CN113329097A

CN113329097A - Content push decision-making method based on big data mining and cloud computing AI (Artificial Intelligence) service system

Info

Publication number: CN113329097A
Application number: CN202110708220.XA
Authority: CN
Inventors: 赵天硕; 黄义宝; 田巧玲
Original assignee: Dongguan Huixuehuiwan Education Technology Co ltd
Current assignee: Dongguan Huixuehuiwan Education Technology Co ltd
Priority date: 2021-06-25
Filing date: 2021-06-25
Publication date: 2021-08-31

Abstract

The embodiment of the application provides a content push decision-making method based on big data mining and a cloud computing AI service system, which can extract validity measurement bitmaps among content response polarity label information, content behavior thermodynamic diagram information and content operation feedback behavior information in content response events as judgment bases for judging whether the content response events in multiple dimensions are validity response behaviors or not, so as to determine reference bases for carrying out decision-making updating on content push rules, and therefore accuracy of content push decision-making updating is improved.

Description

Content push decision-making method based on big data mining and cloud computing AI (Artificial Intelligence) service system

Technical Field

The application relates to the technical field of content push, in particular to a content push decision-making method based on big data mining and a cloud computing AI (AI) service system.

Background

Big data is complex and various, the updating speed is abnormally high, and the scale is larger. But has high analytical new value, so that the research of big data plays a decisive role in the future development trend of the business, such as how to perform information mining to decide the interesting conversation content of the user so as to improve the trend of future business updating.

In the related art, under the promotion of factors such as the current big data technology, the predicted interest point set of the user is mined through the big data to carry out targeted content push, and the method is a content service scheme which is generally used. For the pushed content information, the user also makes relevant content response events in a targeted manner, and in the related art, secondary mining is generally continued for all content response events so as to continuously perform optimized updating on content pushing rules. However, the inventor researches and discovers that the content response events inevitably have some misoperation or noise operation, and if the content response events are dug secondarily, noise characteristics are introduced, so that errors occur in the process of continuously optimizing and updating the content push rules.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, the present application aims to provide a content push decision method and a cloud computing AI service system based on big data mining.

In a first aspect, the present application provides a content push decision method based on big data mining, which is applied to a cloud computing AI service system, where the cloud computing AI service system is in communication connection with a plurality of intelligent service registration terminals, and the method includes:

acquiring a predicted interest point set of a target intelligent service conversation process of an intelligent service registration terminal, and performing initial information push on the intelligent service registration terminal according to the predicted interest point set to acquire a plurality of content response events of initial push content information;

identifying each content response event in the plurality of content response events, and respectively obtaining at least two of corresponding content behavior thermodynamic diagram information, content operation feedback behavior information and content response polarity label information;

if conflict characteristics exist between at least two of the content response polarity label information, the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a first effectiveness metric value corresponding to each content response event from a preset effectiveness metric bitmap;

and determining validity identification information of the content response events based on a plurality of first validity metric values corresponding to the content response events, and performing decision updating on a content push rule of the intelligent service registration terminal according to the validity identification information of the content response events.

In a second aspect, an embodiment of the present application further provides a content push decision system based on big data mining, where the content push decision system based on big data mining includes a cloud computing AI service system and a plurality of intelligent service registration terminals in communication connection with the cloud computing AI service system;

the cloud computing AI service system is used for:

Based on any one of the above aspects, after the plurality of content response events of the initial pushed content information are acquired, each content response event in the plurality of content response events can be identified, and at least two of corresponding content behavior thermodynamic diagram information, content operation feedback behavior information, and content response polarity tag information can be obtained respectively. And if conflict characteristics exist between at least two of the content response polarity label information, the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a first effectiveness metric value corresponding to each content response event from a preset effectiveness metric bitmap. And finally, determining the validity identification information of the content response events based on a plurality of first validity metric values corresponding to the content response events. Therefore, the validity measurement bitmap among the content response polarity label information, the content behavior thermodynamic diagram information and the content operation feedback behavior information in the content response event can be extracted as a judgment basis for judging whether the content response events of multiple dimensions are validity response behaviors or not, so as to determine a reference basis for carrying out decision updating on the content push rules, and improve the accuracy of the content push decision updating.

Drawings

Fig. 1 is a schematic view of an application scenario of a content push decision system based on big data mining according to an embodiment of the present application;

fig. 2 is a schematic flowchart of a content push decision method based on big data mining according to an embodiment of the present application;

fig. 3 is a schematic block diagram of structural components of a cloud computing AI service system for implementing the content push decision method based on big data mining according to an embodiment of the present application.

Detailed Description

The present application will now be described in detail with reference to the drawings, and the specific operations in the method embodiments may also be applied to the apparatus embodiments or the system embodiments.

Fig. 1 is an interaction diagram of a big data mining-based content push decision system 10 according to an embodiment of the present application. The content push decision system 10 based on big data mining may include a cloud computing AI service system 100 and an intelligent business registration terminal 200 communicatively connected to the cloud computing AI service system 100. The big data mining based content push decision system 10 shown in fig. 1 is only one possible example, and in other possible embodiments, the big data mining based content push decision system 10 may also include only at least some of the components shown in fig. 1 or may also include other components.

In an embodiment, the cloud computing AI service system 100 and the intelligent service registration terminal 200 in the big data mining based content push decision system 10 can cooperatively perform the big data mining based content push decision method described in the following method embodiment, and the detailed description of the method embodiment can be referred to in the following steps of the cloud computing AI service system 100 and the intelligent service registration terminal 200.

Fig. 2 is a schematic flow chart of a content push decision method based on big data mining according to an embodiment of the present disclosure, where the content push decision method based on big data mining according to the present disclosure may be executed by the cloud computing AI service system 100 shown in fig. 1, and the content push decision method based on big data mining is described in detail below.

Step S101: the method comprises the steps of obtaining a prediction interest point set of a target intelligent service conversation process of an intelligent service registration terminal, carrying out initial information pushing on the intelligent service registration terminal according to the prediction interest point set, and obtaining a plurality of content response events of initial pushed content information.

Step S102: and identifying each content response event in the plurality of content response events, and respectively obtaining at least two of corresponding content behavior thermodynamic diagram information, content operation feedback behavior information and content response polarity label information.

In a separately implementable embodiment, the content response event may include information of multiple different dimensions, such as, but not limited to, content behavior thermodynamic diagram information, content operation feedback behavior information, and content response polarity tag information, as described above. Further, the content behavior thermodynamic diagram information may be thermal value distribution information of different content behaviors, and the content behavior may be a content browsing behavior, a content forwarding behavior, and the like. The content operation feedback behavior information may be information recorded by content operation feedback behaviors that are correlated with respect to the content information, such as mark behaviors that are not of interest, subscription behaviors that are of interest, and the like. The content response polarity tag information may include a positive polarity tag or a negative polarity tag for different content information, and the like, but is not limited thereto.

It can be understood that, by performing information mining on the content response event to obtain the content behavior thermodynamic diagram information, the content operation feedback behavior information and the content response polarity label information, a data basis is provided for the subsequent analysis of the validity content response event.

Step S103: and if conflict characteristics exist between at least two of the content response polarity label information, the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a first effectiveness metric value corresponding to each content response event from a preset effectiveness metric bitmap.

In an embodiment, the presence of a conflict feature between at least two of the content response polarity tag information, the content behavior thermodynamic diagram information and the content operation feedback behavior information may be understood as a mismatch or a conflict between at least two of the content response polarity tag information, the content behavior thermodynamic diagram information and the content operation feedback behavior information, for example, an abnormality occurs in an information association of the content response polarity tag information, the content behavior thermodynamic diagram information and the content operation feedback behavior information, in which case, it indicates that a corresponding content response event may have a validity problem, and for this reason, a first validity metric value of the corresponding content response event may be determined from a preset validity metric bitmap.

In addition, the first effectiveness metric value can be understood as related analysis value information for providing reference for the effectiveness analysis of the content response event.

In a related embodiment, the first validity metric value corresponding to each content response event may include: a first aggregation effectiveness metric value and a second aggregation effectiveness metric value. Wherein the first aggregate validity metric value may be for local content response events and the second aggregate validity metric value may be for cross-border content response events. Alternatively, the first aggregation effectiveness metric value may also be for a real-time content response event, and the second aggregation effectiveness metric value may also be for a delayed content response event, which is not limited in this embodiment of the present application.

On the basis of the above, if there is a conflict feature between at least two of the content response polarity tag information, the content behavior thermodynamic diagram information, and the content operation feedback behavior information, which are described in step S103, a first validity metric value corresponding to each content response event is obtained from a preset validity metric bitmap, which may be implemented by the following implementation a or implementation B.

Embodiment A: and if the conflict characteristics exist between the content response polarity label information and the content operation feedback behavior information, obtaining a first aggregation effectiveness metric value corresponding to each content response event from a first preset effectiveness metric bitmap.

For embodiment a, if there is a conflict feature between the content response polarity tag information and the content operation feedback behavior information, obtaining a first aggregated validity measure value corresponding to each content response event from a first preset validity measure bitmap includes: and if the content operation feedback behavior information and the content response polarity label information contain a matched first target effectiveness metric bitmap in a first preset effectiveness metric bitmap, indicating that the first aggregation effectiveness metric value corresponding to the first target effectiveness metric bitmap is obtained from the preset effectiveness metric bitmap.

In other words, the corresponding first aggregate validity metric value may also be determined from two dimensions of the content operation feedback behavior information and the content response polarity tag information.

In an embodiment that can be implemented independently, after the step of obtaining the first aggregated validity measure value corresponding to each content response event from the first preset validity measure bitmap if the conflict feature exists between the content response polarity tag information and the content operation feedback behavior information as described in implementation a, the method may further include the following steps: summarizing the label distribution quantity of the first effective content operation feedback behavior information corresponding to the first aggregation effectiveness metric value; and if the distribution quantity of the labels of the first effective content operation feedback behavior information is greater than a first quantity, the effectiveness identification information of the content response event corresponding to the first effective content operation feedback behavior information is an effective content response event.

It will be appreciated that the validity identification information can be determined in combination with the first number by counting the number of tag distributions of the first valid content operation feedback behavior information corresponding to the first aggregate validity metric value. For example, if the distribution number of the tags of the first effective content operation feedback behavior information is greater than the first number, which indicates that the content operation feedback behavior information meets the validity requirement, the validity identification information of the content response event corresponding to the first effective content operation feedback behavior information may be determined as an effective content response event.

Embodiment B: and if the conflict characteristics exist between the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a second aggregation effectiveness metric value corresponding to each content response event from a second preset effectiveness metric bitmap.

For embodiment B, if there is a conflict feature between the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a second aggregated effectiveness metric value corresponding to each content response event from a second preset effectiveness metric bitmap includes: and if the content behavior thermodynamic diagram information and the content operation feedback behavior information contain a matched second target effectiveness metric bitmap in a second preset effectiveness metric bitmap, indicating that the second aggregation effectiveness metric value corresponding to the second target effectiveness metric bitmap is obtained from the preset effectiveness metric bitmap.

In other words, the corresponding first aggregate effectiveness metric value may also be determined based on two reference dimensions of the content behavior thermodynamic diagram information and the content operation feedback behavior information.

In an embodiment, after obtaining the second aggregated effectiveness metric value corresponding to each content response event from the second preset effectiveness metric bitmap if there is a conflict feature between the content behavior thermodynamic diagram information and the content operation feedback behavior information as described in embodiment B, the method further includes: summarizing the label distribution quantity of the second effective content operation feedback behavior information corresponding to the second aggregation effectiveness metric value; and if the distribution quantity of the labels of the second effective content operation feedback behavior information is greater than a second quantity, the effectiveness identification information of the content response event corresponding to the second effective content operation feedback behavior information is an effective content response event.

Step S104: and determining validity identification information of the content response events based on a plurality of first validity metric values corresponding to the content response events, and performing decision updating on a content push rule of the intelligent service registration terminal according to the validity identification information of the content response events.

In a separately implementable embodiment, the validity identification information includes a non-valid content response event and a valid content response event.

In an independently implementable embodiment, after the obtaining of the plurality of content response events of the initial push content information described in step S101, the method further comprises: querying whether the plurality of content response events contain subscription content response characteristics, wherein the subscription content response characteristics comprise: subscription portrait features or interactive portrait features; if each content response event covers the subscription content response feature, obtaining a second validity metric value corresponding to each content response event from a preset subscription content response feature bitmap, thereby determining a plurality of second validity metric values corresponding to a plurality of content response events.

By such design, the second validity metric value can be obtained from the corresponding subscription content response characteristic bitmap query based on the subscription content response characteristic. Based on this, the determining the validity identification information of the plurality of content response events based on the plurality of first validity metric values corresponding to the plurality of content response events described in step S104 may also be implemented by: determining validity identifying information for the plurality of content response events based on the plurality of first validity metric values and the plurality of second validity metric values.

It is understood that, when determining the validity identification information of the plurality of content response events, by performing analysis in combination with the first validity metric value and the second validity metric value, the information of the subscription portrait characteristic or the interaction portrait characteristic dimension can be taken into account, thereby further ensuring the accuracy of the validity identification information.

In an independently implementable embodiment, after the obtaining of the plurality of content response events of the initial push content information described in step S101, the method may further comprise: acquiring a plurality of behavior migration events corresponding to the plurality of content response events; the plurality of behavior migration events are obtained by the cloud computing AI service system making a response after obtaining the plurality of content response events; identifying the plurality of behavior migration events to obtain behavior migration content information corresponding to each behavior migration event; and if the behavior migration content information represents that the migration content information which needs to be shared by the same content object exists, obtaining a third validity metric value corresponding to each content response event from a preset migration sharing metric value bitmap, so as to determine at least one third validity metric value corresponding to a plurality of content response events.

In an independently implementable embodiment, the third measure of effectiveness focuses on the behavioral migration event level. Based on this, the determining the validity identification information of the plurality of content response events based on the plurality of first validity metric values corresponding to the plurality of content response events, which is described in step S104, may include any one of the following manners (1) or (2): (1) determining validity identification information for the plurality of content response events based on the plurality of first validity metric values and the at least one third validity metric value; (2) determining validity identification information for the plurality of content response events based on the plurality of first validity metric values, the plurality of second validity metric values, and the at least one third validity metric value.

In an embodiment that can be implemented independently, the determining the validity identification information of the content response events based on the first validity metric values corresponding to the content response events described in step S104 may include the following: if the first effectiveness metric values corresponding to the content response events contain first target effectiveness metric values larger than the first target metric values, determining that the effectiveness identification information of the content response events corresponding to the first target effectiveness metric values is effective content response events.

In an embodiment, the first target metric value is used for performing validity analysis, and if a first validity metric value greater than the first target metric value exists, the corresponding content response event is indicated to be valid for the reference basis of information pushing. It is to be understood that a content response event may correspond to one or more first validity metric values, and if a first target validity metric value exists in the one or more first validity metric values, the validity identification information of the content response event corresponding to the first target validity metric value may be determined as a valid content response event.

In an embodiment that can be implemented independently, the determining, in step S104, validity identification information of the content response events based on a plurality of first validity metric values corresponding to the content response events may also be implemented through the following steps S1041 to S1043.

Step S1041: and if a plurality of first effectiveness metric values corresponding to the plurality of content response events contain a second target effectiveness metric value which is not larger than the first target metric value, carrying out integral discretization treatment on the second target effectiveness metric value to obtain a first discrete effectiveness metric value.

Step S1042: and if the first discrete effectiveness metric value is not greater than the first discrete target metric value, the effectiveness identification information of the content response event corresponding to the second target effectiveness metric value is a non-effective content response event.

Step S1043: and if the first discrete effectiveness metric value is larger than a first discrete target metric value, the effectiveness identification information of the content response event corresponding to the second target effectiveness metric value is an effective content response event.

In an embodiment that can be implemented independently, the overall discretization process can be understood as weight fusion calculation, and the obtained first discrete validity measure can be understood as a weight fusion calculation result.

In a separately implementable embodiment, the method may further comprise: identifying each content response event in the plurality of content response events, and respectively obtaining the sharing characteristics of the same content object respectively corresponding to the content response events; summarizing statistics of the sharing characteristics of the same content objects corresponding to each content response event; and determining validity identification information of the plurality of content response events according to at least one of a plurality of second validity metric values and the at least one third validity metric value, the plurality of first validity metric values and statistics of the same content object sharing characteristics.

In an embodiment that can be implemented independently, the same content object sharing feature uniquely corresponding to each content response event may be a parameterized feature, and based on this, the same content object sharing features may be summarized, and the determination of the validity identification information may be performed by combining the second validity metric value, the third validity metric value, the first validity metric value, and the same content object sharing feature. For example, the validity identification information may be determined by combining the second validity metric value, the first validity metric value, and the same content object sharing feature, and the validity identification information may be determined by combining the third validity metric value, the first validity metric value, and the same content object sharing feature.

For example, in an embodiment that can be implemented independently, the decision updating of the content push rule of the intelligent service registration terminal according to the validity identification information of the plurality of content response events described in the above step S104 may include the following steps.

Step S1040: and if the effective identification information of the content response events is an effective content response event, extracting the user attention feature of the effective content response event to obtain the user attention feature of the initial push content information, and performing decision updating on the content push rule of the intelligent service registration terminal based on the user attention feature of the initial push content information.

For example, the user attention feature extraction may be performed in combination with a deep learning model, so as to accurately obtain the user attention feature of the initial push content information. Further, an attention feature vector corresponding to the user attention feature of the initial push content information can be analyzed, and the weight of a content push sub-rule corresponding to the content push rule of the intelligent service registration terminal is correspondingly adjusted according to the tendency degree of the attention feature vector corresponding to the user attention feature.

In an embodiment that can be implemented independently, step S101 can be implemented by the following exemplary steps.

Step A110, extracting a conversation intention cluster from a target intelligent business conversation process to be mined.

Research shows that for any intelligent business conversation process, the intelligent business conversation process usually includes a large number of conversation intents, and the large number of conversation intents may include one or more conversation track data; the session trace data is session log information formed by session process records, and further forms session trace big data. And optionally, one or more non-dialog trajectory data may also be included in the large number of dialog intents; by non-dialog trace data is meant dialog log information consisting of other operational behavior related information besides dialog process records, where the other operational behavior related information may include, but is not limited to: the operation behavior is located in the service, the operation behavior source, the operation behavior interaction mode and the like. Accordingly, the non-dialog trace data may be dialog log information composed of the service where the operation behavior is located, dialog log information composed of the source of the operation behavior, and so on.

Because these large numbers of conversation intentions are used to describe the operation behavior information (such as conversation process records, the business where the operation behavior is located, the operation behavior source, etc.) involved in the intelligent business conversation process, and the interest content features of the intelligent business conversation process are used to perform predictive analysis on the relevant content of the main operation behavior (i.e. the key operation behavior) involved in the intelligent business conversation process; it can be seen that the conversation intention in the intelligent business conversation process is connected with the interesting content characteristics of the intelligent business conversation process to a certain extent. Therefore, the embodiment of the application provides a technical solution for determining the content of interest characteristics of the intelligent business conversation process through the conversation intention included in the intelligent business conversation process. Based on the technical scheme, when there is a mining request about the dialog content characteristics of the target intelligent business dialog flow, the cloud computing AI service system 100 may obtain the target intelligent business dialog flow to be mined. Then, a dialogue intention cluster can be extracted from the target intelligent business dialogue process through the step; the dialog intent cluster herein can include a plurality of dialog intents, and the plurality of dialog intents includes at least dialog trajectory data. In an embodiment, the cloud-computing AI service system 100 may perform dialog intention clustering on the target intelligent business dialog process, and match an initial dialog intention cluster obtained by the dialog intention clustering with the at least one intention feature library to match and extract an initial dialog intention contained in the initial dialog intention cluster and located in the at least one dialog intention reference model. And then, determining the conversation intention in the target intelligent business conversation process according to the matched and extracted initial conversation intention, thereby constructing a conversation intention cluster for obtaining the target intelligent business conversation process.

Step A120, an intention point space of the target intelligent business conversation process is constructed by adopting a plurality of conversation intents.

In an independently implementable embodiment, the target intelligent business conversation process may include a plurality of intent points in an intent point space; one intention point reflects one conversation intention, and the conversation intentions reflected by any two intention points with binding relationship have conversation circulation operation in the target intelligent business conversation process. That is, in the target intelligent business conversation process, there are intent points corresponding to two conversation intentions of the conversation circulation operation, and binding relationship exists in the intent point space. Wherein, the dialog circulation operation mentioned here can include any one of the following meanings:

in an embodiment, the above-mentioned dialog circulation operation may refer to: in the process of carrying out conversation circulation on the target intelligent business conversation process by adopting a conversation guide reference service, two conversations intend to simultaneously appear in the circulation process behavior of the conversation guide reference service. The number of quoting times of the dialogue guiding quoting service can be set according to an empirical value or a maximum dialogue number range in a target intelligent business dialogue process; for example, the maximum dialog time range in the target intelligent business dialog process is 20 page active interest dimensions, the reference time can be set to be less than or equal to 20 page active interest dimensions. For example, a dialogue guide reference service with 5 reference times as active interest dimensions of pages is adopted to perform dialogue circulation on a target intelligent business dialogue process; and setting a plurality of dialog intents including: the dialog intentions Q, W, E, R … … assume that the dialog intentions Q and W can be considered to have a dialog flow operation in the target intelligent business dialog flow because they can appear in the dialog guidance reference service at the same time during the dialog flow. Since the dialog intention W and the dialog intention E can simultaneously appear in the dialog guidance referencing service during the dialog flow, the dialog intention W and the dialog intention E can be considered to have a dialog flow operation in the target intelligent business dialog flow. Since the dialog intention E and the dialog intention R cannot simultaneously appear in the dialog guidance reference service, it can be considered that the dialog intention E and the dialog intention R do not have a dialog flow operation in the target intelligent business dialog flow, and so on.

In another embodiment, the above-mentioned dialog flow operation may refer to: in the process of carrying out conversation circulation on a target intelligent business conversation process by adopting a conversation guide reference service, two conversation intents simultaneously appear in the conversation guide reference service, and the relation that the relevant parameters of a conversation application environment between the two conversation intents are larger than the preset relevant parameter values. The dialog application environment related parameters between the two dialog intents can be obtained by calculation according to the interesting dialog track characteristics of the two dialog intents; the dialog application environment related parameter between the two dialog intents can be used for reflecting the dialog application environment feature correlation degree between the two dialog intents, and the dialog application environment related parameter is in direct proportion to the dialog application environment feature correlation degree; that is, the greater the dialog application environment-related parameter between two dialog intents, the greater the dialog application environment feature correlation between the two dialog intents. And, the preset relevant parameter values can be set according to the actual service conditions. For example, let the preset relevant parameter value be K, the dialog application environment relevant parameter between the dialog intention Q and the dialog intention W be kQW, the dialog application environment relevant parameter between the dialog intention W and the dialog intention E be kWE, and the dialog application environment relevant parameter between the dialog intention E and the dialog intention R be kER; and kQW < K, kWE > K, kER < K. Still take over the above example: since the dialog application environment-related parameter (i.e., kQW) between the dialog intention Q and the dialog intention W is smaller than the preset-related parameter value (K), the cloud computing AI service system 100 may consider the dialog intention Q and the dialog intention W not to have a dialog circulation operation in the target intelligent business dialog flow although the dialog intention Q and the dialog intention W may simultaneously appear in the dialog guidance referencing service. Since the dialog application environment-related parameter (i.e., kWE) between the dialog intention W and the dialog intention E is greater than the preset-related parameter value (K), and the dialog intention W and the dialog intention E may simultaneously appear in the dialog guidance reference service, the cloud computing AI service system 100 may consider the dialog intention W and the dialog intention E to have a dialog circulation operation in the target intelligent business dialog flow, and so on. Therefore, when judging whether the two conversation intentions have the conversation circulation operation in the target intelligent business conversation process or not, the embodiment not only considers the distance of the coverage range of the two conversation intentions in the target intelligent business conversation process through the conversation guide reference service, but also considers the conversation application environment characteristic correlation degree between the two conversation intentions, so that the judgment accuracy of the conversation circulation operation can be effectively improved, and the precision of the intention point space is improved.

Based on the above description, in the process of implementing step a120, the cloud computing AI service system 100 may first construct an initial intent point space of the target intelligent business dialog flow by using a plurality of dialog intents; the initial intent point space includes a plurality of intent points, each of which reflects a dialog intent. Next, the cloud computing AI service system 100 may select at least one pair of combinations of streamed dialog intention objects from among the plurality of dialog intents, the combination of streamed dialog intention objects being a combination of dialog intention objects composed of two dialog intents having a dialog stream operation in the target intelligent business dialog flow. Then, the cloud computing AI service system 100 can walk the combination of the dialog intent objects for each pair of flows; for the combination of the currently circulated dialog intention objects which are currently circulated, two intention points for recording two dialog intentions in the combination of the currently circulated dialog intention objects can be respectively connected in the initial intention point space; when the combination of the dialog intention objects of each circulation is wandered away, the intention point space of the target intelligent business dialog process can be obtained. For example, it may be assumed that the plurality of dialog intents includes: dialog intentions Q (recorded with intent point Q), W (recorded with intent point W), E (recorded with intent point E), R (recorded with intent point R), E (recorded with intent point E) … …; and there are a total of 5 pairs of combinations of streamed dialog intent objects from the plurality of dialog intents, which are respectively: (dialog intention Q, dialog intention W), (dialog intention Q, dialog intention R), (dialog intention W, dialog intention E), and (dialog intention R, dialog intention E). Then, the cloud computing AI service system 100 may connect the intention point Q and the intention point W, connect the intention point Q and the intention point R, connect the intention point W and the intention point E, and connect the intention point R and the intention point E, respectively, in the initial intention point space, thereby obtaining the intention point space of the target intelligent business conversation process.

Step A130, based on the dialog circulation information between the various intention points in the intention point space, the dialog intention strength of the dialog intention reflected by the various intention points is calculated.

In an embodiment which can be implemented independently, the dialogue circulation operation refers to the relation that two dialogue intentions appear in one dialogue guide reference service at the same time or refers to the relation that two dialogue intentions appear in one dialogue guide reference service at the same time and a dialogue application environment related parameter between the two dialogue intentions is larger than a preset related parameter value; therefore, it can be known that the higher the occurrence frequency of the dialog intentions with more dialog flow operations in the target intelligent business dialog flow, the more possible the intention interest of the dialog intentions with more dialog flow operations can be considered. Based on this, when the cloud computing AI service system 100 executes the step a130, for example, the number of dialog flow times of the dialog flow operation of each dialog intention may be counted according to the dialog flow information between the respective intention points in the intention point space, and the dialog intention strength of each dialog intention may be determined according to the number of dialog flow times corresponding to each dialog intention on the basis of the principle that the number of dialog flow times and the dialog intention strength are in positive correlation.

In an embodiment which can be implemented independently, the number of dialog circulation corresponding to each dialog intention can be directly used as the dialog intention strength of each dialog intention. Or carrying out regularized conversion on the conversation circulation times corresponding to each conversation intention to obtain the conversation intention strength of each conversation intention; the regularized conversion is processing for mapping the number of times of the dialog flow to a range of 0 to 1. Or, carrying out weight fusion on the conversation circulation times corresponding to each conversation intention by adopting a conversation intention strength parameter to obtain the conversation intention strength of each conversation intention; wherein, the dialogue intention intensity parameter can be set according to the actual business situation. For example, referring to the foregoing example, if the intention point Q is mapped to the intention point W and the intention point Q is mapped to the intention point R, it may be statistically determined that the dialog flow operation of the dialog intention Q reflected by the intention point Q has a dialog flow number of 2; the number of dialog circulation times can be directly used as the dialog intention strength of the dialog intention Q (i.e. the dialog intention strength is 2), or the dialog intention strength parameter (e.g. 1.5) is used to perform weight fusion on the number of dialog circulation times to obtain the dialog intention strength of the dialog intention Q (i.e. the dialog intention strength is 3), and so on.

In another embodiment, which can be implemented independently, it is shown through research that if two dialog intents have a dialog flow operation in the target intelligent business dialog flow, the dialog intention strengths of the two dialog intents will generally affect each other because the two dialog intents are simultaneously present. Based on this, when the cloud computing AI service system 100 executes step a130, for the dialog intention reflected by any arbitrary point, the dialog intention strength of the dialog intention resulting in any arbitrary point can be calculated by combining the dialog intention strength of the dialog intention reflected by the cascaded arbitrary point having a binding relationship with the arbitrary point, so as to improve the accuracy of the dialog intention strength. In an independently implementable embodiment, for a dialog intention reflected by any intention point, at least one cascade intention point having a binding relationship with the any intention point can be determined based on dialog flow information between the intention points in an intention point space; then, the dialogue intention strength of the dialogue intention reflected by any of the cascade intention points is calculated according to the dialogue intention strength of the dialogue intention reflected by each cascade intention point.

The specific implementation manner of calculating the conversation intention strength of the conversation intention reflected by any one of the cascade intention points according to the conversation intention strength of the conversation intention reflected by each cascade intention point can include any one of the following:

the first implementation mode comprises the following steps: the cloud computing AI service system 100 may calculate an initial value of the dialog intention reflected by any one of the intent points according to the number of dialog transitions of the dialog transition operation of the dialog intention reflected by any one of the intent points. Secondly, the times of the dialog intention reflected by any one of the mind points and the dialog intention reflected by each cascade mind point appearing in the dialog guide reference service at the same time can be respectively counted, and the counted times are respectively subjected to regularized conversion to obtain the intention evaluation index of each cascade mind point. For example, for the dialog intention Q reflected by the intention point Q, the intention point Q has two cascade intention points of an intention point W and an intention point R; if the number of times that the dialogue intention Q and the dialogue intention W reflected by the intention point W simultaneously appear in the dialogue guide reference service is 15, the number of times that the dialogue intention Q and the dialogue intention R reflected by the intention point R simultaneously appear in the dialogue guide reference service is 5; the intention evaluation index of the intention point W is 15/(15+5) to 0.75, and the intention evaluation index of the intention point R is 5/(15+5) to 0.25. After the intention evaluation indexes of all the cascade intention points are obtained, the intention evaluation indexes of all the cascade intention points can be adopted to carry out weighted summation on the conversation intention strength of all the cascade intention points; for example, if the dialog intention strength of the intention point W is 0.4 and the dialog intention strength of the intention point R is 0.2, 0.4 × 0.75+0.2 × 0.25 may be executed, or 0.35 may be executed. Then, the numeric value obtained by weighted summation and the initial value of the dialogue intention reflected by any one of the intention points can be summed to obtain the dialogue intention strength of the dialogue intention reflected by any one of the intention points.

The second embodiment: the cloud-computing AI service system 100 may further calculate the conversation intention strength of the conversation intention reflected by any of the cascaded intention points according to the conversation intention strength of the conversation intention reflected by the respective cascaded intention points.

In another embodiment, since the dialog application environment-related parameter can be used to indicate the dialog application environment feature correlation between two dialog intents, research shows that for any dialog intention, the greater the dialog application environment feature correlation between other dialog intents and the dialog intention, the greater the influence of the dialog intention strength of the other dialog intention on the dialog intention strength of the dialog intention will be. Based on this, when the cloud computing AI service system 100 executes step a130, for the dialog intention reflected by any arbitrary point, the dialog intention strength of the dialog intention reflected by any arbitrary point can be calculated by combining the dialog intention strength of the dialog intention reflected by the cascade arbitrary point having a binding relationship with the arbitrary point and the dialog application environment related parameters between the dialog intention reflected by any arbitrary point and the dialog intents reflected by the respective cascade arbitrary points, so as to further improve the precision of the dialog intention strength. In an embodiment, for a dialog intention reflected by any intention point, at least one cascade intention point having a binding relationship with any intention point can be determined based on dialog flow information between the intention points in the intention point space. Then, the dialogue application environment related parameters between the dialogue intention reflected by any one of the mind points and the dialogue intentions reflected by all the cascade mind points can be calculated; and calculating the conversation intention strength of the conversation intention reflected by any one of the cascade intention points according to the calculated conversation application environment related parameters and the conversation intention strength of the conversation intention reflected by each cascade intention point.

According to the calculated conversation application environment related parameters and the conversation intention strength of the conversation intention reflected by each cascade intention point, the specific implementation mode of calculating the conversation intention strength of the conversation intention reflected by any intention point can comprise any one of the following steps:

the first implementation mode comprises the following steps: the cloud computing AI service system 100 may calculate an initial value of the dialog intention reflected by any one of the intent points according to the number of dialog transitions of the dialog intention reflected by any one of the intent points. Secondly, the conversation intention strength of each cascade intention point can be weighted and summed by adopting each conversation application environment related parameter respectively. For example, for the dialog intention Q reflected by the intention point Q, the intention point Q has two cascade intention points of an intention point W and an intention point R; and the intensity of the dialogue intention of the intention point W is 0.4, and the intensity of the dialogue intention of the intention point R is 0.2. If the dialog application environment related parameter between the dialog intention Q and the dialog intention W reflected by the intention point W is kQW, the dialog application environment related parameter between the dialog intention Q and the dialog intention R reflected by the intention point R is kQR; 0.4 × kQW +0.2 × kQR may be performed. Then, the numerical value obtained by weight fusion and the initial value of the dialogue intention reflected by any intention point can be summed to obtain the dialogue intention strength of the dialogue intention reflected by any intention point.

The second embodiment: the cloud computing AI service system 100 may further calculate the conversation intention strength of the conversation intention reflected by any one of the cascade intention points according to the calculated conversation application environment related parameters and the conversation intention strengths of the conversation intents reflected by the respective cascade intention points.

Step A140, selecting interest conversation track data of the target intelligent business conversation process from the conversation intention cluster according to the conversation intention strength of each conversation intention, and constructing an interest prediction feature of the target intelligent business conversation process by using the interest conversation track feature of the interest conversation track data, wherein the interest prediction feature is used for indicating the interest content feature of the target intelligent business conversation process.

Since the interesting content features of the target intelligent business conversation process are used for performing predictive analysis on the relevant content of the main operation behaviors (i.e., key operation behaviors) involved in the target intelligent business conversation process, the cloud computing AI service system 100 can select the conversation track data with the maximum conversation intention strength from the conversation intention clusters according to the conversation intention strengths of the conversation intents after obtaining the conversation intention strengths of the conversation intents. Interest prediction features of the target intelligent business conversation process can then be constructed using the interest conversation track features of the interest conversation track data, the interest prediction features indicating content of interest features of the target intelligent business conversation process.

The specific implementation of selecting the interest dialog track data from the dialog intention cluster by the cloud computing AI service system 100 may be as follows: screening all conversation track data included in the conversation intention cluster, and then selecting the conversation track data with the maximum conversation intention intensity from the screened conversation track data as the interesting conversation track data of the target intelligent business conversation process. Alternatively, the cloud computing AI service system 100 may first select a plurality of election dialog trajectory segments of the target intelligent business dialog flow from the dialog intention cluster according to the dialog trajectory segment segmentation rules and according to the dialog intention strengths of the respective dialog intents. The plurality of election conversation track segments comprise at least one interest conversation track data, and the conversation track segment segmentation rule can be set according to business requirements. For example, a dialog track segment segmentation rule can be set for indicating that dialog intents of a target range are selected from the dialog intention cluster as election dialog track segments in descending order of the dialog intention strength; or a dialog track segment segmentation rule can be set for indicating that from the dialog intention cluster, a dialog intention with a dialog intention strength greater than a dialog intention strength threshold value is selected as an election dialog track segment, and the like.

According to the method and the device, aiming at the target intelligent business conversation process to be mined, a plurality of conversation intents can be extracted from the target intelligent business conversation process, and an intention point space of the target intelligent business conversation process is constructed by adopting the plurality of conversation intents. Because the conversation intentions reflected by any two intention points with binding relationship in the intention point space have conversation circulation operation in the target intelligent business conversation process, the conversation intentions with more conversation circulation operations generally have more intention interest possibility; therefore, the conversation intention strength of the conversation intention reflected by each intention point can be more accurately determined based on the conversation circulation information between the intention points in the intention point space. Then, interest dialog track data of the target intelligent business dialog process can be selected from the dialog intention clusters according to the dialog intention strength of each dialog intention, and interest prediction features used for indicating interest content features of the target intelligent business dialog process are constructed by using the interest dialog track features of the interest dialog track data. Therefore, the precision of the interested dialogue track data can be effectively improved by improving the precision of the dialogue intention strength of each dialogue intention, so that the precision of the interested content characteristics is improved; in addition, the whole interest point prediction process can be automatically completed in real time, the interest content characteristics of the intelligent service conversation process can be automatically identified, and the time cost of interest point prediction is reduced.

Based on the related description of the foregoing embodiments, the present application embodiment further provides another content push decision method based on big data mining in an embodiment that can be implemented independently, and the method can be executed by the aforementioned cloud computing AI service system 100. For example, the content push decision method based on big data mining may include the following steps a210-S290, which are described in detail below.

Step A210, extracting a conversation intention cluster from a target intelligent business conversation process to be mined.

In an embodiment that can be implemented independently, the cloud computing AI service system 100 may perform dialog intention clustering on a target intelligent business dialog process to be mined to obtain an initial dialog intention cluster; the originating dialog intent cluster includes a plurality of originating dialog intents therein, and each originating dialog intent in the originating dialog intent cluster has an intent feature.

After obtaining the initial dialog intention cluster, the cloud computing AI service system 100 may screen a plurality of trigger dialog intents from the initial dialog intention cluster according to at least one dialog intention reference model; the triggering dialog intention referred to herein refers to an initial dialog intention that exists in the at least one dialog intention reference model, i.e., the triggering dialog intention refers to a dialog intention that exists in both the at least one dialog intention reference model and the target intelligent business dialog flow. In an independently implementable embodiment, the cloud computing AI service system 100 can screen a plurality of trigger dialog intents from the starting dialog intent cluster directly according to at least one dialog intent reference model; for example, in an independently implementable embodiment, the cloud computing AI service system 100 can walk around each originating dialog intent in the cluster of originating dialog intents; matching the starting dialogue intention of the current walk with at least one dialogue intention reference model to detect whether the starting dialogue intention of the current walk exists in the at least one dialogue intention reference model; and if so, taking the starting dialog intention of the current wandering as the trigger dialog intention. In another embodiment, which can be implemented independently, since some error intentions may exist in the starting dialog intention cluster, the error intentions are intentions which need to be eliminated and have no practical meaning.

Based on this, when the cloud computing AI service system 100 screens out a plurality of trigger dialog intents from the initial dialog intention cluster according to at least one dialog intention reference model, the initial dialog intention of the target intention feature may be screened out from the initial dialog intention cluster; the target intent characteristics mentioned herein may be specified according to actual business conditions, which may include at least one of: intent trigger behavior, intent destination behavior, and intent data sources. Then, the cloud computing AI service system 100 may screen out a plurality of trigger dialog intents from the initial dialog intents of the target intention feature according to the at least one dialog intention reference model, and the specific implementation thereof is similar to the aforementioned specific implementation of the step of "screening out a plurality of trigger dialog intents from the initial dialog intention cluster directly according to the at least one dialog intention reference model", and the detailed description thereof is omitted here.

After the plurality of trigger dialog intents are screened out, a dialog intention cluster (which can be represented by M') of the target intelligent business dialog flow can be constructed by using the plurality of trigger dialog intents. In an independently implementable embodiment, a plurality of triggering dialog intents can be directly adopted to construct a dialog intention cluster of a target intelligent business dialog process; in this embodiment, the dialog intents in the dialog intent cluster are the triggering dialog intents, and the statistical distribution amount of the dialog intents is equal to the statistical distribution amount of the triggering dialog intents. In another embodiment, some triggering dialogue intents which are adjacent in the target intelligent business dialogue process and have special meanings may exist in the screened triggering dialogue intents; for these triggering dialog intents, they will typically appear in the dialog intent reference model at the same time, and the intents that are composed after their intent contents are associated together are more analytically meaningful than a single triggering dialog intent. For example, for triggering dialog intents "Q" and "W", they would typically appear in the dialog intent reference model at the same time, and "Q + W" is more analytically meaningful than "Q" and "W". In this case, the cloud computing AI service system 100 can associate the content of the triggering dialog intentions together and use the dialog intentions after the content of the intentions is associated as the dialog intentions to improve the accuracy of the subsequent topic identification. Based on this, when the cloud computing AI service system 100 constructs a dialog intention cluster of the target intelligent business dialog flow by using a plurality of trigger dialog intents, it may determine whether there is a trigger dialog intention satisfying the intention content association condition in the plurality of trigger dialog intents; the intention content association condition here may include: the dialog intention coverage in the target intelligent business dialog flow is matched and exists in the same dialog intention reference model. If the triggering dialogue intents meeting the intention content association condition exist in the plurality of triggering dialogue intents, performing intention content association processing on the triggering dialogue intents meeting the intention content association condition; and adding the conversation intention cluster subjected to the intention content association processing and the triggering conversation intention which is not subjected to the intention content association processing into the conversation intention cluster of the target intelligent business conversation process as conversation intents. If there is no trigger dialog intention satisfying the intention content association condition among the plurality of trigger dialog intents, each trigger dialog intention can be added as a dialog intention to a dialog intention cluster of the target intelligent business dialog flow.

Step A220, an intention point space of the target intelligent business conversation process is constructed by adopting a plurality of conversation intents.

In specific implementation, an initial intention point space of a target intelligent business conversation process can be constructed by adopting a plurality of conversation intents; the initial intent point space includes a plurality of intent points, each of which reflects a dialog intent. Secondly, at least one pair of combinations of the circulating conversation intention objects can be selected from the plurality of conversation intentions, and the circulating conversation intention object combination refers to the combination of the conversation intention objects formed by two conversation intentions with a conversation circulating operation in the target intelligent business conversation process. Then, at least one intention point subregion can be determined from the initial intention point space according to the combination of the at least one pair of circulated dialogue intention objects; wherein, any arbitrary point sub-area can include: two intent points of two dialog intents in a combination of a pair of streamed dialog intent objects are recorded, respectively. Then, two intent points in each intent point sub-area can be connected in the initial intent point space respectively, so as to obtain the intent point space of the target intelligent business conversation process.

Step A230, based on the dialog circulation information between the various intention points in the intention point space, calculating the dialog intention strength of the dialog intention reflected by the various intention points.

And step A240, selecting a plurality of election conversation track segments of the target intelligent business conversation process from the conversation intention cluster according to the conversation intention strength of each conversation intention according to the conversation track segment segmentation rule.

In an embodiment, the cloud computing AI service system 100 may select a target range of dialog intents from the dialog intention cluster as a plurality of election dialog trajectory segments of the target intelligent business dialog flow in descending order of the dialog intention strength. In another independently implementable embodiment, the cloud computing AI service system 100 can select a conversation intention from the conversation intention cluster with a conversation intention strength greater than a conversation intention strength threshold as the plurality of election conversation track segments of the target intelligent business conversation process. Wherein the plurality of election dialog track segments include at least one interest dialog track data. For convenience of illustration, the dialog intents of the target range selected from the dialog intention clusters by the plurality of election dialog track segments in descending order of the dialog intention strength will be described as an example.

Step A250, selecting the interest dialogue track data with the maximum dialogue intention intensity from at least one interest dialogue track data as the interest dialogue track data of the target intelligent business dialogue process.

Step A260, constructing an interest prediction feature of the target intelligent business conversation process by using the interest conversation track feature of the interest conversation track data, wherein the interest prediction feature is used for indicating the interest content feature of the target intelligent business conversation process.

In an embodiment, which can be implemented independently, the cloud computing AI service system 100 may first invoke the interest dialog track feature generation model to obtain the interest dialog track features of the interest dialog track data. In an embodiment, the interest dialog track feature of the dialog intention can be determined by combining the contextual data of the dialog intention or by combining the functions of the dialog intention in different dimensions, so that the dialog intention can have the same interest dialog track feature in different dimensions, without limitation.

After the interest conversation track characteristics of the interest conversation track data are obtained, the interest conversation track characteristics of the interest conversation track data can be adopted to construct interest prediction characteristics of the target intelligent business conversation process. In an embodiment which can be independently implemented, related non-conversation track data corresponding to interest conversation track data can be obtained from a conversation intention cluster; the so-called relevant non-conversational trajectory data fulfils the following condition: in the intent point space, intent points for recording related non-dialog track data are mapped with intent points for recording interest dialog track data. Secondly, the interest dialogue track characteristics of the interest dialogue track data and the interest dialogue track characteristics of the related non-dialogue track data can be obtained; it should be noted that any of the interest dialog trajectory features mentioned herein may be generated by the cloud computing AI service system 100 by calling the interest dialog trajectory feature generation model, and any of the interest dialog trajectory features may be a P-dimensional vector; the value of P may be set according to empirical values, for example P200. Then, the interest dialogue track characteristics of the interest dialogue track data and the interest dialogue track characteristics of the related non-dialogue track data can be spliced to obtain the interest prediction characteristics of the target intelligent business dialogue process; in an embodiment, which can be implemented independently, the interest dialogue track features of the interest dialogue track data and the interest dialogue track features of the related non-dialogue track data can be accumulated according to each dimension to obtain the interest prediction features in the target medical treatment text. For example, let the interest conversation track data be "live cable commercial skip data", and its corresponding interest conversation track characteristic be G (l0) ═ l01, l02, …, l 0P; relevant non-conversational trajectory data includes: "Q", "W", "E", and "R", and the corresponding interest dialog track features are as follows in order: g (l1) ═ G (l11, l12, …, l1P), G (l2) ═ l21, l22, …, l2P, G (l3) ═ l31, l32, …, l3P, and G (l4) ═ l41, l42, …, l 4P. Then, the individual interest dialogue track features are added up in each dimension to obtain the interest prediction features g (l01+ l11+ l21+ l31+ l41, l02+ l12+ l22+ l32+ l42, …, l0P + l1P + l2P + l3P + l 4P).

Step A270, obtaining the interest dialogue track characteristics of each election dialogue track segment, and calculating the characteristic correlation degree between the interest dialogue track characteristics and the interest prediction characteristics of each election dialogue track segment.

Step A280, selecting an interest conversation track section of the target intelligent business conversation process from a plurality of election conversation track sections according to the feature correlation degree between the interest conversation track features and the interest prediction features of each election conversation track section.

In the specific implementation process of steps a210-S280, the cloud computing AI service system 100 may first invoke the interest dialog trajectory feature generation model to represent each election dialog trajectory segment as a P-dimensional vector, so as to obtain the interest dialog trajectory features (i.e., P-dimensional vectors) of each election dialog trajectory segment. Secondly, a feature correlation algorithm can be adopted to calculate the feature correlation between the interest conversation track features and the interest prediction features of each election conversation track segment. Then, according to the feature correlation degree between the interest conversation track feature and the interest prediction feature of each election conversation track segment, the election conversation track segment with the feature correlation degree larger than a preset correlation degree threshold value is selected from the multiple election conversation track segments to serve as the interest conversation track segment of the target intelligent business conversation process. It should be noted that the preset correlation threshold mentioned herein may be set according to actual traffic conditions, for example, may be set to 0.85, etc.; also, the interest dialog track segment selected through steps a210-S280 may or may not include the aforementioned interest dialog track data, which is not limited in this respect.

Therefore, the feature correlation between the interest conversation track feature and the interest prediction feature of the interest conversation track segment is greater than the preset correlation threshold through steps a210 to S280, so as to avoid that the election conversation track segment unrelated to the interest content feature is selected as the interest conversation track segment, and if the interest content feature of the target intelligent business conversation process is positioned on the "Q" related operation behavior, the election conversation track segment "W" unrelated to the interest content feature can avoid being selected as the interest conversation track segment of the target intelligent business conversation process. Therefore, the interest dialogue track segment obtained through the steps A210-S280 can reflect the cheating main track of the target intelligent business dialogue process, and the precision of the interest dialogue track segment is improved; and moreover, interest conversation track sections of the target intelligent business conversation process all belong to the same interest content characteristic, and the matching of interest conversation dimensions among the interest conversation track sections is ensured.

Step A290, mapping the target intelligent business conversation process and the interest conversation track segment.

In a specific implementation, after the interest dialog track segment of the target intelligent business dialog flow is selected by the cloud computing AI service system 100 through the above steps a210 to S280, the target intelligent business dialog flow and the interest dialog track segment may be mapped. In an independently implementable embodiment, the cloud computing AI service system 100 can directly add and associate the interest dialog track segment to the dialog track segment list associated with the target intelligent business dialog flow; alternatively, the cloud computing AI service system 100 can add the interest conversation track features of each interest conversation track segment to the list of conversation track segments associated with the target intelligent business conversation process. By mapping the target intelligent business conversation process and the interest conversation track segment, information mining processing can be carried out on the target intelligent business conversation process according to the interest conversation track segment.

It should be noted that the above-mentioned interest prediction features are global deep learning prediction features (i.e. main deep learning prediction features) of the target intelligent business conversation process, and the interest content features are main conversation content features of the target intelligent business conversation process. In an alternative embodiment, when the statistical distribution quantity of the interest dialogue trajectory data extracted from the target intelligent business dialogue process is multiple, the target intelligent business dialogue process is indicated to have multiple dialogue content characteristics. In this case, the cloud computing AI service system 100 may further select interest conversation track data of the target intelligent business conversation process from at least one interest conversation track data, the conversation intention strength of which is less than that of the interest conversation track data. Second, the interest conversation track features of the interest conversation track data can be used to construct deep learning prediction features (i.e., non-essential deep learning prediction features) of the conversation trend of the target intelligent business conversation process, which are used to indicate the conversation content features of the conversation trend of the target intelligent business conversation process. Then, the target intelligent business conversation process, the interest prediction feature and the deep learning prediction feature of the conversation trend can be mapped to the interest trend evaluation application, so that when an interest trend evaluation request exists, the interest trend evaluation processing can be carried out according to the interest prediction feature and the deep learning prediction feature of the conversation trend.

In an independently implementable embodiment, the cloud computing AI service system 100 may further calculate global deep learning prediction features and deep learning prediction features of a conversation trend of other intelligent business conversation processes by using the above method steps, and associate each of the calculated deep learning prediction features to an interest trend evaluation application; the impromptu trend assessment application further includes at least one other intelligent business conversation process, and each other intelligent business conversation process has a corresponding global deep learning prediction feature and a corresponding deep learning prediction feature of a conversation trend. When there is an interest trend evaluation request, the cloud computing AI service system 100 may obtain an interest trend evaluation feature of the interest trend evaluation information carried by the interest trend evaluation request. Secondly, each intelligent business conversation process in the interest trend evaluation application and at least one deep learning prediction characteristic of each intelligent business conversation process can be obtained; each intelligent business conversation process is provided with a business putting weight, and at least one deep learning prediction feature of each intelligent business conversation process comprises a global deep learning prediction feature of each intelligent business conversation process and a deep learning prediction feature of a conversation trend. Then, the feature correlation degree between each deep learning prediction feature and the interest trend evaluation feature of each intelligent business conversation process can be respectively calculated, and the business putting weight of each intelligent business conversation process is updated according to the calculated feature correlation degree. In an embodiment, for any intelligent business conversation process, the deep learning prediction feature with the maximum feature correlation degree can be determined from at least one deep learning prediction feature of any intelligent business conversation process according to the feature correlation degree between each deep learning prediction feature and the interest trend evaluation feature of any intelligent business conversation process. If the deep learning prediction feature with the maximum feature correlation degree is the global deep learning prediction feature of any intelligent business conversation process, the business launching weight of any intelligent business conversation process can be improved so as to update the business launching weight of any intelligent business conversation process; if the deep learning prediction feature with the maximum feature correlation degree is the deep learning prediction feature of the conversation trend of any intelligent business conversation process, the business putting weight of any intelligent business conversation process can be reduced so as to update the business putting weight of any intelligent business conversation process.

Based on the updating principle, after the service delivery weight of each intelligent service conversation process is updated, the cloud computing AI service system 100 can sort each intelligent service conversation process in a descending order according to the updated service delivery weight of each intelligent service conversation process; and selecting the intelligent service conversation process positioned at the head as the intelligent service conversation process to be analyzed to output. Therefore, the intelligent business conversation process with the conversation content characteristics can be subjected to ascending or descending weight sequencing according to the matching condition of the conversation content characteristics, and the accuracy of information mining is improved.

According to the method and the device, aiming at the target intelligent business conversation process to be mined, a plurality of conversation intents can be extracted from the target intelligent business conversation process, and an intention point space of the target intelligent business conversation process is constructed by adopting the plurality of conversation intents. Because the conversation intentions reflected by any two intention points with binding relationship in the intention point space have conversation circulation operation in the target intelligent business conversation process, the conversation intentions with more conversation circulation operations generally have more intention interest possibility; therefore, the conversation intention strength of the conversation intention reflected by each intention point can be more accurately determined based on the conversation circulation information between the intention points in the intention point space. Then, interest dialog track data of the target intelligent business dialog process can be selected from the dialog intention clusters according to the dialog intention strength of each dialog intention, and interest prediction features used for indicating interest content features of the target intelligent business dialog process are constructed by using the interest dialog track features of the interest dialog track data. Therefore, the precision of the interested dialogue track data can be effectively improved by improving the precision of the dialogue intention strength of each dialogue intention, so that the precision of the interested content characteristics is improved; in addition, the whole interest point prediction process can be automatically completed in real time, the interest content characteristics of the intelligent service conversation process can be automatically identified, and the time cost of interest point prediction is reduced. Moreover, after the interest prediction features are obtained, the cloud computing AI service system 100 may further perform a purification process on the dialog track segment of the target intelligent business dialog flow based on the feature correlation between the election dialog track segment and the interest prediction features to obtain an interest dialog track segment with matching of interest dialog dimensions, so that the accuracy of the interest dialog track segment can be effectively improved.

In an embodiment that can be implemented independently, the present application further provides an information pushing method based on smart business big data and interest prediction, which may include the following steps.

Step A310, obtaining a plurality of hotspot interest items under each e-commerce promotion service according to the predicted interest point set of each target intelligent service conversation process under each e-commerce promotion service.

Step A320, determining the interest tendency distribution information of a plurality of hotspot interest items.

The method comprises the steps that a plurality of hot spot interest items exist, at least one interest tendency exists in the plurality of hot spot interest items, and the interest tendency distribution information indicates the statistical distribution quantity of each interest tendency of the plurality of hot spot interest items.

Step A330, clustering the plurality of hot interest items according to the interest tendency distribution information of the plurality of hot interest items to obtain at least two hot interest item clusters, wherein the hot interest item clusters comprise at least one interest tendency hot interest item.

In an embodiment, each hot spot interest item group may include all the hot spot interest items with interest tendency in the plurality of hot spot interest items, and the hot spot interest items with the same interest tendency in each hot spot interest item group may be one or more than one.

Step A340, E-commerce information pushing rule configuration is carried out on at least two hot spot interest item groups, so that the hot spot interest item groups in the at least two hot spot interest item groups located in the same active interest dimension are associated with corresponding information pushing rules, the hot spot interest item groups in the same passive interest dimension are associated with corresponding information pushing rules, and information pushing is carried out based on the configured E-commerce information pushing rules.

In each hot interest item group, different hot interest items can be arranged and distributed according to the passive interest dimension direction. And associating the corresponding information pushing rules with the hot interest item groups positioned in the same active interest dimension.

In an embodiment, the cloud computing AI service system may first perform an initial e-commerce information push rule configuration on the at least two hot spot interest item groups, so as to determine the active interest dimension number configured by the e-commerce information push rule for the at least two hot interest item groups, the statistical distribution quantity of the hot interest item groups of each active interest dimension, and the hot interest item groups of each active interest dimension (i.e. determining the active interest dimension number and the passive interest dimension number), and then, at least one of interest information push crowd and interest information push domains of the at least two hot interest item groups are configured in an associated manner, so that the hot interest item groups in the same active interest dimension are associated with corresponding information push rules, and the hotspot interest item groups positioned in the same passive interest dimension are associated with corresponding information pushing rules.

By means of the design, after the cloud computing AI service system determines the interest tendency distribution information of a plurality of hot interest items, the plurality of hot interest items can be clustered according to the interest tendency distribution information of the plurality of hot interest items to obtain at least two hot interest item groups, and then the at least two hot interest item groups are subjected to E-commerce information pushing rule configuration, so that the hot interest item groups in the at least two hot interest item groups in the same active interest dimension are associated with corresponding information pushing rules, and the hot interest item groups in the same passive interest dimension are associated with corresponding information pushing rules. Therefore, the association configuration process of the hotspot interest items can be simplified, and the information pushing precision is improved.

Fig. 3 is a schematic diagram illustrating a hardware structure of a cloud computing AI service system 100 for implementing the above-mentioned big data mining-based content push decision method according to an embodiment of the present application, where as shown in fig. 3, the cloud computing AI service system 100 may include a processing chip 110 and a machine-readable storage medium 120; the machine-readable storage medium 120 has stored thereon executable code, which when executed by the processing chip 110, causes the processing chip 110 to perform the steps of the above embodiment of the big data mining based content push decision method.

Actually, the cloud computing AI service system may further include a communication interface 140, wherein the processing chip 110, the machine-readable storage medium 120 and the communication interface 140 are connected through the bus 130, and the communication interface 140 is used for communicating with other devices.

In addition, the present application provides a non-transitory machine-readable storage medium having executable code stored thereon, which when executed by a processor of an electronic device, causes the processor to implement at least the steps of an embodiment of a big data mining based content push decision method as described above.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. A content push decision-making method based on big data mining is characterized by being applied to a cloud computing AI service system, wherein the cloud computing AI service system is in communication connection with a plurality of intelligent service registration terminals, and the method comprises the following steps:

acquiring a predicted interest point set of a target intelligent service conversation process of an intelligent service registration terminal, performing initial information push on the intelligent service registration terminal according to the predicted interest point set, and acquiring a plurality of content response events of initial push content information, wherein the predicted interest point set of the target intelligent service conversation process is acquired by mining big data based on a conversation track;

2. The big data mining based content push decision method according to claim 1, wherein after the obtaining of the plurality of content response events of the initial push content information, the method further comprises:

querying whether the plurality of content response events contain subscription content response characteristics, wherein the subscription content response characteristics comprise: subscription portrait features or interactive portrait features;

if each content response event covers the subscription content response feature, obtaining a second validity metric value corresponding to each content response event from a preset subscription content response feature bitmap, so as to determine a plurality of second validity metric values corresponding to a plurality of content response events;

correspondingly, the determining the validity identification information of the plurality of content response events based on a plurality of first validity metric values corresponding to the plurality of content response events includes:

determining validity identifying information for the plurality of content response events based on the plurality of first validity metric values and the plurality of second validity metric values.

3. The big data mining based content push decision method according to claim 1 or 2, wherein after the obtaining of the plurality of content response events of the initial push content information, the method further comprises:

acquiring a plurality of behavior migration events corresponding to the plurality of content response events;

identifying the plurality of behavior migration events to obtain behavior migration content information corresponding to each behavior migration event;

if the behavior migration content information represents that migration content information which needs to be shared by the same content objects exists, obtaining a third validity metric value corresponding to each content response event from a preset migration sharing metric value bitmap, and accordingly determining at least one third validity metric value corresponding to a plurality of content response events;

determining validity identification information for the plurality of content response events based on the plurality of first validity metric values and the at least one third validity metric value;

or determining validity identification information for the plurality of content response events based on the plurality of first validity metric values, the plurality of second validity metric values, and the at least one third validity metric value.

4. The big data mining-based content push decision method according to claim 1, wherein the determining validity identification information of the plurality of content response events based on a plurality of first validity metric values corresponding to the plurality of content response events comprises:

if the first effectiveness metric values corresponding to the content response events contain first target effectiveness metric values larger than the first target metric values, determining that the effectiveness identification information of the content response events corresponding to the first target effectiveness metric values is effective content response events.

5. The big data mining-based content push decision method according to claim 1, wherein the determining validity identification information of the plurality of content response events based on a plurality of first validity metric values corresponding to the plurality of content response events comprises:

if a plurality of first effectiveness metric values corresponding to the plurality of content response events contain a second target effectiveness metric value which is not larger than the first target metric value, carrying out integral discretization treatment on the second target effectiveness metric value to obtain a first discrete effectiveness metric value;

if the first discrete effectiveness metric value is not larger than the first discrete target metric value, the effectiveness identification information of the content response event corresponding to the second target effectiveness metric value is a non-effective content response event;

and if the first discrete effectiveness metric value is larger than a first discrete target metric value, the effectiveness identification information of the content response event corresponding to the second target effectiveness metric value is an effective content response event.

6. The big data mining based content push decision method according to claim 3, further comprising:

identifying each content response event in the plurality of content response events, and respectively obtaining the sharing characteristics of the same content object respectively corresponding to the content response events;

summarizing statistics of the sharing characteristics of the same content objects corresponding to each content response event;

and determining validity identification information of the plurality of content response events according to at least one of a plurality of second validity metric values and the at least one third validity metric value, the plurality of first validity metric values and statistics of the same content object sharing characteristics.

7. The big data mining-based content push decision method according to claim 1, wherein the first validity metric value corresponding to each content response event comprises:

a first aggregation effectiveness metric value and a second aggregation effectiveness metric value;

if a conflict feature exists between at least two of the content response polarity label information, the content behavior thermodynamic diagram information, and the content operation feedback behavior information, obtaining a first validity metric value corresponding to each content response event from a preset validity metric bitmap, including:

if the conflict characteristics exist between the content response polarity label information and the content operation feedback behavior information, obtaining a first aggregation effectiveness metric value corresponding to each content response event from a first preset effectiveness metric bitmap;

or if a conflict feature exists between the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a second aggregation effectiveness metric value corresponding to each content response event from a second preset effectiveness metric bitmap;

correspondingly, if a conflict feature exists between the content response polarity tag information and the content operation feedback behavior information, obtaining a first aggregation validity metric value corresponding to each content response event from a first preset validity metric bitmap includes:

if the content operation feedback behavior information and the content response polarity label information contain a matched first target effectiveness metric bitmap in a first preset effectiveness metric bitmap, indicating to obtain the first aggregation effectiveness metric value corresponding to the first target effectiveness metric bitmap from the preset effectiveness metric bitmap;

correspondingly, if a conflict feature exists between the content behavior thermodynamic diagram information and the content operation feedback behavior information, obtaining a second aggregation validity metric value corresponding to each content response event from a second preset validity metric bitmap includes:

if the content behavior thermodynamic diagram information and the content operation feedback behavior information contain a matched second target effectiveness metric bitmap in a second preset effectiveness metric bitmap, indicating to obtain a second aggregation effectiveness metric value corresponding to the second target effectiveness metric bitmap from the preset effectiveness metric bitmap;

correspondingly, after obtaining a first aggregation validity metric value corresponding to each content response event from a first preset validity metric bitmap if a conflict feature exists between the content response polarity tag information and the content operation feedback behavior information, the method further includes:

summarizing the label distribution quantity of the first effective content operation feedback behavior information corresponding to the first aggregation effectiveness metric value;

and if the distribution quantity of the labels of the first effective content operation feedback behavior information is greater than a first quantity, the effectiveness identification information of the content response event corresponding to the first effective content operation feedback behavior information is an effective content response event.

8. The big data mining-based content push decision method according to claim 7, wherein if there is a conflict feature between the content behavior thermodynamic diagram information and the content operation feedback behavior information, after obtaining a second aggregation validity metric value corresponding to each content response event from a second preset validity metric bitmap, the method further comprises:

summarizing the label distribution quantity of the second effective content operation feedback behavior information corresponding to the second aggregation effectiveness metric value;

and if the distribution quantity of the labels of the second effective content operation feedback behavior information is greater than a second quantity, the effectiveness identification information of the content response event corresponding to the second effective content operation feedback behavior information is an effective content response event.

9. The big data mining-based content push decision method according to any one of claims 1 to 8, wherein the step of obtaining the set of predicted interest points of the target intelligent business conversation process of the intelligent business registration terminal comprises:

extracting a conversation intention cluster from a target intelligent business conversation process to be mined of the intelligent business registration terminal, wherein the conversation intention cluster comprises a plurality of conversation intents, and the plurality of conversation intents at least comprise conversation track data;

adopting the plurality of conversation intentions to construct an intention point space of the target intelligent business conversation process, wherein the intention point space comprises a plurality of intention points; one intention point reflects one conversation intention, and the conversation intentions reflected by any two intention points with binding relationship have conversation circulation operation in the target intelligent business conversation process;

calculating the conversation intention strength of the conversation intention reflected by each intention point based on the conversation circulation information among the intention points in the intention point space;

selecting interest conversation track data of the target intelligent business conversation process from the conversation intention cluster according to the conversation intention strength of each conversation intention, constructing an interest prediction feature of the target intelligent business conversation process by adopting the interest conversation track feature of the interest conversation track data, inputting the interest prediction feature into an interest point prediction network, and obtaining a predicted interest point set of the target intelligent business conversation process, wherein the interest prediction feature is used for indicating an interest content feature of the target intelligent business conversation process;

the step of pushing initial information to the intelligent service registration terminal according to the predicted interest point set and acquiring a plurality of content response events of the initially pushed content information includes:

acquiring a plurality of hot interest items under each electronic commerce promotion service according to the predicted interest point set of each target intelligent service conversation process under each electronic commerce promotion service;

determining interest tendency distribution information of the plurality of hot interest items, wherein at least one interest tendency of the plurality of hot interest items exists, and the interest tendency distribution information indicates the statistical distribution quantity of each interest tendency of the plurality of hot interest items;

clustering the plurality of hot interest items according to the interest tendency distribution information of the plurality of hot interest items to obtain at least two hot interest item clusters, wherein the hot interest item clusters comprise the hot interest items with at least one interest tendency;

e-commerce information push rule configuration is carried out on the at least two hot spot interest item groups, so that the hot spot interest item groups in the at least two hot spot interest item groups in the same active interest dimension are associated with corresponding information push rules, and the hot spot interest item groups in the same passive interest dimension are associated with corresponding information push rules;

the configuring of e-commerce information push rules for the at least two hot interest item groups, so that the hot interest item groups located in the same active interest dimension in the at least two hot interest item groups are associated with corresponding information push rules, and the hot interest item groups located in the same passive interest dimension are associated with corresponding information push rules, includes:

according to a hotspot interest data source of a hotspot interest item included by each of the at least two hotspot interest item groups, performing initial e-commerce information push rule configuration on the at least two hotspot interest item groups;

at least one of an interest information pushing domain and an interest information pushing crowd of each of the at least two hot interest item groups is configured in an associated manner, so that the hot interest item groups in the at least two hot interest item groups located in the same active interest dimension are associated with corresponding information pushing rules, and the hot interest item groups in the same passive interest dimension are associated with corresponding information pushing rules;

the initial e-commerce information push rule configuration is performed on the at least two hot interest item groups according to the hot interest data source of the hot interest item included in each of the at least two hot interest item groups, and includes:

for each of the at least two hot spot interest item groups, determining hot spot progress position information of the hot spot interest item group in a target hot spot distribution network according to the hot spot progress position information of the hot spot interest items in the hot spot interest item groups in the target hot spot distribution network;

and sequencing the at least two hot interest item groups according to the hot progress position information of the at least two hot interest item groups to obtain a hot interest item cluster.

10. A cloud computing AI service system comprising a machine-readable storage medium, a processor; wherein the machine-readable storage medium has stored thereon executable code, which when executed by the processor, causes the processor to perform the big data mining based content push decision method of any of claims 1-9.