CN112784157A - Training method of behavior prediction model, behavior prediction method, device and equipment - Google Patents

Training method of behavior prediction model, behavior prediction method, device and equipment Download PDF

Info

Publication number
CN112784157A
CN112784157A CN202110076021.1A CN202110076021A CN112784157A CN 112784157 A CN112784157 A CN 112784157A CN 202110076021 A CN202110076021 A CN 202110076021A CN 112784157 A CN112784157 A CN 112784157A
Authority
CN
China
Prior art keywords
data
behavior
user
group
behavior prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110076021.1A
Other languages
Chinese (zh)
Inventor
刘宸
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Netease Media Technology Beijing Co Ltd
Original Assignee
Netease Media Technology Beijing Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Netease Media Technology Beijing Co Ltd filed Critical Netease Media Technology Beijing Co Ltd
Priority to CN202110076021.1A priority Critical patent/CN112784157A/en
Publication of CN112784157A publication Critical patent/CN112784157A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Quality & Reliability (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The embodiment of the disclosure provides a training method of a behavior prediction model, a behavior prediction method, a behavior prediction device, electronic equipment and a computer-readable storage medium, and relates to the technical field of computers. The training method of the behavior prediction model comprises the following steps: acquiring behavior data of a plurality of user groups on target information, and generating a tag group of the target information according to the behavior data; inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information; determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group; and adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model. According to the technical scheme of the embodiment of the disclosure, the accuracy of the model can be improved.

Description

Training method of behavior prediction model, behavior prediction method, device and equipment
Technical Field
Embodiments of the present disclosure relate to the field of computer technologies, and in particular, to a training method and apparatus for a behavior prediction model, a behavior prediction method and apparatus, an electronic device, and a computer-readable storage medium.
Background
This section is intended to provide a background or context to the embodiments of the disclosure recited in the claims. The description herein is not admitted to be prior art by inclusion in this section.
With the development of big data, it is important to know the reading preference of the user to determine whether the recommendation result is accurate. Therefore, it is necessary to accurately determine the tendencies of all users with respect to the user behavior of certain target information.
In the related art, when analyzing a user group of user behaviors with respect to target information, calculation is generally performed by a heuristic algorithm and a supervised learning algorithm. The heuristic algorithm is characterized in that a large amount of data is read and researched artificially, rules are summarized from the large amount of data, and a corresponding rule engine is designed to complete analysis of user groups. And the algorithm based on supervised learning learns the mapping relation between the target information and the user group through a large amount of manually marked data.
Disclosure of Invention
However, in the related art, the summarized rule engine is difficult to cover all situations, has certain limitations and is difficult to reuse, and the accuracy of the mapping relation learned through the manually labeled data is low.
For this reason, an improved training method of a behavior prediction model is needed, so as to accurately predict the user behavior of each user group according to the trained behavior prediction model.
In this context, embodiments of the present disclosure desirably provide a training method of a behavior prediction model, a training apparatus of a behavior prediction model, a behavior prediction method, a behavior prediction apparatus, an electronic device, and a computer-readable storage medium.
According to an aspect of the present disclosure, there is provided a training method of a behavior prediction model, including: acquiring behavior data of a plurality of user groups on target information, and generating a tag group of the target information according to the behavior data, wherein the tag group comprises tags corresponding to the user groups; inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information, wherein the user behavior prediction data set comprises user behavior prediction data corresponding to each user group; determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group; and adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model.
In an exemplary embodiment of the disclosure, the generating the tag group of the target information according to the behavior data includes: normalizing the behavior data of the target information by the user groups to obtain normalized behavior data; and generating a label group of the target information according to the normalized behavior data.
In an exemplary embodiment of the present disclosure, the normalizing the behavior data of the target information by the plurality of user groups to obtain normalized behavior data includes: determining a global behavior data maximum value and a global behavior data minimum value in the behavior data of each user group to each sample information; and normalizing the behavior data of the target information by each user group according to the maximum value and the minimum value of the global behavior data to obtain the normalized behavior data.
In an exemplary embodiment of the present disclosure, after normalizing the behavior data of the target information by each user group according to the global behavior data maximum value and the global behavior data minimum value to obtain the normalized behavior data, the method further includes: determining a local behavior data maximum value and a local behavior data minimum value in the behavior data of each user group to the target information; and carrying out normalization processing on the normalized behavior data again according to the local behavior data maximum value and the local behavior data minimum value so as to update the normalized behavior data.
In an exemplary embodiment of the present disclosure, the determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group, and a data distribution state in the user behavior prediction data group includes: determining expected loss items according to the tag groups and the user behavior prediction data groups; determining a regular term according to the data distribution states of a plurality of user groups in the user behavior prediction data group; and logically combining the expected loss term and the regular term to determine the loss function.
In an exemplary embodiment of the present disclosure, the regularization term includes a first regularization term and a second regularization term; the determining a regularization term according to the data distribution states of a plurality of user groups in the user behavior prediction data group includes: determining a first regular term according to the distribution state of the prediction data of the adjacent user groups in the plurality of user groups; determining a second regularization term according to the prediction data of a target user group in the plurality of user groups; the target user group is the user group corresponding to the label with the maximum numerical value in the label group.
In an exemplary embodiment of the present disclosure, the determining a first regularization term according to a distribution state of prediction data of an adjacent user group of the plurality of user groups includes: and calculating a difference value between the prediction data of the adjacent user groups, and determining the first regular term according to the difference value.
According to an aspect of the present disclosure, there is provided a behavior prediction method including: acquiring information to be processed and acquiring a plurality of characteristic information of the information to be processed; inputting the characteristic information into a behavior prediction model, predicting user behaviors of a plurality of user groups aiming at the information to be processed, and obtaining prediction data corresponding to the user groups; the behavior prediction model is obtained by training according to any one of the above training methods of the behavior prediction model.
According to an aspect of the present disclosure, there is provided a training apparatus of a behavior prediction model, including: the tag generation module is used for acquiring behavior data of a plurality of user groups on target information and generating a tag group of the target information according to the behavior data, wherein the tag group comprises tags corresponding to the user groups; the prediction data determining module is used for inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information, wherein the user behavior prediction data set comprises user behavior prediction data corresponding to each user group; the loss function determining module is used for determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group; and the model training module is used for adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model.
In an exemplary embodiment of the present disclosure, the tag generation module includes: the normalization module is used for normalizing the behavior data of the target information by the user groups to obtain normalized behavior data; and the generation control module is used for generating the label group of the target information according to the normalized behavior data.
In an exemplary embodiment of the disclosure, the normalization module includes: the global data determining module is used for determining the maximum value and the minimum value of the global behavior data in the behavior data of each user group to each sample message; and the normalization control module is used for normalizing the behavior data of the target information by each user group according to the global behavior data maximum value and the global behavior data minimum value to obtain the normalized behavior data.
In an exemplary embodiment of the present disclosure, after normalizing the behavior data of the target information by each user group according to the global behavior data maximum value and the global behavior data minimum value to obtain the normalized behavior data, the apparatus further includes: the local data determining module is used for determining a local behavior data maximum value and a local behavior data minimum value in the behavior data of each user group to the target information; and the normalization updating module is used for carrying out normalization processing on the normalization behavior data again according to the local behavior data maximum value and the local behavior data minimum value so as to update the normalization behavior data.
In an exemplary embodiment of the present disclosure, the loss function determination module includes: an expected loss item determining module, configured to determine an expected loss item according to the tag group and the user behavior prediction data group; the regular term determining module is used for determining a regular term according to the data distribution state of a plurality of user groups in the prediction data group; and the logic combination module is used for performing logic combination on the expected loss term and the regular term to determine the loss function.
In an exemplary embodiment of the present disclosure, the regularization term includes a first regularization term and a second regularization term; the regularization term determination module includes: the first determining module is used for determining a first regular term according to the distribution state of the prediction data of the adjacent user groups in the plurality of user groups; the second determining module is used for determining a second regular term according to the prediction data of a target user group in the plurality of user groups; the target user group is the user group corresponding to the label with the maximum numerical value in the label group.
In an exemplary embodiment of the present disclosure, the first determining module includes: and the difference value calculation module is used for calculating the difference value between the prediction data of the adjacent user groups and determining the first regular term according to the difference value.
According to an aspect of the present disclosure, there is provided a behavior prediction apparatus including: the system comprises a characteristic information acquisition module, a processing module and a processing module, wherein the characteristic information acquisition module is used for acquiring information to be processed and acquiring a plurality of characteristic information of the information to be processed; the prediction module is used for inputting the characteristic information into a behavior prediction model, predicting the user behaviors of a plurality of user groups aiming at the information to be processed, and obtaining prediction data corresponding to the user groups; the behavior prediction model is obtained by training according to any one of the above training methods of the behavior prediction model.
According to an aspect of the present disclosure, there is provided an electronic device including: a processor; and a memory for storing executable instructions; wherein the processor is configured to perform the method for training a behavior prediction model of any one of the above items or the method for predicting behavior of any one of the above items via execution of the executable instructions.
According to an aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a method of training a behavior prediction model as described in any one of the above or a method of behavior prediction as described in any one of the above.
According to the training method and device, the behavior prediction method and device, the electronic device and the computer readable storage medium of the behavior prediction model of the embodiment of the disclosure, an accurate tag group can be generated according to behavior data of a user group on target information, modeling between the target information and the user group is completed through feature information of the target information and the user behavior prediction data group, and a loss function of the behavior prediction model is determined according to data distribution states in the tag group, the user behavior prediction data group and the user behavior prediction data group, so that the behavior prediction model can cover all situations, limitation is avoided, multiplexing can be performed, the established behavior prediction model can be more suitable for practical application from multiple dimensions, and accuracy and reliability of the behavior prediction model are improved.
Drawings
The above and other objects, features and advantages of exemplary embodiments of the present disclosure will become readily apparent from the following detailed description read in conjunction with the accompanying drawings. Several embodiments of the present disclosure are illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which:
fig. 1 schematically shows a system architecture diagram of an application scenario of an embodiment of the present disclosure.
Fig. 2 schematically shows a flowchart of a training method of a behavior prediction model in an embodiment of the present disclosure.
Fig. 3 schematically illustrates a flowchart of determining a tag group in the embodiment of the present disclosure.
Fig. 4 schematically shows an architecture diagram of a behavior prediction model in an embodiment of the present disclosure.
Fig. 5 schematically illustrates a flow chart for determining a loss function in an embodiment of the disclosure.
FIG. 6 schematically illustrates a flow chart of a routine implementation of the present disclosure as a predictive method.
Fig. 7 schematically shows a schematic block diagram of a training apparatus of a behavior prediction model according to an embodiment of the present disclosure.
Fig. 8 schematically shows a schematic block diagram of a behavior prediction apparatus of an embodiment of the present disclosure.
Fig. 9 schematically illustrates a block diagram of an electronic device of an embodiment of the disclosure.
In the drawings, the same or corresponding reference numerals indicate the same or corresponding parts.
Detailed Description
The principles and spirit of the present disclosure will be described with reference to a number of exemplary embodiments. It is understood that these embodiments are given solely for the purpose of enabling those skilled in the art to better understand and to practice the present disclosure, and are not intended to limit the scope of the present disclosure in any way. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
As will be appreciated by one skilled in the art, embodiments of the present disclosure may be embodied as a system, apparatus, device, method, or computer program product. Accordingly, the present disclosure may be embodied in the form of: entirely hardware, entirely software (including firmware, resident software, micro-code, etc.), or a combination of hardware and software.
According to an embodiment of the present disclosure, a training method of a behavior prediction model, a training apparatus of a behavior prediction model, a behavior prediction method, a behavior prediction apparatus, an electronic device, and a computer-readable storage medium are provided.
Moreover, any number of elements in the drawings are by way of example and not by way of limitation, and any nomenclature is used solely for differentiation and not by way of limitation.
The principles and spirit of the present disclosure are explained in detail below with reference to several representative embodiments of the present disclosure.
Summary of The Invention
The inventor finds that for heuristic algorithms, the summarized rule engine is difficult to cover all situations, so the algorithms are often not good in generalization performance, and the heuristic algorithm based on the heuristic algorithm is difficult to reuse and consumes a large amount of manpower. And the algorithm based on supervised learning is difficult to obtain high-quality labeled data, so the accuracy is poor.
Based on the above, the basic idea of the present disclosure is: and replacing manually designed labels with behavior data of the user in the system as a basis for modeling and predicting the user behavior.
More specifically, in the present disclosure, behavior data of a plurality of user groups on target information is acquired, and a tag group of the target information is generated according to the behavior data, where the tag group includes tags corresponding to the user groups; inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information, wherein the user behavior prediction data set comprises user behavior prediction data corresponding to each user group; determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group; and adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model. Therefore, according to the embodiment of the disclosure, the accuracy and reliability of the training of the behavior prediction model can be improved.
Having described the general principles of the present disclosure, various non-limiting embodiments of the present disclosure are described in detail below.
Application scene overview
It should be noted that the following application scenarios are merely illustrated to facilitate understanding of the spirit and principles of the present disclosure, and embodiments of the present disclosure are not limited in this respect. Rather, embodiments of the present disclosure may be applied to any scenario where applicable.
Referring first to fig. 1, fig. 1 shows a system architecture diagram of one exemplary application scenario of an embodiment of the present disclosure. The training method of the behavior prediction model can be used for browsing application scenes such as commodities, articles, videos and the like. As shown in fig. 1, the system architecture 100 includes a terminal 101 network 102 and a server 103. The average click rate of the target information of the plurality of user groups extracted from the terminal 101 is analyzed by the server 103, so that the server predicts the tendency of the user behavior of each user group in the target information, and further recommends appropriate content for each user group according to the tendency. Those skilled in the art will appreciate that the schematic framework shown in fig. 1 is merely one example in which embodiments of the present disclosure may be implemented. The scope of applicability of the disclosed embodiments is not limited in any way by this framework.
It should be noted that the servers 103 may be local servers or remote servers, and in addition, the servers 103 may also be other products capable of providing a storage function or a processing function, such as a cloud server, and the embodiments of the present disclosure are not limited specifically herein. The server may also be a terminal device with fast computing capability or an in-vehicle device, and the like, which is not limited herein. The terminal 101 may be any device capable of generating user behavior data, such as a smart phone, a tablet computer, and a computer.
It should be understood that, in an application scenario of the present disclosure, the actions of the embodiments of the present disclosure may be performed by the server 103, or may be performed by a terminal having computing capability, and the present disclosure is not limited in any way in terms of the execution subject as long as the actions disclosed by the embodiments of the present disclosure are performed. The following description will be made by taking the execution subject as a server.
Exemplary method
In the following, in connection with the application scenario of fig. 1, a training method of a behavior prediction model according to an exemplary embodiment of the present disclosure is described with reference to fig. 2. It should be noted that the above application scenarios are merely illustrated for the convenience of understanding the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. Rather, embodiments of the present disclosure may be applied to any scenario where applicable.
FIG. 2 shows a flow diagram of a method of training a behavior prediction model according to an embodiment of the present disclosure. Referring to fig. 2, the training method of the behavior prediction model may include the following steps:
in step S210, behavior data of a plurality of user groups for target information is acquired, and a tag group of the target information is generated according to the behavior data, where the tag group includes tags corresponding to the user groups;
in step S220, inputting the feature information of the target information into a behavior prediction model to be trained, to obtain a user behavior prediction data set of the target information, where the user behavior prediction data set includes user behavior prediction data corresponding to each user group;
in step S230, determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group, and a data distribution state in the user behavior prediction data group;
in step S240, model parameters of the behavior prediction model are adjusted according to the loss function to train the behavior prediction model.
According to the technical scheme, modeling between the target information and the user group is completed through the characteristic information of the target information and the user behavior prediction data group, and the loss function of the behavior prediction model is determined according to the data distribution states in the tag group, the user behavior prediction data group and the user behavior prediction data group, so that the behavior prediction model can cover all situations, limitation is avoided, multiplexing can be performed under multiple scenes, the behavior prediction model is more consistent with practical application, and the accuracy and reliability of the behavior prediction model are improved.
Next, a method for training a behavior prediction model in the embodiment of the present disclosure is explained in detail with reference to the drawings.
In step S210, behavior data of a plurality of user groups for target information is acquired, and a tag group of the target information is generated according to the behavior data, where the tag group includes tags corresponding to the user groups.
In the embodiment of the present disclosure, the target information may be any type of information on the target platform. The target platform may be a news platform, an interactive platform, or a shopping platform, and the like, and the news platform is taken as an example for illustration. The target information may be text information or picture information, etc., and the text information is taken as an example for explanation here. When the target platform is a news platform, the target information may be, for example, any article on the news platform, such as article 1. The plurality of user groups refers to a group determined by dividing all users according to age distribution, and the number of the specific divided user groups can be determined according to actual requirements.
For example, a batch of seed users with age information on a certain news platform is selected, and the batch of users are divided into n user groups according to the ages of the batch of seed users. Wherein any user can be represented by u, an
Figure BDA0002907644380000091
Figure BDA0002907644380000092
Age as the representative age groupiThe user group of (1).
The behavior data refers to data generated by user behaviors of each user group on the target information, the user behaviors can be reading behaviors and can be expressed as operation behaviors such as clicking, browsing, collecting, forwarding, commenting and the like, and the user behaviors are described by clicking operation. The behavior data can be used for describing the behavior characteristics of each user group to the target information, and the behavior data of each user group can be the same or different. In the embodiment of the present disclosure, for each piece of target information, behavior data of each user group for the target information may be obtained, and the behavior data may be represented by an average click rate. The average click rate refers to an average value of click rates of the respective user groups for the target information. In the embodiment of the present disclosure, the average click rate of the target information by the plurality of user groups may be represented as follows:
Figure BDA0002907644380000101
wherein
Figure BDA0002907644380000102
Age as the representative age groupiUser group of
Figure BDA0002907644380000103
Average click rate on target information. When the target information (an article) is fully exposed in each age group, the reading tendency of each age group of the target information is approximately equal to the average click rate of the target information in each age group. The click rate refers to the index of the number of times that the specified content on a website or an application platform is clicked and exposed, and the click rate is usually an important measurement index in a recommendation system. Thus, this behavior data may be used to describe the reading tendency or reading tendency value of the user population.
In the embodiment of the disclosure, for the target information, a plurality of characteristic information and behavior data thereof can be determined. A plurality of feature information may be used to describe the content of the target information, and the plurality of feature information may include, for example, but not limited to, a title, a keyword, a category, and the like. Further, the characteristic information and the behavior data may be combined in a fixed format.
For example, for the target information dociOrganizing the characteristic information and the behavior data into the following data format, which is expressed as follows:
Figure BDA0002907644380000104
wherein the content of the first and second substances,
Figure BDA0002907644380000105
as target information dociTitle of, wiIs the ith word in the title, and t is the length of the title.
Figure BDA0002907644380000106
As target information dociA set of keywords of, wherein wordiK is the number of keywords of the ith keyword of the target information.
Figure BDA0002907644380000107
Is the object information dociThe category (2).
Figure BDA0002907644380000108
Is age groupjTarget information doc of user groupiAverage click rate above.
After the behavior data of the plurality of user groups for the target information is acquired, a tag group may be generated according to the behavior data of the plurality of user groups, where the tag group may include a tag corresponding to each user group, and the tag here may be used to represent a fitting object of the behavior prediction model, that is, a target that is desired to be output. Because manual labeling is not carried out, the generated label group can be used as pseudo labeling data, so that the labeling accuracy is improved, and the model is conveniently processed.
A flow chart for determining a tag group is schematically shown in fig. 3, and referring to fig. 3, mainly includes the following steps:
in step S310, the behavior data of the target information is normalized by the plurality of user groups, so as to obtain normalized behavior data.
In this step, the click rate data of some age groups on the target information is higher, and the click rate data of other age groups on the target information is lower. In order to avoid the influence of the factor that the behavior data benchmarks of the user groups of different ages are inconsistent, the behavior data can be normalized for multiple times to obtain normalized behavior data. Specifically, the behavior data of the user group of the same age group on the plurality of sample information (different articles) can be normalized, and the normalization on the age group is realized, so that the behavior data of the user group of different age groups on the same sample information is comparable, the reading tendency conditions of the user group of different age groups can be more approximate, and the accuracy and the authenticity of the behavior data are improved.
The specific steps of normalizing the behavior data of the target information of a plurality of user groups comprise: determining a global behavior data maximum value and a global behavior data minimum value in the behavior data of each user group to each sample information; and normalizing the behavior data of the target information by each user group according to the maximum value and the minimum value of the global behavior data to obtain the normalized behavior data. Wherein each sample information refers to all information of the target platform, such as article 1, article 2, article 3, and so on. The target information may be any one of a plurality of sample information. The global behavior data maximum value refers to behavior data with the largest value among behavior data of a plurality of pieces of sample information for each user group, for example, a maximum value determined from average click rates of articles 1, 2, and 3 for each user group. The global behavior data minimum value refers to behavior data with the smallest value among behavior data of a plurality of pieces of sample information for each user group, for example, the minimum value determined from average click rates of articles 1, 2, and 3 for each user group. Further, the behavior data of the target information of each user group can be normalized according to the maximum value and the minimum value of the global behavior data, so that normalized behavior data is obtained, that is, the average click rate of each user group on the target information is normalized, so that the normalized click rate is obtained. The normalization herein refers to maximum and minimum normalization, and can be specifically expressed by formula (1):
Figure BDA0002907644380000111
wherein the content of the first and second substances,
Figure BDA0002907644380000112
the behavior data, i.e., average click-through rate,
Figure BDA0002907644380000113
for the minimum value of the global behavior data,
Figure BDA0002907644380000114
is the global behavior data maximum.
By carrying out normalization processing on the behavior data, the average click rate of the user group can be closer to the real reading tendency of the user group in the age group, and the accuracy of the behavior data is improved.
After the normalized behavior data is obtained, the absolute value of the behavior data of the articles with similar subjects and semantics may be inconsistent due to the factors of the recommendation system. Therefore, the normalized behavior data after normalization over the age group is normalized again with respect to the target information (inside the article) to update the resultant normalized behavior data.
When the normalized behavior data is updated, the local behavior data maximum value and the local behavior data minimum value can be determined in the behavior data of each user group to the target information, and the normalized behavior data is further normalized again according to the local behavior data maximum value and the local behavior data minimum value so as to update the normalized behavior data. The local behavior data minimum value refers to behavior data with the minimum value in the behavior data of each user group to the target information, namely, the minimum value determined from the average click rate of each user group to the article 1. The local behavior data maximum value refers to behavior data with the largest value in behavior data of each user group for the target information, for example, a maximum value determined from the average click rate of each user group for the article 1. Further, the normalized behavior data may be normalized again according to the local behavior data maximum value and the local behavior data minimum value to update the normalized behavior data, and the normalization may be performed again by equation (2):
Figure BDA0002907644380000121
wherein the content of the first and second substances,
Figure BDA0002907644380000122
for the minimum of the local behavior data,
Figure BDA0002907644380000123
for the maximum value of the local behavior data,
Figure BDA0002907644380000124
behavior data for each user group for the target information.
In this way, for the target information, the reading tendency of the user group of each age group to the target information can be analyzed by using the updated normalized behavior data.
In step S320, a tag group of the target information is generated according to the normalized behavior data.
In the embodiment of the present disclosure, after obtaining the normalized behavior data, the updated normalized behavior data may be determined as the tags corresponding to each user group, so that a tag group of the target information is formed according to the tags corresponding to each user group, and the tag group is used to describe the reading tendencies of a group of user groups of different ages. And representing a fitting object of the model by using the label group as pseudo-labeling data.
In the embodiment of the disclosure, the pseudo-label data between the target information and the age group is generated through the behavior data of the target information of the user group, and the pseudo-label data is used as the label group, so that the generated label group is more accurate relative to the label manually labeled and better conforms to the actual situation.
In step S220, the feature information of the target information is input into a behavior prediction model to be trained, so as to obtain a user behavior prediction data set of the target information, where the user behavior prediction data set includes user behavior prediction data corresponding to each user group.
In the embodiment of the present disclosure, the feature information of the target information includes, but is not limited to, a keyword, a title, and a category. The behavior prediction model to be trained may be any suitable model, and the behavior prediction model to be trained is exemplified as the BERT model. After the feature information of the target information is acquired, the feature information can be used as input data and input into the behavior prediction model to be trained, so that the behavior prediction model to be trained processes the feature information to obtain a user behavior prediction data set of the target information. It should be noted that the user behavior prediction data set may include user behavior prediction data corresponding to each user group.
For example, the feature information of the target information and the label set may be determined as a training sample of the behavior prediction model to be trained for model training, and the format of the training sample may be expressed as
Figure BDA0002907644380000131
Wherein the title
Figure BDA0002907644380000132
Keyword
Figure BDA0002907644380000133
And categories
Figure BDA0002907644380000134
May be used as input to the behavior prediction model to be trained.
Figure BDA0002907644380000135
And the normalized label group of the target information of each user group. And the final output result of the behavior prediction model to be trained is the user behavior prediction data set of the target information. The user behavior prediction data set may be expressed as
Figure BDA0002907644380000136
Wherein
Figure BDA0002907644380000137
Representing at the target information dociAbove, the behavior prediction model to be trained is for ageThe segment is agejThe reading tendency of the user group of (1).
Fig. 4 schematically shows an architecture diagram of a behavior prediction model to be trained, and referring to fig. 4, feature information of target information may be used as an input of the behavior prediction model to be trained, an output result of the model is obtained after passing through an embedding layer and a transmission layer, and a final prediction result is further obtained through a softmax layer, where the prediction result is a user behavior prediction data set of the target information. Namely, the whole treatment process is as follows: the title, the keywords and the category of the target information are converted into a plurality of text vectors, the text vectors and the delimiter vectors are spliced to determine input vectors, and prediction is performed according to the input vectors to obtain user behavior prediction data of a plurality of user groups in the target information.
Specifically, referring to fig. 4, the input in the input layer of the BERT model is made by the title, keyword, and category of the target information, three parts being spliced after passing through the embedding layer. It should be noted that the titles, the keywords, the categories, and the like are all in text form, the embedding layer of the model is to convert the feature information into a vector with a fixed dimension, that is, a plurality of text vectors, and the plurality of text vectors and the feature information are in one-to-one correspondence. And then splicing the plurality of text vectors and the separator vectors to obtain a vector representation representing the target information as an input vector.
In the embodiment of the disclosure, the input vector of the behavior prediction model to be trained can be represented as bertinput＝concat(Ecls,Etitle,Esep,Ecate,Esep,Ekeyword). Wherein E isclsThe method is a special acquisition mark vector in the BERT model and is used for gathering all information input by the BERT model, namely, fusion title, category and keyword information. EsepAre markers, i.e. separator vectors, for separating different information. EtitleAs a title vector in the text vector, EcateAs category vectors in text vectors, EkeywordEmbedding the keyword vector in the text vector, namely title, category and keyword through embedding layer of BERT modelAnd (3) shown in the specification. After the input vector passes through the transmission layer, the text vectors can be fused to obtain a vector fused with all input information, and further obtain an output vector. Referring to FIG. 4, an input vector bertinputAfter passing through the transmission layer of the BERT model, an output vector BERT is obtainedoutput＝[Tcls,Ttitle,Tsep,Tcate,Tsep,Tkeyword]. Wherein, TclsAll the characteristic information is fused, and other vectors correspond to the vectors in the input vectors one by one.
Further, a user behavior prediction data group including user behavior prediction data corresponding to each user group may be obtained according to the output vector. That is, the flag vector in which all the feature information is fused in the output vector may be output to the softmax layer for processing to determine that a final prediction result is obtained as the user behavior prediction data set. The user behavior prediction data set is used for describing the prediction value of the reading behavior of the user group of each age group on the target information. The user behavior prediction data may be determined by equation (3):
mod elpredict＝soft max(Matrixsoftmax×Tcls) Formula (3)
Wherein the content of the first and second substances,
Figure BDA0002907644380000141
is an n x 1 vector, n is the number of all age groups, each dimension
Figure BDA0002907644380000142
Age of representative ageiThe reading tendency of the user group. MatrixsoftmaxIs a matrix of n x h, h is the embedding dimension of the hidden vector in the BERT model, TclsIs an h x 1 vector, and softmax is a common normalization function in machine learning. Referring to FIG. 4, the predicted value may be any value between 0 and 1.
In the embodiment of the disclosure, the user behavior prediction data set of the target information is obtained through the characteristic information of the target information, so that the tendency of each user group to the user behavior of the target information can be predicted from multiple dimensions, the accuracy of the user behavior prediction data set is improved, and a basis is further provided for model training.
In step S230, a loss function of the behavior prediction model is determined according to the tag group, the user behavior prediction data group, and a data distribution state in the user behavior prediction data group.
In the embodiment of the present disclosure, the data distribution state in the user behavior prediction data set is used to describe a specific state of the prediction data of each user group in the user behavior prediction data set, such as a data change trend, a data size, and the like. Further, a loss function of the behavior prediction model may be determined based on the tag group and the user behavior prediction data group, and a data distribution state in the user behavior prediction data group to construct the behavior prediction model.
A flow chart for determining the loss function is schematically shown in fig. 5, and with reference to fig. 5, mainly comprises the following steps:
in step S510, determining expected loss items according to the tag group and the user behavior prediction data group;
in step S520, determining a regular term according to data distribution states of a plurality of user groups in the user behavior prediction data group;
in step S530, the expected loss term and the regularization term are logically combined to determine the loss function.
In the embodiment of the present disclosure, first, an expected loss term may be determined according to the tag group and the user behavior prediction data group, and the expected loss term may be used to describe a distribution value of the pseudo tag data, and the expected loss term may be calculated by a mean square error. Specifically, the difference between the prediction data of the user group of each age group and the label (updated normalized behavior data) of the user group of the age group in the user behavior prediction data group may be calculated, and specifically, the expected loss term may be calculated by formula (4):
Figure BDA0002907644380000151
secondly, in order to make the prediction of the behavior prediction model more in line with objective rules, a regular term is required to be combined to guide the training of the behavior prediction model. The regularization term is used to describe regularization, which refers to a process that introduces additional information to account for overfitting. During the optimization process of machine learning, the regularization term is often added to the objective function. The regularization term herein may include a first regularization term and a second regularization term, and the first regularization term and the second regularization term are of different types. The first regularization term is used to describe the state of change of the age group. The second regularization term is used to describe prediction accuracy. In the embodiment of the present disclosure, the first regularization term and the second regularization term may be determined according to a data distribution state of a plurality of user groups in the user behavior prediction data group.
Since the age groups are continuously varied, the difference in preference of the user groups of the adjacent age groups is small, and the difference in preference of the user groups of the non-adjacent age groups is large. For the user population of each age group, the reading tendency (i.e. the prediction data) should be a smoothly varying curve, not a wavy line with violent shaking. Therefore, when the first regularization term is determined according to the data distribution states of a plurality of user groups in the user behavior prediction data group, the first regularization term may be determined according to the distribution states of prediction data of adjacent user groups in the plurality of user groups. The adjacent user group may be any two adjacent user groups. The difference between the prediction data of the adjacent user groups in the user behavior prediction data group can be calculated, and then the first regular term is determined according to the difference, which can be specifically determined by formula (5):
Figure BDA0002907644380000161
in addition, the application process also pays attention to whether the prediction data of the user group with the highest tendency is accurate. Based on this, when the second regular term is determined according to the data distribution state of the plurality of user groups in the user behavior prediction data group, the second regular term may be determined according to an evaluation parameter of prediction data of a target user group in the plurality of user groups, and the evaluation parameter may be, for example, accuracy. The target user group refers to the user group corresponding to the label with the maximum value in the label group. That is, the age group may be an age group corresponding to the maximum value of the normalized behavior data in the updated normalized behavior data. The second regularization term may be a head regularization term to increase a penalty according to a prediction error condition of the second regularization term. Therefore, the second regularization term may be determined according to the accuracy of the prediction data of the user group of the age group with the highest attention tendency degree, and specifically may be calculated according to formula (5):
Figure BDA0002907644380000162
after the first regular term and the second regular term are obtained, the expected loss term, the first regular term and the second regular term may be logically combined according to the corresponding weights to determine the loss function. The logical combination may be to perform an addition operation. The loss function of the behavior prediction model may be determined according to equation (6):
loss＝0.7*lmse+0.2*lcon+0.1*ltopformula (6)
The weight of the expected loss term in the loss function is the largest, and the weight corresponding to the weight of the second regular term is the smallest.
Through the technical scheme in fig. 5, the regularization term can be added on the basis of the expected loss term, so that the loss function of the behavior prediction model is more consistent with the actual situation, and the accuracy and reliability of the loss function are improved.
In step S240, model parameters of the behavior prediction model are adjusted according to the loss function to train the behavior prediction model.
In the embodiment of the present disclosure, after determining the loss function of the behavior prediction model, the model parameters of the behavior prediction model may be adjusted according to the loss function, so as to train the behavior prediction model. The loss function indicates a distribution state to which the prediction data should conform, and the tag group is a fitting object, i.e., a value desired to be output. The output of the behavior prediction model is a user behavior prediction data set representing the distribution of age group tendencies. The specific process of model training may include: the method comprises the steps of taking a plurality of pieces of characteristic information such as titles, keywords and categories corresponding to target information as input of a model to obtain prediction data, taking behavior data of a plurality of user groups to the target information as labels of the model, adjusting model parameters of the behavior prediction model according to a loss function to train the model, fitting the behavior prediction model, and enabling the prediction data obtained by the behavior prediction model to be matched with label groups according to the weight of the loss function.
According to the technical scheme, a behavior prediction model of the mapping relation from the text semantics of the target information to the reading tendencies of the user groups of different ages can be established, the loss function can be determined through the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group, the behavior prediction model is trained based on the loss function, and the accuracy, reliability and stability of the model are improved. The method avoids the limitation of the model in the related technology, can be reused, and increases the flexibility and the application range. According to the idea of weak supervised learning, the modeling from the target information to the user groups corresponding to the age groups is completed by using the characteristic information of the title, the keywords, the category and the like of the target information and the behavior data of the user groups of all the age groups on the target platform to the target information, so that the established behavior prediction model can accurately fuse the rules of the age groups, and the accuracy of the model is improved. Furthermore, two regular terms, namely a smooth constraint term and a head constraint term, are added to generate a loss function to train the model, so that the trained model is closer to practical application and more accurate.
The embodiment of the disclosure also provides a behavior prediction method, which can be implemented by the behavior prediction model described in the above embodiment. A flow chart of a behavior prediction method, which may be performed by a server, which may be the server 103 shown in fig. 1, is schematically shown in fig. 6. Referring to fig. 6, the behavior prediction method at least includes steps S610 to S620, which are described in detail as follows:
in step S610, information to be processed is acquired, and a plurality of feature information of the information to be processed is acquired.
In the embodiment of the present disclosure, the information to be processed may be the same as the type of the target information, for example, any article needing to be processed on the news platform. The plurality of characteristic information of the information to be processed may include, but is not limited to, a title, a keyword, a category, and the like. For example, information to be processed shown on the terminal, such as an article 2 on the target platform, may be obtained. Furthermore, semantic analysis can be performed on the article 2 to obtain the title, the keyword and the category of the information to be processed.
In step S620, the feature information is input to a behavior prediction model, and user behaviors of a plurality of user groups with respect to the information to be processed are predicted, so as to obtain prediction data corresponding to the user groups.
In the embodiment of the present disclosure, a plurality of feature information corresponding to information to be processed may be input to the behavior prediction model obtained by training according to the above embodiment, so that the behavior prediction model obtains, according to the input feature information, prediction data of user behavior of each user group on the information to be processed. The user behavior here may be reading behavior. The prediction data may be a tendency or degree of tendency of a group of users of each age group to read behavior of the information to be processed. The behavior prediction method can be applied to prediction of news reading behaviors, and can also be applied to other aspects, such as prediction of commodity clicking during commodity recommendation and the like. The tendency may be a value between 0 and 1, with the tendency being positively correlated with the value. The closer to 1, the higher the tendency of the user group representing the age group to read the information to be processed, and conversely, the closer to 0, the lower the tendency of the user group representing the age group to read the information to be processed.
Furthermore, after the tendency of each user group to the reading behavior of the information to be processed is determined, appropriate information can be recommended to each user group based on the prediction result, the recommendation condition that the age group and the recommended information are not matched is reduced, the recommendation accuracy and pertinence are improved, and the reading experience and recommendation effect of the user are improved.
Exemplary devices
Next, a training device of a behavior prediction model according to an exemplary embodiment of the present disclosure will be described with reference to fig. 7. As shown in fig. 7, the training apparatus 700 for the behavior prediction model may include:
the tag generation module 701 is configured to obtain behavior data of a plurality of user groups for target information, and generate a tag group of the target information according to the behavior data, where the tag group includes tags corresponding to the user groups;
a prediction data determining module 702, configured to input the feature information of the target information into a behavior prediction model to be trained, to obtain a user behavior prediction data set of the target information, where the user behavior prediction data set includes user behavior prediction data corresponding to each user group;
a loss function determining module 703, configured to determine a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group, and a data distribution state in the user behavior prediction data group;
a model training module 704, configured to adjust model parameters of the behavior prediction model according to the loss function, so as to train the behavior prediction model.
In an exemplary embodiment of the present disclosure, the tag generation module includes: the normalization module is used for normalizing the behavior data of the target information by the user groups to obtain normalized behavior data; and the generation control module is used for generating the label group of the target information according to the normalized behavior data.
In an exemplary embodiment of the disclosure, the normalization module includes: the global data determining module is used for determining the maximum value and the minimum value of the global behavior data in the behavior data of each user group to each sample message; and the normalization control module is used for normalizing the behavior data of the target information by each user group according to the global behavior data maximum value and the global behavior data minimum value to obtain the normalized behavior data.
In an exemplary embodiment of the present disclosure, after normalizing the behavior data of the target information by each user group according to the global behavior data maximum value and the global behavior data minimum value to obtain the normalized behavior data, the apparatus further includes: the local data determining module is used for determining a local behavior data maximum value and a local behavior data minimum value in the behavior data of each user group to the target information; and the normalization updating module is used for carrying out normalization processing on the normalization behavior data again according to the local behavior data maximum value and the local behavior data minimum value so as to update the normalization behavior data.
In an exemplary embodiment of the present disclosure, the loss function determination module includes: an expected loss item determining module, configured to determine an expected loss item according to the tag group and the user behavior prediction data group; the regular term determining module is used for determining a regular term according to the data distribution state of a plurality of user groups in the prediction data group; and the logic combination module is used for performing logic combination on the expected loss term and the regular term to determine the loss function.
In an exemplary embodiment of the present disclosure, the regularization term includes a first regularization term and a second regularization term; the regularization term determination module includes: the first determining module is used for determining a first regular term according to the distribution state of the prediction data of the adjacent user groups in the plurality of user groups; the second determining module is used for determining a second regular term according to the prediction data of a target user group in the plurality of user groups; the target user group is the user group corresponding to the label with the maximum numerical value in the label group.
In an exemplary embodiment of the present disclosure, the first determining module includes: and the difference value calculation module is used for calculating the difference value between the prediction data of the adjacent user groups and determining the first regular term according to the difference value.
In addition, fig. 8 schematically shows a block diagram of a behavior prediction apparatus according to an exemplary embodiment of the present disclosure. Referring to fig. 8, the behavior prediction apparatus 800 may include:
a characteristic information obtaining module 801, configured to obtain information to be processed and obtain a plurality of characteristic information of the information to be processed;
the prediction module 802 is configured to input the plurality of feature information into a behavior prediction model, and predict user behaviors of a plurality of user groups for the information to be processed to obtain prediction data corresponding to the plurality of user groups; wherein, the behavior prediction model is a model obtained by training according to the training method of the behavior prediction model of any one of claims 1 to 7.
It should be noted that the specific details of each module of the training apparatus for behavior prediction models have been described in detail in the step of the training method for the corresponding behavior prediction models, and the specific details of each module of the behavior prediction apparatus have been described in detail in the step of the corresponding behavior prediction method, so that the details are not repeated here.
Exemplary electronic device
An electronic device 900 according to this embodiment of the disclosure is described below with reference to fig. 9. The electronic device 900 shown in fig. 9 is only an example and should not bring any limitations to the functionality or scope of use of the embodiments of the present disclosure.
As shown in fig. 9, the electronic device 900 is embodied in the form of a general purpose computing device. Components of electronic device 900 may include, but are not limited to: the at least one processing unit 910, the at least one memory unit 920, a bus 930 connecting different system components (including the memory unit 920 and the processing unit 109), and a display unit 940. The bus 930 may include a data bus, an address bus, and a control bus.
Wherein the storage unit stores program code that is executable by the processing unit 910 to cause the processing unit 910 to perform steps according to various exemplary embodiments of the present disclosure described in the above section "exemplary method" of the present specification. For example, the processing unit 910 may perform the steps as shown in fig. 2 or fig. 6.
The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM)9201 and/or a cache memory unit 9202, and may further include a read only memory unit (ROM) 9203.
Storage unit 920 may also include a program/utility 9204 having a set (at least one) of program modules 9205, such program modules 9205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which, or some combination thereof, may comprise an implementation of a network environment.
The electronic device 900 may also communicate with one or more external devices 1000 (e.g., keyboard, pointing device, bluetooth device, etc.). Such communication may occur via input/output (I/O) interface 950. Also, the electronic device 900 may communicate with one or more networks (e.g., a Local Area Network (LAN), a Wide Area Network (WAN) and/or a public network, such as the Internet) via the network adapter 960. As shown, the network adapter 960 communicates with the other modules of the electronic device 900 via the bus 930. It should be appreciated that although not shown, other hardware and/or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, among others.
Exemplary Medium
Next, a computer-readable storage medium of an exemplary embodiment of the present disclosure will be explained.
In some possible embodiments, various aspects of the disclosure may also be implemented in the form of a program product including program code for causing a terminal device to perform the steps in the method for training a behavior prediction model according to various exemplary embodiments of the disclosure described in the "exemplary methods" section above of this specification when the program product is run on the terminal device, for example, the processing unit may perform the steps as shown in fig. 2. The steps of the behavior prediction method as described in fig. 6 may also be performed.
The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
A program product for training of a behavior prediction model according to an embodiment of the present disclosure may employ a portable compact disc read only memory (CD-ROM) and include program code, and may be run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto, and in this document, a readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A readable signal medium may include a propagated data signal with readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A readable signal medium may also be any readable medium that is not a readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code on a readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Program code for carrying out operations for the present disclosure may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user computing device, partly on a remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device over any kind of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or may be connected to an external computing device (e.g., over the internet using an internet service provider).
It should be noted that although several modules or sub-modules of the electronic device are mentioned in the above detailed description, such division is merely illustrative and not mandatory. Indeed, the features and functionality of two or more of the modules described above may be embodied in one module, in accordance with embodiments of the present disclosure. Conversely, the features and functions of one module described above may be further divided into embodiments by a plurality of modules.
Further, while the operations of the disclosed methods are depicted in the drawings in a particular order, this does not require or imply that these operations must be performed in this particular order, or that all of the illustrated operations must be performed, to achieve desirable results. Additionally or alternatively, certain steps may be omitted, multiple steps combined into one step execution, and/or one step broken down into multiple step executions.
While the spirit and principles of the present disclosure have been described with reference to several particular embodiments, it is to be understood that the present disclosure is not limited to the particular embodiments disclosed, nor is the division of aspects, which is for convenience only as the features in such aspects may not be combined to benefit. The disclosure is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.

Claims (10)

1. A method for training a behavior prediction model, comprising:
acquiring behavior data of a plurality of user groups on target information, and generating a tag group of the target information according to the behavior data, wherein the tag group comprises tags corresponding to the user groups;
inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information, wherein the user behavior prediction data set comprises user behavior prediction data corresponding to each user group;
determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group;
and adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model.
2. The method for training a behavior prediction model according to claim 1, wherein the generating the tag set of the target information according to the behavior data comprises:
normalizing the behavior data of the target information by the user groups to obtain normalized behavior data;
and generating a label group of the target information according to the normalized behavior data.
3. The method for training the behavior prediction model according to claim 2, wherein the normalizing the behavior data of the target information by the plurality of user groups to obtain normalized behavior data includes:
determining a global behavior data maximum value and a global behavior data minimum value in the behavior data of each user group to each sample information;
and normalizing the behavior data of the target information by each user group according to the maximum value and the minimum value of the global behavior data to obtain the normalized behavior data.
4. The training method of the behavior prediction model according to claim 3, wherein after normalizing the behavior data of the target information by each user group according to the maximum value and the minimum value of the global behavior data to obtain the normalized behavior data, the method further comprises:
determining a local behavior data maximum value and a local behavior data minimum value in the behavior data of each user group to the target information;
and carrying out normalization processing on the normalized behavior data again according to the local behavior data maximum value and the local behavior data minimum value so as to update the normalized behavior data.
5. The method for training a behavior prediction model according to claim 1, wherein the determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group, and the data distribution state in the user behavior prediction data group comprises:
determining expected loss items according to the tag groups and the user behavior prediction data groups;
determining a regular term according to the data distribution states of a plurality of user groups in the user behavior prediction data group;
and logically combining the expected loss term and the regular term to determine the loss function.
6. A training method for a behavior prediction model according to claim 5, characterized in that the regularization term comprises a first regularization term and a second regularization term; the determining a regularization term according to the data distribution states of a plurality of user groups in the user behavior prediction data group includes:
determining a first regular term according to the distribution state of the prediction data of the adjacent user groups in the plurality of user groups;
determining a second regularization term according to the prediction data of a target user group in the plurality of user groups; the target user group is the user group corresponding to the label with the maximum numerical value in the label group.
7. The method for training a behavior prediction model according to claim 6, wherein the determining a first regularization term according to the distribution states of the prediction data of the neighboring user groups of the plurality of user groups comprises:
and calculating a difference value between the prediction data of the adjacent user groups, and determining the first regular term according to the difference value.
8. An apparatus for training a behavior prediction model, comprising:
the tag generation module is used for acquiring behavior data of a plurality of user groups on target information and generating a tag group of the target information according to the behavior data, wherein the tag group comprises tags corresponding to the user groups;
the prediction data determining module is used for inputting the characteristic information of the target information into a behavior prediction model to be trained to obtain a user behavior prediction data set of the target information, wherein the user behavior prediction data set comprises user behavior prediction data corresponding to each user group;
the loss function determining module is used for determining a loss function of the behavior prediction model according to the tag group, the user behavior prediction data group and the data distribution state in the user behavior prediction data group;
and the model training module is used for adjusting the model parameters of the behavior prediction model according to the loss function so as to train the behavior prediction model.
9. An electronic device, comprising:
a processor; and
a memory for storing executable instructions;
wherein the processor is configured to perform the method of training a behavioral prediction model of any one of claims 1 to 7 via execution of the executable instructions.
10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out a method of training a behaviour prediction model according to any one of claims 1 to 7.
CN202110076021.1A 2021-01-20 2021-01-20 Training method of behavior prediction model, behavior prediction method, device and equipment Pending CN112784157A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110076021.1A CN112784157A (en) 2021-01-20 2021-01-20 Training method of behavior prediction model, behavior prediction method, device and equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110076021.1A CN112784157A (en) 2021-01-20 2021-01-20 Training method of behavior prediction model, behavior prediction method, device and equipment

Publications (1)

Publication Number Publication Date
CN112784157A true CN112784157A (en) 2021-05-11

Family

ID=75757351

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110076021.1A Pending CN112784157A (en) 2021-01-20 2021-01-20 Training method of behavior prediction model, behavior prediction method, device and equipment

Country Status (1)

Country Link
CN (1) CN112784157A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113965805A (en) * 2021-10-22 2022-01-21 北京达佳互联信息技术有限公司 Prediction model training method and device and target video editing method and device
CN115102779A (en) * 2022-07-13 2022-09-23 中国电信股份有限公司 Prediction model training and access request decision method, device and medium

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113965805A (en) * 2021-10-22 2022-01-21 北京达佳互联信息技术有限公司 Prediction model training method and device and target video editing method and device
CN115102779A (en) * 2022-07-13 2022-09-23 中国电信股份有限公司 Prediction model training and access request decision method, device and medium
CN115102779B (en) * 2022-07-13 2023-11-07 中国电信股份有限公司 Prediction model training and access request decision method, device and medium

Similar Documents

Publication Publication Date Title
CN110968695A (en) Intelligent labeling method, device and platform based on active learning of weak supervision technology
CN110110233B (en) Information processing method, device, medium and computing equipment
CN109871491A (en) Forum postings recommended method, system, equipment and storage medium
US11276099B2 (en) Multi-perceptual similarity detection and resolution
CN111950593A (en) Method and device for recommending model training
CN108121814B (en) Search result ranking model generation method and device
CN110708285B (en) Flow monitoring method, device, medium and electronic equipment
US20220284067A1 (en) Method for pushing information, electronic device
JP2020149693A (en) Generation device, generation method, and generation program
CN111754278A (en) Article recommendation method and device, computer storage medium and electronic equipment
CN112784157A (en) Training method of behavior prediction model, behavior prediction method, device and equipment
CN112966701A (en) Method and device for classifying objects
CN117011737A (en) Video classification method and device, electronic equipment and storage medium
CN112328889A (en) Method and device for determining recommended search terms, readable medium and electronic equipment
CN116680481B (en) Search ranking method, apparatus, device, storage medium and computer program product
CN116738081B (en) Front-end component binding method, device and storage medium
CN111340605A (en) Method and device for training user behavior prediction model and user behavior prediction
CN111597430A (en) Data processing method and device, electronic equipment and storage medium
CN113220994B (en) User personalized information recommendation method based on target object enhanced representation
CN112200602B (en) Neural network model training method and device for advertisement recommendation
JP2022087842A (en) Computer mounting method for selecting suitable machine learning pipeline for processing new data set, computing system, and computer program (item recommendation having application to automated artificial intelligence)
CN113094584A (en) Method and device for determining recommended learning resources
CN113762992A (en) Method and device for processing data
CN116823407B (en) Product information pushing method, device, electronic equipment and computer readable medium
CN114579757A (en) Knowledge graph assistance-based text processing method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination