CN114969504A

CN114969504A - Big data processing method and system combining user interest analysis

Info

Publication number: CN114969504A
Application number: CN202210334635.XA
Authority: CN
Inventors: 任国明
Original assignee: Individual
Current assignee: Individual
Priority date: 2022-03-31
Filing date: 2022-03-31
Publication date: 2022-08-30

Abstract

The embodiment of the disclosure discloses a big data processing method and a big data processing system combined with user interest analysis, which ensure that a user interest analysis network can fully mine the session state of a service session in digital service interaction data through user activity description contents by determining the user activity description contents of the digital service interaction data to be subjected to the user interest analysis, ensure that a user interest mining result obtained by mining the user interest analysis network can be better aligned with the service session of the digital service interaction data, and ensure the user interest mining precision and the reliability. The user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained through a user activity description identification network different from the user interest analysis network, and is processed through different networks, so that the high-quality user activity description content can be obtained, and the relevance of the user interest mining result and the service session of the digital service interaction data is improved.

Description

Big data processing method and system combining user interest analysis

Technical Field

The present disclosure relates to the field of big data technologies, and in particular, to a big data processing method and system in combination with user interest analysis.

Background

With the rapid development of the internet and information technology, network services such as digital services and the like are more and more popular, and a large amount of information is gathered to form 'user big data'. In the early development stage of digital services, users can obtain required information through a search engine, but in the big data era, it is becoming more and more difficult for users to quickly and accurately find required information, and the problem of information overload becomes an important problem in the internet technology. In order to improve the problem, a recommendation system is developed, and unlike a search engine, the recommendation system does not need a user to provide specific requirements, but actively recommends information meeting their interests and requirements to the user by analyzing the historical behaviors of the user, but the quality of analysis results obtained by the current user analysis technology is difficult to guarantee.

Disclosure of Invention

It is an object of the present disclosure to provide a big data processing method and system in conjunction with user interest analysis.

The technical scheme of the disclosure is realized by at least some of the following embodiments.

A big data processing method in conjunction with user interest analysis, the method being performed by a big data processing system, the method comprising: determining digital service interaction data to be subjected to user interest analysis; determining user activity description content of the digital service interaction data to be subjected to user interest analysis through a user activity description recognition network; and mining the user interest of the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network by combining the user activity description content, and obtaining a user interest mining result of the digital service interaction data to be subjected to the user interest analysis.

The method and the device are applied to the embodiment, the user activity description content of the digital service interaction data to be subjected to the user interest analysis is determined, the user interest analysis network can fully mine the session state of the business session in the digital service interaction data to be subjected to the user interest analysis through the user activity description content, the user interest mining result obtained by mining the user interest analysis network can be better aligned with the business session of the digital service interaction data to be subjected to the user interest analysis, and the user interest mining precision and the reliability are guaranteed to a certain extent. Meanwhile, the user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained through a user activity description identification network different from the user interest analysis network, and the user activity description content with high quality can be obtained through processing through different networks, so that the correlation between the user interest mining result and the service session of the digital service interaction data to be subjected to the user interest analysis is improved.

Under some independently designed ideas, the user interest mining result comprises a project attention heat mining result; the method for mining the user interest of the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network by combining the user activity description content to obtain the user interest mining result of the digital service interaction data to be subjected to the user interest analysis comprises the following steps: processing the digital service interaction data to be subjected to user interest analysis through a user interest analysis network to obtain service session thermodynamic index data of the digital service interaction data to be subjected to user interest analysis; and acquiring a project attention heat mining result of the digital service interaction data to be subjected to the user interest analysis by combining the service session heat index data and the user activity description content.

The method and the device are applied to the embodiment, and the accuracy of user interest mining of the user interest analysis network under the service session of the changeable thermodynamic indexes can be guaranteed by determining the service session thermodynamic index data of the digital service interaction data aiming at the user interest analysis to be performed.

Under some independently designed ideas, the service session thermodynamic index data is user activity description guidance distribution of user activity description guidance characteristics of different service interaction behavior events carrying digital service interaction data to be subjected to user interest analysis, and the user activity description content is user activity description distribution of user activity description of different service interaction behavior events carrying digital service interaction data to be subjected to user interest analysis. The method for obtaining the project attention heat mining result of the digital service interaction data to be subjected to the user interest analysis by combining the service session heat index data and the user activity description content comprises the following steps: and carrying out binary operation on the user activity description guide distribution and the user activity description distribution to obtain a project attention heat mining result of the digital service interaction data to be subjected to user interest analysis.

The method and the device are applied to the embodiment, and the thermodynamic index data of the service session can be flexibly determined through the user activity description guide distribution, so that the accuracy of user interest mining of the user interest analysis network under the service session of the changeable thermodynamic index is guaranteed.

Under some independently designed ideas, the user interest analysis network comprises a non-exclusive key content mining layer and a project attention heat translation layer. The method for processing the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network to obtain the service session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis comprises the following steps: performing key content mining on digital service interaction data to be subjected to user interest analysis through a non-exclusive key content mining layer to obtain service interaction key content distribution, and splicing the service interaction key content distribution and key content distribution of a first service session state derived by a user activity description mining layer of a user activity description recognition network to obtain first spliced key content distribution; and translating the first splicing key content distribution through the project attention heat translation layer to obtain service session heat index data of the digital service interaction data to be subjected to user interest analysis.

The method is applied to the embodiment, and the key content distribution of the first service session state derived by the user activity description mining layer of the service interaction key content distribution and user activity description recognition network is spliced by the non-independent key content mining layer, so that the user interest analysis network can ensure the user interest mining precision and credibility to a certain extent through the key content of the first service session state key content distribution.

Under some independently designed ideas, the non-exclusive key content mining layer comprises X key content mining modules which are cascaded, and each key content mining module comprises a user activity description optimization program. The splicing of the key content distribution of the service interaction key content and the key content distribution of the first service session state derived by the user activity description mining layer of the user activity description recognition network to obtain the first spliced key content distribution comprises the following steps: the method comprises the steps of distributing and loading service interaction key content to a first key content mining module; for each key content mining module: splicing the key content distribution derived by the upstream key content mining module and the key content distribution of the first service session state through a user activity description optimization program to obtain second spliced key content distribution matched with the key content mining module; the content diversity evaluation in the service session state key content distribution matched with each key content mining module is different; and combining the second splicing key content distribution of the last key content mining module to obtain the first splicing key content distribution.

The method is applied to the embodiment, the service session state key content distribution derived by the user activity description recognition network and the service interaction key content distribution obtained by the user interest analysis network by mining the key content of the digital service interaction data to be subjected to the user interest analysis are spliced through the user activity description optimization program, so that the service session state data of the service session in the digital service interaction data to be subjected to the user interest analysis in the service session state key content distribution can be used by the user interest analysis network, and the beneficial effect of sharing the recognition result obtained by the user activity description recognition network to the user interest analysis network for use is realized.

Under some independently designed ideas, before the key content distribution derived by the upstream key content mining module and the business session state key content distribution are spliced by the user activity description optimization program to obtain a second spliced key content distribution matched with the key content mining module, the method further includes: performing content simplification operation on the key content distribution derived by the upstream key content mining module; and/or, splicing the key content distribution derived by the key content mining module at the upstream and the key content distribution of the service session state through a user activity description optimization program to obtain a second spliced key content distribution matched with the key content mining module, wherein the second spliced key content distribution comprises the following steps: describing the optimization program execution by user activity: and changing the business session state key content distribution into a business session state key content distribution with a specified scale, processing the changed business session state key content distribution and the key content distribution derived by the upstream key content mining module, and obtaining a second spliced key content distribution matched with the key content mining module.

By applying the method and the device to the embodiment, the key content distribution derived by the key content mining module at the upstream can be simplified through content simplification operation. Meanwhile, the user activity description optimization program processes the key content distribution of the service session state and the key content distribution derived by the upstream key content mining module, and the splicing of the key content distribution of the service session state and the key content distribution derived by the upstream key content mining module is accurately realized.

Under some independently designed ideas, the translating the first splicing key content distribution through the project attention heat translation layer to obtain service session thermodynamic index data of the digital service interaction data to be subjected to user interest analysis, including: and translating the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs through a project attention heat translation layer to obtain service session thermodynamic index data of the digital service interaction data to be subjected to user interest analysis.

The project attention heat translation layer can accurately and completely obtain the business session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis through the first splicing key content distribution and the second splicing key content distribution derived by the user activity description optimization program.

Under some independently designed ideas, the user interest analysis network further comprises a service demand scoring translation layer; in combination with the user activity description content, performing user interest mining on the digital service interaction data to be subjected to the user interest analysis through a user interest analysis network to obtain a user interest mining result of the digital service interaction data to be subjected to the user interest analysis, and further comprising: and translating the first spliced key content distribution through a service demand scoring translation layer to obtain a service demand scoring mining result of the digital service interaction data to be subjected to user interest analysis.

The service demand scoring translation layer can accurately and completely obtain the service demand scoring excavation result of the digital service interaction data to be subjected to the user interest analysis through the first splicing key content distribution.

Under some independently designed ideas, the service demand scoring mining result of the digital service interaction data to be subjected to the user interest analysis is obtained by translating the first spliced key content distribution through a service demand scoring translation layer, and the method comprises the following steps: and translating the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs through a service demand scoring translation layer to obtain a service demand scoring mining result of the digital service interaction data to be subjected to user interest analysis.

The service demand scoring translation layer can accurately and completely obtain the service demand scoring mining result of the digital service interaction data to be subjected to the user interest analysis through the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs.

Under some independently designed ideas, the user activity description recognition network comprises a user activity description mining layer, a user activity description translation layer and a user activity description optimization layer. The method for determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis through the user activity description recognition network comprises the following steps of: mining digital service interaction data to be subjected to user interest analysis through a user activity description mining layer to obtain key content distribution of a first service session state; translating the key content distribution of the first service session state through a user activity description translation layer to obtain translation key content distribution; and splicing the key content distribution and the translation key content distribution of the first service session state through the user activity description optimization layer to obtain the user activity description content of the digital service interaction data to be subjected to user interest analysis.

The embodiment is applied, the digital service interaction data to be subjected to the user interest analysis is processed through the user activity description mining layer, the user activity description translation layer and the user activity description optimization layer of the user activity description recognition network, and the user activity description content of the digital service interaction data to be subjected to the user interest analysis can be accurately and completely obtained.

Under some independently designed ideas, mining the digital service interaction data to be subjected to the user interest analysis through a user activity description mining layer to obtain the key content distribution of the first service session state, including: and carrying out a plurality of stages of user activity description mining on the digital service interaction data to be subjected to user interest analysis through a user activity description mining layer to obtain the key content distribution of each stage of matched first service session state, wherein the content diversity evaluation in the key content distribution of each stage of matched first service session state is different, and the key content distribution of the first service session state matched by the key content mining layer of the last stage is loaded to a user activity description translation layer. The method for splicing the key content distribution and the translation key content distribution of the first service session state through the user activity description optimization layer to obtain the user activity description content of the digital service interaction data to be subjected to user interest analysis comprises the following steps: and through the implementation of a user activity description optimization layer, fusing the key content distribution of the matched first service session state of each stage to obtain key content distribution of a second service session state, fusing the key content distribution of the second service session state and the key content distribution of translation to obtain key content distribution of a third service session state, and obtaining user activity description content of the digital service interaction data to be subjected to user interest analysis by combining the key content distribution of the third service session state.

The embodiment is applied, and the richer identification results can be mined in stages by performing the mining of the user activity description of a plurality of stages on the digital service interaction data to be subjected to the user interest analysis, so that the state content of the service session for obtaining the digital service interaction data to be subjected to the user interest analysis can be more accurate.

Under some independently designed ideas, the user activity description recognition network and the user interest analysis network are independently configured.

In some independently designed concepts, before determining, by the user activity description recognition network, the user activity description content of the digitized service interaction data to be subjected to the user interest analysis, the method further includes: obtaining a user activity description identification network through a first authentication example queue configuration, wherein the digital service interaction data in the first authentication example queue is annotated with user activity description content; and determining authenticated user activity description content of the digital service interaction data in the second authentication example queue through the configured user activity description recognition network, and configuring the user interest analysis network through the second authentication example queue and the authenticated user activity description content.

The method and the device are applied to the embodiment, and the user activity description identification network can be configured only through the authenticated template of the user activity description by performing targeted configuration on the user activity description identification network, so that the mining quality of the user interest mining result is improved, and the interference on the mining quality of the user interest mining result caused by the lack of the authenticated template of the user interest mining result is improved.

Under some independently designed concepts, the second authentication instance queue includes a first local authentication instance queue and a second local authentication instance queue, and the configuring the user interest analysis network through the second authentication instance queue and the authenticated user activity description content includes: configuring the user interest analysis network through the first local authentication example queue and the authenticated user activity description content matched with the first local authentication example queue so as to change variables of a non-exclusive key content mining layer and a project attention heat translation layer in the user interest analysis network; and configuring the user interest analysis network through the second local authentication example queue and the authenticated user activity description content matched with the second local authentication example queue so as to change the variables of a non-exclusive key content mining layer and a service demand scoring translation layer in the user interest analysis network.

When the method is applied to the embodiment, the non-independent key content mining layer, the project attention heat translation layer, the non-independent key content mining layer and the service demand scoring translation layer are respectively configured, so that the project attention heat distribution and the service demand scoring distribution with higher accuracy and reliability can be obtained when the user interest analysis network carries out user interest mining on the digital service interaction data to be subjected to user interest analysis.

A topic processing server comprising: a memory for storing an executable computer program, a processor for implementing the above method when executing the executable computer program stored in the memory.

A computer-readable storage medium, on which a computer program is stored which, when executed, performs the above-described method.

According to one embodiment of the disclosure, by determining the user activity description content of the digital service interaction data to be subjected to user interest analysis, the user interest analysis network is ensured to fully mine the session state of the service session in the digital service interaction data to be subjected to user interest analysis through the user activity description content, the user interest mining result obtained by mining the user interest analysis network is ensured to be better aligned with the service session of the digital service interaction data to be subjected to user interest analysis, and the user interest mining precision and the reliability are ensured to a certain extent; meanwhile, the user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained through a user activity description identification network different from the user interest analysis network, and the high-quality user activity description content can be obtained through processing of different networks, so that the correlation between the user interest mining result and the service session of the digital service interaction data to be subjected to the user interest analysis is improved.

Drawings

FIG. 1 is a schematic diagram illustrating one communication configuration of a large data processing system in which embodiments of the present disclosure may be implemented.

FIG. 2 is a flow diagram illustrating a big data processing method in conjunction with user interest analysis, in which embodiments of the present disclosure may be implemented.

FIG. 3 is an architectural diagram illustrating an application environment for a big data processing method in conjunction with user interest analysis in which embodiments of the present disclosure may be implemented.

Detailed Description

The technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the drawings in the embodiments of the present disclosure. For the purpose of making the purpose, technical solutions and advantages of the present disclosure clearer, the present disclosure will be described in further detail with reference to the accompanying drawings, the described embodiments should not be construed as limiting the present disclosure, and all other embodiments obtained by a person of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present disclosure. In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. The terminology used herein is for the purpose of describing embodiments of the disclosure only and is not intended to be limiting of the disclosure.

FIG. 1 is a block diagram illustrating one communication configuration of a big data processing system 100 in which embodiments of the present disclosure may be implemented, the big data processing system 100 comprising a memory 101 for storing executable computer programs, a processor 102 for implementing the big data processing method in conjunction with user interest analysis in embodiments of the present disclosure when executing the executable computer programs stored in said memory 101.

Fig. 2 is a flowchart illustrating a big data processing method combined with user interest analysis, which may implement an embodiment of the present disclosure, and the big data processing method combined with user interest analysis may be implemented by the big data processing system 100 illustrated in fig. 1, and further may include technical solutions described in the following related steps.

Process 11: and determining the digital service interaction data to be subjected to the user interest analysis.

For the embodiment of the present disclosure, the digital service interaction data to be analyzed for the user interest may be understood as the basic digital service interaction data for user interest mining to determine the matching user interest mining result. The digital service interaction data to be analyzed for the user interest can be digital office service interaction data or VR/AR/MR service interaction data. The digital service interaction data can be an interaction log or an interaction record, and can be recorded in a form of videotext, voice and the like.

Process 12: and determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis through the user activity description recognition network.

For the embodiment of the present disclosure, the user activity description recognition network is an intelligent model deployed in combination with machine learning ML, and is used to determine a recognition result (feature information) of the digital service interaction data to be subjected to the user interest analysis, so as to obtain the user activity description content of the digital service interaction data to be subjected to the user interest analysis. The user activity description recognition network determines recognition results for the digital service interaction data to be analyzed for user interest, and a plurality of key content distributions (feature matrices or graphs) can be obtained. The user activity description content may be user activity description content (how large and small description features) of each service interaction behavior event in the digital service interaction data to be subjected to the user interest analysis, and session state (such as related service scene data) in the underlying digital service interaction data may be obtained through the user activity description content, and may be state data of a service session in the digital service interaction data to be subjected to the user interest analysis.

For some possible examples, the user activity description recognition network may be an LSTM (long short term memory network), which may splice multi-dimensional (different activity description numbers, different digital service interaction data feature recognizability) key content distribution, so as to obtain a user interest mining result with greater feature recognition, more comprehensive content, and more credible user interest differentiation in the digital service interaction data.

Process 13: and mining the user interest of the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network by combining the user activity description content, and obtaining a user interest mining result of the digital service interaction data to be subjected to the user interest analysis.

For the embodiment of the present disclosure, after obtaining the user activity description content of the digital service interaction data to be subjected to the user interest analysis, the user interest analysis network can perform user interest mining on the basic digital service interaction data through the user activity description content. For example, the user interest analysis network may perform user interest mining on the digital service interaction data to be subjected to the user interest analysis in combination with the user activity description content of each service interaction behavior event in the user activity description content and the state data of the service session included in the user activity description content to obtain a user interest mining result, which may be, for example, a project attention popularity mining result and a service demand scoring mining result. For some possible examples, the user interest analysis network may also be an LSTM.

The method is applied to the Process11-Process13, and ensures that the user interest analysis network can fully mine the session state of the business session in the digital service interaction data to be subjected to the user interest analysis through the user activity description content by determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis, ensures that the user interest mining result obtained by mining the user interest analysis network can be better aligned with the business session of the digital service interaction data to be subjected to the user interest analysis, and ensures the user interest mining precision and credibility to a certain extent; meanwhile, the user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained through a user activity description identification network different from the user interest analysis network, and the user activity description content with high quality can be obtained through processing through different networks, so that the correlation between the user interest mining result and the service session of the digital service interaction data to be subjected to the user interest analysis is improved.

Under other independent design ideas, the big data processing method combined with the user interest analysis can be realized through the following contents.

Process 21: and determining the digital service interaction data to be subjected to the user interest analysis.

Process 22: and determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis through the user activity description recognition network.

For the embodiment of the present disclosure, the user activity description content is a user activity description distribution of user activity descriptions of different service interaction behavior events carrying digital service interaction data to be subjected to user interest analysis, for example, each service interaction behavior event in the digital service interaction data to be subjected to user interest analysis has a matched user activity description.

For the disclosed embodiments, the user activity description recognition network includes a user activity description mining layer, a user activity description translation layer, and a user activity description optimization layer. The user activity description mining layer can mine key contents of digital service interaction data to be subjected to user interest analysis, the user activity description translation layer can translate related activity descriptions and then generate key content distribution, and the user activity description optimization layer can further improve the generation result of the translation layer.

In other embodiments, which can be implemented independently, the user activity description content of the digitized service interaction data to be analyzed by the user activity description recognition network can cover the following Process221-Process 223.

The Process 221: and mining the digital service interaction data to be subjected to user interest analysis through a user activity description mining layer to obtain the key content distribution of the first service session state.

For the embodiment of the disclosure, the digital service interaction data to be subjected to the user interest analysis may be mined by the user activity description mining layer of the user activity description recognition network, so as to determine a recognition result in the digital service interaction data to be subjected to the user interest analysis. The identification result obtained by mining the digital service interaction data to be subjected to the user interest analysis by the user activity description mining layer may be state key content of a service session in the digital service interaction data to be subjected to the user interest analysis, where the state key content includes, for example, state topic information and interest distinguishing information. Further, the user activity description mining layer may determine a first business session state key content distribution, which may also be, for example, a state key content distribution for a business session in the digital service interaction data to be subjected to the user interest analysis.

For the embodiment of the disclosure, on the basis that the user activity description mining layer has a multi-level subnet, the user activity description mining layer may perform a plurality of levels of user activity description mining (for example, performing key content mining) on the digital service interaction data to be subjected to the user interest analysis, and the key content distribution obtained by each level of key content mining layer is the first service session state key content distribution. Illustratively, when the key content mining layer comprises four key content mining units for mining, the first-order key content mining units can mine the digital service interaction data to be subjected to user interest analysis, and then generate the first service session state key content distribution. And the second-order key content mining unit continues mining by taking the first service session state key content distribution derived by the first-order key content mining unit as a reference, and then generates the matched first service session state key content distribution. Meanwhile, differences exist in content diversity evaluation in the first service session state key content distribution derived from each stage of the user activity description mining layer. The content diversity evaluation may cover the feature recognition degree of the first service session state key content distribution and the dimension of the recognition result. In this case, the first service session state key content distribution of the last order match would be loaded to the user activity description translation layer. By carrying out a plurality of stages of user activity description mining on the digital service interaction data to be subjected to the user interest analysis, richer identification results can be mined in stages, so that the state content of the service session of the digital service interaction data to be subjected to the user interest analysis can be more accurate.

The Process 222: and translating the key content distribution of the first service session state through the user activity description translation layer to obtain translation key content distribution.

For the embodiment of the disclosure, after the digital service interaction data to be subjected to the user interest analysis is mined by the user activity description mining layer and then the first business session state key content distribution is generated, the first business session state key content distribution is translated by the user activity description translation layer to obtain the translation key content distribution. When the user activity description translation layer translates (decodes) the first service session state key content distribution, the recognition result extracted by the user activity description mining layer may be translated, and the translation key content distribution with the specified size and the specified feature recognition degree is determined. Illustratively, the dimension of the recognition result in the translation key content distribution may be 32 dimensions, and the feature recognition degree is 0.25 of the digital service interaction data to be subjected to the user interest analysis.

For some possible examples, on the basis that the user activity description translation layer has a multi-level subnet, the user activity description translation layer still performs multi-level translation on the key content distribution of the first service session state, and the first-level translation layer performs translation on the key content distribution of the first service session state, so as to generate a matched pre-translation key content distribution. And the second-level translation layer translates the translation key content distribution derived by the first-level translation layer, and then generates a matched pre-translation key content distribution. The pre-translation key distribution derived at the last stage may be, for example, a translation key distribution.

The Process 223: and splicing the key content distribution and the translation key content distribution of the first service session state through the user activity description optimization layer to obtain the user activity description content of the digital service interaction data to be subjected to user interest analysis.

For some possible examples, after the translation is performed by the user activity description mining layer, in order to facilitate deep optimization of the recognition result derived by the user activity description mining layer and obtain more accurate service session state data of the digital service interaction data to be subjected to the user interest analysis, the first service session state key content distribution and the translation key content distribution may be spliced by the user activity description optimization layer to obtain the user activity description content of the digital service interaction data to be subjected to the user interest analysis. Illustratively, the recognition result in the first business session state key content distribution and the recognition result of the translation key content distribution can be spliced together to obtain the user activity description content of the digital service interaction data to be subjected to the user interest analysis. Illustratively, the recognition result in the first service session state key content distribution and the recognition result in the translation key content distribution are both 32-dimensional, and the user activity description content obtained after splicing may be 64-dimensional.

For some possible examples, the user activity description content is a user activity description distribution of user activity descriptions of different service interaction behavior events carrying digital service interaction data to be subjected to user interest analysis, for example, there is a matching user activity description for each service interaction behavior event in the digital service interaction data to be subjected to user interest analysis. For some possible examples, on the basis that the user activity description mining layer has a multi-level subnet, the user activity description optimization layer may merge the first service session state key content distribution matched at each level to obtain a second service session state key content distribution, and merge the second service session state key content distribution with the translation key content distribution to obtain a third service session state key content distribution.

For some possible examples, a first business session state key content distribution derived for a partial key content mining subnet of the user activity description mining layer may also be deployed for fusion. In addition, the key content distribution of the first business conversation state derived by each stage of key content mining layer can be processed through the user activity description optimization layer, so that the description layer number and the feature recognition degree of each second business conversation state key content distribution are consistent. After obtaining the distribution of the key content of the third service session state, the user activity description optimization layer may perform further translation in combination with the recognition result of the distribution of the key content of the third service session state to obtain the user activity description content of the digital service interaction data to be subjected to the user interest analysis, where the user activity description content may be the user activity description distribution.

For some possible examples, the user activity description identifying Network _ a includes: a user activity description mining layer Part _ A01, a user activity description translation layer Part _ A02, and a user activity description optimization layer Part _ A03. The user activity description mining layer Part _ a01 is determined by 1 base moving average unit a011 and four key content mining units a 012. The basic moving average unit a011 can perform basic description mining on the digital service interaction data to be subjected to the user interest analysis, and input the key content distribution to the key content mining unit a 012. The key content mining unit a012 reduces the feature recognition degree of key content distribution to the set proportion of the initial digital service interaction data on the basis of mining the description of larger dimension in order. Each key content mining unit a012 can output the first service session state key content distribution, and the first service session state key content distribution derived by the last key content mining unit a012 is loaded to the user activity description translation layer Part _ a 02. The user activity description translation layer Part _ a02 is determined by 1 moving average unit a021 and four translation units a022, and the four translation units a022 orderly translate the relevant activity descriptions and determine the translation key content distribution with dimension 32 and feature recognition degree of the digitized service interaction data 0.25 to be subjected to the user interest analysis. The user activity description optimization layer Part _ a03 includes four translation units a031 and four moving average units a 032. Further, the first traffic session key content distribution obtained by the key content mining unit a012 abstraction is merged using the bridging and translation unit a031, which may be obtained, for example, for the second traffic session state key content distribution. And then merging the key content distribution of the second service session state and the key content distribution of the translation to obtain the key content distribution of the third service session state. And then, the four moving average units a032 perform the sequential translation, so as to obtain the user activity description content of the digitized service interaction data to be subjected to the user interest analysis, such as the user activity description distribution.

It can be understood that, after the user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained, the user interest mining can be performed on the digital service interaction data to be subjected to the user interest analysis through the obtained user activity description content, so as to obtain a user interest mining result of the digital service interaction data to be subjected to the user interest analysis.

Under some design ideas which can be independently implemented, when a project attention heat mining result needs to be obtained, the step of 'performing user interest mining on digital service interaction data to be subjected to user interest analysis through a user interest analysis network by combining user activity description contents to obtain a user interest mining result of the digital service interaction data to be subjected to user interest analysis' can be realized through the following contents.

Process 23: and processing the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network to obtain service session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis.

For some possible examples, the user interest analysis network may be an LSTM. The user interest analysis network can perform key content mining operation on the digital service interaction data to be subjected to user interest analysis so as to obtain service session thermodynamic index data of the digital service interaction data to be subjected to user interest analysis. Illustratively, the service session thermodynamic indicator data can be understood as the activity condition of the service session in the digital service interaction data to be analyzed for user interest. Illustratively, the business session thermodynamic indicator data is a user activity description guidance distribution of user activity description guidance features carrying different service interaction behavior events of digitized service interaction data to be subjected to user interest analysis. The user activity description bootstrapping distribution can be used to construct liveness conditions for the business session. Furthermore, the user activity description guidance feature may also be understood as an adaptive description of the user activity description. The user activity description guided distribution may be understood as an adaptive description set of user activity descriptions.

For some possible examples, the user interest analysis network includes a non-exclusive key content mining layer, a project attention heat translation layer, and a service demand scoring translation layer. The digital service interaction data to be subjected to the user interest analysis is processed through the user interest analysis network, so as to obtain the service session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis, which can exemplarily include the following contents.

The Process 231: the method comprises the steps of performing key content mining on digital service interaction data to be subjected to user interest analysis through a non-independent key content mining layer to obtain service interaction key content distribution, and splicing the service interaction key content distribution and key content distribution of a first service session state derived by a user activity description mining layer of a user activity description recognition network to obtain first spliced key content distribution.

For some possible examples, the non-exclusive key content mining layer means that the recognition result determined by the key content mining layer can be used as the mining result for obtaining the item attention heat degree and the mining result for the service demand score at the same time. The first spliced key content distribution obtained after splicing may include the state key content of the service session in the digital service interaction data to be subjected to the user interest analysis and other identification results.

For some possible examples, the non-exclusive key content mining layer includes a cascade of X key content mining modules, each key content mining module including a user activity description optimization program (such as an intelligent description adjustment program/feature adjuster).

Under other design ideas which can be independently implemented, the service interaction key content distribution and the service session state key content distribution derived by the user activity description mining layer of the user activity description recognition network are spliced to obtain a first spliced key content distribution, which exemplarily can cover the following processes 2311 to 2313.

Process 2311: the service interaction key content distribution is loaded to a first key content mining module.

For example, other key content mining layers of the user interest analysis network may first perform key content mining on the digital service interaction data to be subjected to the user interest analysis to obtain a service interaction key content distribution. And then the service interaction key content is distributed and loaded to the first key content mining module. And performing subsequent processing on the service interaction key content distribution by using a key content mining module.

Process 2312: and each key content mining module splices the key content distribution and the business session state key content distribution derived by the upstream key content mining module through a user activity description optimization program to obtain a second spliced key content distribution matched with the key content mining module.

In the related steps, there are differences in the evaluation of description diversity in the distribution of the business session state key content matched by each key content mining module. Further, after the key content mining module determines the service interaction key content distribution, the key content distribution derived by the key content mining module at the upstream and the business session state key content distribution can be spliced through the user activity description optimization program, and a second spliced key content distribution matched with the key content mining module is obtained. Further, the business conversation state key content distribution matched by each key content mining module has different content diversity evaluation. The differences in content diversity evaluation can understand the differences in feature recognition degree and diversity of recognition results of key content distribution of the business session state.

For some possible examples, as far as the first key content mining module is concerned, it determines the service interaction key content distribution after key content mining by other moving average units. For the second key content mining module, the determined service interaction key content distribution is the second splicing key content distribution derived by the previous key content mining module.

For another example, when the user activity description mining layer of the user activity description recognition network has only one level, it indicates that the user activity description mining layer generates only one first service session state key content distribution, and in this case, all the key content mining modules may splice the first service session state key content distribution with the key content distribution derived by the previous key content mining module. When the user activity description mining layer of the user activity description recognition network has multiple levels, the key content distribution of the first service session state derived by the multi-level user activity description mining layer can be spliced with the key content distribution derived by the key content mining module respectively.

Illustratively, the key content distribution of the first business session state obtained by the user activity description mining layer of the first order is loaded to the first key content mining module, and the key content distribution of the first business session state obtained by the user activity description mining layer of the second order is loaded to the second key content mining module, so that the second key content mining module can splice the key content distribution derived by the previous key content mining module and the key content distribution of the first business session state obtained by the user activity description mining layer of the second order.

For some possible examples, the user activity description optimization program splices the key content distribution derived by the key content mining module at the upstream and the business session state key content distribution to obtain a second spliced key content distribution matched with the key content mining module, and specifically includes: the user activity description optimization program changes the business session state key content distribution into the business session state key content distribution with the specified scale, and can change the feature recognition degree and the diversity of recognition results of the business session state key content distribution. And the user activity description optimization program processes the changed business session state key content distribution and key content distribution derived by an upstream key content mining module to obtain second spliced key content distribution matched with the key content mining module.

For example, the user activity description optimization program of the second key content mining module may process the second splicing key content distribution derived by the first key content mining module and the business session state key content distribution input by the first key content mining module, so as to obtain a second splicing key content distribution matched with the second key content mining module.

In this way, the user activity description optimization program also processes the key content distribution of the service session state and the key content distribution derived by the upstream key content mining module, and realizes accurate and reliable splicing of the key content distribution of the service session state and the key content distribution derived by the upstream key content mining module.

For some possible examples, for each key content mining module, content reduction may be performed on the key content distribution derived by the key content mining module upstream before the key content distribution derived by the key content mining module upstream and the business session state key content distribution are spliced by the user activity description optimization program to obtain a second spliced key content distribution matched with the key content mining module. Illustratively, the second key content mining module performs a content compaction operation (pooling) on the derived second splicing key content distribution. Through content reduction operation, the distribution of the second splicing key content can be reduced, so that the distribution of the second splicing key content reaches the standard.

Process 2313: and combining the second splicing key content distribution of the last key content mining module to obtain the first splicing key content distribution.

For some possible examples, on the basis that the last stage of the non-exclusive key content mining layer of the user interest analysis network is not the last key content mining module, namely, after the key content mining module indicating that the non-exclusive key content mining layer is at the tail, a plurality of key content mining units are further arranged, and after the last key content mining module outputs the second splicing key content distribution, the key content mining units are used for mining again to process the spliced identification result. The key content distribution derived after the last stage processing of the non-exclusive key content mining layer may be, for example, the first-stitched key content distribution. For example, a content reduction operation may be performed on the second splicing key content distribution output by the last key content mining module, so as to further reduce the second splicing key content distribution, and then mining is continued through the key content mining unit to determine the identification result, where the derived key content distribution may be, for example, the first splicing key content distribution. For some possible examples, the second splicing key content may also be distributed directly as the first splicing key content distribution.

Therefore, the business conversation state key content distribution derived by the user activity description recognition network and the service interaction key content distribution obtained by mining the key content of the digital service interaction data to be subjected to the user interest analysis by the user interest analysis network are spliced by the user activity description optimization program, so that the business conversation state data of the business conversation in the digital service interaction data to be subjected to the user interest analysis in the business conversation state key content distribution can be used by the user interest analysis network, the recognition result obtained by the user activity description recognition network is shared to the user interest analysis network for use, and the mining quality of the user interest mining result can be guaranteed.

It can be understood that, after the first splicing key content distribution is obtained, user interest mining can be continuously performed on the digital service interaction data to be subjected to the user interest analysis through the first splicing key content distribution to obtain a user interest mining result.

Process 232: and translating the first splicing key content distribution through the project attention heat translation layer to obtain service session heat index data of the digital service interaction data to be subjected to user interest analysis.

In view of the fact that the first splicing key content distribution comprises the state key content of the business conversation and other recognition results of the digital service interaction data to be subjected to the user interest analysis, the first splicing key content distribution can be translated through the item attention heat translation layer to obtain business conversation thermal index data of the digital service interaction data to be subjected to the user interest analysis, and user activity description guidance distribution of user activity description guidance characteristics of each service interaction behavior event in the digital service interaction data to be subjected to the user interest analysis can be obtained.

For some possible examples, the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs may be translated by the project attention heat translation layer, so as to obtain business session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis. The project attention heat translation layer can simultaneously determine first splicing key content distribution derived from the last stage of a non-exclusive key content mining layer of the user interest analysis network and second splicing key content distribution of X user activity description optimization programs, and translates the two key content distributions to obtain business session heat index data of the digital service interaction data to be subjected to user interest analysis. When the last stage of the non-exclusive key content mining layer is the key content mining module, the translation can be performed by determining a first splicing key content distribution derived by the key content mining module at the end and a second splicing key content distribution derived by a user activity description optimization program of other key content mining modules.

For some possible examples, there are multiple key content mining modules of the non-exclusive key content mining layer, and the project attention heat translation layer may determine a second stitched key content distribution derived by the multiple key content mining modules at the same time for translation.

For example, the project attention heat translation layer determines second splicing key content distributions derived by three key content mining modules having upstream and downstream relationships, and then three moving average layers having upstream and downstream relationships may be deployed in the project attention heat translation layer to determine the second splicing key content distributions derived by the three key content mining modules, respectively, for translation.

For example, a first sliding average layer of the project attention heat translation layer may determine a first stitched key content distribution derived by the non-exclusive key content mining layer and a second stitched key content distribution derived by the first user activity description optimization program to translate, and then generate a key content distribution. The second moving average layer of the project attention heat translation layer can translate through the key content distribution derived by the previous moving average layer and the second spliced key content distribution derived by the second user activity description optimization program.

For some possible examples, after the moving average layer of the project attention heat translation layer is translated through the first splicing key content distribution and the second splicing key content distribution, the moving average layer is translated through a plurality of moving average layers to change the project attention heat distribution derived by the project attention heat translation layer. For example, the item attention heat distribution may be a result of summarizing attention heats of different items, and the attention heat may be a user attention heat or a third party merchant attention heat.

Therefore, by determining the service session thermodynamic index data of the digital service interaction data aiming at the user interest analysis to be performed, the user activity description guidance distribution of the user activity description guidance features of each service interaction behavior event can be obtained, the service session thermodynamic index data can be flexibly determined, and the accuracy of user interest mining of the user interest analysis network under the service session of the changeable thermodynamic index (the situation that the change of the thermodynamic index is complicated) is improved.

Process 24: and acquiring a project attention heat mining result of the digital service interaction data to be subjected to the user interest analysis by combining the service session heat index data and the user activity description content.

For some possible examples, after obtaining the service session thermodynamic index data of the service session in the digital service interaction data to be subjected to the user interest analysis, the user interest mining can be performed on the digital service interaction data to be subjected to the user interest analysis by combining the service session thermodynamic index data and the user activity description content derived by the user activity description recognition network, so as to obtain an item attention heat mining result of the digital service interaction data to be subjected to the user interest analysis. Illustratively, the project attention heat mining results of the digitized service interaction data to be analyzed for user interest can be obtained with a user activity description guidance distribution and a user activity description distribution.

For some possible examples, the user activity description guidance distribution and the user activity description distribution may be subjected to binary operation to obtain item attention heat mining results of the digital service interaction data to be subjected to the user interest analysis.

Therefore, the user activity description guidance distribution fully considers the state topic information and the interest distinguishing information in the service session state key content provided by the user activity description recognition network, ensures that the project attention heat mining result obtained by the user interest analysis network mining can avoid noise interference, improves the project distinguishing degree, and enables the service session of the service demand scoring mining result to be better aligned (correlation matching) with the service session of the digital service interaction data to be subjected to the user interest analysis.

For some possible examples, the user interest analysis network further includes a service demand scoring translation layer. Since the identification results extracted by the non-exclusive key content mining layer can be used to obtain service requirement scoring mining results. In view of this, after the non-exclusive key content mining layer performs key content mining on the digital service interaction data to be subjected to the user interest analysis, for example, after the Process231, the following may be continuously executed: and translating the first spliced key content distribution through a service requirement scoring translation layer to obtain a service requirement scoring mining result of the digital service interaction data to be subjected to the user interest analysis.

It can be understood that the final stage of the non-exclusive key content mining layer derives a first splicing key content distribution containing the key content of the service session state of the service session in the digital service interaction data to be analyzed for user interest. In view of this, the first splicing key content distribution may be translated through the service demand scoring translation layer, and a service demand scoring mining result of the digitized service interaction data to be subjected to the user interest analysis is obtained.

For some possible examples, the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs may be translated through the service demand scoring translation layer, so as to obtain a service demand scoring mining result of the digital service interaction data to be subjected to the user interest analysis. The service demand scoring translation layer can simultaneously determine first splicing key content distribution derived from the last stage of the non-exclusive key content mining layer of the user interest analysis network and second splicing key content distribution of the X user activity description optimization programs, and translates the two key content distributions to obtain a service demand scoring mining result of the digital service interaction data to be subjected to user interest analysis. When the last stage of the non-exclusive key content mining layer is the key content mining module, the translation can be performed by determining a first splicing key content distribution derived by the key content mining module at the end and a second splicing key content distribution derived by a user activity description optimization program of other key content mining modules.

For some possible examples, there are multiple key content mining modules of the non-exclusive key content mining layer, and the service demand scoring translation layer may determine a second stitched key content distribution derived by the multiple key content mining modules to translate at the same time. For example, the service demand scoring translation layer determines second splicing key content distribution derived by three cascaded key content mining modules, and then three cascaded moving average layers may be deployed in the service demand scoring translation layer to determine second splicing key content distribution derived by three key content mining modules, respectively, for translation.

For example, a first sliding average layer of the service requirement scoring translation layer may determine a first stitched key content distribution derived by the non-exclusive key content mining layer and a second stitched key content distribution derived by the first user activity description optimizer to translate, and then generate a key content distribution. The second moving average of the service requirement scoring translation layer may translate through a key content distribution derived by the previous moving average and a second stitched key content distribution derived by a second user activity description optimizer.

For some possible examples, after the moving average layer of the service demand scoring translation layer translates through the first splicing key content distribution and the second splicing key content distribution, the moving average layer translates through a plurality of moving average layers to change the service demand scoring distribution finally derived by the service demand scoring translation layer.

In view of this, the user interest mining is performed on the digital service interaction data to be subjected to the user interest analysis through the first splicing key content distribution including the service session state key content of the service session in the digital service interaction data to be subjected to the user interest analysis, and service requirement scores which are matched as much as possible are added to different items of the service session in the digital service interaction data to be subjected to the user interest analysis through the service session state key content, so that the user interest mining precision and the user interest mining reliability are guaranteed to a certain extent.

For some possible examples, the user interest analysis network LSTM _ C includes: a non-exclusive key content mining layer LSTM _ C1, a project attention heat translation layer LSTM _ C2, and a service demand scoring translation layer LSTM _ C3.

Illustratively, the non-exclusive key content mining layer LSTM _ C1 includes a moving average cell LSTM _ C11 and a plurality of key content mining modules LSTM _ C12. The key content mining module LSTM _ C12 includes a user activity description optimizer Software _ C121. The user activity description optimizer Software _ C121 may identify network local connections with the user activity description. The item attention heat translation layer LSTM _ C2 includes a plurality of moving average cells LSTM _ C21. The service demand score translation layer LSTM _ C3 includes a plurality of moving average units LSTM _ C31. Wherein the user activity description optimizer Software _ C121 bridges the partial moving average unit LSTM _ C21 of the project attention heat translation layer LSTM _ C2 and the partial moving average unit LSTM _ C31 of the service demand scoring translation layer LSTM _ C3, respectively.

For some possible examples, the user interest analysis network LSTM _ C may process the digital service interaction data to be subjected to the user interest analysis, and obtain the service session thermodynamic indicator data of the digital service interaction data to be subjected to the user interest analysis. The user interest analysis network LSTM _ C may also identify user activity description content derived by the network in combination with the service session thermodynamic indicator data and the user activity description, to obtain an item attention heat mining result of the digitized service interaction data to be subjected to user interest analysis. Meanwhile, the user interest analysis network LSTM _ C may also determine the service demand score mining result.

For some possible examples, the non-independent key content mining layer LSTM _ C1 may exemplarily perform key content mining on digital service interaction data to be subjected to user interest analysis to obtain a service interaction key content distribution, and concatenate the service interaction key content distribution and a first business session state key content distribution derived by the user activity description mining layer of the user activity description recognition network to output a first concatenated key content distribution.

Illustratively, the moving average unit LSTM _ C111 preceding the key content mining module LSTM _ C12 may perform key content mining on the digitized service interaction data to be subjected to the user interest analysis to obtain the service interaction key content distribution described above. The key content mining module LSTM _ C12 may splice the key content distribution derived by the upstream key content mining module and the business session state key content distribution derived by the key content mining layer of the user activity description recognition network through the user activity description optimization program Software _ C121 to obtain a second spliced key content distribution matching the key content mining module. The moving average unit LSTM _ C112, which is located after the key content mining module LSTM _ C12, may mine the second splicing key content distribution derived by the last key content mining module to determine the first splicing key content distribution. The key content mining module LSTM _ C12 may also include a pooling moving average unit LSTM _ C012 for content compaction operations by key content distributions derived by upstream key content mining modules. The project attention heat translation layer LSTM _ C2 may cover 5 moving average cells. Further, the last-order moving average unit LSTM _ C21 outputs traffic session thermal index data of the digitized service interaction data to be analyzed for user interest, which may be a guided profile for user activity description. The service demand scoring translation layer LSTM _ C3 includes 5 moving average units LSTM _ C31, the moving average unit LSTM _ C31 performs ordered translation on the first concatenation state key content distribution derived by the non-exclusive key content mining layer, and the last-order moving average unit LSTM _ C31 determines the service demand scoring mining result.

In some design concepts that can be implemented independently, the user activity description recognition network and the user interest analysis network may be configured before the user activity description content of the digital service interaction data to be analyzed for user interest is determined by the user activity description recognition network.

Illustratively, since the user activity description recognition network comprises a user activity description mining layer, a user activity description translation layer and a user activity description optimization layer which are not mutually influenced, the personalized configuration of the user activity description recognition network and the personalized configuration of the user interest analysis network can be realized, and information intersection and confusion caused by joint configuration are avoided.

In this regard, for some possible examples, the user activity description recognition network and the user interest analysis network are configured independently. The user activity description recognition network and the user interest analysis network may be configured, for example, in configuring the user activity description recognition network and the user interest analysis network, both pointedly and asynchronously.

It can be understood that the user activity description mining layer, the user activity description translation layer and the user activity description optimizing layer which are not mutually influenced are deployed for the user activity description recognition network, so that the user activity description recognition network can be pertinently configured, and the user activity description recognition network can be configured only through the authenticated user activity description template, so that the mining quality of the user interest mining result is improved, and the interference on the mining quality of the user interest mining result caused by the lack of the authenticated user interest mining result template is improved.

For some possible examples, in configuring the user activity description recognition network, the user activity description recognition network may be obtained through a first authentication example queue configuration, wherein the digitized service interaction data in the first authentication example queue is annotated with user activity description content. The user activity description content may be a matching user activity description for each service interaction behavior event in the digital service interaction data. The first authentication instance queue may be a sample data set.

For some possible examples, after configuring the user activity description recognition network, the authenticated user activity description content of the digital service interaction data in the second authentication example queue may be determined by the configured user activity description recognition network, and the user interest analysis network may be configured through the second authentication example queue and the authenticated user activity description content. The digitized service interaction data of the second authentication instance queue may be annotated with a project attention heat distribution label (real label) and a service demand score distribution label. The second authentication instance queue may be a different format of sample than the first authentication instance queue.

For some possible examples, the second authentication instance queue includes a first local authentication instance queue and a second local authentication instance queue. The digitized service interaction data of the first local authentication example queue may be annotated with a project attention heat distribution tag and the digitized service interaction data of the second local authentication example queue may be annotated with a service requirement score distribution tag.

In configuring the user interest analysis network with the second authentication instance queue and the authenticated user activity description, the following may be implemented as an example: configuring the user interest analysis network through the first local authentication example queue and the authenticated user activity description content matched with the first local authentication example queue so as to change variables of a non-exclusive key content mining layer and a project attention heat translation layer in the user interest analysis network; and configuring the user interest analysis network through the second local authentication example queue and the authenticated user activity description content matched with the second local authentication example queue so as to change the variables of a non-exclusive key content mining layer and a service demand scoring translation layer in the user interest analysis network.

Illustratively, the user activity description content matched by the first partial authentication example queue is obtained through a configured user activity description recognition network. By annotating the first local authentication instance queue with project attention heat distribution tags, configuration of a non-exclusive key content mining layer and a project attention heat translation layer in a user interest analysis network can be achieved. The user activity description content matched by the second local authentication example queue is obtained through the configured user activity description recognition network.

It can be understood that after configuring the non-exclusive key content mining layer and the project attention heat translation layer in the user interest analysis network through the first local authentication example queue annotated with the project attention heat distribution tag, the non-exclusive key content mining layer and the service requirement scoring translation layer in the user interest analysis network can be further configured. The configuration of the non-exclusive key content mining layer and the service demand scoring translation layer in the network may be analyzed for user interest, illustratively by annotating a second local authentication example queue with service demand scoring distribution tags. In the configuration process of the user activity description recognition network and the user interest analysis network, the configuration quality can be judged according to the related network quality cost (model loss), and the corresponding network variable is changed according to the quantized value of the model loss, so that the configuration is completed. Therefore, by respectively configuring the non-exclusive key content mining layer, the project attention heat translation layer, the non-exclusive key content mining layer and the service demand scoring translation layer, when the user interest analysis network carries out user interest mining on digital service interaction data to be subjected to user interest analysis, project attention heat distribution and service demand scoring distribution with higher accuracy and reliability can be obtained.

It can be understood that by determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis, the user interest analysis network is ensured to fully mine the session state of the service session in the digital service interaction data to be subjected to the user interest analysis through the user activity description content, the user interest mining result obtained by mining the user interest analysis network is ensured to be better aligned with the service session of the digital service interaction data to be subjected to the user interest analysis, and the user interest mining precision and the reliability are ensured to a certain extent; meanwhile, the user activity description content of the digital service interaction data to be subjected to the user interest analysis is obtained through a user activity description identification network different from the user interest analysis network, and the user activity description content with high quality can be obtained through processing through different networks, so that the correlation between the user interest mining result and the service session of the digital service interaction data to be subjected to the user interest analysis is improved.

Under some design ideas which can be independently implemented, in the user interest mining result of obtaining the digital service interaction data to be subjected to the user interest analysis, the method can further comprise the following steps: customizing a pushing scheme according to the user interest mining result; and carrying out differentiated information recommendation based on the push scheme.

For some possible examples, different and personalized push schemes can be determined according to the user interest mining result, then flexible information recommendation is performed according to the push schemes, the information recommendation is guaranteed to meet the tastes and preferences of different users as far as possible, the quality of the information recommendation is improved on the side, and resource waste caused by a large amount of information push is avoided.

Under some design ideas which can be independently implemented, a pushing scheme is customized according to the user interest mining result, and the method can be implemented by the following implementation modes: determining preference subject feature distribution and demand scene feature distribution in the user interest mining result; associating the preference theme feature distribution and the demand scene feature distribution in the user interest mining result by combining a feature distribution correlation coefficient between the preference theme feature distribution and the demand scene feature distribution in the user interest mining result to obtain a feature distribution association result; determining the demand scene feature distribution with abnormal association as candidate demand scene feature distribution, and determining push tendency prediction information matched with the candidate demand scene feature distribution through feature distribution commonality scores between the demand scene feature distribution and the candidate demand scene feature distribution in the feature distribution association result; associating the pushing tendency prediction information matched with the candidate demand scene feature distribution to obtain a pushing tendency association result; determining a core requirement in the user interest mining result and pushing tendency prediction information corresponding to the core requirement through the pushing tendency correlation result and the feature distribution correlation result; and determining a pushing scheme according to the core requirements and the pushing tendency prediction information. Therefore, the core requirements can be mined by sinking to the angle of the preferred theme and the angle of the requirement scene, so that the core requirements and the corresponding pushing tendency prediction information are accurately determined, and the determined pushing scheme can be matched with the key requirements of the user as much as possible.

Under some independently implementable design ideas, the determining of the preference topic feature distribution and the demand scene feature distribution in the user interest mining result comprises: determining at least two preference subject feature vectors and at least two demand scene feature vectors in the user interest mining result; determining preference subject feature vector commonality scores and preference subject feature vector differences between the at least two preference subject feature vectors, and determining demand scene feature vector commonality scores and demand scene feature vector differences between the at least two demand scene feature vectors; sorting the at least two preference theme feature vectors through the preference theme feature vector commonality score and the preference theme feature vector difference to obtain preference theme feature distribution in the user interest mining result; one preferred topic feature distribution comprises X preferred topic feature vectors; sorting the at least two demand scene feature vectors through the demand scene feature vector commonality score and the demand scene feature vector difference to obtain demand scene feature distribution in the user interest mining result; a demand scene feature distribution comprising X demand scene feature vectors; x is a positive integer. In this way, the integrity of the preference topic feature distribution and the demand scene feature distribution can be guaranteed.

Fig. 3 is an architecture diagram illustrating an application environment of a big data processing method combined with user interest analysis, in which a big data processing system 100 and a service client 200 communicating with each other may be included, in which an embodiment of the present disclosure may be implemented. Based on this, the big data processing system 100 and the service user side 200 implement or partially implement the big data processing method combining the user interest analysis of the embodiment of the present disclosure at runtime.

The above description is only for the preferred embodiment of the present disclosure, and is not intended to limit the scope of the present disclosure.

Claims

1. A big data processing method in conjunction with user interest analysis, wherein the method is performed by a big data processing system, the method comprising:

determining digital service interaction data to be subjected to user interest analysis; determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis through a user activity description identification network;

and mining the user interest of the digital service interaction data to be subjected to the user interest analysis through a user interest analysis network by combining the user activity description content to obtain a user interest mining result of the digital service interaction data to be subjected to the user interest analysis.

2. The method of claim 1, wherein the user interest mining results comprise item attention popularity mining results; the mining of the user interest of the digital service interaction data to be subjected to the user interest analysis through a user interest analysis network in combination with the user activity description content to obtain the user interest mining result of the digital service interaction data to be subjected to the user interest analysis comprises the following steps:

processing the digital service interaction data to be subjected to the user interest analysis through a user interest analysis network to obtain service session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis;

and acquiring the item attention heat mining result of the digital service interaction data to be subjected to the user interest analysis by combining the service session heat index data and the user activity description content.

3. The method according to claim 2, wherein the service session thermodynamic indicator data is a user activity description guidance distribution of user activity description guidance characteristics of different service interaction behavior events carrying the digitized service interaction data to be subjected to the user interest analysis, and the user activity description content is a user activity description distribution of user activity description of different service interaction behavior events carrying the digitized service interaction data to be subjected to the user interest analysis;

the obtaining of the item attention heat mining result of the digital service interaction data to be subjected to the user interest analysis by combining the service session thermodynamic index data and the user activity description content comprises the following steps: and carrying out binary operation on the user activity description guide distribution and the user activity description distribution to obtain a project attention heat mining result of the digital service interaction data to be subjected to user interest analysis.

4. The method of claim 2 or 3, wherein the user interest analysis network comprises a non-exclusive key content mining layer and a project attention heat translation layer; the processing the digital service interaction data to be subjected to the user interest analysis through the user interest analysis network to obtain the service session thermodynamic index data of the digital service interaction data to be subjected to the user interest analysis includes:

performing key content mining on the digital service interaction data to be subjected to user interest analysis through the non-exclusive key content mining layer to obtain service interaction key content distribution, and splicing the service interaction key content distribution and key content distribution of a first service session state derived by a user activity description mining layer of the user activity description recognition network to obtain first spliced key content distribution;

and translating the first splicing key content distribution through the project attention heat translation layer to obtain service session heat index data of the digital service interaction data to be subjected to the user interest analysis.

5. The method of claim 4, wherein the non-exclusive key content mining layer comprises a concatenation of X key content mining modules, each of the key content mining modules comprising a user activity description optimizer, X being a positive integer; the splicing the service interaction key content distribution and the first service session state key content distribution derived by the user activity description mining layer of the user activity description recognition network to obtain a first spliced key content distribution includes:

loading the service interaction key content distribution to a first one of the key content mining modules;

for each key content mining module, splicing the key content distribution derived by the key content mining module at the upstream and the key content distribution of the first service session state through the user activity description optimization program to obtain a second spliced key content distribution matched with the key content mining module; the content diversity evaluation in the service session state key content distribution matched by each key content mining module is different;

and combining the second splicing key content distribution of the key content mining module at the tail end to obtain the first splicing key content distribution.

6. The method of claim 5, wherein before said splicing the key content distribution derived by the key content mining module upstream and the business session state key content distribution by the user activity description optimization program to obtain a second spliced key content distribution matched by the key content mining module, the method further comprises: performing content reduction operation on the key content distribution derived by the key content mining module at the upstream;

and/or, the obtaining a second spliced key content distribution matched with the key content mining module by splicing the key content distribution derived by the key content mining module at the upstream and the key content distribution of the business session state through the user activity description optimization program includes: and changing the business session state key content distribution into business session state key content distribution of a specified scale through the user activity description optimization program, and processing the changed business session state key content distribution and key content distribution derived by the key content mining module at the upstream to obtain second spliced key content distribution matched with the key content mining module.

7. The method of claim 6, wherein the translating, by the item attention heat translation layer, the first splicing key content distribution to obtain business session thermodynamic indicator data of the digitized service interaction data to be subjected to the user interest analysis comprises:

and translating the first splicing key content distribution and the second splicing key content distribution of the X user activity description optimization programs through the project attention heat translation layer to obtain service session thermodynamic index data of the digital service interaction data to be subjected to user interest analysis.

8. The method of claim 7, wherein the user interest analysis network further comprises a service demand scoring translation layer; the mining of the user interest of the digital service interaction data to be subjected to the user interest analysis is performed through a user interest analysis network in combination with the user activity description content to obtain a user interest mining result of the digital service interaction data to be subjected to the user interest analysis, and the mining method further comprises the following steps:

translating the first splicing key content distribution through the service demand scoring translation layer to obtain a service demand scoring mining result of the digital service interaction data to be subjected to user interest analysis;

the step of translating the first splicing key content distribution through the service demand scoring translation layer to obtain a service demand scoring mining result of the digital service interaction data to be subjected to the user interest analysis includes: translating the first splicing key content distribution and a second splicing key content distribution of the X user activity description optimization programs through the service demand scoring translation layer to obtain a service demand scoring mining result of the digital service interaction data to be subjected to user interest analysis;

the user activity description recognition network comprises a user activity description mining layer, a user activity description translation layer and a user activity description optimization layer; the determining the user activity description content of the digital service interaction data to be subjected to the user interest analysis through the user activity description recognition network comprises the following steps: mining the digital service interaction data to be subjected to user interest analysis through the user activity description mining layer to obtain key content distribution of a first service session state; translating the key content distribution of the first service session state through the user activity description translation layer to obtain translation key content distribution; and splicing the key content distribution of the first service session state and the translation key content distribution through the user activity description optimization layer to obtain the user activity description content of the digital service interaction data to be subjected to user interest analysis.

9. The method of claim 1, wherein the user activity description recognition network and the user interest analysis network are independently configured;

wherein, before the determining, by the user activity description recognition network, the user activity description content of the digital service interaction data to be subjected to the user interest analysis, the method further comprises: obtaining the user activity description identification network through a first authentication example queue configuration, wherein the digital service interaction data in the first authentication example queue is annotated with user activity description content; determining authenticated user activity description content of the digital service interaction data in a second authentication example queue through the configured user activity description recognition network, and configuring the user interest analysis network through the second authentication example queue and the authenticated user activity description content;

wherein the second authentication instance queue comprises a first local authentication instance queue and a second local authentication instance queue, and the configuring the user interest analysis network through the second authentication instance queue and the authenticated user activity description content comprises: configuring the user interest analysis network through the first local authentication example queue and the authenticated user activity description content matched with the first local authentication example queue so as to change variables of a non-exclusive key content mining layer and a project attention heat translation layer in the user interest analysis network; configuring the user interest analysis network through the second local authentication example queue and the authenticated user activity description content matched with the second local authentication example queue to change variables of a non-exclusive key content mining layer and a service demand scoring translation layer in the user interest analysis network.

10. A topic processing server, comprising:

a memory for storing an executable computer program, a processor for implementing the method of any one of claims 1-9 when executing the executable computer program stored in the memory.