CN114048387B

CN114048387B - Content recommendation method based on big data and AI prediction and artificial intelligence cloud system

Info

Publication number: CN114048387B
Application number: CN202111397916.1A
Authority: CN
Inventors: 赵运柱
Original assignee: Zhongshan Mingyuan Cloud Technology Co ltd
Current assignee: Zhongshan Mingyuan Cloud Technology Co ltd
Priority date: 2021-11-23
Filing date: 2021-11-23
Publication date: 2023-03-24
Anticipated expiration: 2041-11-23
Also published as: CN114048387A

Abstract

The embodiment of the disclosure provides a content recommendation method based on big data and AI prediction and an artificial intelligence cloud system, which can acquire cooperative interaction behavior data of a service user of a service request terminal and other service users aiming at personalized service content, predict the cooperative interaction behavior data based on a pre-configured cooperative portrait prediction model, acquire a cooperative portrait sequence corresponding to the cooperative interaction behavior data, and recommend corresponding cooperative personalized service content to the service user and other service users according to the cooperative portrait sequence. Therefore, by predicting collaborative interaction behavior data with collaborative interest preference characteristics among a plurality of service users and then pushing content based on the collaborative portrait sequence corresponding to the collaborative interaction behavior data, characteristic omission which possibly exists when pushing optimization is carried out on collaborative service information can be reduced, and pushing accuracy is improved.

Description

Content recommendation method based on big data and AI prediction and artificial intelligence cloud system

Technical Field

The disclosure relates to the technical field of internet content services, and exemplarily relates to a content recommendation method based on big data and AI prediction and an artificial intelligence cloud system.

Background

With the maturity of big data and artificial intelligence technology, intelligent recommendation is already applied to various fields of the internet, and a plurality of internet service providers can provide intelligent recommendation of content related to products in respective product forms based on the idea of machine learning. In the related art, only the interest portrait of a single user is generally considered when content is pushed based on a user interest preference portrait, and then a great feature omission exists when push optimization is performed on some service information which may need to be coordinated.

Disclosure of Invention

In order to overcome at least the above disadvantages in the prior art, an object of the present disclosure is to provide a content recommendation method based on big data and AI prediction and an artificial intelligence cloud system.

In a first aspect, the present disclosure provides a content recommendation method based on big data and AI prediction, which is applied to an artificial intelligence cloud system, where the artificial intelligence cloud system is in communication connection with a plurality of service request terminals, and the method includes:

generating personalized service content of the service request terminal according to the service operation big data of the service request terminal;

acquiring cooperative interaction behavior data of the service user of the service request terminal and other service users aiming at the personalized service content;

predicting the collaborative interaction behavior data based on a pre-configured collaborative portrait prediction model to obtain a collaborative portrait sequence corresponding to the collaborative interaction behavior data;

and recommending corresponding collaborative personalized service contents for the business user and the other business users according to the collaborative portrait sequence.

In a second aspect, an embodiment of the present disclosure further provides a content recommendation system based on big data and AI prediction, where the content recommendation system based on big data and AI prediction includes an artificial intelligence cloud system and a plurality of service request terminals communicatively connected to the artificial intelligence cloud system;

the artificial intelligence cloud system is used for:

According to any one of the aspects, in the embodiment provided by the present disclosure, collaborative interaction behavior data of a service user of a service request terminal and other service users for personalized service content may be obtained, the collaborative interaction behavior data is predicted based on a preconfigured collaborative portrait prediction model, a collaborative portrait sequence corresponding to the collaborative interaction behavior data is obtained, and then, corresponding collaborative personalized service content is recommended for the service user and other service users according to the collaborative portrait sequence. Therefore, by predicting collaborative interaction behavior data with collaborative interest preference characteristics among a plurality of service users and then pushing content based on the collaborative portrait sequence corresponding to the collaborative interaction behavior data, characteristic omission which possibly exists when pushing optimization is carried out on collaborative service information can be reduced, and pushing accuracy is improved.

Drawings

To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings that need to be called in the embodiments are briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present disclosure and therefore should not be considered as limiting the scope, and those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.

Fig. 1 is a schematic view of an application scenario of a content recommendation system based on big data and AI prediction according to an embodiment of the present disclosure;

FIG. 2 is a schematic flow chart of a content recommendation method based on big data and AI prediction according to an embodiment of the disclosure;

FIG. 3 is a functional block diagram of a content recommendation device based on big data and AI prediction according to an embodiment of the disclosure;

fig. 4 is a block diagram illustrating a structure of an artificial intelligence cloud system for implementing the above-described content recommendation method based on big data and AI prediction according to an embodiment of the present disclosure.

Detailed Description

The present disclosure is described in detail below with reference to the drawings, and the specific operation methods in the method embodiments can also be applied to the device embodiments or the system embodiments.

Fig. 1 is a schematic application scenario diagram of a content recommendation system 10 based on big data and AI prediction according to an embodiment of the present disclosure. The big data and AI prediction based content recommendation system 10 may include an artificial intelligence cloud system 100 and a service request terminal 200 communicatively connected to the artificial intelligence cloud system 100. The big data and AI prediction based content recommendation system 10 shown in FIG. 1 is but one possible example, and in other possible embodiments, the big data and AI prediction based content recommendation system 10 may also include only at least some of the components shown in FIG. 1 or may also include other components.

In an independently implementable embodiment, the artificial intelligence cloud system 100 and the service request terminal 200 in the content recommendation system 10 based on big data and AI prediction may cooperatively perform the content recommendation method based on big data and AI prediction described in the following method embodiments, and for the specific steps of the artificial intelligence cloud system 100 and the service request terminal 200, reference may be made to the detailed description of the following method embodiments.

In order to solve the technical problems in the background art, the content recommendation method based on big data and AI prediction provided by the present embodiment may be performed by the artificial intelligence cloud system 100 shown in fig. 1, and the content recommendation method based on big data and AI prediction is described in detail below.

Step S101, generating personalized service content of the service request terminal according to the service operation big data of the service request terminal.

Step S102, obtaining the cooperative interaction behavior data of the service user of the service request terminal and other service users aiming at the personalized service content.

And step S103, predicting the collaborative interaction behavior data based on a preconfigured collaborative portrait prediction model to obtain a collaborative portrait sequence corresponding to the collaborative interaction behavior data.

And step S104, recommending corresponding collaborative personalized service contents for the service user and other service users according to the collaborative portrait sequence.

In this embodiment, after the personalized service content of the service request terminal is generated, the personalized service content may be pushed to the service request terminal in real time or according to a preset period, so that the service user of the service request terminal may perform collaboration and interaction with respect to the personalized service content in other service users, for example, in the interaction process, collaborative interaction behavior data for the personalized service content may be generated, such as collaborative comment behavior data, collaborative forwarding behavior data, collaborative completion behavior data, collaborative content tagging behavior data, and the like, but not limited thereto. The inventor of the present application finds that the collaborative interaction behavior data can reflect interest preference portraits of collaborative existence of business users and other business users, and in the related art, only interest portraits of single users are considered when content push is performed based on the user interest preference portraits, so that great feature omission exists when push optimization is performed on some service information which may need to be collaborated. Based on the method, the collaborative portrait prediction model can predict collaborative interaction behavior data based on the preconfigured collaborative portrait prediction model, obtain a collaborative portrait sequence corresponding to the collaborative interaction behavior data, and recommend corresponding collaborative personalized service content for the business user and other business users according to the collaborative portrait sequence, for example, the collaborative personalized service content matched with portrait features of each collaborative portrait in the collaborative portrait sequence can be obtained and used as the corresponding collaborative personalized service content recommended for the business user and other business users. Therefore, by predicting collaborative interaction behavior data with collaborative interest preference characteristics among a plurality of service users and then pushing content based on the collaborative portrait sequence corresponding to the collaborative interaction behavior data, characteristic omission which possibly exists when pushing optimization is carried out on collaborative service information can be reduced, and pushing accuracy is improved.

In one embodiment, the embodiments of the present disclosure are applicable to various online application services, where the online application services refer to online application services in which an intention association relationship may exist between different session flows.

The artificial intelligence cloud system 100 may obtain at least two business conversation processes of the online application service, and obtain conversation process intention data and conversation process interaction data of each business conversation process, where the conversation process intention data is data obtained by responding to or identifying a conversation process intention (such as continuously paying attention to a certain object or shielding a certain object) between the business conversation process and another business conversation process, and the conversation process interaction data is data obtained by an interaction activity (such as office collaborative interaction and intelligent medical collaborative interaction) for the business conversation process. In one embodiment, the artificial intelligence cloud system 100 may obtain, from a service process of a session flow (e.g., the service process 102a of the session flow, the service process 102b of the session flow, the service process 102c of the session flow, and the like), session flow intention data and session flow interaction data of a business session flow corresponding to the service process of the session flow, generate the session flow intention data of the business session flow according to the obtained session flow intention data, and generate the session flow interaction data of the business session flow according to the obtained session flow interaction data.

The artificial intelligence cloud system 100 inputs the session flow intention data and the session flow interaction data of each business session flow into the trained session flow intention artificial intelligence model, and generates data association distribution information, where the data association distribution information includes data association distribution of each business session flow, and the data association distribution refers to characteristics of the session flow intention data that can reflect both the session flow interaction data of the corresponding business session flow and the corresponding business session flow. When the artificial intelligence cloud system 100 obtains a session flow response request (or a session flow optimization request) of a first session flow, the artificial intelligence cloud system 100 may obtain a session flow node sequence of the first session flow, obtain a first data association distribution of the first session flow from data association distribution information, and obtain a second data association distribution of a second session flow included in the session flow node sequence, where the first session flow may be any one of at least two business session flows.

The artificial intelligence cloud system 100 may obtain correlation parameters of the first data correlation distribution and the second data correlation distribution, and when the correlation parameters are greater than preset correlation parameters, it may be considered that an intention tendency value between the first session flow and the second session flow is large, and the second session flow has an intention demand relationship, and then the second session flow is determined as a demand session flow; when the correlation parameter is less than or equal to the preset correlation parameter, the intention tendency value between the first conversation process and the second conversation process is considered to be low, and the second conversation process is a conversation process in which the intention requirement does not exist in the first conversation process. The data association distribution is a characteristic of the corresponding business conversation process after the conversation process interactive data and the conversation process intention data are combined, so that the larger the correlation parameter between the data association distribution is, the higher the possibility that the intention association relation between the corresponding business conversation processes needs to be kept is, and through the characteristic fusion, the accuracy of identifying the intention demand relation between the business conversation processes and the business conversation processes is improved. By intelligently identifying the intention requirement relationship between the service conversation process and the service conversation process based on the characteristics of the conversation process (which is equivalent to identifying the intention requirement relationship between the service conversation process and the service conversation process based on artificial intelligence), the accuracy in the content pushing process is improved.

In an embodiment, if there are a service session flow 2011, a service session flow 2012, a service session flow 2013, and the like, the artificial intelligence cloud system 100 obtains session flow interaction data and session flow intention data of the service session flow 2011, obtains session flow interaction data and session flow intention data of the service session flow 2012, obtains session flow interaction data and session flow intention data of the service session flow 2013, and transfers and fuses session flow interaction data of different service session flows based on the session flow intention data of each service session flow to obtain data association distribution information 202, where the data association distribution information 202 includes data association distributions of each service session flow, such as data association distribution 2021 of the service session flow 2011, data association distribution 2022 of the service session flow 2012, and data association distribution 2023 of the service session flow 2013. The data association distribution is obtained by fusing the conversation process interaction data corresponding to the business conversation process, the conversation process intention data and the conversation process interaction data transmitted by other business conversation processes, so that the data association distribution can more comprehensively represent the conversation process characteristics corresponding to the business conversation process, and the intention requirement relationship identification accuracy can be improved.

After the artificial intelligence cloud system 100 obtains the data association distribution information, the intention requirement relationship between any two service session flows can be identified based on the data association distribution information. For example, the artificial intelligence cloud system 100 obtains the data association distribution 2021 of the business conversation process 2011 and the data association distribution 2022 of the business conversation process 2012 from the data association distribution information, obtains a correlation parameter between the data association distribution 2021 and the data association distribution 2022, and may determine the intention requirement relationship between the business conversation process 2011 and the business conversation process 2012 based on the correlation parameter. Through the feature recognition, the accuracy in the content pushing process is improved.

The following description is made in detail with reference to a possible embodiment, which should be understood as an independently implementable embodiment, and the specific implementation method of the above step S101 is described in detail with reference to the embodiment.

Step A101, obtaining session flow interactive data and session flow intention data of at least two service session flows in service operation big data of a service request terminal.

For example, the artificial intelligence cloud system 100 obtains session flow interaction data and session flow intention data of at least two business session flows. The session process interactive data refers to interactive activity data of a corresponding service session process, and includes click activity data, shared activity data, collaborative activity data or multimedia activity data. The session flow intention data refers to data obtained by responding or identifying according to a session flow intention (such as continuously paying attention to a certain object or shielding a certain object) between the service session flow and other service session flows, such as session attention object data, session interaction heat data, session sharing data, session collection data and the like.

The artificial intelligence cloud system 100 may obtain the session flow interaction data and the session flow intention data of each service session flow, and collate the session flow interaction data and the session flow intention data to obtain the session flow interaction data and the session flow intention data of each service session flow. For example, the session flow interaction data may include click activity data, share activity data, collaborative activity data, multimedia activity data, and the like.

The session flow intention data may be directional intention data or undirected intention data. Taking the data of the session attention object as an example, the total number of the intention objects generated between the service session process 1 and the service session process 2 includes the number of the intention objects sent to the service session process 2 by the service session process 1, and the number of the intention objects sent to the service session process 1 by the service session process 2. When the session flow intention data is directed intention data, the number of intention objects sent to the service session flow 1 to the service session flow 2 is used as session attention object data of the service session flow 1, and the number of intention objects sent to the service session flow 1 by the service session flow 2 is used as session attention object data of the service session flow 2; when the session flow intention data is undirected intention data, the session attention object data of the business session flow 1 and the business session flow 2 are the total intention object number between the business session flow 1 and the business session flow 2. For example, if the number of the intention objects of the message sent by the service session flow 1 to the service session flow 2 is 4, and the number of the intention objects of the message sent by the service session flow 2 to the service session flow 1 is 5, when the session flow intention data is directed intention data, the session attention object data of the service session flow 1 is 4, and the session attention object data of the service session flow 2 is 5; when the session flow intention data is undirected intention data, the session attention object data of the business session flow 1 is 9, and the session attention object data of the business session flow 2 is 9.

Taking the session interaction heat data as an example, the session interaction heat data may be a daily heat, a temporal heat, or the like. Taking the session interaction heat data as the daily heat as an example, if the service session flow 1 sends 35 messages to the service session flow 2 within 7 days, the service session flow 2 sends 21 messages to the service session flow 1 within 7 days. When the session flow intention data is directional intention data, the session interaction heat data of the service session flow 1 is (35/7 = 5), and the session interaction heat data of the service session flow 2 is (21/7 = 3); when the session flow intention data is undirected intention data, the session interaction heat data of business session flow 1 is { (35 + 21)/7 =8}, and the session interaction heat data of business session flow 2 is 8.

Step A102, determining an intention sharing conversation process corresponding to each business conversation process according to the conversation process intention data of each business conversation process, and performing data association on the conversation process interaction data of the intention sharing conversation process to the conversation process interaction data of the corresponding business conversation process to obtain data association distribution information.

For example, the artificial intelligence cloud system 100 may obtain data association distribution information according to the session flow interaction data and the session flow intention data of at least two service session flows, where the data association distribution information includes data association distribution of each service session flow. For example, interactive data distribution is generated according to conversation process interactive data of at least two business conversation processes, intention data distribution is generated according to conversation process intention data of at least two business conversation processes, and intention sharing conversation processes corresponding to each business conversation process are determined based on the intention data distribution; and transmitting the conversation process interactive data of the intention sharing conversation process to the conversation process interactive data of the business conversation process corresponding to the intention sharing conversation process according to the model transmission layer, the interactive data distribution and the intention data distribution in the intention artificial intelligence model to obtain data association distribution information. If the number of the at least two business conversation processes is N, the dimension of the conversation process intention data corresponding to each business conversation process is N, intention data distribution is generated according to the conversation process intention data of the at least two business conversation processes, the intention data distribution is the distribution of N × N, wherein the intention data distribution is marked as a, the intention data distribution a has N × N elements, and Ai is the element in the ith row in the intention data distribution a and is used for indicating conversation process intention data between the business conversation process 1i and other business conversation processes. Wherein N and i are positive integers, and i is less than or equal to N.

For example, the intent data distribution is generated as follows:

when the conversation process intention data of each business conversation process is undirected intention data, intention data generated according to the conversation process intention data of at least two business conversation processes are distributed symmetrically; when the conversation process intention data of each business conversation process is directional intention data, intention data generated according to the conversation process intention data of at least two business conversation processes are distributed in an asymmetric mode.

In one embodiment, the artificial intelligence cloud system 100 may obtain session flow intention data of each service session flow, and generate an intention distribution according to the session flow intention data, where each distribution unit of the intention distribution corresponds to one service session flow, and a connection attribute between two distribution units is used to indicate session flow intention data of the two distribution units respectively corresponding to the session flows participating in the session flows. And obtaining conversation process intention data of each business conversation process according to the intention distribution so as to obtain intention data distribution according to the conversation process intention data.

When generating the intention data distribution according to the session process intention data of the at least two service session processes, the artificial intelligence cloud system 100 may generate an initial intention data distribution according to the session process intention data of the at least two service session processes; generating intention attribute distribution of at least two service conversation processes according to the initial intention data distribution; sharing and configuring intention feedback distribution in the initial intention data distribution, and regularizing the shared and configured initial intention data distribution based on intention attribute distribution to generate regularized intention data distribution; and generating the intention data distribution according to the regularized intention data distribution. The intention feedback distribution is a unit distribution, which may be a square matrix, and is a distribution in which elements on a diagonal line (i.e., a main diagonal line) from an upper left corner to a lower right corner are all 1, and elements on other positions are all 0, and is used to share and configure session flow intention data of the service session flow itself in the initial intention data distribution.

Further, the conversation process intention data comprises at least two intention unit data, and the regularized intention data distribution comprises at least two regularized intention data sub-distributions corresponding to the intention unit data respectively. When generating the intention data distribution according to the regularized intention data distribution, the artificial intelligence cloud system 100 may obtain intention behavior influence parameters corresponding to at least two intention unit data, respectively, and fuse the regularized intention data sub-distributions based on the intention behavior influence parameters to generate the intention data distribution. For example, the number of the at least two intention unit data is the number of the sub-data of the session flow intention included in the session flow intention data, that is, the session flow intention data includes m sub-data of the session flow intention, and the artificial intelligence cloud system 100 may generate m corresponding intention unit data according to the m sub-data of the session flow intention, where each intention unit data corresponds to one intention data sub-distribution. Wherein m is a positive integer. Where an initial intent data distribution is denoted B, which includes m initial intent data subdivisions, which may be denoted B = { B1, B2, \8230;, bm }, where each initial intent data subdistribution may be considered a N x N distribution, where N is the number of at least two traffic session flows. For example, if the session flow intention data includes 4 session flow intention sub-data such as session attention object data, session interaction heat data, session sharing data, and session collection data, 4 intention unit data may be generated, where each intention unit data corresponds to one initial intention data sub-distribution, and in this case, m =4 and the initial intention data distribution B = { B1, B2, B3, B4}.

Further, if the session flow intention data includes session attention object data, session interaction heat data, session sharing data and session collection data, extracting features of the session attention object data to obtain intention unit data 1, and generating an initial intention data sub-distribution 1 according to the intention unit data 1. From the initial intent data sub-distribution 1, a sub-intent attribute distribution 1 is generated. Based on the child intention attribute distribution 1, the initial intention data child distribution 1 sharing the configuration intention feedback distribution is regularized to generate a regularized intention data child distribution 1. Similarly, extracting the intention unit data 2 of the conversation interaction heat data, obtaining an initial intention data sub-distribution 2 according to the intention unit data 2, obtaining a sub-intention attribute distribution 2, and regularizing the initial intention data sub-distribution 2 based on the sub-intention attribute distribution 2 to obtain a regularized intention data sub-distribution 2; extracting intention unit data 3 of the session sharing data, obtaining initial intention data sub-distribution 3 according to the intention unit data 3, obtaining sub-intention attribute distribution 3, and regularizing the initial intention data sub-distribution 3 based on the sub-intention attribute distribution 3 to obtain regularized intention data sub-distribution 3; extracting intention unit data 4 of the conversation collection data, obtaining an initial intention data sub-distribution 4 according to the intention unit data 4, obtaining a sub-intention attribute distribution 4, and regularizing the initial intention data sub-distribution 4 based on the sub-intention attribute distribution 4 to obtain a regularized intention data sub-distribution 4. The method comprises the steps of sharing a configuration intention behavior influence parameter W1 for a regularized intention data sub-distribution 1, sharing a configuration intention behavior influence parameter W2 for a regularized intention data sub-distribution 2, sharing a configuration intention behavior influence parameter W3 for a regularized intention data sub-distribution 3, sharing a configuration intention behavior influence parameter W4 for a regularized intention data sub-distribution 4, and fusing to obtain intention data distribution.

For another example, if the session flow intention data includes session attention object data, there are a service session flow 1, a service session flow 2, a service session flow 3, and a service session flow 4. The number of the intention objects sent to the service session flow 2 by the service session flow 1 is 4, and the number of the intention objects sent to the service session flow 4 is 10; the number of the intention objects sent to the service conversation process 1 by the service conversation process 2 is 5, and the number of the intention objects sent to the service conversation process 3 is 4; the number of the intention objects sent to the service session flow 2 by the service session flow 3 is 15; the number of the intention objects sent to the business conversation process 1 by the business conversation process 4 is 5. And extracting the characteristics of the acquired session flow intention data to obtain initial intention data distribution. If the initial intention data distribution is asymmetric distribution, determining the initial intention data distribution as regular intention data distribution by obtaining intention attribute distribution according to the initial intention data distribution B and normalizing the initial intention data distribution B based on intention attribute distribution D; if the initial intention data distribution is a symmetric distribution, the initial intention data distribution is determined such that an intention attribute distribution is obtained from the initial intention data distribution B, and the initial intention data distribution B is regularized based on an intention attribute distribution D, so that a regularized intention data distribution can be obtained. Since, in this example, it is assumed that the session flow intention data includes one session flow intention sub-data of the session attention object data, the regularized intention data distribution may be regarded as the intention data distribution a.

For example, the artificial intelligence cloud system 100 may obtain the session process intent sub-data of m at least two service session processes, which is equivalent to obtaining m × N data (N is the number of at least two service session processes), perform feature extraction on the session process intent sub-data of each at least two service session processes, to obtain the regularized intent data distribution 601, where the regularized intent data distribution 601 includes m regularized intent data sub-distributions, input the m regularized intent data sub-distributions into the intent artificial intelligence model, and fuse the m regularized intent data sub-distributions based on the intent artificial intelligence model, to obtain the intent data distribution, where the m regularized intent data sub-distributions respectively correspond to the intent behavior influence parameters W1, W2, to Wm. The artificial intelligence cloud system 100 performs feature extraction on the acquired session flow interaction data to obtain an interaction data distribution 603, and performs feature propagation on the intention data distribution 602 and the interaction data distribution 603 to obtain data association distribution information 604.

The process of generating data association distribution information based on the intention data distribution and the interaction data distribution is as follows:

and taking the interactive data distribution as initial data association distribution information to obtain the transmission influence parameter distribution of the intention artificial intelligence model. And taking the distribution of the transmission influence parameters as configuration information of a model transmission layer in the intention artificial intelligence model, and performing weighted iteration on the initial data association distribution information based on the model transmission layer with the configuration information and the intention data distribution to obtain data association distribution information.

The interactive data distribution may be denoted as X, where the interactive data distribution X is a distribution of N × F, N is the number of at least two service session flows, and F is a dimension of the interactive data of the session flow, that is, the interactive data of the session flow is the number of the interactive sub-data of the session flow included in the interactive data of the session flow. For example, if the session flow interaction data includes 4 pieces of session flow interaction subdata, such as click activity data, sharing activity data, collaborative activity data, or multimedia activity data, F is 4. And carrying out multiple propagation on the conversation process interactive data of each business conversation process based on a model transfer layer of the intention artificial intelligence model to obtain data association distribution information.

Further, the intention distribution 701 may be formed according to intention data distribution and interaction data distribution, where the intention distribution 701 includes N distribution units, each distribution unit corresponds to one service session flow, each distribution unit carries session flow interaction data of the service session flow corresponding to the node, and a connection attribute in the intention distribution 701 is used to represent session flow intention data of each service session flow. Wherein the interactive data distribution is composed of session process interactive data of each service session process. This interactive data distribution is regarded as initial data associated distribution information, and the initial data associated distribution information is denoted as H0, that is, H0= X.

And determining an intention sharing conversation process corresponding to each business conversation process based on an intention artificial intelligence model, and performing data association on conversation process interaction data of the intention sharing conversation process to the conversation process interaction data of the corresponding business conversation process. Among them, in the intention distribution 701, other business conversation flows having edges with each business conversation flow may be regarded as the intention sharing conversation flow of the business conversation flow. For example, taking the session flow intention data as undirected intention data as an example, if an edge exists between a node corresponding to the service session flow 1 and a node corresponding to the service session flow 2, the service session flow 1 is an intention sharing session flow of the service session flow 2, and the service session flow 2 is an intention sharing session flow of the service session flow 1; taking the session flow intention data as the directional intention data as an example, if the node corresponding to the service session flow 1 has a unidirectional edge pointing to the node corresponding to the service session flow 2, the service session flow 2 is the intention sharing session flow of the service session flow 1. Inputting the intention distribution 701 into a first propagation layer in an intention artificial intelligence model, performing data association on conversation process interaction data carried by each distribution unit in the intention distribution 701 based on a model transfer layer, and processing the conversation process interaction data after the data association based on an activation function 702 in the model transfer layer to obtain output of the first propagation layer, namely first layer data association distribution information, which is recorded as H1.

In the intention artificial intelligence model, data association distribution map 703 is output by performing data association k times on session flow interaction data of each service session flow, each distribution unit in the data association distribution map 703 carries data association distribution of a corresponding service session flow, and the data association distribution of the corresponding service session flow carried by each distribution unit constitutes data association distribution information.

Step a103, obtaining a first data association distribution of the first session flow and a second data association distribution of the second session flow from the data association distribution information, and identifying an intention requirement relationship between the first session flow and the second session flow based on a correlation parameter of the first data association distribution and the second data association distribution.

For example, the data association distribution of any two service session flows is obtained from the data association distribution information, and the intention requirement relationship between the two service session flows is identified according to the correlation parameter between the data association distributions of the two service session flows and the correlation parameter.

For example, the artificial intelligence cloud system 100 may obtain a first data association distribution of a first session flow and a second data association distribution of a second session flow, obtain a characteristic intention tendency reference condition of the first data association distribution and the second data association distribution, obtain shared characteristic information of the first data association distribution and the second data association distribution, and fuse the characteristic intention tendency reference condition and the shared characteristic information to generate a flow related characteristic between the first session flow and the second session flow. And acquiring the correlation parameters of the first data correlation distribution and the second data correlation distribution based on the process correlation characteristics.

In one embodiment, the first data association distribution is first data association distribution information, and the second data association distribution is second data association distribution information. When acquiring the reference condition of the characteristic intention tendency of the first data association distribution and the second data association distribution and acquiring the shared characteristic information of the first data association distribution and the second data association distribution, the artificial intelligence cloud system 100 takes the coincidence ratio of the first data association distribution information and the second data association distribution information as the reference condition of the characteristic intention tendency of the first data association distribution and the second data association distribution; and performing same feature extraction on the first data association distribution information and the second data association distribution information to obtain shared feature information of the first data association distribution and the second data association distribution. If the correlation parameter is greater than the preset correlation parameter, determining that the intention demand relationship between the first conversation process and the second conversation process is an intention interest relationship; and if the correlation parameter is less than or equal to the preset correlation parameter, determining that the intention demand relationship between the first conversation process and the second conversation process is an intention separation relationship.

Recording first data association distribution information as Hi, recording second data association distribution information as Hj, acquiring a characteristic intention tendency reference condition of the first data association distribution and the second data association distribution, and determining the characteristic intention tendency reference condition as | Hi-Hj |; and acquiring shared characteristic information of the first data association distribution and the second data association distribution, and determining that the shared characteristic information is obtained by fusing the characteristic intention tendency reference condition and the shared characteristic information to generate flow related characteristics between the first conversation flow and the second conversation flow. And acquiring the correlation parameters of the first data correlation distribution and the second data correlation distribution based on the process correlation characteristics.

In one embodiment, the artificial intelligence cloud system 100 obtains a session flow optimization request of a target session flow, and obtains a session flow node sequence of the target session flow, where the session flow node sequence includes an associated session flow of the target session flow. Acquiring target data association distribution of a target session flow from the data association distribution information, acquiring association data association distribution of an association session flow in the session flow node sequence, acquiring a correlation parameter between the target data association distribution and the association data association distribution, and determining the association session flow with the correlation parameter larger than a preset correlation parameter as a required session flow of the target session flow. And sending the demand session flow to the target session flow so that the target session flow can determine the demand content based on the demand session flow.

Further, in an independently implementable embodiment, the training process for the intended artificial intelligence model can be as follows:

and acquiring reference conversation process data, and acquiring reference interest conversation process data and reference separation conversation process data of the reference conversation process data. The method comprises the steps of obtaining reference conversation process intention data and reference conversation process interactive data of reference conversation process data, obtaining reference intention interest characteristics and reference interactive interest characteristics of the reference interest conversation process data, and obtaining reference intention separation characteristics and reference interactive separation characteristics of the reference separation conversation process data. Taking the reference conversation process intention data, the reference conversation process interactive data, the reference intention interest characteristics and the reference interactive interest characteristics as reference positive training data; and taking the reference conversation process intention data, the reference conversation process interactive data, the reference intention separation characteristic and the reference interactive separation characteristic as reference negative training data. And training the initial artificial intelligence model based on the reference positive training data and the reference negative training data to generate an intention artificial intelligence model. The reference positive training data and the reference negative training data may be referred to as reference training data.

The annotation information may include an interest annotation and a separation annotation, where the reference positive training data carries the interest annotation and the reference negative training data carries the separation annotation. When an initial artificial intelligence model is trained based on reference positive training data and reference negative training data to generate an intention artificial intelligence model, specifically, the reference training data is input into the initial artificial intelligence model, a sample correlation parameter corresponding to the reference training data is obtained, a loss function between the sample correlation parameter and labeling information is obtained, and the initial artificial intelligence model is optimized based on the loss function to obtain the intention artificial intelligence model. If the reference forward training data are input into the initial artificial intelligence model, obtaining a reference forward training data correlation parameter corresponding to the reference forward training data, obtaining a forward loss function between the reference forward training data correlation parameter and the interest label, and optimizing the initial artificial intelligence model based on the forward loss function; if the reference negative training data are input into the initial artificial intelligence model, obtaining a reference negative training data correlation parameter corresponding to the reference negative training data, obtaining a negative loss function between the reference negative training data correlation parameter and the separation label, and optimizing the initial artificial intelligence model based on the negative loss function. And determining the optimized initial artificial intelligence model as an intention artificial intelligence model.

The method comprises the steps of obtaining session flow interactive data and session flow intention data of at least two service session flows, determining intention sharing session flows corresponding to each service session flow according to the session flow intention data of each service session flow, and performing data association on the session flow interactive data of the intention sharing session flows to the session flow interactive data of the corresponding service session flows to obtain data association distribution information, wherein the data association distribution information comprises data association distribution of each service session flow; the method comprises the steps of obtaining first data correlation distribution of a first conversation process from data correlation distribution information, obtaining correlation parameters of the first data correlation distribution and the second data correlation distribution with second data of a second conversation process, and identifying an intention requirement relation between the first conversation process and the second conversation process based on the correlation parameters, wherein the first conversation process and the second conversation process belong to at least two business conversation processes. Through the process, the session flow intention data and the session flow interaction data of each service session flow are integrated to obtain data association distribution of each service session flow, wherein the data association distribution is obtained based on the distribution of the connection between the service session flow and the service session flow, the session flow interaction data of each service session flow is subjected to data association, and one data association distribution is obtained according to the result of the data association, and combines the session flow interaction data and the session flow intention data of the corresponding service session flow and the session flow interaction data of the intention sharing session flow of the corresponding service session flow, so that when the correlation parameter between the two service session flows is large, the intention requirement between the two service session flows is obvious, and the difference of the common characteristics of the two service session flows or the characteristics of the respectively associated session flows is small, so that the possibility that the intention requirement relationship exists between the two service session flows is high. Therefore, in the embodiment of the present disclosure, the correlation parameter between the two service session flows can be used to measure the intention tendency reference condition of the two service session flows, so that the intention requirement relationship between the two service session flows can be obtained through the correlation parameter between the two service session flows, and the matching accuracy between the subsequent service content and the actual user intention is improved. Meanwhile, the intention requirement relation among different business conversation processes is intelligently identified based on the conversation process interactive data and the conversation process intention data of the business conversation processes, and the accuracy in the content pushing process is improved.

In an embodiment, the following exemplary steps may be implemented in the process of generating the personalized service content pushed to the service request terminal according to the intention requirement relationship between the first session flow and the second session flow.

Step C110, when it is determined that the intention requirement relationship between the first conversation process and the second conversation process is an intention interest relationship, obtaining target conversation process interest data having the intention interest relationship between the first conversation process and the second conversation process, and inputting the target conversation process interest data into a conversation process image prediction network.

In this embodiment, the conversation process image prediction network may be a deep learning network trained in advance, and the conversation process image prediction network is used to perform conversation process image prediction on the target conversation process interest data so as to ensure the reliability of subsequent trend object mining.

Further, the embodiment of the present disclosure further provides a training process for the session flow image prediction network, where the session flow image prediction network is obtained by training based on reference session flow interest data and reference convergence parameter information, and the reference session flow interest data is a session flow interest data sequence in which the number of positive session interest topics is inconsistent with the number of negative session interest topics; the reference convergence parameter information is determined based on the conversation process portrait classification information and the target conversation process portrait information.

Furthermore, the target session flow image information is the target session flow image information corresponding to each reference session flow interest data segment in the reference session flow interest data, the session flow image classification information is the session flow image classification information corresponding to the reference session flow interest data segment acquired by using the session flow image prediction network, and the reference convergence parameter information includes first convergence parameter information, second convergence parameter information, and convergence parameter optimization node information.

Based on the above, before step C110, the session flow sketch prediction network may be trained in advance, and the training process for the session flow sketch prediction network includes the following steps a and b.

Step a, acquiring the reference conversation process interest data and target conversation process image information corresponding to each reference conversation process interest data fragment in the reference conversation process interest data.

And b, training a reference conversation process image prediction network according to the reference conversation process interest data and the target conversation process image information to obtain the conversation process image prediction network.

On the basis of the above, the session flow interest data sequence includes a plurality of reference session flow interest data segments, the reference session flow image prediction network includes a reference feature extraction structure and a reference feature prediction structure, and step b can also be implemented by: performing feature extraction and portrait component output on each reference conversation process interest data fragment through the reference feature extraction structure to obtain a reference conversation process portrait component corresponding to each reference conversation process interest data fragment; performing conversation process portrait prediction on the reference conversation process portrait component through the reference characteristic prediction structure to obtain conversation process portrait classification information; and determining the reference convergence parameter information according to the conversation process image classification information and the target conversation process image information corresponding to each reference conversation process interest data fragment, and adjusting the network convergence reference information of the reference conversation process image prediction network according to the reference convergence parameter information until the floating change of the reference convergence parameter information is smaller than the set floating change or the training of the set times is completed.

On the basis of the above, the determining the reference convergence parameter information according to the session flow image classification information and the target session flow image information corresponding to each reference session flow interest data segment includes: determining first network convergence reference information according to conversation process image classification information corresponding to each reference conversation process interest data fragment, conversation process image floating data in the target conversation process image information and second preset conversation process image information; determining second network convergence reference information according to the delayed session flow image of the first network convergence reference information; and generating the reference convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data, the reference change data of the forward conversation interest topic, the reference fixed data and the convergence parameter optimization node information.

In this embodiment, the network convergence reference information may be understood as network parameter information, the reference change data may be weight information with real-time dynamic change, the reference fixed data may be weight information without real-time dynamic change, and the floating data of the session flow image may be used to represent information of influence of different session flow images on other session flow images.

Further, the generating the reference convergence parameter information according to the second network convergence reference information, the session flow image classification information, the session flow image floating data, the reference change data of the forward session interest topic, the reference fixed data, and the convergence parameter optimization node information includes: generating the first convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data and the reference change data of the forward conversation interest topic; generating second convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data, the reference change data of the forward conversation interest topic and the reference fixed data; and generating the reference convergence parameter information according to the first convergence parameter information, the second convergence parameter information and the convergence parameter optimization node information.

In this way, by implementing the contents described in the above steps a and b, it is possible to implement training of the session flow image prediction network in advance, thereby ensuring model performance of the session flow image prediction network.

And step C120, performing session flow portrait prediction on the target session flow interest data through the session flow portrait prediction network to obtain a session flow portrait corresponding to the target session flow interest data.

In this embodiment, there are a plurality of session flow images corresponding to the target session flow interest data, for example, the session flow image 1, the session flow image 2, or the session flow image 3, but not limited thereto, it is understood that the key session activity events may be different in different session flow images, and different key session activity events can be distinguished as much as possible by analyzing different session flow images of the target session flow interest data, so as to comprehensively implement analysis and mining of the distribution information of the topics of interest.

In this embodiment, the session flow image prediction network includes a feature extraction structure and a feature prediction structure, where the feature extraction structure and the feature prediction structure may be functional network layers in the session flow image prediction network, and further, step C120 may be implemented by the following steps: inputting the target conversation process interest data into the feature extraction structure for feature extraction and portrait component output so as to obtain conversation process portrait components corresponding to the target conversation process interest data; inputting the conversation process image component into the feature prediction structure to predict the conversation process image so as to obtain conversation process image information in a conversation interaction process; and determining a conversation process image corresponding to the target conversation process interest data according to first preset conversation process image information and the conversation process image information in the conversation interaction process.

In this embodiment, the conversation process image component may be an image component formed by time-series filtering. By means of the design, different conversation process images corresponding to the target conversation process interest data can be accurately and completely determined through the mutual matching of the feature extraction structure and the feature prediction structure.

It is to be understood that the feature extraction structure may further include a plurality of functional layers having a cascade relationship, for example, the feature extraction structure may further include a first feature extraction structure, a second feature extraction structure layer, a third feature extraction structure, and so on, based on which the inputting of the target conversation process interest data to the feature extraction structure for feature extraction and portrait component output to obtain a conversation process portrait component corresponding to the target conversation process interest data described above includes: identifying each conversation process interest point feature in the target conversation process interest data as an interest feature component through the first feature extraction structure; performing interest trend component analysis on the interest data of the target session process through the second feature extraction structural layer, and performing trend dimension feature extraction on trend attribute features corresponding to the obtained interest trend component analysis to obtain trend dimension features; performing feature extraction on interest feature components and trend dimension features corresponding to the interest point features of each conversation process through the third feature extraction structure to obtain conversation process image components corresponding to the interest point features of each conversation process; and determining conversation process image components corresponding to the target conversation process interest data according to the conversation process image components corresponding to all the conversation process interest point characteristics in the target conversation process interest data.

In the above, the conversation process image label is used to distinguish different conversation process images. In actual implementation, the conversation process image components can be further determined by screening the conversation process interest point characteristics, then performing the conversation process image prediction of the target conversation process interest data in parallel and further acquiring the corresponding trend dimension characteristics, wherein the conversation process image components can be realized based on a Kmeans clustering algorithm. By the design, the independence of the conversation process image components can be ensured.

Step C130, acquiring a corresponding key session activity event from the target session process interest data according to the session process image, and generating an attention topic distribution information according to the session process image and the key session activity event, so as to determine the personalized service content of the service request terminal according to the attention topic distribution information.

In this embodiment, the acquiring of the corresponding key session activity event from the target session flow interest data according to the session flow image may include the following contents described in steps C131 to C133.

Step C131, obtaining a first session activity event and a second session activity event corresponding to the target session process interest data according to the portrait matching template corresponding to the session process image, where the first session activity event includes a session activity event that does not carry session permission attributes in the target session process interest data, and the second session activity event includes a session activity event that carries session permission attributes in the target session process interest data. In this embodiment, the session authority attribute may be used to distinguish different session activity events, and the session activity events are used to characterize independent events or event groups specifically describing session behaviors.

In this embodiment, the obtaining a first session activity event and a second session activity event corresponding to the target session flow interest data according to the sketch matching template corresponding to the session flow image further includes: segmenting the target conversation process interest data according to an image matching template corresponding to the conversation process image to obtain a first conversation activity event which does not carry conversation authority attribute in the target conversation process interest data, and clustering the first conversation activity event in the target conversation process interest data aiming at a conversation service label to serve as the first conversation activity event; and acquiring a second session activity event carrying session authority attributes in the target session process interest data according to the first session activity event, and clustering the second session activity event in the target session process interest data aiming at session service labels to serve as the second session activity event.

Step C132, performing frequent item mining on the first session activity event to obtain the non-dynamic frequent item characteristics corresponding to the first session activity event; and performing frequent item mining on the second session activity event to obtain a dynamic frequent item characteristic corresponding to the second session activity event.

In this embodiment, the mining the frequent items of the first session activity event to obtain the non-dynamic frequent item feature corresponding to the first session activity event includes: calling a first frequent item feature extraction layer in a preset frequent item mining model, and performing frequent item mining on the first session activity event to obtain the non-dynamic frequent item feature corresponding to the first session activity event. The mining the frequent items of the second session activity event to obtain the dynamic frequent item characteristics corresponding to the second session activity event includes: and calling a second frequent item feature extraction layer in the preset frequent item mining model, and performing frequent item mining on the second session activity event to obtain the dynamic frequent item feature corresponding to the second session activity event.

Step C133, performing feature fusion based on the frequent item classification probability on the dynamic frequent item feature and the non-dynamic frequent item feature to obtain an interest fusion feature corresponding to the target session process interest data; clustering the interest points of the interest fusion characteristics to obtain clustering information corresponding to the interest data of the target session process; and when the clustering information meets a preset clustering feedback condition, acquiring a session activity event matched with the clustering category attribute from the target session process interest data through the clustering category attribute indicated by the clustering information as the key session activity event.

In this embodiment, the performing feature fusion based on frequent item classification probability on the dynamic frequent item feature and the non-dynamic frequent item feature to obtain an interest fusion feature corresponding to the target session flow interest data includes: calling a feature fusion layer in the preset frequent item mining model, and performing feature fusion based on frequent item classification probability on the dynamic frequent item features and the non-dynamic frequent item features to obtain interest fusion features corresponding to the target session process interest data.

In this embodiment, when the clustering information described in step C133 meets a preset clustering feedback condition, the session activity event matching the clustering category attribute is acquired from the target session flow interest data through the clustering category attribute indicated by the clustering information as the key session activity event, and further, the method may include the following contents described in steps C1331 to C1334.

Step C1331, obtaining key session activity behavior data of the clustering information; and respectively carrying out dynamic track analysis and static track analysis on the activity behavior objects of the plurality of session activity behavior data in the key session activity behavior data to obtain dynamic track analysis information and static track analysis information.

Step C1332, performing dynamic track expansion processing on the dynamic track analysis information through a preset dynamic track expansion mode to obtain a dynamic session activity behavior data set comprising a dynamic track; and performing static track expansion processing on the static track analysis information in a preset static track expansion mode to obtain a static session activity behavior data set comprising a static track.

Step C1333, analyzing the conversation response frequency based on the dynamic conversation activity behavior data set and the static conversation activity behavior data set to obtain a conversation activity behavior object matched with the target conversation response index in the key conversation activity behavior data; the target session response indicator includes at least one of a dynamic trajectory and a static trajectory.

In this embodiment, the session activity behavior object is used to perform session subscription volume analysis on the key session activity behavior data, so as to achieve accurate acquisition of the key session activity event.

Step C1334, performing session subscription volume analysis on the key session activity behavior data according to the session activity behavior object to obtain a session subscription volume analysis result, and if the session subscription volume analysis result indicates that the key session activity behavior data corresponds to a subscription volume increase trend state, acquiring a session activity event matched with the dynamic track from target session process interest data in a subscription volume increase trend state corresponding to the cluster category attribute indicated by the cluster information as the key session activity event.

Further, in order to quickly and flexibly generate the personalized service content of the service request terminal, the step C130 of generating the interested topic distribution information according to the session flow image and the key session activity event to determine the personalized service content of the service request terminal according to the interested topic distribution information may include the following steps: acquiring positive tendency information and negative tendency information in the key session activity event according to the session attribute label information corresponding to the session flow image; on the basis of the trend proportion change between the positive trend information and the negative trend information in the key session activity event, trend object mining is carried out on the positive trend information and the negative trend information in the key session activity event to obtain trend object mining information; determining the negative tendency information with abnormity in tendency object mining as reference negative tendency information, and determining a personalized tag matched with the reference negative tendency information according to the similarity between the negative tendency information in the tendency object mining information and the reference negative tendency information; carrying out tendency object mining on the personalized label matched with the reference negative tendency information and the reference negative tendency information to obtain key tendency object information; according to the key tendency object information and the tendency object mining information, determining concerned subject distribution information in the key session activity event and labeling characteristics corresponding to the concerned subject distribution information; wherein the tagged features comprise different user attention attributes corresponding to the attention topic distribution information; and according to the concerned subject distribution information and the corresponding labeling characteristics thereof, carrying out personalized service content pushing on the service request terminal by adopting a preset service content data source to obtain the personalized service content.

Further, inputting the acquired target session flow interest data into a session flow image prediction network to obtain a corresponding session flow image, and implementing personalized service content push to the service request terminal according to the session flow image to obtain personalized service content, which may include: acquiring target conversation process interest data, and inputting the target conversation process interest data into a conversation process image prediction network; performing session flow portrait prediction on the target session flow interest data through the session flow portrait prediction network to obtain a session flow portrait corresponding to the target session flow interest data; and acquiring a corresponding key session activity event from the target session process interest data according to the session process image, generating concerned subject distribution information according to the session process image and the key session activity event, and determining personalized service content of the service request terminal according to the concerned subject distribution information.

Fig. 3 is a schematic functional block diagram of a content recommendation device 300 based on big data and AI prediction according to an embodiment of the disclosure, and the functions of the functional blocks of the content recommendation device 300 based on big data and AI prediction are described in detail below.

The first obtaining module 310 is configured to generate personalized service content of the service request terminal according to the service operation big data of the service request terminal.

The second obtaining module 320 is configured to obtain collaborative interaction behavior data of the service user of the service request terminal and other service users for the personalized service content.

The prediction module 330 is configured to predict the collaborative interaction behavior data based on a preconfigured collaborative portrait prediction model, and obtain a collaborative portrait sequence corresponding to the collaborative interaction behavior data.

And the recommending module 340 is configured to recommend corresponding collaborative personalized service content to the service user and the other service users according to the collaborative portrait sequence.

Fig. 4 illustrates a hardware structural diagram of an artificial intelligence cloud system 100 for implementing the big data and AI prediction based content recommendation method described above according to an embodiment of the present disclosure, and as shown in fig. 4, the artificial intelligence cloud system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In a specific implementation process, at least one processor 110 executes computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 may execute the content recommendation method based on big data and AI prediction according to the above method embodiment, the processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 may be configured to control transceiving actions of the communication unit 140, so as to perform data transceiving with the aforementioned service request terminal 200.

For a specific implementation process of the processor 110, reference may be made to the above-mentioned various method embodiments executed by the artificial intelligence cloud system 100, which implement principles and technical effects similar to each other, and details of this embodiment are not described herein again.

In addition, the embodiment of the disclosure also provides a readable storage medium, which is preset with computer-executable instructions, and when a processor executes the computer-executable instructions, the content recommendation method based on big data and AI prediction is implemented as above.

Finally, it should be understood that the examples in this specification are only intended to illustrate the principles of the examples in this specification. Other variations are also possible within the scope of this description. Accordingly, by way of example, and not limitation, alternative configurations of the embodiments of the specification can be seen as matching the teachings of the specification. Accordingly, the embodiments of the present description are not limited to only those embodiments explicitly described and depicted herein.

Claims

1. A content recommendation method based on big data and AI prediction is applied to an artificial intelligence cloud system, wherein the artificial intelligence cloud system is in communication connection with a plurality of service request terminals, and the method comprises the following steps:

recommending corresponding collaborative personalized service contents for the business user and the other business users according to the collaborative portrait sequence;

the step of generating the personalized service content of the service request terminal according to the service operation big data of the service request terminal includes:

acquiring session flow interactive data and session flow intention data of at least two service session flows in the service operation big data of the service request terminal;

determining an intention sharing conversation process corresponding to each business conversation process according to the conversation process intention data of each business conversation process, and performing data association on the conversation process interactive data of the intention sharing conversation process to the conversation process interactive data of the corresponding business conversation process to obtain data association distribution information; the data association distribution information comprises data association distribution of each service session process;

acquiring a first data association distribution of a first session flow from the data association distribution information, acquiring a correlation parameter of the first data association distribution and a second data association distribution of a second session flow, and identifying an intention demand relationship between the first session flow and the second session flow based on the correlation parameter; the first session process and the second session process belong to the at least two service session processes;

generating personalized service content pushed to the service request terminal according to the intention requirement relationship between the first session flow and the second session flow so as to push the personalized service content to the service request terminal;

the step of generating personalized service content pushed to the service request terminal according to the intention requirement relationship between the first session flow and the second session flow so as to push the personalized service content to the service request terminal includes:

when the intention requirement relation between the first conversation process and the second conversation process is determined to be an intention interest relation, target conversation process interest data with the intention interest relation between the first conversation process and the second conversation process are obtained, and the target conversation process interest data are input into a conversation process image prediction network;

performing session flow portrait prediction on the target session flow interest data through the session flow portrait prediction network to obtain a session flow portrait corresponding to the target session flow interest data;

acquiring a corresponding key session activity event from the target session process interest data according to the session process image, generating concerned subject distribution information according to the session process image and the key session activity event, and determining personalized service content of the service request terminal according to the concerned subject distribution information;

the conversation process image prediction network comprises a feature extraction structure and a feature prediction structure; the predicting the conversation process portrait of the target conversation process interest data through the conversation process portrait predicting network to obtain the conversation process portrait corresponding to the target conversation process interest data includes:

inputting the target conversation process interest data into the feature extraction structure for feature extraction and portrait component output so as to obtain conversation process portrait components corresponding to the target conversation process interest data;

inputting the conversation process image component into the feature prediction structure to predict the conversation process image so as to obtain the conversation process image information of the conversation interaction process;

determining a conversation process image corresponding to the target conversation process interest data according to first preset conversation process image information and conversation process image information of the conversation interaction process;

the feature extraction structure comprises a first feature extraction structure, a second feature extraction structural layer and a third feature extraction structure; the inputting the target conversation process interest data into the feature extraction structure for feature extraction and portrait component output to obtain conversation process portrait components corresponding to the target conversation process interest data includes:

identifying each conversation process interest point feature in the target conversation process interest data as an interest feature component through the first feature extraction structure;

performing interest trend component analysis on the interest data of the target session process through the second feature extraction structural layer, and performing trend dimension feature extraction on trend attribute features corresponding to the obtained interest trend component analysis to obtain trend dimension features;

performing feature extraction on interest feature components and trend dimension features corresponding to the interest point features of each conversation process through the third feature extraction structure to obtain conversation process image components corresponding to the interest point features of each conversation process;

determining conversation process image components corresponding to the target conversation process interest data according to the conversation process image components corresponding to all the conversation process interest point characteristics in the target conversation process interest data;

the acquiring the corresponding key session activity event from the target session process interest data according to the session process image comprises:

segmenting the target conversation process interest data according to an image matching template corresponding to the conversation process image to obtain a first conversation activity event which does not carry conversation authority attribute in the target conversation process interest data, and clustering the first conversation activity event in the target conversation process interest data aiming at a conversation service label to serve as the first conversation activity event; according to the first session activity event, acquiring a second session activity event carrying session permission attributes in the target session process interest data, clustering the second session activity event in the target session process interest data aiming at session service tags to serve as the second session activity event, wherein the first session activity event comprises the session activity event not carrying the session permission attributes in the target session process interest data, and the second session activity event comprises the session activity event carrying the session permission attributes in the target session process interest data;

performing frequent item mining on the first session activity event to obtain the non-dynamic frequent item characteristics corresponding to the first session activity event;

performing frequent item mining on the second session activity event to obtain a dynamic frequent item characteristic corresponding to the second session activity event;

performing feature fusion based on frequent item classification probability on the dynamic frequent item features and the non-dynamic frequent item features to obtain interest fusion features corresponding to the target session process interest data;

clustering the interest points of the interest fusion characteristics to obtain clustering information corresponding to the interest data of the target session process;

when the clustering information meets a preset clustering feedback condition, acquiring a session activity event matched with the clustering category attribute from the target session process interest data through the clustering category attribute indicated by the clustering information as the key session activity event;

generating concerned subject distribution information according to the session flow image and the key session activity event so as to determine personalized service content of the service request terminal according to the concerned subject distribution information, wherein the concerned subject distribution information comprises the following steps:

acquiring positive tendency information and negative tendency information in the key session activity event according to the session attribute label information corresponding to the session flow image; on the basis of the trend proportion change between the positive trend information and the negative trend information in the key session activity event, trend object mining is carried out on the positive trend information and the negative trend information in the key session activity event to obtain trend object mining information;

determining the negative tendency information with abnormity in tendency object mining as reference negative tendency information, and determining a personalized tag matched with the reference negative tendency information according to the similarity between the negative tendency information in the tendency object mining information and the reference negative tendency information; carrying out tendency object mining on the personalized label matched with the reference negative tendency information and the reference negative tendency information to obtain key tendency object information; according to the key tendency object information and the tendency object mining information, determining concerned subject distribution information in the key session activity event and labeling characteristics corresponding to the concerned subject distribution information; wherein the tagged features comprise different user attention attributes corresponding to the attention topic distribution information;

and according to the concerned subject distribution information and the corresponding labeling characteristics thereof, carrying out personalized service content pushing on the service request terminal by adopting a preset service content data source to obtain the personalized service content.

2. The big data and AI prediction based content recommendation method according to claim 1, wherein said conversational flow image prediction network is trained based on reference conversational flow interest data and reference convergence parameter information, said reference conversational flow interest data being a conversational flow interest data sequence where the number of positive conversational interest topics is inconsistent with the number of negative conversational interest topics; the reference convergence parameter information is determined according to session flow portrait classification information and target session flow portrait information, wherein the target session flow portrait information is target session flow portrait information corresponding to each reference session flow interest data fragment in the reference session flow interest data, the session flow portrait classification information is session flow portrait classification information corresponding to the reference session flow interest data fragment acquired by using the session flow portrait prediction network, and the reference convergence parameter information includes first convergence parameter information, second convergence parameter information and convergence parameter optimization node information, and the method further includes:

acquiring the reference conversation process interest data and target conversation process image information corresponding to each reference conversation process interest data fragment in the reference conversation process interest data;

training a reference conversation process image prediction network according to the reference conversation process interest data and the target conversation process image information to obtain the conversation process image prediction network;

the conversation process interest data sequence comprises a plurality of reference conversation process interest data fragments, and the reference conversation process image prediction network comprises a reference feature extraction structure and a reference feature prediction structure; the training of the reference conversation process image prediction network according to the conversation process interest data sequence and the target conversation process image information to obtain the conversation process image prediction network comprises the following steps:

performing feature extraction and portrait component output on each reference conversation process interest data fragment through the reference feature extraction structure to obtain a reference conversation process portrait component corresponding to each reference conversation process interest data fragment;

performing conversation process portrait prediction on the reference conversation process portrait component through the reference characteristic prediction structure to obtain conversation process portrait classification information;

and determining the reference convergence parameter information according to the conversation process image classification information and the target conversation process image information corresponding to each reference conversation process interest data fragment, and adjusting the network convergence reference information of the reference conversation process image prediction network according to the reference convergence parameter information until the floating change of the reference convergence parameter information is smaller than the set floating change or the training of the set times is completed.

3. The big data and AI prediction based content recommendation method of claim 2, wherein said determining said reference convergence parameter information based on session flow image classification information and target session flow image information corresponding to each of said reference session flow interest data tiles comprises:

determining first network convergence reference information according to conversation process image classification information corresponding to each reference conversation process interest data fragment, conversation process image floating data in the target conversation process image information and second preset conversation process image information;

determining second network convergence reference information according to the delayed session flow image of the first network convergence reference information;

and generating the reference convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data, the reference change data of the forward conversation interest topic, the reference fixed data and the convergence parameter optimization node information.

4. The big data and AI prediction based content recommendation method according to claim 3, wherein said generating said reference convergence parameter information based on said second network convergence reference information, said session flow image classification information, said session flow image floating data, reference variation data of forward session interest topic, reference fixed data and said convergence parameter optimized node information comprises:

generating the first convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data and the reference change data of the forward conversation interest topic;

generating second convergence parameter information according to the second network convergence reference information, the conversation process image classification information, the conversation process image floating data, the reference change data of the forward conversation interest topic and the reference fixed data;

and generating the reference convergence parameter information according to the first convergence parameter information, the second convergence parameter information and the convergence parameter optimization node information.

5. An artificial intelligence cloud system comprising a processor and a machine-readable storage medium having at least one instruction, at least one program, a set of codes, or a set of instructions stored therein, the at least one instruction, the at least one program, the set of codes, or the set of instructions being loaded and executed by the processor to implement the big data and AI prediction based content recommendation method of any one of claims 1-4.