based on the time-series direction of the online conversation activity data:
acquiring each conversation intention text cluster, wherein each conversation intention text cluster comprises: a first conversation intention text in a current template conversation behavior and a second conversation intention text in a next template conversation behavior;
calculating first association parameters corresponding to the conversation intention text clusters respectively, wherein each first association parameter is used for indicating an association parameter between the corresponding first conversation intention text and the corresponding second conversation intention text;
and determining target conversation intention text clusters with corresponding first associated parameters more than or equal to the first set associated parameters from each conversation intention text cluster, and determining the conversation intention linkage information of two associated conversation behaviors according to each determined target conversation intention text cluster.
In the embodiment of the present application, the first correlation parameter may be represented by, but is not limited to, a P2 feature distance, and the P2 feature distance may also be referred to as a euclidean correlation metric value. The smaller the P2 feature distance, the larger the first correlation parameter.
In this embodiment, each session intention text cluster may be directly obtained, the first associated parameter corresponding to each session intention text cluster is calculated, and then, from each session intention text cluster, a target session intention text cluster whose corresponding first associated parameter is greater than or equal to the first set associated parameter is determined, and session intention linkage information of two associated session behaviors is determined according to each determined target session intention text cluster.
In this embodiment, for the current template conversation behavior, second conversation intention texts respectively associated with each first conversation intention text may be respectively determined for each first conversation intention text in the current template conversation behavior in sequence, and then conversation intention linkage information of the two associated conversation behaviors is determined according to the second conversation intention texts respectively associated with each first conversation intention text, where each first conversation intention text and each associated second conversation intention text are determined as each target conversation intention text cluster. And if the number of the second conversation intention texts of which the correlation metric value is more than or equal to the first set correlation parameter is more than one, selecting the second conversation intention text which is the largest in the first correlation parameter and is relative to the first conversation intention text as the second conversation intention text which is relative to the first conversation intention text. The second session intent text associated with the first session intent text may be considered to be a manifestation of the first session intent text in the next online session behavior, when both are two texts in the same session mining feature.
For example, if the template online conversation behavior data includes online conversation behavior 1 and online conversation behavior 2, online conversation behavior 1 includes conversation intention text a, conversation intention text B, and conversation intention text C, and online conversation behavior 2 includes conversation intention text D, conversation intention text E, and conversation intention text F.
For online conversation behavior 1, the current template conversation behavior is online conversation behavior 1, the next template conversation behavior is online conversation behavior 2, for conversation intention text a, each conversation intention text cluster is obtained, each conversation intention text cluster comprises conversation intention text cluster 1 (conversation intention text a, conversation intention text D), conversation intention text cluster 2 (conversation intention text a, conversation intention text E), conversation intention text cluster 3 (conversation intention text a, conversation intention text F), then, an association parameter between conversation intention text a and conversation intention text D is calculated as a first association parameter 1 corresponding to conversation intention text cluster 1, an association parameter between conversation intention text a and conversation intention text E is calculated as a first association parameter 2 corresponding to conversation intention text cluster 2, an association parameter between conversation intention text a and conversation intention text F is calculated as a first association parameter 3 corresponding to conversation intention text cluster 3, if the first set association parameter is 90%, the first association parameter 1, the first association parameter 2, the first association parameter 3 is 90%, the first association parameter 3 is 80%, the next template conversation behavior is determined from text cluster 1, the conversation intention text cluster is online conversation intention cluster 1, and the conversation intention text cluster 3, and the conversation cluster is the same as the conversation intention cluster.
Likewise, assume that for the conversation intention text B, from among the conversation intention text cluster 4 (conversation intention text B, conversation intention text D), the conversation intention text cluster 5 (conversation intention text B, conversation intention text E), the conversation intention text cluster 6 (conversation intention text B, conversation intention text F), the target conversation intention text cluster is determined to be the conversation intention text cluster 5 (conversation intention text B, conversation intention text E), and for the conversation intention text C, from among the conversation intention text cluster 7 (conversation intention text C, conversation intention text D), the conversation intention text cluster 8 (conversation intention text C, conversation intention text E), the conversation intention text cluster 9 (conversation intention text C, conversation intention text F), the target conversation intention text cluster is determined to be the conversation intention text cluster 9 (conversation intention text C, conversation intention text F).
Therefore, the conversation intention linkage information of the two correlated conversation behaviors is determined based on the first correlation parameters corresponding to the conversation intention text clusters, the precision of the conversation intention linkage information of the two correlated conversation behaviors is improved, the precision of the conversation mining characteristics is further improved, and the online conversation behavior data searching effect is improved.
Step S10322, determining one or more template conversation mining characteristics corresponding to the template online conversation behavior data according to the conversation emotion intention contact information and the conversation intention linkage information of the two associated conversation behaviors.
The conversation emotion intention contact information comprises characteristic contact information between conversation emotion texts and conversation intention texts in the online conversation behavior data of each template, namely the characteristic contact information between the conversation emotion texts and the conversation intention texts in each online conversation behavior. The conversation intention linkage information of the two correlated conversation behaviors comprises characteristic contact information between the conversation intention texts in the front and back frames and the conversation intention texts. One or more template session mining features can be determined according to the session emotional intention contact information and the associated session intention linkage information of the two session behaviors, and each template session mining feature comprises a plurality of session intention knowledge vectors and a plurality of session emotional knowledge vectors of one online session user.
For example, assume that the session emotion intention linkage information includes a session emotion knowledge vector a and a session intention knowledge vector a in online session behavior 1, a session emotion knowledge vector D and a session intention knowledge vector D in online session behavior 2, and a session emotion knowledge vector G and a session intention knowledge vector G in online session behavior 3, the session intention linkage information of the two associated session behaviors includes feature linkage information between the session intention knowledge vector a and the session intention knowledge vector D, the session intention knowledge vector D and the session intention knowledge vector G, the template session mining feature 1 is determined according to the session emotion intention linkage information and the session intention linkage information of the two associated session behaviors, the template session mining feature 1 is a session mining feature of searching template 1 in online session behavior 1, online session behavior 2, and online session behavior 3, the template session mining feature 1 includes session emotion knowledge vector a and session intention vector a in online session behavior 1, session emotion knowledge vector D and session intention vector D in online session behavior 2, and session emotion knowledge vector G and session intention vector G in online session behavior 3. Correspondingly, according to the conversation emotion intention contact information and the associated conversation intention linkage information of the two conversation behaviors, a template conversation mining feature 2 and a template conversation mining feature 3 are determined, the template conversation mining feature 2 is a conversation mining feature of the search template 2 in the online conversation behavior 1 and the online conversation behavior 2, and the template conversation mining feature 3 is a conversation mining feature of the search template 3 in the online conversation behavior 1, the online conversation behavior 2 and the online conversation behavior 3.
Therefore, the conversation mining characteristics contained in the online conversation behavior data can be determined through the conversation emotion intention contact information and the associated conversation intention linkage information of the two conversation behaviors, and the weight of each conversation intention text in the data searching process can be improved when the online conversation behavior data is searched through the conversation mining characteristics in the following process, so that errors in data searching caused by single characteristics are avoided.
In some exemplary design concepts, in an implementation flow of determining session intention linkage information of two associated session behaviors, a feature vector sequence corresponding to each session intention text is recorded. For example, each session intention text included in the first template session behavior of the plurality of template session behaviors may be used as an initial text, and an initialization feature vector sequence corresponding to each template session mining feature may be recorded. The initialized feature vector sequence comprises the following information: the method comprises the steps of session mining feature ID, the time sequence of the session intention text in the session mining feature, the location of a session node of the session intention text, a session intention knowledge vector, an affiliated text ID and an affiliated online session behavior data ID. The initialized value of the session mining feature ID and the time sequence of the session intention text in the session mining feature is 1.
Taking a conversation intention text A in an online conversation behavior 1 as an example, recording a characteristic vector sequence 1 corresponding to a template conversation mining characteristic 1, wherein in the characteristic vector sequence 1, a conversation mining characteristic ID is 1, the time sequence of the conversation intention text in the conversation mining characteristic is 1, the conversation node position of the conversation intention text is (2, 2), the conversation intention knowledge vector is a conversation intention knowledge vector A, the belonging text ID is 1, and the belonging online conversation behavior data ID is online conversation behavior data 1.
When the first template conversation behavior starts, after each target conversation intention text cluster is determined, recording a corresponding feature vector sequence according to a second conversation intention text contained in the determined target conversation intention text cluster.
For example, the conversation intention text A is associated with the conversation intention text D, a feature vector sequence 2 corresponding to the template conversation mining feature 1 is recorded, in the feature vector sequence 2, the conversation mining feature ID is 1, the time sequence of the conversation intention text in the conversation mining feature is 2, the conversation node position of the conversation intention text is (2, 3), the conversation intention knowledge vector is the conversation intention knowledge vector D, the belonging text ID is 2, and the belonging online conversation behavior data ID is online conversation behavior data 1.
The feature vector sequence also comprises session emotion text information, wherein the session emotion text information comprises an ID of the session emotion text associated with the feature text and a corresponding session emotion knowledge vector. The ID of the conversation emotion text can be increased from 0 to 1 every new increment.
Still taking the session intention text a in the online session behavior 1 as an example, the feature vector sequence 1 corresponding to the template session mining feature 1 further includes a session mining feature ID, a time sequence of the session intention text in the session mining feature, a session node position of the session intention text, a session intention knowledge vector, a session emotion knowledge vector, an ID of the session emotion text, an ID of the text to which the text belongs, and an ID of online session behavior data to which the text belongs, where the session emotion knowledge vector is the session emotion knowledge vector a, and the ID of the session emotion text is 0.
If a certain session intention knowledge vector does not have an associated session intention knowledge vector in the next online session behavior, the corresponding feature vector sequence does not need to be recorded.
For any template conversation behavior, if a conversation intention text which is not associated with the last conversation intention text exists, namely a new conversation intention exists in the template conversation behavior, the conversation intention text is used as an updated conversation intention text, and the initialized feature vector sequence of the updated template conversation mining features is recorded.
In some exemplary design approaches, if a plurality of template session mining features exist, second association parameters between a plurality of session emotion features covered by the plurality of template session mining features are determined, and feature contact information between the plurality of template session mining features is determined according to the determined second association parameters. Wherein, the second correlation parameter can also be represented by an L2 characteristic distance. Correspondingly, in the implementation process of determining the target online conversation behavior data from the candidate online conversation behavior data according to the determined template conversation mining characteristics and the target conversation mining characteristics corresponding to the candidate online conversation behavior data in the cloud online behavior big data of the cloud online service, the target online conversation behavior data can be determined from the candidate online conversation behavior data according to the determined template conversation mining characteristics, the target conversation mining characteristics corresponding to the candidate online conversation behavior data, the feature contact information among the template conversation mining characteristics and the feature contact information among the target conversation mining characteristics.
Illustratively, when determining second association parameters among a plurality of session emotion features covered by a plurality of template session mining features, for any two template session mining features in the plurality of template session mining features, determining second association parameters among session emotion knowledge vectors covered by any two template session mining features, and correspondingly, when determining feature association information among the plurality of template session mining features according to the determined second association parameters, determining feature association information among any two template session mining features according to the determined second association parameters.
When determining the feature contact information among the plurality of template session mining features according to the determined second associated parameters, aiming at any two template session mining features covered in the plurality of template session mining features, the following steps are carried out:
calculating a plurality of conversation emotion knowledge vectors covered in one template conversation mining feature, respectively calculating second association parameters between the plurality of conversation emotion knowledge vectors covered in the other template conversation mining feature, and determining target second association parameters of which corresponding values are greater than or equal to preset second set association parameters from the calculated second association parameters; and when any two template session mining characteristics belong to the same online session user according to the determined second association parameters of each target, configuring the characteristic contact information between any two template session mining characteristics in the characteristic contact information.
And when the number of the target second correlation parameters is larger than the threshold number, determining that any two template session mining characteristics belong to the same online session user. Or when the ratio of the number of the second target associated parameters to the total number of the conversation emotion texts is greater than or equal to the threshold ratio, determining that any two template conversation mining features belong to the same online conversation user, wherein the total number of the conversation emotion texts is as follows: the number of the session emotion knowledge vectors covered in one template session mined feature, or the number of the session emotion knowledge vectors covered in the other template session mined feature, or the minimum value of the number of the session emotion knowledge vectors covered in the two template session mined features.
Illustratively, taking template session mining feature 1 and template session mining feature 2 as examples, template session mining feature 1 includes session emotion knowledge vector a, session emotion knowledge vector D and session emotion knowledge vector G of search template 1, template session mining feature 2 includes session emotion knowledge vector B and session emotion knowledge vector F of search template 1, second association parameters between session emotion knowledge vector a and session emotion knowledge vector B, session emotion knowledge vector a and session emotion knowledge vector F, session emotion knowledge vector D and session emotion knowledge vector B, session emotion knowledge vector D and session emotion knowledge vector F, session emotion knowledge vector G and session emotion knowledge vector B are calculated respectively, then, from the calculated second association parameters, a target second association parameter with a corresponding value greater than or equal to a preset second set association parameter is determined, if the ratio is 0.5, the second set association parameter is 0.2, the number of target second association parameters is 2, the number of texts is greater than or equal to the preset second set association parameter, and the total number of the second association parameters is greater than or equal to the preset second set association parameter, and the total number of the second association parameter is determined as 0.5, and the total number of the second set emotion knowledge vector is equal to the preset emotion knowledge vector 1.
If the two template session mining features are associated with each other, the session emotion IDs in the two template session mining features can be marked as the same ID, the same ID can be any one of the session emotion IDs corresponding to the two template session mining features, and the session emotion ID with the smallest value in the session emotion IDs corresponding to the two template session mining features can also be adopted.
Therefore, on the basis that a plurality of template session mining features exist, feature contact information among the plurality of template session mining features is determined according to the determined second associated parameters, and when online session behavior data searching is carried out on the basis of the template session mining features, the session mining features with the association can be quickly searched on the basis of the feature contact information, so that the data searching speed is improved. Furthermore, the template conversation mining characteristics belonging to the same search template can be logically connected, so that the subsequent data search speed is increased.
In some exemplary design concepts, in order to improve the accuracy in the online conversation behavior data search process, when step S104 is executed, the following embodiments may be implemented:
step S1041, determining feature association parameters between each target session mining feature and each template session mining feature according to each template session mining feature and each target session mining feature corresponding to each candidate online session behavior data in the cloud online behavior big data of the cloud online service.
For example, when step S1041 is executed, the following embodiments may be implemented:
step S10411, determining a session intention knowledge vector association parameter between each object session mining feature and each template session mining feature according to the session intention knowledge vector covered by each template session mining feature and according to the session intention knowledge vector covered by each object session mining feature.
Taking the template session mining feature a and the target session mining feature B as an example, the template session mining feature a is any one of the template session mining features, and the target session mining feature B is any one of the target session mining features.
For example, template session mining feature A comprises a session intention knowledge vector A1, a session intention knowledge vector A2, a session intention knowledge vector B8230, a session intention knowledge vector BM calculating a session intention knowledge vector A1, a session intention knowledge vector A2, a session intention knowledge vector B8230, a session intention knowledge vector B2, a session intention knowledge vector B8230, AN AN associating parameter between the session intention knowledge vector B1, the session intention knowledge vector B2, a session intention knowledge vector B8230, a session intention associating parameter between BM and BM.
Step S10412, determining a session emotion knowledge vector association parameter between each target session mining feature and each template session mining feature according to the session emotion knowledge vector covered by each template session mining feature and the session emotion knowledge vector covered by each target session mining feature.
For example, if template session mining feature A includes session emotion knowledge vector A1, session emotion knowledge vector A2, .... Session emotion knowledge vector AN, and target session mining feature B includes session emotion knowledge vector B1, session emotion knowledge vector B2, .... Session emotion knowledge vector BM, respectively, session emotion knowledge vector A1, session emotion knowledge vector A2, .... 8230;. Session emotion knowledge vector AN respectively includes session emotion knowledge vector B1, session emotion knowledge vector B2,. 8230;. Session intention association parameters between session emotion knowledge vectors BM are calculated.
Step S10413, determining feature association parameters between each target session mining feature and each template session mining feature according to the determined session intention knowledge vector association parameters and each session emotion knowledge vector association parameters.
Wherein, the feature association parameter between the template session mining feature a and the target session mining feature B can be determined based on at least one of the following information: the number of similar conversation intention knowledge vectors in the template conversation mining characteristic A and the target conversation mining characteristic B, and the number of similar conversation emotion knowledge vectors in the template conversation mining characteristic A and the target conversation mining characteristic B.
When the conversation intention knowledge vector in the template conversation mining feature A and the conversation intention knowledge vector in the target conversation mining feature B are larger than the set conversation intention association parameter, the conversation intention knowledge vector in the template conversation mining feature A and the conversation intention knowledge vector in the target conversation mining feature B are similar conversation intention knowledge vectors. Illustratively, the session intention association parameter is set to 0.2.
Correspondingly, when the conversation emotion knowledge vector in the template conversation mining feature A and the conversation emotion association parameter between the conversation emotion knowledge vector in the target conversation mining feature B are greater than the preset conversation emotion setting association parameter, the conversation emotion knowledge vector in the template conversation mining feature A and the conversation intention knowledge vector in the target conversation mining feature B are similar conversation emotion knowledge vectors.
Therefore, according to the conversation intention knowledge vector association parameters and the conversation emotion knowledge vector association parameters, the characteristic association parameters between the target conversation mining characteristics and the template conversation mining characteristics are determined, in this way, the feature association parameters include the session intention knowledge vector association parameters and the session emotion knowledge vector association parameters, so that the feature association parameters are more accurate.
Step S1042, determining data association parameters between the candidate online conversation behavior data and the template online conversation behavior data according to the determined feature association parameters.
And when the number of the similar conversation intention knowledge vectors in the template conversation mining feature A and the target conversation mining feature B is larger than a first number threshold value and/or the number of the similar conversation emotion knowledge vectors in the template conversation mining feature A and the target conversation mining feature B is larger than a second number threshold value, determining that the template conversation mining feature A and the target conversation mining feature B are the same conversation mining feature.
Wherein the first number threshold is determined based on a first preset weight and a first number of session intention knowledge vectors, wherein the first number of session intention knowledge vectors may be determined based on at least one of: the number of session intention knowledge vectors covered in the template session mining feature a, the number of session intention knowledge vectors covered in the target session mining feature B, and the minimum of the number of session intention knowledge vectors covered in the template session mining feature a and the target session mining feature B, but are not limited thereto.
For example, if the number of the session intention knowledge vectors covered in the template session mining feature a is 40, and the number of the session intention knowledge vectors covered in the target session mining feature B is 30, for example, the first preset weight is 1/3, and the number of the first session intention knowledge vectors is 30, the first number threshold is 1/3 × 30=10.
The second quantity threshold is determined based on a second preset weight and a second number of the conversation emotion knowledge vectors, wherein the second number of the conversation emotion knowledge vectors can be determined based on at least one of the following information: the number of the session emotion knowledge vectors covered in the template session mining feature a, the number of the session emotion knowledge vectors covered in the target session mining feature B, and the minimum value of the number of the session emotion knowledge vectors covered in the template session mining feature a and the target session mining feature B are not limited to these.
For example, the number of the session emotion knowledge vectors covered in the template session mining feature a is 40, the number of the session emotion knowledge vectors covered in the target session mining feature B is 30, and if the first preset weight is 1/4 and the number of the first session intention knowledge vectors is 40, the first number threshold is 1/4 × 40=10.
For example, taking a candidate online conversation behavior data and a template online conversation behavior data as an example, according to the determined feature association parameters, each target conversation mining feature in the candidate online conversation behavior data can be determined, and according to the number of the same conversation mining features existing in each template conversation mining feature, the data association parameters between the candidate online conversation behavior data and the template online conversation behavior data are determined. For example, the data association parameter may also be referred to as online session behavior repetition.
For example, the value of the data association parameter between the candidate online session behavior data and the template online session behavior data is a ratio of the number of the same session mining features included in the candidate online session behavior data and the template online session behavior data to a third session mining feature number, where the third trajectory number may be determined based on at least one of the following information: the number of target session mining features covered in the candidate online session activity data, the number of template session mining features covered in the template online session activity data, or the minimum of the foregoing two, but is not limited thereto.
And S1043, determining target online conversation behavior data meeting the search requirement according to data association parameters between the candidate online conversation behavior data and the template online conversation behavior data.
In some exemplary design concepts, the step S1043 may be executed in the following manners, but is not limited to:
for example, target online session behavior data meeting the search requirement is determined according to data association parameters between the candidate online session behavior data and the template online session behavior data, the candidate online session behavior data is ranked, and a certain amount of target online session behavior data is determined from the candidate online session behavior data.
For another example, the candidate online conversation behavior data with one or more data association parameters with the template online conversation behavior data larger than the preset online conversation behavior data set association parameters is determined from the candidate online conversation behavior data, and the determined one or more candidate online conversation behavior data is used as the target online conversation behavior data.
When the template session mining feature a and the target session mining feature B are the same session mining feature, the candidate online session behavior data corresponding to the template session mining feature a may be used as the target online session behavior data, and the template session mining feature a, the data association parameter between the candidate online session behavior data corresponding to the template session mining feature a and the template online session behavior data, and the online session behavior data ID of the candidate online session behavior data corresponding to the template session mining feature a may be obtained.
The target online session behavior data may be at least one, and the number of the same session mining features may also be at least one.
Next, a description will be given of a model weight parameter update flow of the conversational emotion encoding model.
In this embodiment, the session emotion encoding model before the model weight parameter updating process is referred to as an initial session emotion encoding model, and the session emotion encoding model satisfying the model convergence condition is referred to as a target session emotion encoding model.
The updating process of the model weight parameters of the initial session emotion encoding model comprises two stages: the method comprises an example conversation behavior data collection stage and a model updating stage, wherein the example conversation behavior data collection stage is used for acquiring a data sequence to be learned, and the model updating stage is used for training an initial conversation emotion coding model according to the data sequence to be learned.
In the example conversation behavior data collection stage, firstly, each association model preparation data combination is determined according to a first reference association parameter between each example model preparation data in the example model preparation data sequence, secondly, the data sequence to be learned is constructed according to each association model preparation data combination, each piece of data to be learned comprises at least three example conversation behavior data, and the association model preparation data combination corresponding to one example conversation behavior data in the at least three example conversation behavior data is different from the association model preparation data combinations corresponding to other example conversation behavior data.
Each association model preparation data combination comprises at least two example conversation behavior data, and each example conversation behavior data in the association model preparation data combination is an association model preparation data combination. The associated model preparation data set may also be referred to herein as a positive example model preparation data set.
In the implementation procedure for determining each combination of the associated model preparation data according to the first reference associated parameter between each of the example model preparation data in the example model preparation data sequence, the following methods can be adopted, but are not limited to:
a: performing session emotion analysis on each sample session behavior data in the sample model preparation data sequence, determining session emotion texts corresponding to each sample session behavior data, then performing training calibration on every two obtained session emotion texts in each sample session emotion text, determining whether every two session emotion texts are associated model preparation data combinations, and determining each set of associated model preparation data combinations according to the training calibration result. The training and calibration of the conversation emotion text can be a training and calibration model adopting pre-training.
B: see the following scheme:
step S1101, performing session emotion analysis on each sample session behavior data in the sample model preparation data sequence, and determining a session emotion text corresponding to each sample session behavior data. The example model preparation data sequence may include at least one online conversation activity data and may also include at least one text.
In some exemplary design concepts, in order to increase the search speed of the example conversation behavior data, before performing the conversation emotion analysis on each example conversation behavior data in the example model preparation data sequence, a certain amount of example model preparation data may be extracted from the example model preparation data sequence based on a preset data extraction interval, and accordingly, when performing the conversation emotion analysis on each example conversation behavior data in the example model preparation data sequence, the conversation emotion analysis may be performed on the extracted example model preparation data, so as to increase the data collection speed.
For example, the example model preparation data sequence is online conversation behavior data 1, the online conversation behavior data 1 includes online conversation behavior 1 to online conversation behavior N, and if one extraction interval is 10, the online conversation behavior 1, the online conversation behavior 11, the online conversation behavior 21, and the like are extracted from the online conversation behavior 1 to online conversation behavior N.
In the implementation process of step S1101, when performing session emotion analysis on the example model preparation data, the session emotion text included in the example model preparation data may be analyzed by using the semantic neural network model, and the location of the session node of the session emotion text may be recorded.
And step S1102, determining a session emotion knowledge vector corresponding to each session emotion text according to the obtained session emotion texts.
Illustratively, when step S1102 is executed, the obtained session emotion texts are loaded into the pre-trained session emotion encoding model, and the session emotion knowledge vectors corresponding to the session emotion texts are determined.
And S1103, clustering the obtained emotion knowledge vectors of each session, and determining each cluster.
The clustering number is determined based on the total number of the session emotion texts included in each example session behavior data, for example, a ratio of the total number to a preset number is calculated, and an integer value of the ratio is determined as the clustering number.
For example, if the preset number is 20 and the total number of the session emotion texts included in each sample session behavior data is 200, the number of clusters is 200/20=10, that is, each obtained session emotion knowledge vector is clustered, and 10 clusters are determined. For another example, if the preset number is 20, and the total number of the session emotion texts included in each sample session behavior data is 270, the number of clusters is [270/20] =14, that is, each obtained session emotion knowledge vector is clustered, and 14 clusters are determined.
And step S1104, determining the session emotion knowledge vectors meeting the set clustering requirements from each cluster according to the correlation metric values among the session emotion knowledge vectors in each cluster, and determining each association model preparation data combination according to the session emotion text corresponding to the determined session emotion knowledge vectors.
For example, when step S1104 is executed, for each session emotion knowledge vector in each cluster, a clustering weight corresponding to the session emotion knowledge vector is determined according to a correlation metric between the session emotion knowledge vector and other session emotion knowledge vectors in the same cluster except the session emotion knowledge vector, and then a session emotion knowledge vector meeting a set clustering requirement is determined from each cluster according to a clustering weight corresponding to each session emotion knowledge vector in each cluster.
Taking a session emotion knowledge vector ki in the cluster k as an example, the cluster k is any one of the clusters, each session emotion knowledge vector is contained in the cluster k, and the session emotion knowledge vector ki is any one of the session emotion knowledge vectors contained in the cluster k.
In some exemplary design concepts, the clustering weight corresponding to the conversation emotion knowledge vector ki may be determined by the following embodiments:
and A1, calculating correlation measurement values between the session emotion knowledge vector ki and other session emotion knowledge vectors except ki in the clustering k. The correlation metric between the session emotion knowledge vector ki and other session emotion knowledge vectors may be represented in, but not limited to, a P2 feature distance.
For example, assuming that cluster a includes session emotion knowledge vector 1 to session emotion knowledge vector 20, correlation metric P1 between session emotion knowledge vector 1 and session emotion knowledge vector 2 is calculated for session emotion knowledge vector 1, correlation metric P2 between session emotion knowledge vector 1 and session emotion knowledge vector 3 is calculated, and correlation metrics P1 to P19 between session emotion knowledge vector 1 and session emotion knowledge vector 2 to session emotion knowledge vector 20 are calculated correspondingly.
And step A2, determining a preset number of conversation emotion knowledge vectors from other conversation emotion knowledge vectors based on the calculated correlation metric values.
Illustratively, based on the calculated correlation metric values, a session emotion knowledge vector with the minimum correlation metric value is determined based on a preset number from other session emotion knowledge vectors.
For example, the correlation metric values between the conversation emotion knowledge vector 1 and the conversation emotion knowledge vectors 2-20 are respectively P1-P19, if the values of P1-P19 are P19, P18, ...:P 1 in sequence based on descending order, and if the preset number is 5, P19, P18, P17, P16 and P15 are determined from other conversation emotion knowledge vectors.
And step A3, determining clustering weight of the session emotion knowledge vector ki according to the determined correlation metric value between each session emotion knowledge vector and the session emotion knowledge vector ki.
Illustratively, the average value of the determined correlation metric values corresponding to the session emotion knowledge vectors is used as the clustering weight of the session emotion knowledge vector ki.
For example, the average of P19, P18, P17, P16, and P15 may be used as the clustering weight of the session emotion knowledge vector 1.
According to the clustering weight corresponding to each conversation emotion knowledge vector in each cluster, in the implementation process of determining the conversation emotion knowledge vector meeting the set clustering requirement from each cluster, determining the conversation emotion knowledge vector with the corresponding clustering weight smaller than the set clustering weight from each conversation emotion knowledge vector in each cluster, and taking the determined conversation emotion knowledge vector as the conversation emotion knowledge vector meeting the set clustering requirement.
If the clustering weight of the session emotion knowledge vector ki is smaller than the set clustering weight, the session emotion knowledge vector ki is reserved, namely if the clustering weight of the session emotion knowledge vector ki is smaller than the set clustering weight, the session emotion knowledge vector ki is the session emotion knowledge vector meeting the set clustering requirement. And if the clustering weight of the session emotion knowledge vector ki is more than or equal to the set clustering weight, removing the session emotion knowledge vector ki from the cluster k.
And setting clustering weight, wherein the clustering weight is determined based on the clustering weight corresponding to each conversation emotion knowledge vector in the clustering k. Illustratively, based on values, an intermediate clustering weight is determined from clustering weights corresponding to session emotion knowledge vectors in the cluster k, and the intermediate clustering weight is used as a set clustering weight.
For example, if the clustering weight is set to 10 and the clustering weight of the session emotion knowledge vector 1 is greater than the set clustering weight, the session emotion knowledge vector 1 is deleted from the cluster 1.
Correspondingly, clustering weights corresponding to the conversation emotion knowledge vector 1 to the conversation emotion knowledge vector 20 are respectively calculated for the cluster 1, if the clustering weight is set to be 10, and the clustering weights corresponding to the conversation emotion knowledge vector 1 to the conversation emotion knowledge vector 20 are respectively 1-20, the conversation emotion knowledge vector 10 to the conversation emotion knowledge vector 20 are removed from the cluster 1, and at the moment, the cluster 1 comprises the conversation emotion knowledge vector 1 to the conversation emotion knowledge vector 9.
If the number of session emotion knowledge vectors covered in a cluster is less than the set number of clusters, the cluster may be discarded. For example, the cluster data is set to 5, and when the number of the session emotion knowledge vectors covered in one cluster is less than 5, the cluster is discarded.
In some exemplary design ideas, in an implementation process of preparing a data combination according to each association model and constructing a data sequence to be learned, the implementation process can be implemented by the following embodiments:
in step S1401, a corresponding association model preparation data combination is extracted from each of the identified association model preparation data combinations.
Wherein the association model preparation data combination comprises: the association model preparation data combination 1 comprises example model preparation data 1a,1b, 1c and the like, wherein the example model preparation data 1a,1b and 1c are conversation emotion texts of the same user, the association model preparation data combination 2 comprises example model preparation data 2a,2b, 2c and the like, the association model preparation data combination co comprises example model preparation data coa, cob, coc and the like, the association model preparation data combination 1 extracts two example conversation behavior data 1a and 1b, determines the association model preparation data combination 1 (1a and 1b), the association model preparation data combination 2 extracts two example conversation behavior data 2a and 2b, determines the association model preparation data combination 2 (2a and 2b), and correspondingly extracts two example conversation behavior data coa and cobs from the association model preparation data combination co and determines the association model preparation data combination 2 (2a and 2b).
Step S1402 is to perform the following steps for each of the association model preparation data combinations:
firstly, taking example conversation behavior data contained in one association model preparation data combination as reference example conversation behavior data, extracting corresponding other example conversation behavior data from other association model preparation data combinations respectively, and calculating second reference association parameters between the extracted other example conversation behavior data and the reference example conversation behavior data respectively;
then, one or more target example conversation behavior data are determined from the other example conversation behavior data according to the calculated second reference correlation parameters, and one or more data to be learned are determined according to the one or more target example conversation behavior data and a correlation model preparation data combination.
When the data to be learned includes three example session behavior data, the data to be learned may also be a triple, which is represented as (e, j, g), e represents an anchor point, j represents positive example session behavior data, g represents negative example session behavior data, e and j form a positive example session behavior data combination, and e and g form a negative example session behavior data combination. In this embodiment, the reference example session behavior data in the association model preparation data set is referred to as anchor point, the other example session behavior data in the association model preparation data set is referred to as positive example session behavior data, and the determined target example session behavior data is referred to as negative example session behavior data.
When extracting corresponding one of the other example session behavior data from the other association model preparation data combinations, one of the example session behavior data may be randomly selected from each of the other association model preparation data combinations. The second reference correlation parameter may employ, but is not limited to, an L2 feature distance.
Taking the correlation model preparation data set 1 (1a, 1b) as an example, the example model preparation data 1a in the correlation model preparation data set 1 (1a, 1b) is used as reference example session behavior data, the example model preparation data 2a, .... The example model preparation data coa is extracted from the correlation model preparation data set 2 (2a, 2b), and the reference example session behavior data 1a is calculated and is respectively associated with the example model preparation data 2a, the example model preparation data 3b, \ .... And second reference correlation parameters between the example model preparation data coa.
In the implementation process of determining one or more target example session behavior data from various other example session behavior data according to the calculated second reference correlation parameter, the following methods may be adopted, but are not limited to:
a: and sequencing the other example session behavior data based on the descending order according to the calculated value of the second reference correlation parameter, and sequentially selecting target example session behavior data with preset target example session behavior data quantity from the other example session behavior data based on the descending order according to the sequencing result.
The reference example conversation behavior data 1a and the example conversation behavior data 2a, the example conversation behavior data 3b, ..., respectively, second reference association parameters between the example conversation behavior data coa are P2a, P3b, ..., and Pcoa, respectively, and according to the calculated values of the second reference association parameters, the other example conversation behavior data are sorted based on the descending order, and the sorting result is: p2a, P3b, .... Pcoa, if the number of the preset target example conversation behavior data is 20, according to the sorting result, sequentially selecting 20 example conversation behavior data from each other example conversation behavior data: example session behavior data 2a, example session behavior data 3b, ..., and example session behavior data 20a as target example session behavior data.
Illustratively, the example conversation behavior data of each other example is sorted based on the descending order according to the calculated value of the second reference correlation parameter, and the example conversation behavior data of the preset quantity of the example conversation behavior data is deleted from the conversation behavior data of each other example based on the descending order according to the sorting result, and the target example conversation behavior data of the preset quantity of the target example conversation behavior data is sequentially selected from the deleted conversation behavior data of each other example.
For example, the reference example session behavior data 1a and the example session behavior data 2a, the example session behavior data 3b, ..., respectively, the second reference association parameters between the example session behavior data coa are P2a, P3b, ..., pcoa, respectively, and the respective other example session behavior data are sorted based on the descending order according to the calculated values of the second reference association parameters, and the sorting result is: p2a, P3b, \8230- \8230, pcoa, if the deletion quantity of the preset example session behavior data is 5 and the quantity of the preset target example session behavior data is 20, deleting the example session behavior data 2a, the example session behavior data 3b, the example session behavior data 4a, the example session behavior data 5b and the example session behavior data 6a from each other example session behavior data according to the sorting result, and sequentially selecting 20 example session behavior data from the example session behavior data 7a, the example session behavior data 8b, the example session behavior data 8230- \8230and the example session behavior data coa: example session behavior data 7a, example session behavior data 8b, ..., and example session behavior data 27a are targeted example session behavior data.
In the implementation process of determining one or more data to be learned according to the combination of one or more target example session behavior data and one association model preparation data, one association model preparation data combination is respectively combined with one or more target example session behavior data to determine one or more data to be learned.
For example, if the target example conversation activity data is example conversation activity data 7a, example conversation activity data 8b, ..., example conversation activity data 27a with respect to reference example conversation activity data 1a, association model preparation data combination 1 (1a, 1b) is combined with example conversation activity data 7a, example conversation activity data 8b, .... Example conversation activity data 27a, respectively, to determine data to be learned, each of which includes: (1a, 1b, 7a), (1a, 1b, 8b), (...(1a, 1b, 27a).
Exemplarily, in order to improve the update effect of the model weight parameter, in the update process of the model weight parameter, a process of constructing a data sequence to be learned is described below, and the following steps are performed in each traversal loop stage by performing traversal loop update on the initial session emotion encoding model:
step S1801, a data sequence to be learned corresponding to each training round is constructed according to each association model preparation data combination covered in the example model preparation data sequence and a training round load amount set in advance, each data to be learned includes at least three example session behavior data, and an association model preparation data combination corresponding to one example session behavior data of the at least three example session behavior data is different from association model preparation data combinations corresponding to other example session behavior data.
After the association model preparation data combinations are acquired in the example session behavior data collection stage, the data sequences to be learned corresponding to the training rounds can be determined according to the association model preparation data combinations in the example session behavior data collection stage, or the data sequences to be learned corresponding to the training rounds can be determined according to the association model preparation data combinations in the model update stage, and the acquisition method of the data sequences to be learned refers to steps S1401-S1402.
And step S1802, loading the constructed data sequences to be learned to the initial session emotional coding model in training turns, determining the corresponding overall session emotional coding cost value, updating the model weight parameters of the initial session emotional coding model according to the obtained overall session emotional coding cost value, and outputting the target session emotional coding model when the model convergence condition is met.
In some exemplary design concepts, in the following application embodiments, the above method may further include the following steps.
Step S100, acquiring a user interaction window established by the second user and the pushed first user, and acquiring user interaction text data in the user interaction window;
step S200, mining an intentional knowledge point of progress on the user interactive text data to generate an intentional knowledge point of progress;
step S300, carrying out content pushing on the user groups of the first user and the second user based on the advanced intention knowledge point.
In some exemplary design ideas, the step of performing advanced intention knowledge point mining on the user interaction text data to generate an advanced intention knowledge point includes:
analyzing an interactive semantic keyword directed graph from the user interactive text data, wherein the interactive semantic keyword directed graph comprises W groups of interactive semantic keywords with semantic relation, and W is an integer not less than 1;
acquiring an attention keyword directed graph by using the interactive semantic keyword directed graph, wherein the attention keyword directed graph comprises W groups of attention keywords with semantic relation;
mining a first topic feature vector sequence by utilizing the interactive semantic keyword directed graph and utilizing a first topic feature analysis unit in a neural network model, wherein the first topic feature vector sequence comprises W first topic feature vectors;
mining a second topic feature vector sequence by using a second topic feature analysis unit in the neural network model by using the attention keyword directed graph, wherein the second topic feature vector sequence comprises W second topic feature vectors;
determining intention knowledge point prediction information corresponding to the interactive semantic keyword directed graph by using the first topic feature vector sequence and the second topic feature vector sequence and an intention knowledge point prediction unit in the neural network model;
and determining the advanced intention knowledge point of the interactive semantic keyword directed graph by using the intention knowledge point prediction information.
In some exemplary design ideas, determining intention knowledge point prediction information corresponding to the interactive semantic keyword directed graph by using an intention knowledge point prediction unit in the neural network model by using the first topic feature vector sequence and the second topic feature vector sequence includes:
acquiring W first interactive knowledge point representations by using the first topic feature vector sequence and a first interactive knowledge point classification unit in the neural network model, wherein each first interactive knowledge point representation corresponds to one first topic feature vector;
acquiring W second interactive knowledge point representations by using the second topic feature vector sequence and a second interactive knowledge point classification unit in the neural network model, wherein each second interactive knowledge point representation corresponds to one second topic feature vector;
combining the W first interactive knowledge point representations and the W second interactive knowledge point representations to obtain W target interactive knowledge point representations, wherein each target interactive knowledge point representation comprises a first interactive knowledge point representation and a second interactive knowledge point representation;
and utilizing the W target interaction knowledge point representations and utilizing the intention knowledge point prediction unit in the neural network model to determine intention knowledge point prediction information corresponding to the interactive semantic keyword digraph.
In some exemplary design approaches, obtaining W first interactive knowledge point characterizations by using the first interactive knowledge point classification unit in the neural network model using the first topic feature vector sequence includes:
regarding each group of first topic feature vectors in the first topic feature vector sequence, acquiring first staged coding information by utilizing staged coding units in the first interactive knowledge point classification unit, wherein the first interactive knowledge point classification unit belongs to the neural network model; for each set of first topic feature vectors in the first sequence of topic feature vectors: acquiring first integral coding information by utilizing an integral coding unit in the first interactive knowledge point classification unit; acquiring first option information by utilizing the first staged coding information and the first overall coding information and utilizing a characteristic item selection branch based on expected cross entropy in the first interactive knowledge point classification unit;
acquiring a first interaction knowledge point representation by utilizing a first integral coding unit in the first interaction knowledge point classification unit by utilizing the first option information and the first topic feature vector;
acquiring W second interactive knowledge point representations by using the second topic feature vector sequence and using a second interactive knowledge point classification unit in the neural network model, wherein the acquiring comprises the following steps:
for each set of second topic feature vectors in the second sequence of topic feature vectors: acquiring second phased encoding information by utilizing a phased encoding unit in the second interactive knowledge point classification unit, wherein the second interactive knowledge point classification unit belongs to the neural network model; acquiring second integral coding information by utilizing an integral coding unit in the second interactive knowledge point classification unit; acquiring second option information by utilizing the second staged coding information and the second overall coding information and utilizing a characteristic item selection branch based on expected cross entropy in the second interactive knowledge point classification unit;
and acquiring a second interactive knowledge point representation by utilizing a second integral coding unit in the second interactive knowledge point classification unit by utilizing the second option information and the second theme feature vector.
FIG. 2 schematically illustrates an artificial intelligence system 100 that can be used to implement various embodiments described herein.
For one embodiment, FIG. 2 illustrates an artificial intelligence system 100, the artificial intelligence system 100 having a plurality of processors 102, a control module (chipset) 104 coupled to at least one of the processor(s) 102, a memory 106 coupled to the control module 104, a non-volatile memory (NVM)/storage 108 coupled to the control module 104, a plurality of input/output devices 110 coupled to the control module 104, and a network interface 112 coupled to the control module 106.
The processor 102 may include a plurality of single-core or multi-core processors, and the processor 102 may include any combination of general-purpose or special-purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In some alternative embodiments, the artificial intelligence system 100 can be a server device such as a gateway described in the embodiments of the present application.
In some alternative embodiments, the artificial intelligence system 100 can include a plurality of computer-readable media (e.g., memory 106 or NVM/storage 108) having instructions 114 and a plurality of processors 102 in combination with the plurality of computer-readable media and configured to execute the instructions 114 to implement modules to perform the actions described in this disclosure.
For one embodiment, control module 104 may include any suitable interface controllers to provide any suitable interface to one or more of processor(s) 102 and/or any suitable device or component in communication with control module 104.
The memory 106 may be used, for example, to load and store data and/or instructions 114 for the artificial intelligence system 100. For one embodiment, memory 106 may comprise any suitable volatile memory, such as suitable DRAM. In some alternative embodiments, the memory 106 may comprise a double data rate type four synchronous dynamic random access memory (DDR 4 SDRAM).
For one embodiment, control module 104 may include multiple input/output controllers to provide an interface to NVM/storage 108 and input/output device(s) 110.
For example, NVM/storage 108 may be used to store data and/or instructions 114. NVM/storage 108 may include any suitable non-volatile memory (e.g., flash memory) and/or may include any suitable non-volatile storage device(s) (e.g., multiple Hard Disk Drives (HDDs), multiple Compact Disc (CD) drives, and/or multiple Digital Versatile Disc (DVD) drives).
The NVM/storage 108 may include storage resources that are physically part of the device on which the artificial intelligence system 100 is installed, or it may be accessible by the device and may not be necessary as part of the device. For example, NVM/storage 108 may be accessible via input/output device(s) 110 in connection with a network.
The input/output device(s) 110 may provide an interface for the artificial intelligence system 100 to communicate with any other suitable device, and the input/output device(s) 110 may include a communications component, a pinyin component, a sensor component, and so forth. The network interface 112 may provide an interface for the artificial intelligence system 100 to communicate in accordance with a plurality of networks, and the artificial intelligence system 100 may wirelessly communicate with a plurality of components of a wireless network in accordance with any of a plurality of wireless network standards and/or protocols, such as access to a communication standard-based wireless network, such as WiFi, 2G, 3G, 4G, 5G, etc., or a combination thereof.
For one embodiment, at least one of the processor(s) 102 may be packaged together with logic for multiple controllers (e.g., memory controller modules) of the control module 104. For one embodiment, at least one of the processor(s) 102 may be packaged together with logic for multiple controllers of the control module 104 to form a System In Package (SiP). For one embodiment, at least one of the processor(s) 102 may be integrated on the same die with the logic of the multiple controllers of the control module 104. For one embodiment, at least one of the processor(s) 102 may be integrated on the same die with logic for multiple controllers of the control module 104 to form a system on a chip (SoV).
In various embodiments, the artificial intelligence system 100 can be, but is not limited to being: a server, a desktop computing device, or a mobile computing device (e.g., a laptop computing device, a handheld computing device, a tablet, a netbook, etc.), among other terminal devices. In various embodiments, the artificial intelligence system 100 can have more or fewer components and/or different architectures. For example, in some alternative embodiments, the artificial intelligence system 100 includes a plurality of cameras, a keyboard, a liquid crystal display (LVD) screen (including a touch screen display), a non-volatile memory port, a plurality of antennas, a graphics chip, an application specific integrated circuit (ASIV), and a speaker.
The present application is described in detail above, and the principles and embodiments of the present application are described herein by using specific examples, which are only used to help understand the method and the core idea of the present application; meanwhile, for a person skilled in the art, according to the idea of the present application, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present application.
Claims (10)
1. A big data analysis method applied to cloud online service is characterized in that the big data analysis method is applied to the artificial intelligence system, and the method comprises the following steps:
according to an online session scene designated in cloud online service, performing session emotion analysis on a plurality of template session behaviors contained in template online session behavior data respectively, and determining one or more session emotion texts and corresponding session emotion knowledge vectors contained in the plurality of template session behaviors;
according to an online conversation scene designated in the cloud online service, respectively carrying out conversation intention analysis on the template conversation behaviors, and determining one or more conversation intention texts and corresponding conversation intention knowledge vectors covered by the template conversation behaviors;
determining one or more template session mining characteristics corresponding to the template online session behavior data according to the determined session emotion texts and the determined session intention texts, wherein each template session mining characteristic comprises a plurality of session intention knowledge vectors and a plurality of session emotion knowledge vectors of an online session user;
according to the determined conversation mining characteristics of each template and target conversation mining characteristics corresponding to each candidate online conversation behavior data in cloud online behavior big data of cloud online services, target online conversation behavior data are determined from each candidate online conversation behavior data, and a first user corresponding to each template online conversation behavior data is pushed to a second user corresponding to the corresponding determined target online conversation behavior data.
2. The big data analysis method applied to the cloud online service, according to claim 1, wherein the determining one or more template session mining characteristics corresponding to the template online session behavior data according to the determined session emotion texts and the determined session intention texts comprises:
determining conversation emotion intention contact information between each conversation emotion text and each conversation intention text according to each determined conversation emotion text and each determined conversation intention text;
and determining one or more template session mining characteristics corresponding to the template online session behavior data according to the session emotion intention contact information and the first correlation parameters between the session intention knowledge vectors corresponding to the session intention texts.
3. The big data analysis method applied to the cloud online service, according to claim 2, wherein the determining of the session emotion intention contact information between each session emotion text and each session intention text according to each determined session emotion text and each determined session intention text comprises:
calculating semantic association degrees between one or more conversation intention texts covered in one template conversation behavior and one or more conversation emotion texts covered in the one template conversation behavior respectively aiming at each template conversation behavior in the plurality of template conversation behaviors;
logically connecting the conversation emotion texts with the conversation intention texts, wherein the corresponding semantic association degrees of the conversation intention texts and the conversation emotion texts are greater than or equal to a preset semantic association degree;
determining one or more template session mining characteristics corresponding to the template online session behavior data according to the session emotion intention contact information and the first association parameters between the session intention knowledge vectors corresponding to the session intention texts respectively, wherein the method comprises the following steps:
calculating first association parameters between the session intention knowledge vectors corresponding to the session intention texts respectively, and determining the session intention linkage information of two associated session behaviors corresponding to the template online session behavior data according to the calculated first association parameters;
determining one or more template conversation mining characteristics corresponding to the template online conversation behavior data according to the conversation emotion intention contact information and the associated conversation intention linkage information of the two conversation behaviors;
the calculating first association parameters between the session intention knowledge vectors corresponding to the session intention texts respectively, and determining the session intention linkage information of the two associated session behaviors corresponding to the template online session behavior data according to the calculated first association parameters includes:
aiming at the plurality of template conversation behaviors, sequentially acquiring each conversation intention text cluster based on the time sequence direction of the online conversation behavior data, wherein each conversation intention text cluster comprises: a first conversation intention text in a current template conversation behavior and a second conversation intention text in a next template conversation behavior;
calculating first association parameters corresponding to the conversation intention text clusters respectively, wherein each first association parameter is used for indicating an association parameter between the corresponding first conversation intention text and the corresponding second conversation intention text;
and determining target conversation intention text clusters with corresponding first associated parameters more than or equal to first set associated parameters from the conversation intention text clusters, and determining the conversation intention linkage information of the two associated conversation behaviors according to the determined target conversation intention text clusters.
4. The big data analysis method applied to the cloud online service according to claim 2, wherein after determining one or more template session mining features corresponding to the template online session behavior data according to the determined session emotion texts and the determined session intention texts, the method further comprises:
if a plurality of template session mining features exist, determining a second association parameter between a plurality of session emotional features covered by the plurality of template session mining features;
and determining feature contact information among the plurality of template session mining features according to the determined second correlation parameters.
5. The big data analysis method applied to the cloud online service, according to claim 4, wherein the determining feature contact information among the plurality of template session mining features according to the determined second associated parameters comprises:
calculating a plurality of conversation emotion knowledge vectors covered in one template conversation excavation feature aiming at any two template conversation excavation features covered in the plurality of template conversation excavation features, respectively relating second parameters with a plurality of conversation emotion knowledge vectors covered in another template conversation excavation feature, and determining target second relating parameters of which corresponding numerical values are larger than or equal to preset second set relating parameters from the calculated second relating parameters;
and when the two template session mining characteristics belong to the same online session user according to the determined second association parameters of each target, configuring the characteristic contact information between the two template session mining characteristics in the characteristic contact information.
6. The big data analysis method applied to the cloud online service according to any one of claims 1 to 5, wherein the determining of the target online conversation behavior data from the candidate online conversation behavior data according to the determined template conversation mining characteristics and the target conversation mining characteristics corresponding to the candidate online conversation behavior data in the cloud online behavior big data of the cloud online service comprises:
determining feature association parameters between each target session mining feature and each template session mining feature according to each template session mining feature and each target session mining feature corresponding to each candidate online session behavior data in cloud online behavior big data of cloud online service;
determining data association parameters between the candidate online conversation behavior data and the template online conversation behavior data according to the determined feature association parameters;
determining one or more candidate online conversation behavior data with data association parameters between the one or more candidate online conversation behavior data and the template online conversation behavior data larger than preset online conversation behavior data set association parameters from the candidate online conversation behavior data, and taking the determined one or more candidate online conversation behavior data as target online conversation behavior data;
determining feature association parameters between each target session mining feature and each template session mining feature according to each template session mining feature and target session mining features corresponding to each candidate online session behavior data in cloud online behavior big data of cloud online services, wherein the feature association parameters comprise:
determining conversation intention knowledge vector association parameters between each target conversation excavation feature and each template conversation excavation feature according to conversation intention knowledge vectors covered by each template conversation excavation feature and conversation intention knowledge vectors covered by each target conversation excavation feature;
determining session emotion knowledge vector association parameters between the target session mining features and the template session mining features according to the session emotion knowledge vectors covered by the template session mining features and the session emotion knowledge vectors covered by the target session mining features;
and determining feature association parameters between the target session mining features and the template session mining features according to the determined session intention knowledge vector association parameters and the determined session emotion knowledge vector association parameters.
7. The big data analysis method applied to the cloud online service according to any one of claims 1 to 5, wherein the determining one or more conversation emotion texts and corresponding conversation emotion knowledge vectors covered by the template conversation behaviors comprises:
loading one or more conversation emotion texts covered by the template conversation behaviors into a target conversation emotion coding model meeting a model convergence condition respectively, and determining a corresponding conversation emotion knowledge vector, wherein the target conversation emotion coding model is generated by updating a traversal loop model weight parameter, and the following steps are performed in each traversal loop stage:
according to each association model preparation data combination covered in the example model preparation data sequence and the load quantity of the training round set in advance, constructing a data sequence to be learned corresponding to each training round, wherein each data to be learned comprises at least three example conversation behavior data, and the association model preparation data combination corresponding to one example conversation behavior data in the at least three example conversation behavior data is different from the association model preparation data combination corresponding to other example conversation behavior data;
loading each constructed data sequence to be learned to an initial session emotion coding model in training turns, determining corresponding overall session emotion coding cost values, updating model weight parameters of the initial session emotion coding model according to the obtained overall session emotion coding cost values, and outputting a target session emotion coding model when the model convergence conditions are met;
the method for constructing the data sequence to be learned corresponding to each training round according to each associated model preparation data combination covered in the example model preparation data sequence and the load of the training round set in advance comprises the following steps:
extracting associated model preparation data combinations corresponding to the training rounds from the associated model preparation data combinations covered in the example model preparation data sequence according to the load quantity of the training rounds set in advance;
for each association model preparation data combination, respectively taking one example conversation behavior data contained in one association model preparation data combination as reference example conversation behavior data, respectively extracting corresponding one other example conversation behavior data from other association model preparation data combinations, and calculating second reference association parameters between the extracted other example conversation behavior data and the reference example conversation behavior data;
determining one or more target example session behavior data from the other example session behavior data according to the calculated second reference correlation parameters, and determining one or more data to be learned according to the one or more target example session behavior data and the correlation model preparation data combination;
the association model preparation data combination is determined based on the following steps:
performing session emotion analysis on each sample session behavior data in the sample model preparation data sequence, and determining a session emotion text and a corresponding session emotion knowledge vector which respectively correspond to each sample session behavior data;
clustering the obtained emotion knowledge vectors of each session, and determining each cluster;
determining the session emotion knowledge vectors meeting the set clustering requirements from each cluster according to the correlation metric values among the session emotion knowledge vectors in each cluster, and determining preparation data combinations of each association model according to the session emotion texts corresponding to the determined session emotion knowledge vectors;
determining the session emotion knowledge vectors meeting the set clustering requirements from each cluster according to the correlation metric values among the session emotion knowledge vectors in each cluster, comprising:
determining clustering weights corresponding to the conversation emotion knowledge vectors in each cluster according to the correlation metric values among the conversation emotion knowledge vectors in each cluster;
determining the conversation emotion knowledge vectors meeting the set clustering requirements from each cluster according to the clustering weight corresponding to each conversation emotion knowledge vector in each cluster;
determining the clustering weight corresponding to each conversation emotion knowledge vector in each cluster according to the correlation metric value among the conversation emotion knowledge vectors in each cluster, including:
aiming at each conversation emotion knowledge vector in each cluster, the following steps are carried out:
calculating a session emotion knowledge vector in a cluster, respectively calculating correlation metric values between the session emotion knowledge vector and other session emotion knowledge vectors except the session emotion knowledge vector in the cluster, and determining a preset number of session emotion knowledge vectors from the other session emotion knowledge vectors based on the calculated correlation metric values;
determining clustering weight of the session emotion knowledge vector according to the determined correlation metric value between each session emotion knowledge vector and the session emotion knowledge vector;
determining the session emotion knowledge vectors meeting the set clustering requirements from each cluster according to the clustering weight corresponding to each session emotion knowledge vector in each cluster, including:
determining a set clustering weight corresponding to each cluster according to a clustering weight corresponding to each session emotion knowledge vector in each cluster;
and determining the corresponding session emotion knowledge vector with the clustering weight smaller than the set clustering weight from the session emotion knowledge vectors in the clusters respectively, and taking the determined session emotion knowledge vector as the session emotion knowledge vector meeting the set clustering requirement.
8. The big data analysis method applied to the cloud online service, according to any one of claims 1 to 5, wherein the method further comprises:
acquiring a user interaction window established by the second user and the pushed first user, and acquiring user interaction text data in the user interaction window;
mining an intentional knowledge point of progress on the user interactive text data to generate an intentional knowledge point of progress;
and performing content pushing on the user group where the first user and the second user are located based on the advanced intention knowledge point.
9. The big data analysis method applied to the cloud online service, according to claim 8, wherein the step of performing advanced knowledge point mining on the user interaction text data to generate an advanced knowledge point comprises:
analyzing an interactive semantic keyword directed graph from the user interactive text data, wherein the interactive semantic keyword directed graph comprises W groups of interactive semantic keywords with semantic relation, and W is an integer not less than 1;
acquiring an attention keyword directed graph by using the interactive semantic keyword directed graph, wherein the attention keyword directed graph comprises W groups of attention keywords with semantic relation;
mining a first topic feature vector sequence by utilizing the interactive semantic keyword directed graph and utilizing a first topic feature analysis unit in a neural network model, wherein the first topic feature vector sequence comprises W first topic feature vectors;
mining a second topic feature vector sequence by using a second topic feature analysis unit in the neural network model by using the attention keyword directed graph, wherein the second topic feature vector sequence comprises W second topic feature vectors;
determining intention knowledge point prediction information corresponding to the interactive semantic keyword directed graph by using the first topic feature vector sequence and the second topic feature vector sequence and an intention knowledge point prediction unit in the neural network model;
and determining the advanced intention knowledge point of the interactive semantic keyword directed graph by using the intention knowledge point prediction information.
10. An artificial intelligence system, comprising a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions, and the machine-executable instructions are loaded and executed by the processor to implement the big data analysis method applied to the cloud online service of any one of claims 1 to 9.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310140645.4A CN115982323A (en) | 2023-02-21 | 2023-02-21 | Big data analysis method and artificial intelligence system applied to cloud online service |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310140645.4A CN115982323A (en) | 2023-02-21 | 2023-02-21 | Big data analysis method and artificial intelligence system applied to cloud online service |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115982323A true CN115982323A (en) | 2023-04-18 |
Family
ID=85961659
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310140645.4A Pending CN115982323A (en) | 2023-02-21 | 2023-02-21 | Big data analysis method and artificial intelligence system applied to cloud online service |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115982323A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116450671A (en) * | 2023-06-12 | 2023-07-18 | 大白熊大数据科技(常熟)有限公司 | Intelligent interaction session big data analysis method and big data server |
CN117591662A (en) * | 2024-01-19 | 2024-02-23 | 川投信息产业集团有限公司 | Digital enterprise service data mining method and system based on artificial intelligence |
-
2023
- 2023-02-21 CN CN202310140645.4A patent/CN115982323A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116450671A (en) * | 2023-06-12 | 2023-07-18 | 大白熊大数据科技(常熟)有限公司 | Intelligent interaction session big data analysis method and big data server |
CN116450671B (en) * | 2023-06-12 | 2023-08-22 | 大白熊大数据科技(常熟)有限公司 | Intelligent interaction session big data analysis method and big data server |
CN117591662A (en) * | 2024-01-19 | 2024-02-23 | 川投信息产业集团有限公司 | Digital enterprise service data mining method and system based on artificial intelligence |
CN117591662B (en) * | 2024-01-19 | 2024-03-29 | 川投信息产业集团有限公司 | Digital enterprise service data mining method and system based on artificial intelligence |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US10958748B2 (en) | Resource push method and apparatus | |
CN110366734B (en) | Optimizing neural network architecture | |
US20190354810A1 (en) | Active learning to reduce noise in labels | |
CN111582453B (en) | Method and device for generating neural network model | |
CN115982323A (en) | Big data analysis method and artificial intelligence system applied to cloud online service | |
CN109376267B (en) | Method and apparatus for generating a model | |
CN107947951A (en) | Groups of users recommends method, apparatus and storage medium and server | |
CN110427560A (en) | A kind of model training method and relevant apparatus applied to recommender system | |
CN110413867B (en) | Method and system for content recommendation | |
CN111353303B (en) | Word vector construction method and device, electronic equipment and storage medium | |
CN113505272B (en) | Control method and device based on behavior habit, electronic equipment and storage medium | |
CN111159563A (en) | Method, device and equipment for determining user interest point information and storage medium | |
US20230252070A1 (en) | Method and apparatus for training retrieval model, retrieval method and apparatus, device and medium | |
CN112395396A (en) | Question-answer matching and searching method, device, system and storage medium | |
CN113821657A (en) | Artificial intelligence-based image processing model training method and image processing method | |
CN113742580B (en) | Recall method and device for target type data, electronic equipment and storage medium | |
WO2024093578A1 (en) | Voice recognition method and apparatus, and electronic device, storage medium and computer program product | |
US11976931B2 (en) | Method and apparatus for guiding voice-packet recording function, device and computer storage medium | |
CN116383521B (en) | Subject word mining method and device, computer equipment and storage medium | |
CN113190696A (en) | Training method of user screening model, user pushing method and related devices | |
CN110569447B (en) | Network resource recommendation method and device and storage medium | |
CN111639199A (en) | Multimedia file recommendation method, device, server and storage medium | |
CN116070027A (en) | Online pushing topic analysis method based on artificial intelligence and big data system | |
CN113157582B (en) | Test script execution sequence determining method and device | |
CN112905885B (en) | Method, apparatus, device, medium and program product for recommending resources to user |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |