Data processing method, device, electronic device and storage medium

Abstract

The embodiment of the present application proposes a data processing method, device, electronic device and storage medium. Among them, the method is: obtaining the historical behavior data of the user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is the behavior data within a first preset time period, the second behavior data is the behavior data within a second preset time period, and the first preset time period is greater than the second preset time period; extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data; fusing the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determining the user's interest information based on the fused feature information. Through this application, accurate user interest information can be obtained based on the user's historical behavior data.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a data processing method, device, electronic device and storage medium. Background Art

[0002] With the continuous development of Internet technology, it is becoming more and more common to recommend corresponding content to users based on their interests. Common application scenarios include advertising recommendation, video recommendation, news recommendation, e-commerce recommendation, etc. In the content recommendation scenario, it is necessary to model user interests so that the recommended content can be accepted by users.

[0003] At present, the existing modeling method usually models the user's long-term interests and the user's short-term interests separately. This modeling method does not take into account the correlation between the user's long-term interests and short-term interests, resulting in the inability to accurately and comprehensively obtain the user's interests. Summary of the invention

[0004] The embodiments of the present invention provide a data processing method, device, electronic device and storage medium, which can more accurately determine the user's interest information based on the user's historical behavior data.

[0005] The present application provides a data processing method, which specifically includes:

[0006] Acquire historical behavior data of the user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period;

[0007] Extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data;

[0008] The feature information of the first behavior data and the feature information of the second behavior data are fused to obtain fused feature information, and the interest information of the user is determined according to the fused feature information.

[0009] The embodiment of the present application provides a data processing device, which has the function of implementing the above-mentioned data processing method. The function can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes:

[0010] an acquisition unit, configured to acquire historical behavior data of a user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period;

[0011] a feature extraction unit, configured to extract features from the first behavior data and the second behavior data respectively, to obtain feature information of the first behavior data and feature information of the second behavior data;

[0012] A processing unit is used to fuse the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determine the user's interest information according to the fused feature information.

[0013] An embodiment of the present application provides an electronic device, which includes a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, and the computer program includes program instructions, and the processor is configured to call the program instructions to execute the operations involved in the above-mentioned data processing method.

[0014] An embodiment of the present application provides a computer-readable storage medium for storing computer program instructions for use in an electronic device, which includes a program for executing the above-mentioned data processing method.

[0015] The embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned data processing method.

[0016] Through the embodiments of the present application, the user's historical behavior data is divided into first behavior data and second behavior data, and feature extraction is performed on the first behavior data to obtain feature information of the first behavior data, and feature extraction is performed on the second behavior data to obtain feature information of the second behavior data, and the feature information of the first behavior data and the feature information of the second behavior data are fused, so as to obtain more accurate user interest information. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0018] Figure 1 It is a schematic diagram of the architecture of a data processing system provided in an embodiment of the present application;

[0019] Figure 2 is a model schematic diagram of an interest identification network model provided in an embodiment of the present application;

[0020] Figure 3 is a flow chart of a data processing method provided in an embodiment of the present application;

[0021] Figure 4 is a schematic diagram of a pooling process provided in an embodiment of the present application;

[0022] Figure 5 is a flow chart of another data processing method provided in an embodiment of the present application;

[0023] Figure 6 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0024] Figure 7 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] It should be noted that the descriptions of "first", "second", etc. involved in the embodiments of the present application are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the technical features defined as "first" or "second" may explicitly or implicitly include at least one of the features.

[0027] Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool, which can be used on demand and is flexible and convenient. Cloud computing technology will become an important support. The backend services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the rapid development and application of the Internet industry, in the future, each item may have its own identification mark, which needs to be transmitted to the backend system for logical processing. Data of different levels will be processed separately. All kinds of industry data need strong system backing support, which can only be achieved through cloud computing.

[0028] Big data refers to a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time frame. It is a massive, high-growth, and diverse information asset that requires new processing models to have stronger decision-making power, insight discovery, and process optimization capabilities. With the advent of the cloud era, big data has also attracted more and more attention. Big data requires special technologies to effectively process large amounts of data within a tolerable time frame. Technologies applicable to big data include large-scale parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems.

[0029] The embodiment of the present application performs cloud computing on big data through cloud technology. The big data here refers to the historical behavior data of users on social products within a specified time, where the historical behavior data can specifically be the user's click behavior and browsing behavior on XX video playback software within a year, or the user's click behavior and browsing behavior on XX news software within a year, etc.

[0030] At present, the existing modeling method usually models the user's long-term interests and the user's short-term interests separately. This modeling method does not take into account the correlation between the user's long-term interests and short-term interests, resulting in the inability to more accurately and comprehensively reflect the user's interest in the product. Based on this, the embodiment of the present application proposes a data processing method, which specifically comprises: obtaining the user's historical behavior data, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is the behavior data within a first preset time period, the second behavior data is the behavior data within a second preset time period, and the first preset time period is greater than the second preset time period; performing feature extraction on the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data; performing fusion processing on the feature information of the first behavior data and the feature information of the second behavior data to obtain fusion feature information, and determining the user's interest information according to the fusion feature information. Through the data processing method provided by the embodiment of the present application, by fusing the user's long-term behavior data and short-term behavior data, the user's long-term and short-term fusion behavior data is obtained, so that the user's interests can be determined more comprehensively and accurately, and the accuracy of the recommendation algorithm can be improved.

[0031] It should be noted that this application is applicable to related fields such as natural language processing and deep neural networks based on artificial intelligence, and can be applied to products such as Tencent Advertising, Tencent Video, and Tencent News to improve the accuracy of the product's recommendation algorithm. The specific application scenarios can be: based on the user's historical behavior data over a period of time, information that the user is interested in can be recommended to the user in a more targeted manner.

[0032] The electronic device involved in the embodiments of the present application is an entity for receiving or transmitting signals. Common electronic devices include, for example, mobile phones, tablet computers, laptop computers, PDAs, mobile Internet devices (MID), vehicles, roadside devices, aircraft, wearable devices, such as smart watches, smart bracelets, pedometers, and other smart devices with data analysis and data processing functions, but the embodiments of the present application are not limited thereto.

[0033] In order to better understand the data processing method provided in the embodiment of the present application, the system architecture diagram applicable to the embodiment of the present application is first described below.

[0034] See also Figure 1 , Figure 1 A schematic diagram of the architecture of a data processing system provided in an embodiment of the present application is shown in FIG. Figure 1As shown, the system architecture diagram includes: electronic device 100, network 200 and server 300. The electronic device 100 is connected to the server 300 via the network 200, and the network 200 can be a wide area network or a local area network, or a combination of the two.

[0035] In one possible implementation, the server 300 stores the user's historical behavior data, and the electronic device 100 can obtain the user's historical behavior data within a specified time period from the server 300 through the network 200. Specifically, the electronic device 100 obtains the user's browsing behavior on XX news software within three months, or the user's clicking behavior on XX advertising software, or the user's browsing behavior on XX video software, etc.

[0036] In one possible implementation, the electronic device 100 includes XX news software, XX advertising software, and XX video software. The electronic device 100 directly obtains the user's browsing behavior on the XX news software within three months, or the user's clicking behavior on the XX advertising software, or the user's browsing behavior on the XX video software.

[0037] In one implementation, the electronic device 100 performs feature extraction on the acquired historical behavior data. Specifically, the electronic device 100 performs feature extraction on the acquired first behavior data and the acquired second behavior data to obtain feature information of the first behavior data and feature information of the second behavior data. The first behavior data may be the user's behavior data within three months, and the second behavior data may be the user's behavior data within one month.

[0038] In one implementation, the electronic device 100 fuses the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determines the user's interest information based on the fused feature information. The electronic device 100 can send the obtained user's interest information to the server 300 via the network 200. The server 300 can store the received user's interest information locally so as to make interest recommendations to the user based on the user's behavior. The electronic device 100 can also make recommendations to the user in real time based on the obtained user's interest information. The recommendation interface is as follows: Figure 1 shown.

[0039] It can be understood that the system architecture diagram described in the embodiment of the present application is for more clearly illustrating the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided in the embodiment of the present application. A person of ordinary skill in the art can know that with the evolution of the system architecture and the emergence of new business scenarios, the technical solution provided in the embodiment of the present application is also applicable to similar technical problems.

[0040] See also Figure 2 , Figure 2 is a schematic diagram of an interest identification network model provided in an embodiment of the present application. Figure 2 As shown, the interest recognition network model may include a word embedding processing module 210 , an attention module 220 and a splicing module 230 .

[0041] In a possible implementation, the word embedding processing module 210 performs word embedding processing on the user's historical behavior data, attribute information, and third behavior data to obtain word vectors of the historical behavior data, word vectors of the attribute information, and word vectors of the third behavior data. The historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, the first preset time period is greater than the second preset time period, the third behavior data is behavior data within a third preset time period, and the third preset time period is no longer than the preset time period from the current time. The word embedding processing module 210 converts the words in the text into word vectors determined by their context. Specifically, the word embedding processing module 210 obtains the user's historical behavior data and processes the user's historical behavior data. The user's historical behavior data contains at least one word, so at least one word embedding matrix is ​​obtained. Then, at least one word embedding matrix is ​​truncated and completed to obtain the word vectors of the user's historical behavior data, specifically the word vectors of the first behavior data and the word vectors of the second behavior data. Similarly, the word embedding processing module 210 processes the attribute information of the user to obtain at least one word embedding matrix, and then truncates and completes the at least one word embedding matrix to obtain the word vector of the attribute information of the user. The word embedding processing module 210 processes the third behavior data to obtain at least one word embedding matrix, and then truncates and completes the at least one word embedding matrix to obtain the word vector of the third behavior data.

[0042] In a possible implementation, first, the attention mechanism module 220 receives the word vector of the first behavior data, the word vector of the second behavior data, and the word vector of the attribute information sent by the word embedding processing module 210, and performs average pooling processing on the word vector of the first behavior data to obtain the feature information of the first behavior data, and performs cumulative pooling processing on the word vector of the second behavior data to obtain the feature information of the second behavior data. Then, the attention mechanism module 220 fuses and splices the feature information of the second behavior data with the feature information of the first behavior data through the attention mechanism to obtain fused feature information. Finally, the attention mechanism module 220 splices the fused feature information, the feature information of the second behavior data, and the feature information of the attribute information to obtain spliced ​​feature information.

[0043] In one possible implementation, the splicing module 230 receives the splicing feature information output from the attention mechanism module 220, and uses the splicing feature information as the input of the attention module of the interest recognition network model, and outputs the user's interest information through processing of several fully connected layers and activation functions.

[0044] In a possible implementation, the user's interest information and the feature information of the third behavior feature are processed through an attention mechanism to obtain the accuracy of the user's interest information. The accuracy is used to adjust the parameters of the interest recognition network model. Specifically, if the accuracy is greater than a preset threshold, the user's interest information is determined to be accurate. If the accuracy is less than the preset threshold, the user's interest information is determined to be inaccurate, and the activation function can be replaced or the number of network channels can be increased.

[0045] See also Figure 3 , Figure 3 is a flow chart of a data processing method provided in an embodiment of the present application. The method is applied to electronic devices, such as Figure 3 As shown, the data processing method may include steps S310 to S330. Among them:

[0046] Step S310: Obtain the user's historical behavior data, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period.

[0047] The historical behavior data includes first behavior data and second behavior data. The first behavior data is the behavior data within a first preset time period, and the second behavior data is the behavior data within a second preset time period. The first preset time period is greater than the second preset time period. Specifically, the first behavior data is the user's behavior data within 90 days, and the second behavior data is the user's behavior data within 7 days. The behavior data may have different definitions depending on the specific application scenario. For example, in a video application scenario, the behavior data may be the user's playback behavior, in an advertising application scenario, the behavior data may be the user's click behavior, and in a news application scenario, the behavior data may be the user's browsing behavior, etc. The present invention does not limit this.

[0048] Step S320: extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data.

[0049] In one possible implementation, first, the electronic device performs word embedding processing on the first behavior data to obtain a word vector of the first behavior data. The first behavior data includes multiple words, and the electronic device performs word embedding processing on the first behavior data through a word embedding model to obtain at least one word embedding matrix, and truncates and completes at least one word embedding matrix to obtain a word vector of the user's first behavior data. It should be noted that the electronic device can also perform word embedding processing on the first behavior data by means of embedding data processing, which is not limited by the present invention.

[0050] For example, the electronic device performs word embedding processing on the first behavior data through the Word2vec model, wherein Word2vec can quickly and effectively express a word into a vector form through an optimized training model based on a given corpus, providing a new tool for applied research in the field of natural language processing. Word2vec relies on skip-grams or continuous bag of words (CBOW) to establish neural word embedding. Then the electronic device uses the Word2vec model to convert the words in the first behavior data into word vectors determined by their context. Therefore, the probability distribution of each keyword and the combination of different words in the first behavior data can be obtained through neural network learning, and the word vector matrix corresponding to each word in the first behavior data can be obtained. Finally, multiple word vector matrices are truncated and completed to obtain the word vector of the user's first behavior data.

[0051] In a possible implementation, the electronic device performs average pooling processing on the word vectors of the first behavior data to obtain feature information of the first behavior data. Figure 4 As shown, Figure 4 A schematic diagram of a pooling process provided for an embodiment of the present application, the pooling process may include average pooling (avg-pooling) processing and cumulative pooling (sum-pooling) processing, and the purpose of the pooling process is to reduce the dimension. Since the long-term behavior data of users tend to be stable, the word vector of the first behavior data is averaged and pooled. For example, the word vector of the user's behavior data within 90 days is averaged and pooled, and the characteristic information of the user's behavior data within 90 days can be obtained. Specifically, the user is a sports fan, so the characteristic information of the user's behavior data within 90 days includes sports label information. In this way, the key data in the first behavior data can be refined and the redundant data can be removed, thereby reducing the hardware cost of data processing and reducing the processing time.

[0052] In a possible implementation, the electronic device performs word embedding processing on the second behavior data by using a word embedding model or embedding data processing to obtain a word vector of the second behavior data. The word vector of the second behavior data is accumulated and pooled to obtain feature information of the second behavior data. Figure 4 As shown, after the behavior data is pooled, the characteristic information of the obtained behavior data is reduced in both dimension and parameter quantity. Since the short-term behavior data of users is often greatly affected by temporary factors such as hot spots, and the data fluctuates significantly in the short term, the word vector of the second behavior data is accumulated and pooled to obtain the characteristic information of the second behavior data. For example, the word vector of the user's behavior data within 7 days is accumulated and pooled to obtain the characteristic information of the user's behavior data within 7 days. Specifically, the user recently had a baby, so he began to pay attention to maternal and child information, so the characteristic information of the user's behavior data within 7 days includes maternal and child label information. In this way, the user's short-term behavior data is accumulated and pooled, so that the changes in the user's interests caused by short-term fluctuations will not be ignored because of the user's long-term behavior data, and the user's interest information can be reflected more comprehensively in real time, thereby improving the accuracy of the user's interest information.

[0053] In one possible implementation, the feature information of the first behavior data includes multiple first unit feature information, and the feature information of the second behavior data includes multiple second unit feature information. For example, the first behavior data is the user's behavior data within the past three months, and further, the first behavior data includes the user's behavior data within 90 days and the user's behavior data within 30 days; the second behavior data is the user's behavior data within the past week, and further, the second behavior data includes the user's behavior data within 7 days and the user's behavior data within 3 days. Specifically, the electronic device performs word embedding processing on the user's behavior data within 90 days. In this way, the first behavior data and the second behavior data are further divided, and the user's behavior data within a specified time period can be further analyzed, thereby improving the accuracy of the user's interest information.

[0054] For example, the electronic device performs word embedding processing on the user's behavior data within 90 days to obtain the word vector of the behavior data within 90 days, and performs average pooling processing on the word vector of the behavior data within 90 days to obtain the feature information of the behavior data within 90 days, which is the first unit feature information of the feature information of the first behavior data. The electronic device performs word embedding processing on the user's behavior data within 30 days to obtain the word vector of the behavior data within 30 days, and performs average pooling processing on the word vector of the behavior data within 30 days to obtain the feature information of the behavior data within 30 days, which is the first unit feature information of the feature information of the first behavior data. For another example, the electronic device performs word embedding processing on the user's behavior data within 7 days to obtain the word vector of the behavior data within 90 days, and performs average pooling processing on the word vector of the behavior data within 7 days to obtain the feature information of the behavior data within 7 days, which is the second unit feature information of the feature information of the second behavior data. The electronic device performs word embedding processing on the user's behavior data within 3 days to obtain the word vector of the behavior data within 3 days, and performs average pooling processing on the word vector of the behavior data within 3 days to obtain the feature information of the behavior data within 3 days, which is the second unit feature information of the feature information of the second behavior data. It should be noted that the first behavior data mentioned in the embodiment of the present application includes two first unit feature information and the second behavior data includes two second unit feature information for illustrative purposes only. The present invention does not specifically limit the number of unit feature information into which the first behavior data and the second behavior data are specifically divided. In addition, the time window lengths between the unit feature information included in the first behavior data and the second behavior data may be the same or different.

[0055] Step S330: Fusing the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determining the user's interest information based on the fused feature information.

[0056] In a possible implementation, the feature information of the first behavior data includes a plurality of first unit feature information, and the feature information of the second behavior data includes a plurality of second unit feature information. Then, the electronic device fuses the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, including: first, the electronic device fuses each second unit feature information of the plurality of second unit feature information to each first unit feature information of the plurality of first unit feature information through an attention mechanism to obtain a plurality of unit fused feature information corresponding to the plurality of first unit feature information; then, the electronic device splices the plurality of unit fused feature information to obtain fused feature information.

[0057] For example, the first behavior data is divided into two groups of behavior data (behavior data within 90 days and behavior data within 30 days), and the second behavior data is divided into two groups of behavior data (behavior data within 7 days and behavior data within 3 days) as an example for specific description. The electronic device uses the attention mechanism to take the feature information of the first behavior data as the key and the feature information of the second behavior data as the query. Specifically, the electronic device performs attention calculation on the feature information of the behavior data within 7 days and the feature information of the behavior data within 3 days to the feature information of the behavior data within 90 days, and obtains the feature information of the behavior data within 90 days after processing (the second unit feature information); the electronic device performs attention calculation on the feature information of the behavior data within 7 days and the feature information of the behavior data within 3 days to the feature information of the behavior data within 30 days, and obtains the feature information of the behavior data within 30 days after processing (the second unit feature information). Then, the electronic device performs splicing processing on the feature information of the behavior data within 90 days after processing and the feature information of the behavior data within 30 days after processing to obtain the fused feature information.

[0058] In a possible implementation, the electronic device obtains attribute information of the user, and performs word embedding processing on the attribute information of the user to obtain feature information of the attribute information of the user, wherein the attribute information of the user refers to the basic attributes of the user, such as age, gender, etc.

[0059] In a possible implementation, the electronic device determines the user's interest information based on the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information. Specifically, the electronic device performs splicing processing on the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information to obtain spliced ​​feature information; the electronic device inputs the spliced ​​feature information into the interest recognition network model to output the user's interest information.

[0060] For example, the electronic device uses the splicing feature information as the input of the interest recognition network model, and finally outputs the user's interest information through processing by the interest recognition network model. Among them, the interest recognition network model can be based on a deep learning neural network model, specifically a feed forward neural network (FNN) model or a recurrent neural network (RNN) model or a long shortterm memory (LSTM) model and other neural network models with natural language processing and deep learning capabilities, and the present invention is not limited to this.

[0061] Among them, the interest recognition network model may include a hidden layer and an output layer. The hidden layer is characterized by being fully connected and black box. The interest recognition network model uses a nonlinear activation function to prevent overfitting during training and improve generalization ability. In a possible implementation, relu can alleviate the saturation that occurs when the value of y is too large, and at the same time prevent overfitting and accelerate training. Therefore, the relu function is selected as the activation function of this solution; in order to prevent the occurrence of overfitting (that is, the error of the test set increases due to being too close to the true distribution of the training set), a regularization norm penalty term is added. Considering the stability and easy saturation characteristics of the sigmoid function, the output layer of this solution can select the sigmoid function as the output function.

[0062] Through the data processing method provided in the embodiment of the present application, the user's historical behavior data is divided into first behavior data and second behavior data, and feature extraction processing is performed on the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data. Then, the feature information of the first behavior data and the feature information of the second behavior data are fused through the attention mechanism, which can enhance the correlation between the user's long-term and short-term behavior data, more comprehensively reflect the user's interest information, and thus obtain more accurate user interest information.

[0063] See also Figure 5 , Figure 5 is a flow chart of another data processing method provided in an embodiment of the present application. The method is applied to electronic devices, such as Figure 5 As shown, the data processing method may include steps S510 to S560. Among them:

[0064] Step S510: Obtain the user's historical behavior data, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period.

[0065] Step S520: extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data.

[0066] Step S530: Fusing the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determining the user's interest information based on the fused feature information.

[0067] It should be noted that, for details of steps S510-S530 in the embodiment of the present application, reference may be made to the execution process of the electronic device of steps S310-S330 in the above embodiment, and the embodiment of the present application will not be repeated here.

[0068] Step S540: obtaining the third behavior data of the user, where the third behavior data is behavior data within a third preset time period, and the time length from the third preset time period to the current time period does not exceed the preset time length.

[0069] Specifically, the user's third behavior data refers to the user's behavior data in a short period of time, and the third behavior data is not included in the first behavior data and the second behavior data. Among them, the preset time length can be 3 days, and the third time period can be 1 day or 2 days. For example, the user's third behavior data is the behavior data within 1 day.

[0070] Step S550: extracting features from the third behavior data to obtain feature information of the third behavior data.

[0071] In one possible implementation, first, the electronic device performs word embedding processing on the third behavior data to obtain a word vector of the third behavior data; then, the electronic device performs accumulation pooling processing on the word vector of the third behavior data to obtain feature information of the third behavior data.

[0072] For example, assuming that the third behavior data is behavior data within 1 day, first, the electronic device performs word embedding processing on the user's behavior data within 1 day through a word embedding model or through embedding data processing to obtain the word vector of the behavior data within 1 day; then, since the third behavior data is shorter-term behavior data, the electronic device performs pooling processing on the third behavior data through cumulative pooling. Specifically, the electronic device performs cumulative pooling processing on the word vector of the behavior data within 1 day to obtain feature information of the behavior data within 1 day.

[0073] Step S560: Determine the accuracy of the interest information according to the interest information and the characteristic information of the third behavior data, and the accuracy is used to adjust the parameters of the interest recognition network model.

[0074] In one possible implementation, first, the electronic device fuses the interest information and the feature information of the third behavior data through an attention mechanism to obtain target feature information; then, the electronic device calculates the interaction probability between the feature information of the first behavior data and the feature information of the second behavior data and the target feature information based on the target feature information, the feature information of the first behavior data, the feature information of the second behavior data, and the feature information of the third behavior data; finally, the electronic device determines the accuracy of the interest information based on the interaction probability.

[0075] For example, first, the electronic device uses the attention mechanism to take the user's interest information as the key and the feature information of the third behavior data as the query. Specifically, the electronic device performs attention calculation on the feature information of the user's behavior data within 1 day and the user's interest information within 90 days to obtain the target feature information. Then, assuming that the attention calculation result (i.e., the target feature information) is v, the feature information of the user's first behavior data is x1, the feature information of the user's second behavior data is x2, and the feature information of the third behavior data is y, then the interaction probability between the behavior objects based on the user's first behavior data, second behavior data, and third behavior data is P = exp(v·y) / (exp(v·x1)+exp(v·x2)). Finally, the electronic device determines the accuracy of the interest information based on the interaction probability P, loss = -logP.

[0076] Among them, the accuracy is used to adjust the parameters of the interest recognition network model. Specifically, if the accuracy is greater than the preset degree threshold, it is determined that the user's interest information is accurate. If the accuracy is less than the preset degree threshold, it is determined that the user's interest information is inaccurate, and the activation function can be replaced or the number of network channels can be increased. For example, when the calculated accuracy is greater than 80%, it is determined that the user's interest information is accurate, and the user's interest information obtained by the interest recognition network model can accurately describe the user's interest. Later, the interest information can be stored locally or used as training data for the recommendation algorithm model. If the calculated accuracy is less than 80%, it is determined that the user's interest information is inaccurate, and the user's interest information obtained by the interest recognition network model cannot accurately describe the user's interest. Therefore, the interest recognition network model needs to be optimized, which can be specifically to obtain the user's historical behavior data in a longer period of time, such as obtaining the user's historical behavior data within one year; it can also be to divide the user's historical behavior data more finely, such as dividing the first behavior data into 5 groups of first unit behavior data and the second behavior data into 5 groups of second unit behavior data; it can also be to increase the number of channels of the hidden layer in the interest recognition network model, etc.

[0077] In one possible implementation, after the electronic device determines that the user's interest information is accurate, it obtains preset feature data in a preset database and calculates the similarity between the preset feature data and the interest information, wherein the category corresponding to the preset feature data is a preset category; if the similarity between the preset feature data and the user's interest information is greater than a preset degree threshold, the user's category is determined to be the preset category.

[0078] For example, the preset database stores interest tag information (preset feature data) and population packages corresponding to the interest tag information (preset feature data). After the electronic device determines that the interest information is accurate, the electronic device calculates the similarity between the interest information and the preset feature data. If the similarity between the interest information and the preset feature data is greater than 80%, it is determined to classify the user into the population package corresponding to the preset feature data. In this way, users with similar interest information can be clustered and divided into user population packages.

[0079] Through the method provided in the embodiment of the present application, the user's third behavior data and the user's interest information are processed through the attention mechanism, the accuracy of the user's interest information can be obtained, and the parameters of the interest recognition network model are adjusted according to the accuracy, thereby further improving the accuracy of the user's interest information.

[0080] See also Figure 6 , Figure 6 is a schematic diagram of the structure of a data processing device provided in an embodiment of the present application. The data processing device is used to execute Figure 3 to Figure 5 In the steps performed by the electronic device in the corresponding method embodiment, the data processing device may include:

[0081] An acquisition unit 610 is configured to acquire historical behavior data of a user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period;

[0082] A feature extraction unit 620, configured to extract features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data;

[0083] The processing unit 630 is used to fuse the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, and determine the user's interest information according to the fused feature information.

[0084] In a possible implementation, the processing unit 630 determines the user's interest information according to the fused feature information, including:

[0085] Acquire attribute information of the user, and perform word embedding processing on the attribute information of the user to obtain feature information of the attribute information of the user;

[0086] The user's interest information is determined according to the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information.

[0087] In a possible implementation, the processing unit 630 determines the user's interest information according to the fusion feature information, the feature information of the second behavior data, and the feature information of the user's attribute information, including:

[0088] Inputting the fusion feature information, the feature information of the second behavior data, and the feature information of the user's attribute information into a splicing module of an interest recognition network model for splicing processing to obtain splicing feature information;

[0089] The spliced ​​feature information is input into the attention module of the interest recognition network model to output the user's interest information.

[0090] In a possible implementation, the feature information of the first behavior data includes a plurality of first unit feature information, and the feature information of the second behavior data includes a plurality of second unit feature information; the processing unit 630 fuses the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, including:

[0091] By using an attention mechanism, each second unit feature information in the plurality of second unit feature information is fused with each first unit feature information in the plurality of first unit feature information to obtain a plurality of unit fused feature information corresponding to the plurality of first unit feature information;

[0092] The plurality of unit fusion feature information are concatenated to obtain fusion feature information.

[0093] In a possible implementation, the feature extraction unit 620 extracts features from the first behavior data and the second behavior data respectively, and obtains feature information of the first behavior data and feature information of the second behavior data including:

[0094] Performing word embedding processing on the first behavior data to obtain word vectors of the first behavior data, and performing average pooling processing on the word vectors of the first behavior data to obtain feature information of the first behavior data;

[0095] Perform word embedding processing on the second behavior data to obtain word vectors of the second behavior data, and perform accumulation pooling processing on the word vectors of the second behavior data to obtain feature information of the second behavior data.

[0096] In a possible implementation manner, after the processing unit 630 determines the user's interest information according to the fused feature information, the processing further includes:

[0097] The acquisition unit 610 acquires third behavior data of the user, where the third behavior data is behavior data within a third preset time period, and the time length from the third preset time period to the current time period does not exceed a preset time length;

[0098] The feature extraction unit 620 extracts features from the third behavior data to obtain feature information of the third behavior data;

[0099] The processing unit 630 determines the accuracy of the interest information according to the interest information and the characteristic information of the third behavior data, and the accuracy is used to adjust the parameters of the interest identification network model.

[0100] In a possible implementation, the processing unit 630 determines the accuracy of the interest information according to the interest information and the characteristic information of the third behavior data, including:

[0101] The interest information and the feature information of the third behavior data are fused through an attention mechanism to obtain target feature information;

[0102] Calculating the interaction probability between the feature information of the first behavior data, the feature information of the second behavior data, and the target feature information according to the target feature information, the feature information of the first behavior data, the feature information of the second behavior data, and the feature information of the third behavior data;

[0103] The accuracy of the interest information is determined according to the interaction probability.

[0104] Through the data processing device provided in the embodiment of the present application, the user's historical behavior data is divided into first behavior data and second behavior data, and feature extraction processing is performed on the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data. Then, the feature information of the first behavior data and the feature information of the second behavior data are fused through the attention mechanism, which can enhance the correlation between the user's long-term and short-term behavior data, more comprehensively reflect the user's interest information, and thus obtain more accurate user interest information.

[0105] See also Figure 7 , Figure 7 is a schematic diagram of a structure of an electronic device provided in an embodiment of the present application, the electronic device is used to perform Figure 3 to Figure 5The steps performed by the electronic device in the corresponding method embodiment include: one or more processors 710; one or more input devices 720, one or more output devices 730 and a memory 740. The processor 710, the input device 720, the output device 730 and the memory 740 are connected via a bus 750. The memory 720 is used to store a computer program, and the computer program includes program instructions. The processor 710 is used to execute the program instructions stored in the memory 740 to perform the following operations:

[0106] Acquire historical behavior data of the user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period;

[0107] Extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data;

[0108] The feature information of the first behavior data and the feature information of the second behavior data are fused to obtain fused feature information, and the interest information of the user is determined according to the fused feature information.

[0109] In a possible implementation manner, the processor 710 determines the user's interest information according to the fused feature information, including:

[0110] Acquire attribute information of the user, and perform word embedding processing on the attribute information of the user to obtain feature information of the attribute information of the user;

[0111] The user's interest information is determined according to the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information.

[0112] In a possible implementation manner, the processor 710 determines the interest information of the user according to the fusion feature information, the feature information of the second behavior data, and the feature information of the attribute information of the user, including:

[0113] Inputting the fusion feature information, the feature information of the second behavior data, and the feature information of the user's attribute information into a splicing module of an interest recognition network model for splicing processing to obtain splicing feature information;

[0114] The spliced ​​feature information is input into the attention module of the interest recognition network model to output the user's interest information.

[0115] In a possible implementation, the feature information of the first behavior data includes a plurality of first unit feature information, and the feature information of the second behavior data includes a plurality of second unit feature information; the processor 710 fuses the feature information of the first behavior data with the feature information of the second behavior data to obtain fused feature information, including:

[0116] By using an attention mechanism, each second unit feature information in the plurality of second unit feature information is fused with each first unit feature information in the plurality of first unit feature information to obtain a plurality of unit fused feature information corresponding to the plurality of first unit feature information;

[0117] The plurality of unit fusion feature information are concatenated to obtain fusion feature information.

[0118] In a possible implementation, the processor 710 performs feature extraction on the first behavior data and the second behavior data respectively, and obtains feature information of the first behavior data and feature information of the second behavior data including:

[0119] Performing word embedding processing on the first behavior data to obtain word vectors of the first behavior data, and performing average pooling processing on the word vectors of the first behavior data to obtain feature information of the first behavior data;

[0120] Perform word embedding processing on the second behavior data to obtain word vectors of the second behavior data, and perform accumulation pooling processing on the word vectors of the second behavior data to obtain feature information of the second behavior data.

[0121] In a possible implementation manner, after the processor 710 determines the user's interest information according to the fused feature information, the further step includes:

[0122] Acquire third behavior data of the user, where the third behavior data is behavior data within a third preset time period, and the distance between the third preset time period and the current time period does not exceed a preset time period;

[0123] Extracting features from the third behavior data to obtain feature information of the third behavior data;

[0124] The accuracy of the interest information is determined according to the interest information and the characteristic information of the third behavior data, and the accuracy is used to adjust the parameters of the interest identification network model.

[0125] In a possible implementation manner, the processor 710 determines the accuracy of the interest information according to the interest information and the characteristic information of the third behavior data, including:

[0126] The interest information and the feature information of the third behavior data are fused through an attention mechanism to obtain target feature information;

[0127] Calculating the interaction probability between the feature information of the first behavior data, the feature information of the second behavior data, and the target feature information according to the target feature information, the feature information of the first behavior data, the feature information of the second behavior data, and the feature information of the third behavior data;

[0128] The accuracy of the interest information is determined according to the interaction probability.

[0129] Through the electronic device provided by the embodiment of the present application, the user's historical behavior data is divided into first behavior data and second behavior data, and feature extraction processing is performed on the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data. Then, the feature information of the first behavior data and the feature information of the second behavior data are fused through the attention mechanism, which can enhance the correlation between the user's long-term and short-term behavior data, more comprehensively reflect the user's interest information, and thus obtain more accurate user interest information.

[0130] According to one aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the operations involved in the above-mentioned data processing method.

[0131] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the process of the embodiment of the above-mentioned data processing method. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0132] The above disclosure is only part of the embodiments of the present invention, which certainly cannot be used to limit the scope of the rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire historical behavior data of the user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period; Extracting features from the first behavior data and the second behavior data respectively to obtain feature information of the first behavior data and feature information of the second behavior data; the feature information of the first behavior data includes a plurality of first unit feature information, and the feature information of the second behavior data includes a plurality of second unit feature information; By using an attention mechanism, each second unit feature information in the plurality of second unit feature information is fused with each first unit feature information in the plurality of first unit feature information to obtain a plurality of unit fused feature information corresponding to the plurality of first unit feature information; The plurality of unit fusion feature information are spliced ​​to obtain fusion feature information, and the user's interest information is determined according to the fusion feature information; Acquire third behavior data of the user, and perform feature extraction on the third behavior data to obtain feature information of the third behavior data; the third behavior data is behavior data within a third preset time period, and the third preset time period is no longer than a preset time period from the current time period; The interest information and the feature information of the third behavior data are fused through an attention mechanism to obtain the accuracy of the interest information.

2. The method according to claim 1, characterized in that The determining the user's interest information according to the fused feature information includes: Acquire attribute information of the user, and perform word embedding processing on the attribute information of the user to obtain feature information of the attribute information of the user; The user's interest information is determined according to the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information.

3. The method according to claim 2, characterized in that The determining the user's interest information according to the fused feature information, the feature information of the second behavior data, and the feature information of the user's attribute information includes: Inputting the fusion feature information, the feature information of the second behavior data, and the feature information of the user's attribute information into a splicing module of an interest recognition network model for splicing processing to obtain splicing feature information; The spliced ​​feature information is input into the attention module of the interest recognition network model to output the user's interest information.

4. The method according to claim 1, characterized in that The extracting features of the first behavior data and the second behavior data to obtain feature information of the first behavior data and feature information of the second behavior data respectively includes: Performing word embedding processing on the first behavior data to obtain word vectors of the first behavior data, and performing average pooling processing on the word vectors of the first behavior data to obtain feature information of the first behavior data; Perform word embedding processing on the second behavior data to obtain word vectors of the second behavior data, and perform accumulation pooling processing on the word vectors of the second behavior data to obtain feature information of the second behavior data.

5. The method according to claim 3, characterized in that: The extracting features from the third behavior data to obtain feature information of the third behavior data includes: Perform word embedding processing on the third behavior data to obtain word vectors of the third behavior data, and perform accumulation pooling processing on the word vectors of the third behavior data to obtain feature information of the third behavior data.

6. The method according to claim 5, characterized in that The fusing the interest information and the feature information of the third behavior data through the attention mechanism to obtain the accuracy of the interest information includes: The interest information and the feature information of the third behavior data are fused through an attention mechanism to obtain target feature information; Calculating the interaction probability between the feature information of the first behavior data, the feature information of the second behavior data, and the target feature information according to the target feature information, the feature information of the first behavior data, the feature information of the second behavior data, and the feature information of the third behavior data; The accuracy of the interest information is determined according to the interaction probability, and the accuracy is used to adjust parameters of the interest identification network model.

7. A data processing device, characterized in that: include: an acquisition unit, configured to acquire historical behavior data of a user, wherein the historical behavior data includes first behavior data and second behavior data, the first behavior data is behavior data within a first preset time period, the second behavior data is behavior data within a second preset time period, and the first preset time period is greater than the second preset time period; a feature extraction unit, configured to extract features from the first behavior data and the second behavior data respectively, to obtain feature information of the first behavior data and feature information of the second behavior data; the feature information of the first behavior data includes a plurality of first unit feature information, and the feature information of the second behavior data includes a plurality of second unit feature information; a processing unit, configured to fuse each second unit feature information in the plurality of second unit feature information with each first unit feature information in the plurality of first unit feature information through an attention mechanism, to obtain a plurality of unit fused feature information corresponding to the plurality of first unit feature information; The processing unit is further used to perform splicing processing on the plurality of unit fusion feature information to obtain fusion feature information, and determine the user's interest information according to the fusion feature information; The processing unit is further used to obtain third behavior data of the user, and perform feature extraction on the third behavior data to obtain feature information of the third behavior data; the third behavior data is behavior data within a third preset time period, and the third preset time period is no longer than a preset time period from the current time period; The processing unit is further used to fuse the interest information and the feature information of the third behavior data through an attention mechanism to obtain the accuracy of the interest information.

8. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a set of program codes, and the processor calls the program codes stored in the memory to execute any one of the methods described in 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is enabled to perform the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises computer instructions, and the computer instructions are suitable for being loaded by a processor and executing the data processing method according to any one of claims 1 to 6.

Images