CN112288554A

CN112288554A - Commodity recommendation method and device, storage medium and electronic device

Info

Publication number: CN112288554A
Application number: CN202011352471.0A
Authority: CN
Inventors: 钟子宏
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-11-27
Filing date: 2020-11-27
Publication date: 2021-01-29
Anticipated expiration: 2040-11-27
Also published as: CN112288554B

Abstract

The invention discloses a commodity recommendation method and device based on artificial intelligence, a storage medium and an electronic device. Wherein, the method comprises the following steps: acquiring target account data of a target account in a target time period, wherein the current probability parameter of the target account in the current time period and the first probability parameter of the target account in a first time period are acquired; inputting the target account data, the first probability parameter and the current probability parameter into a target neural network model, wherein the target neural network model is used for outputting a recognition result, and the recognition result is used for indicating the probability that a target commodity is clicked by a target account under the condition that the target commodity is pushed to the target account within a target time period; and determining to push the target commodity to the target account in the target time period or forbidding to push the target commodity to the target account in the target time period according to the identification result. The invention solves the technical problem of low accuracy of recommending commodities.

Description

Commodity recommendation method and device, storage medium and electronic device

Technical Field

The invention relates to the field of computers, in particular to a commodity recommendation method and device, a storage medium and an electronic device.

Background

In the prior art, in the process of determining a commodity to be pushed to a user, a training sample is usually used to train a neural network model, and then user data is input into the trained neural network model, and a result of prediction is given by the neural network model to determine whether to push the commodity to the user.

However, in the above process, the influence of the multi-stage state transition on the prediction result is not considered in the training and using process of the neural network model, so that the problem of inaccurate commodity recommendation is caused.

In view of the above problems, no effective solution has been proposed.

Disclosure of Invention

The embodiment of the invention provides a commodity recommendation method and device, a storage medium and an electronic device, and at least solves the technical problem of low commodity recommendation accuracy.

According to an aspect of an embodiment of the present invention, there is provided a commodity recommendation method including: acquiring target account data of a target account in a target time period, wherein the target account data comprises a current probability parameter of the target account in the current time period and a first probability parameter of the target account in a first time period, the target account data comprises target behavior characteristics of the target account and target commodity information of a target commodity which is expected to be pushed to the target account, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through the first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period; inputting the target account data, the first probability parameter, and the current probability parameter into a target neural network model, wherein the target neural network model is a model trained using sample data, the sample data includes a second probability parameter of the target account, the first probability parameter, the current account data, and the first account data, the second probability parameter is obtained from second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used to indicate a probability that the target account clicks on the target commodity when the target commodity is pushed to the target account in the target time period; and determining to push the target commodity to the target account in the target time period or forbidding to push the target commodity to the target account in the target time period according to the identification result.

According to another aspect of the embodiments of the present invention, there is also provided a commodity recommending apparatus including: a first obtaining unit, configured to obtain target account data of a target account in a target time period, where the target account data includes a current probability parameter of the target account in the current time period and a first probability parameter of the target account in a first time period, where the target account data includes target behavior characteristics of the target account and target commodity information of a target commodity to be pushed to the target account, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period; an input unit configured to input the target account data, the first probability parameter, the current probability parameter, and the target neural network model into a target neural network model, where the target neural network model is a model trained using sample data, the sample data includes a second probability parameter of the target account, the first probability parameter, the current account data, and the first account data, the second probability parameter is obtained from second account data of the target account in a second time period, the second time period is one time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used to indicate a probability that the target account clicks the target commodity when the target commodity is pushed to the target account in the target time period; a first determining unit, configured to determine, according to the recognition result, whether to push the target product to the target account within the target time period or to prohibit pushing the target product to the target account within the target time period.

As an optional example, the apparatus further includes: a third obtaining unit, configured to obtain a target probability parameter of the target account if the target commodity is pushed to the target account within the target time period after determining, according to the recognition result, that the target commodity is pushed to the target account within the target time period or after prohibiting the target commodity from being pushed to the target account within the target time period, where the target probability parameter is obtained according to a target number of times that the target account clicks on the target commodity within the target time period and a total number of times that the target commodity is pushed to the target account within the target time period; a predicting unit, configured to predict, using the target probability parameter, a target result of pushing the target product to the target account within a third time period after the target time period; a fifth determining unit, configured to determine, according to the target result, whether to push the target product to the target account within the third time period or to prohibit the target product from being pushed to the target account within the third time period.

According to still another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the above-mentioned article recommendation method when running.

According to still another aspect of the embodiments of the present invention, there is also provided an electronic device, including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the above-mentioned article recommendation method through the computer program.

In the embodiment of the present invention, target account data of a target account in a target time period is obtained, where the target account data includes a current probability parameter of the target account in the current time period and a first probability parameter of the target account in a first time period, where the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained by first account data of the target account in the first time period, the current probability parameter is obtained by current account data of the target account in the current time period, and the target account data, the first probability parameter, and the current probability parameter are input into a target neural network model, the target neural network model is a model trained by using sample data, the sample data includes a second probability parameter of the target account, the first probability parameter, the current account data, and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used to indicate a probability that the target account clicks the target commodity when the target commodity is pushed to the target account in the target time period; and determining a method for pushing the target commodity to the target account in the target time period or forbidding pushing the target commodity to the target account in the target time period according to the identification result. According to the method, in the process of training the neural network model, the influence of the multi-stage state transition on the pushed commodity is considered, and meanwhile, in the using process of the model, the influence of the multi-stage state transition on the pushed commodity is also considered, so that the identification accuracy of the model is improved, the commodity recommendation accuracy is further improved, and the technical problem of low commodity recommendation accuracy is solved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the invention without limiting the invention. In the drawings:

FIG. 1 is a schematic diagram of an application environment of an alternative merchandise recommendation method according to an embodiment of the invention;

FIG. 2 is a schematic diagram of an application environment of an alternative merchandise recommendation method according to an embodiment of the invention;

FIG. 3 is a flow chart of an alternative merchandise recommendation method according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a user interface of an alternative merchandise recommendation method according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an algorithm structure of a model of an alternative merchandise recommendation method according to an embodiment of the present invention;

FIG. 6 is a block diagram of an alternative merchandise recommendation method according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of an alternative merchandise recommendation device according to an embodiment of the present invention;

FIG. 8 is a schematic diagram of another alternative merchandise recommendation device according to an embodiment of the present invention

FIG. 9 is a schematic structural diagram of another alternative merchandise recommendation device according to an embodiment of the present invention

Fig. 10 is a schematic structural diagram of an alternative electronic device according to an embodiment of the invention.

Detailed Description

In order to make the technical solutions of the present invention better understood, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

It should be noted that the terms "first," "second," and the like in the description and claims of the present invention and in the drawings described above are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the invention described herein are capable of operation in sequences other than those illustrated or described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

Artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

The embodiment of the application relates to artificial intelligence Machine Learning, wherein Machine Learning (ML) is a multi-field cross subject and relates to a plurality of subjects such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

According to an aspect of the embodiments of the present invention, there is provided a behavior recognition method, which may be applied to, but not limited to, the environment shown in fig. 1 as an optional implementation manner.

As shown in fig. 1, the terminal device 102 includes a memory 104 for storing various data generated during the operation of the terminal device 102, a processor 106 for processing and operating the various data, and a display 108 for displaying recommended goods. An application client may be run on the terminal device 102. Terminal device 102 may interact with server 112 via network 110. The terminal device 102 may send each item of acquired data of the target account to the server 112 through S102, and the server 112 may configure the target neural network model, give a recognition result through S104 by the target neural network model, and then issue the recognition result to the terminal device 102 through S106, so that the terminal device 102 determines whether to push the target product. The server 112 includes a database 114 for storing various data and a processing engine 116 for deploying the above-described target neural network model. The terminal device 102 may recommend display of the target commodity in the target time period or not display the target commodity according to the recognition result.

As an alternative embodiment, the behavior recognition method described above may be applied, but not limited to, in an environment as shown in fig. 2.

As shown in fig. 2, the terminal device 202 includes a memory 204 for storing various data generated during the operation of the terminal device 202, a processor 206 for processing and operating the various data, and a display 208 for recommending and displaying the target product. The terminal device 202 may obtain the target account data, the current probability parameter, and the first probability parameter of the user through steps S302 to S306, input the data into the target neural network model arranged in the terminal device 202, obtain an output result of the target neural network model, and recommend to display the target commodity or not to display the target commodity according to the output result.

Optionally, in this embodiment, the terminal device 102 or the terminal device 202 may be a terminal device configured with a target client, and may include, but is not limited to, at least one of the following: mobile phones (such as Android phones, iOS phones, etc.), notebook computers, tablet computers, palm computers, MID (Mobile Internet Devices), PAD, desktop computers, smart televisions, etc. The target client may be a video client, an instant messaging client, a browser client, an educational client, etc. Such networks may include, but are not limited to: a wired network, a wireless network, wherein the wired network comprises: a local area network, a metropolitan area network, and a wide area network, the wireless network comprising: bluetooth, WIFI, and other networks that enable wireless communication. The server may be a single server, a server cluster composed of a plurality of servers, or a cloud server. The above is merely an example, and this is not limited in this embodiment.

Optionally, as an optional implementation manner, as shown in fig. 3, the method for recommending a commodity includes:

s302, acquiring target account data of a target account in a target time period, wherein the target account data comprises target behavior characteristics of the target account and target commodity information of a target commodity which is expected to be pushed to the target account, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through the first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period;

s304, inputting the target account data, a first probability parameter and a current probability parameter into a target neural network model, wherein the target neural network model is a model trained by using sample data, the sample data comprises a second probability parameter of the target account, the first probability parameter, the current account data and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is used for outputting a recognition result, and the recognition result is used for indicating the probability that the target account clicks a target commodity under the condition that the target commodity is pushed to the target account in the target time period;

s306, the target commodity is pushed to the target account within the target time period or the target commodity is forbidden to be pushed to the target account within the target time period according to the identification result.

Alternatively, the above-mentioned product recommendation method may be, but is not limited to, applied to a process of recommending a product. The specific application scenario is not limiting. For example, when the user opens the application program or uses the application program, the product recommendation method in the embodiment of the present application may be used to recommend a product. The application type is not limited. For example, the system can be a shopping application, or a mobile banking application, or even an application playing multimedia resources. Account data of the target account of the user can be acquired, and whether to recommend the target commodity to the target account in the third time period is determined.

It should be noted that the target product in the embodiment of the present application is not limited to the entity product that can be recommended on the network, the product information or the link of the entity product may be recommended when recommending, and the target product may also be a resource, such as any one or more of a fund security, a coupon, a membership, an exemption eligibility, and a gift. If the target commodity is the membership of an application program, the user can download the corresponding application program to obtain the membership. That is, the present application does not limit the specific contents of the target product.

Alternatively, the click may be to view details or to download or jump to a link or to purchase or join a shopping cart, etc. If the user clicks on the goods, it indicates that the user is interested in the goods, and if the user does not click on the recommended goods, it may indicate that the user is not interested in the goods.

Optionally, in this embodiment of the application, the current time period is a time period in which the user is using the target account, the current time period may include a first time period and a second time period before the current time period, and the durations of any two time periods of the first time period, the second time period, the current time period, and the target time period may be the same or different. For example, the first time period may be one day and the second time period may be one hour. Or the first time period may be one day, the second time period one hour, and the current time period one day. The specific duration of the time period can be flexibly configured.

Optionally, the target account data may include behavior characteristics of the target account and commodity information of a target commodity pushed to the target account within a target time period, where the behavior characteristics may include an operation behavior of the target account, and the target account data may also include commodity information of the target commodity pushed to the target account within the target time period. The merchandise information may include identification, type, value, etc. information of the merchandise.

The first commodity information, the second commodity information, the current commodity information and the target commodity information refer to commodity information of commodities recommended to the target account in each time period (the target commodity information is pre-recommended and is not yet recommended), and the commodities recommended to the target account in each time period can be the same or different.

For example, in the process of recommending the coupon to the target account, the information of the coupon pushed to the target account in the current time period, such as 2020/11/1/12:00:00 to 2020/11/1/14:00:00, and the information of the target account in the current time period may be acquired. The probability of the user clicking on the recommended commodity in the last time period, such as 2020/11/1/8:00:00 to 2020/11/1/12:00:00, and the probability of the user clicking on the recommended commodity in the current time period are obtained, and then the obtained two probabilities and the behavior characteristics and the coupon information of the target account in the current time period are input into the target neural network model. And giving a recognition result by the target neural network model, wherein the recognition result is a numerical value, and predicting the target probability of the target account clicking the target commodity if the target commodity is pushed to the target account in the target time period according to the numerical value.

And judging whether to recommend the target commodity to the user in the target time period according to the target probability. Because the result of the recommendation is not good if the user is not interested. In the process, the original neural network model is trained by using the sample data in advance to obtain the target neural network model, and the click probability parameters of the first time period and the second time period before the current time period are used in the training sample, so that the training effect of the target neural network model is improved, and the identification accuracy of the target neural network model is improved. In the prediction process, the target account data of the target account in the target time period is acquired, and meanwhile, the click probability parameter of the current time period and the click probability parameter of the first time period are also acquired, so that the accuracy of the prediction result can be improved, and the accuracy of recommending the target commodity is improved.

As an optional example, before inputting the target account data and the first probability parameter, the current probability parameter into the target neural network model, the method further comprises:

acquiring historical data of a target account;

determining account behavior characteristics of a target account in the historical data within a current time period as current behavior characteristics, and determining commodity information pushed to the target account within the current time period as current commodity information;

determining behavior characteristics of a target account in the historical data in a first time period as first behavior characteristics, and determining information of commodities pushed to the target account in the first time period as first commodity information;

acquiring a first probability parameter and a second probability parameter;

and training the original neural network model by using the current behavior characteristic, the current commodity information, the first behavior characteristic, the first commodity information, the first probability parameter and the second probability parameter to obtain a target neural network model.

Optionally, the history record may include behavior characteristics of the target account in each time period, commodity information of the commodity pushed by the target account in each time period, and a click record of the pushed commodity in each time period, where the click record may record whether the pushed commodity is clicked or not. After the history is obtained, the current behavior feature, the current commodity information, the first behavior feature, the first commodity information, the first parameter probability and the second parameter probability are obtained.

Optionally, the first probability parameter may be used to indicate a probability that the user clicks on the pushed commodity when the user clicks on the pushed commodity within the first time period, for example, the probability that the user may click on the pushed commodity within the first time period is 80%. The second probability parameter may be used to indicate a probability that the user clicks on the pushed commodity when pushing the commodity within the second time period.

And training an original neural network model through the obtained current behavior characteristics, the current commodity information, the first behavior characteristics, the first commodity information, the first probability parameters and the second probability parameters to obtain a target neural network model.

In the method, in the process of training the model, the original neural network model is trained by using the current commodity information and the current behavior characteristics in the current account data of the target account, and meanwhile, the original neural network model is trained by using the first probability parameter of the first time period before the current time period and the second probability parameter of the second time period, so that the identification accuracy of the trained target neural network model is improved.

As an alternative example, obtaining the first probability parameter and the second probability parameter includes:

acquiring a first number of times that a target account clicks a target commodity pushed to the target account within a first time period and a second number of times that the target account clicks the target commodity pushed to the target account within a second time period

And determining a first probability parameter according to the first times and a first total number of times of pushing the target commodities to the target account in a first time period, and determining a second probability parameter according to a second time and a second total number of times of pushing the target commodities to the target account in a second time period.

Optionally, the first total number of times or the second total number of times is a total number of times of commodities pushed to the target account within the first time period or the second time period. If the commodity is pushed once, the same commodity is pushed for multiple times, and multiple times of pushing is carried out. For example, taking the first total number as an example, in the first time period, two commodities are pushed for the first time, and three commodities are pushed for the second time, so that the first total number is counted as 6 times.

The first frequency or the second frequency is the frequency of clicking the pushed target commodity by the target account. Taking the first number as an example, the multiple clicks on the same product are counted as the product pushed by the multiple clicks. If one commodity is pushed three times and clicked twice in the first time period, the first time number is 2. And clicking a plurality of commodities pushed once respectively, and counting for a plurality of times for the first time. If three commodities are pushed at a time and two commodities are clicked, the first output is 2. In the first time period, three commodities are pushed for the first time, two commodities are pushed for the second time, two commodities are clicked by the first target account, and one commodity is clicked by the second target account. The first input is 3 and the first total number is 5.

By determining the first probability parameter and the second probability parameter through the method, the probability that the user clicks the recommended commodity in the first time period and the probability that the user clicks the recommended commodity in the second time period can be accurately determined.

Optionally, the first probability parameter may be further divided into a first sub-probability parameter and a second sub-probability parameter, where the first sub-probability parameter represents a probability that the user clicks on the pushed commodity in the first time period, the second sub-probability parameter represents a probability that the user does not click on the pushed commodity in the first time period, the second probability parameter may be further divided into a third sub-probability parameter and a fourth sub-probability parameter, the third sub-probability parameter represents a probability that the user clicks on the pushed commodity in the second time period, and the fourth sub-probability parameter represents a probability that the user does not click on the pushed commodity in the second time period.

In the process of training the original neural network model by using the sample data to obtain the target neural network model, the method further comprises the following steps:

dividing the sample data into training data and test data according to a preset proportion;

performing the following operations on the original neural network model until the recognition accuracy of the original neural network model is greater than a first threshold:

training an original neural network model by using training data, determining a first model parameter of the original neural network model, testing the original neural network model under the first model parameter by using test data to obtain a test result output by the original neural network model under the first model parameter, and adjusting the first model parameter under the condition that the test result indicates that the recognition accuracy is less than or equal to a first threshold value.

Optionally, the division of the training data and the test data is randomly divided, and the division is not specifically divided according to the difference of the data. For example, there are 10 pieces of test data, and the test data may be divided according to a predetermined ratio, for example, into 8 pieces of training data and 2 pieces of test data, or into 6 pieces of training data and 4 pieces of test data. The specific content of the sample data does not influence the result of the division.

After dividing the sample data and the test data, the original neural network model may be trained using the sample data to obtain a first model parameter of the original neural network model. If it is desired to see if the original neural network model at this point has been trained, the trained original neural network model may be tested using the test data. And inputting the test data into the trained original neural network model to obtain a test result output by the original neural network model. The recognition accuracy of the trained original neural network model can be calculated through the output test result. If the recognition accuracy is less than or equal to the first threshold, such as 95%, the training of the original neural network model is required to be continued. Until the recognition accuracy of the original neural network model obtained by testing the test data is more than 95%. Through the embodiment, the identification accuracy of the target neural network model can be ensured, and the accuracy of recommended commodities is further improved.

As an optional example, after obtaining the sample data, the method further includes:

dividing the sample data into a first type sample and a second type sample according to a preset condition, wherein the first type sample is a single specific feature, and the second type sample is a continuous feature;

coding the first type sample to obtain a coded first type sample;

performing decorrelation processing, normalization processing and feature discretization processing on the second type sample to obtain a processed second type sample;

and determining the encoded first type samples and the processed second type samples as new sample data.

Alternatively, the sample data may be subjected to a data preprocessing operation, and then the original neural network model is trained by using the sample data after the preprocessing operation.

The pre-processing first includes classification of the sample data. The sample data can be divided into a first type sample and a second type sample according to the data type. The first type of sample may be a sparse type sample, the sample features are sparse type features, the second type of sample may be a dense type sample, and the sample features are dense type features. And (3) encoding the sparse type sample to obtain an encoded sample, performing decorrelation processing and normalization processing on the dense type sample, and finally performing characteristic discretization processing to obtain a processed sample. And training the original neural network model by using the processed samples. The embodiment can preprocess the sample data to improve the accuracy of training the original neural network model.

As an optional implementation manner, the determining, according to the recognition result, to push the target commodity to the target account within the target time period or to prohibit the target commodity from being pushed to the target account within the target time period includes:

acquiring an identification result;

determining to push the target commodity to the target account within the target time period under the condition that the identification result is larger than the target threshold value;

and determining that the target commodity is prohibited from being pushed to the target account within the target time period when the identification result is smaller than or equal to the target threshold value.

Alternatively, the recognition result may be a number, and may belong to 0-1, where a closer 1 indicates a higher probability that the user clicks on the pushed target product, and a closer 0 indicates a lower probability that the user clicks on the pushed product. By obtaining the recognition result of the target neural network model, it can be determined whether to recommend a target commodity to the target account in the target time period.

As an optional implementation manner, after determining that the target commodity is pushed to the target account within the target time period or is prohibited from being pushed to the target account within the target time period according to the recognition result, the method further includes:

under the condition that a target commodity is pushed to a target account within a target time period, acquiring a target probability parameter of the target account, wherein the target probability parameter is obtained according to the target times of the target account for clicking the target commodity within the target time period and the total times of pushing the target commodity to the target account within the target time period;

predicting a target result of pushing the target commodity to the target account within a third time period after the target time period by using the target probability parameters;

and determining to push the target commodity to the target account in the third time period or forbidding to push the target commodity to the target account in the third time period according to the target result.

Optionally, the third time period is a time period after the target time period. If the target commodity is recommended to the target account in the determined target time period, a target probability parameter of the target account, that is, the probability of the target account clicking the target commodity, needs to be acquired. May be obtained using the number of clicks compared to the total number of recommendations. After the target probability parameter is obtained, the parameter, the current probability parameter of the current time period, and the target account data of the target time period can be used to predict the click probability of the target account when the target commodity is pushed to the target account in the third time period. And determining whether the target commodity is pushed in the third time period according to the click probability.

The above-described product recommendation method is explained below with reference to specific examples.

Interpretation of terms: and (3) state transition: the state of this stage is often the result of the state of the previous stage and the decision of the previous stage. If the state Sk of the K stage and the decision uk (Sk) are given, the state Sk +1 of the K +1 stage is completely determined; first order state transition: the state of the stage is determined only by the state of the previous stage and the decision result of the previous stage, namely the state Sk +1 of the K +1 is only influenced by the state Sk of the K stage and the decision uk (Sk); second-order state transition: the state of the K +1 stage is only determined by the state of the K, K-1 stage and the result of the decision, i.e. the state Sk +1 of the K +1 is influenced by the states Sk and Sk-1 of the K, K-1 stage and the decisions uk (Sk), uk-1 (Sk-1); bayes second-order state transition: the state Sk +1 of K +1 is influenced by the states Sk, Sk-1 of stage K, K-1 and decisions uk (Sk), uk-1(Sk-1), and can be expressed as the following expression:

（1）

wherein α and β are parameters.

Sigmoid function: one class is defined as a function of the form:

（2）

wherein z is a parameter.

Two-classification LR algorithm: also called a logistic regression model, the linear regression model is converted into a probability prediction model by introducing a Sigmoid function into the linear regression model and mapping the continuous output values of the uncertain range of the linear regression into the range of (0, 1).

Gradient descent method of two-stage state transition: namely, gradient descent under the influence of second-order state transition probability in the gradient descent solving process; unbiased estimation: the mathematical expectation of the estimator is equal to the true value of the estimated parameter. And (3) recall ratio: predicting the proportion of correct regular data to real regular data; precision ratio: the proportion of the correct positive data to the positive data is predicted; AUC: the area under the ROC curve is a standard for judging the quality of the two-classification prediction model. Wherein, the abscissa of the ROC curve is the false case rate, and the ordinate is the true case rate.

And (4) estimating the CTR: click Through Rate (CTR), a term commonly used for internet advertisement, refers to the Click arrival Rate of web advertisement (photo advertisement/text advertisement/keyword advertisement/ranking advertisement/video advertisement, etc.), i.e. the actual number of clicks of the advertisement (strictly speaking, the number of hits to a target page) is divided by the advertisement presentation amount (Show content).

User characteristics: and (4) recording the behavior of the user in the business and refining the data of the behavior. The method comprises the following steps: clicking, collecting, paying amount, paying times, active duration, active days and the like of the user in the service.

The article is characterized in that: and (4) refining the attribute of the article and the data thereof. The method comprises the following steps: item click-through rate, collection, average payment amount (total payment amount per item/number of paid people), average active duration (total active time per item/number of active people), etc.

And (3) downloading a label: and clicking and downloading the coupons on the activity page by the user, wherein the clicking and downloading marks are 1, and the clicking non-downloading mark is 0.

According to the commodity recommendation method, the influence of the second-order state transition on the classification result is considered, the problem that the influence of the state transition on the classification effect is not considered in the traditional two-classification algorithm is solved, and samples with concentrated probability are further separated. The influence of the second-order state transition on the classification result is considered, so that the defect that only the first-order state transition is considered in the state fusion optimization classification algorithm is solved, and the classification effect is further improved. Moreover, the second-order state transition probability is applied to the training process of gradient reduction of the model, the trained model is fused into the second-order state transition probability, and not only is the state probability added to the probability to recalculate the binary probability when the model training is finished for testing and predicting. After the model is trained, the classification probability using state is transferred to be further optimized and adjusted after the model is trained, and the discrimination of the model is ensured.

The commodity recommendation method is generally divided into two major steps. The first is the training of the original neural network model (the target neural network model is obtained by training), and the second is the use of the target neural network model (the input data gives a recognition result, and whether the target commodity is pushed or not is determined according to the recognition result).

Taking today as the current time period, yesterday as the first time period, the previous day as the second time period, and tomorrow as the target time period as an example, as shown in fig. 4, today recommends a voucher, and yesterday and previous day recommendations have both ended. The recommendation for tomorrow has not yet begun. It should be noted that one or more recommendations may be made per day, and one or more recommendations may be made per recommendation. The present embodiment is not limited. Whether coupons are to be recommended to the user tomorrow can be predicted from data of today, yesterday, and the past day.

The training of the original neural network model is divided into the steps of obtaining a sample and training the sample.

Obtaining a sample: and acquiring historical data of the target account. The historical data of the target account number comprises the behavior characteristics of the target account number in each time period, namely the behavior of the account number for pushing the voucher. The historical data also comprises commodity information of commodities pushed by the target account in each time period, such as the voucher pushed in each time period, the value, the type and the total pushing times of the voucher. The present embodiment is explained by taking a voucher as an example, and the pushed goods are not limited to a voucher. The historical data also includes the result of the user logging into the account clicking on the pushed voucher. May include whether to click, number of clicks, etc.

After sample data is acquired, the sample data needs to be preprocessed. The preprocessing is divided into the construction of training samples and test samples of the samples. And constructing user sample data by using the user behavior characteristics (user characteristics), commodity characteristics (item characteristics) and user tags (click results) in the K-1 period to obtain a plurality of pieces of sample data. And distinguishing the multiple samples into sparse features and dense features. Wherein, the sparse feature is subjected to onehot coding processing, and the dense feature is subjected to principal component analysis to be subjected to decorrelation processing, normalization (standardization) processing, feature discretization processing and the like. Dividing the processed sparse features, dense features and user classification labels into training samples (the proportion is a) and testing samples (the proportion is 1-a) at random according to a certain proportion, wherein a is a parameter, for example, dividing the samples into the training samples at random: test sample = 8:2 (i.e., training and test samples are randomly sliced at an 8:2 ratio).

A prediction sample construction phase. And constructing user prediction sample data by using the user characteristic, the item characteristic and the user tag in the K period, and distinguishing the prediction sample into a sparse characteristic and a dense characteristic. Wherein, the sparse type characteristic is subjected to onehot processing, and the dense type characteristic is subjected to PCA decorrelation processing, normalization (standardization) processing, characteristic discretization processing and the like. That is, the user characteristics and item characteristics of the user in the K +1 phase are predicted by using the data in the K phase. Since the user characteristic and the item characteristic are basically unchanged or do not change greatly, the predicted user characteristic and item characteristic of the K +1 period are accurate.

The user behavior characteristic user characteristic in the K-1 stage mainly comprises the following steps: basic attribute data such as user gender, age, region and the like; active attribute data such as active days, active duration, active function quantity, interval of registration time and current time days and the like; recharging attribute data such as recharging amount, consumption amount, recharging times, recharging days, interval between the first recharging and the current time days and the like; user function clicks, user pick-up gift bag/gift certificate type (quantity, number, value), use gift bag/gift certificate type (quantity, value), expired gift bag/gift certificate type (quantity, value), etc. User classification label (label) in stage K [ coupon click scenario label is: 1-represents click (positive sample), 0-represents no click (negative sample) ]. If the coupon is a scene requiring coupon clicking and downloading, after clicking in the period K-1, downloading in the period K with the label of 1 as a positive sample, and after not clicking in the period K-1, not downloading in the period K with the label of 0 as a negative sample.

After the data in the historical data are obtained, the click probability of the user after the commodity is pushed to the user can be determined according to the data of the previous day and the yesterday. If the push is carried out 300 times in the previous day and the user clicks 30 times, the click probability is 10% (the first sub-probability parameter), the no-click probability is 90% (the second sub-probability parameter), the push is carried out 20 times yesterday, and the click probability is 75% (the third sub-probability parameter) and the no-click probability is 25% (the fourth sub-probability parameter).

Model training: using the behavior characteristics of the target account number at present and commodity information of the target commodity pushed at present, and the click result of the user at present, and the click probability of the user yesterday being 10% and the no click probability being 90%, the click probability of the user last day being 75% and the no click probability being 25%, inputting the results into the original neural network model, and training parameters of the original neural network model to obtain a model parameter W. The method specifically comprises the following steps: calculating the proportion of positive and negative samples (the proportion of the negative samples is as follows) by using the user label (Yk-1) in the K-1 stage

The positive sample ratio is

) As first order state transition probability (

，

) Unbiased estimation of (2); calculating the proportion of positive and negative samples (the proportion of the negative samples is as follows) by using the user label (Yk-2) in the K-2 stage

The positive sample ratio is

) As first order state transition probability (

，

) Unbiased estimation of (d). Calculating the proportion of positive samples to negative samples by using a K-phase user label (Yk) (the proportion of the negative samples is:

the positive sample fraction is:

) As first order state transition probability (

，

) Unbiased estimation of (2); calculating the proportion of positive and negative samples (the proportion of the negative samples is as follows) by using the user label (Yk-1) in the K-1 stage

The positive sample ratio is

) As first order state transition probability (

，

) Unbiased estimation of (d).

And a model training phase under the second-order state transition. And (4) carrying out Bayes second-order state transition probability model which is derived by mathematics on the training samples.

Substituting into K-2 and K-1 order state transition probabilities

，

，

，

By gradient descent under second-order state transition (II)The gradient descent method of the stage state transition, namely the gradient descent solving process is all influenced by the second-order state transition probability) to obtain the model weight W (the model weight is used for measuring the contribution of the characteristic feature X to the click probability or the label Y).

And after the model is trained to obtain a first model parameter, namely the weight W, in a model test evaluation stage under second-order state transition. Testing the trained model W by using the test sample, and substituting the test sample into the K-2 and K-1 order state transition probabilities

，

，

，

And calculating classification probability and evaluation indexes (indexes such as recall ratio, precision ratio and AUC) under the test sample through a formula M3, and if the evaluation indexes achieve the evaluation effect, storing the model W. If the model evaluation is not passed, the model is repeatedly trained until the model reaches the evaluation.

And training to obtain the target neural network model. Followed by the use of a model.

Model prediction phase under second order state transition. Using a trained model W, using a prediction sample, namely the predicted account behavior characteristics and commodity information in the K +1 stage, and substituting the prediction sample into K-1 and K-order state transition probabilities

，

，

And calculating the downloading probability of the user through the model. The model may be:

（3）

where K denotes today, K-1 denotes yesterday, K-2 denotes the previous day, K +1 denotes tomorrow (in this example only, actual K is the current time period, K-1 is the first time period, K-2 is the second time period, and K +1 is the target time period) and the K-1 phase state is denoted as

The probability of taking 0 or 1 state respectively is

，

0 means no click, 1 means click; the state of the K phase is expressed as

The probability of 0 or 1 state is taken as

，

. Here, the behavior characteristics are represented by X, the product characteristics are represented by I, and the click probability (probability parameter) is represented by Y. In the above formula, exp () is a language function.W _XIn order to be the parameters of the model,X _kfor the purpose of the behavior characteristics of the users of today,

is the click probability of yesterday. i is 0 or 1, and j is 0 or 1.

The result obtained by the formula is the target probability parameter of the predicted target time period, namely the probability that the coupon is pushed to the user in the tomorrow and the user clicks. The coupon may be recommended to the user on the next day if the recognition result of the target neural network model is greater than 0.5, which may be bounded by 0.5. And if the identification result of the target neural network model is less than or equal to 0.5, prohibiting the coupon from being recommended to the user in the next day.

The algorithm structure is shown in fig. 5. In fig. 5, the account behavior feature, the commodity feature, the probability parameter (click label, which is used to indicate whether the user clicks the recommended commodity or the number of clicks or the probability) of K-1, the probability parameter (state transition probability) of K-1, and the probability parameter (state transition probability) of K-2 are used to train the hyperopia neural network model, so as to obtain the usable target neural network model. Then, the use of the model is performed. In the using process of the model, the account behavior characteristics of K +1, the commodity characteristics of K +1 and the probability parameters of K and K-1 are obtained, and the probability parameter of K +1 is obtained. I.e. a third probability parameter for a third time period is obtained. The specific frame diagram is shown in fig. 6.

Table (1) shows the coupon download scenario effect comparing the traditional two-classification algorithm, the state fusion two-classification algorithm and the Bayesian second-order state transition two-classification algorithm. Compared with the prior art, the Bayesian second-order state transition two-classification algorithm in the embodiment of the application obviously improves the identification accuracy of the model.

Watch (1)

And if the fact that the coupon is pushed to the user in the tomorrow is determined, the coupon is pushed to the user in the tomorrow, a click result of the user is obtained, a third probability parameter is obtained, and whether the user clicks the coupon if the coupon is pushed in the tomorrow or the click probability is predicted by using the third probability parameter.

It should be noted that, for simplicity of description, the above-mentioned method embodiments are described as a series of acts or combination of acts, but those skilled in the art will recognize that the present invention is not limited by the order of acts, as some steps may occur in other orders or concurrently in accordance with the invention. Further, those skilled in the art should also appreciate that the embodiments described in the specification are preferred embodiments and that the acts and modules referred to are not necessarily required by the invention.

According to another aspect of the embodiment of the invention, a commodity recommending device for implementing the commodity recommending method is also provided. As shown in fig. 7, the apparatus includes:

a first obtaining unit 702, configured to obtain target account data of a target account in a target time period, where the target account data includes target account information of the target account and target commodity information of a target commodity to be pushed to the target account in advance, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through the first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period;

the input unit 704 is configured to input target account data, a first probability parameter and a current probability parameter into a target neural network model, where the target neural network model is a model trained using sample data, the sample data includes a second probability parameter of a target account, the first probability parameter, the current account data and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used to indicate a probability that the target account clicks a target commodity when the target commodity is pushed to the target account in the target time period;

a first determining unit 706, configured to determine, according to the recognition result, to push the target product to the target account within the target time period or prohibit the target product from being pushed to the target account within the target time period.

As an alternative embodiment, as shown in fig. 8, the first determining unit 706 includes:

a second obtaining module 802, configured to obtain an identification result;

the second determining module 804 is configured to determine that the target commodity is pushed to the target account within the target time period when the identification result is greater than the target threshold;

and a third determining module 806, configured to determine that the target product is prohibited from being pushed to the target account within the target time period, if the recognition result is less than or equal to the target threshold.

As an alternative embodiment, as shown in fig. 9, the apparatus further includes:

a third obtaining unit 902, configured to obtain a target probability parameter of the target account when it is determined, according to the recognition result, that the target commodity is pushed to the target account within the target time period or the target commodity is prohibited from being pushed to the target account within the target time period, where the target probability parameter is obtained according to the target frequency of the target account clicking the target commodity within the target time period and the total frequency of the target commodity being pushed to the target account within the target time period;

a predicting unit 904, configured to predict, using the target probability parameter, a target result of pushing the target commodity to the target account within a third time period after the target time period;

a fifth determining unit 906, configured to determine, according to the target result, to push the target product to the target account within the third time period or prohibit pushing the target product to the target account within the third time period.

For other examples of this embodiment, please refer to the above examples, which are not described herein again.

According to another aspect of the embodiment of the present invention, there is also provided an electronic device for implementing the above commodity recommendation method, where the electronic device may be a terminal device or a server shown in fig. 10. The present embodiment takes the electronic device as a server as an example for explanation. As shown in fig. 10, the electronic device comprises a memory 1002 and a processor 1004, the memory 1002 having stored therein a computer program, the processor 1004 being arranged to execute the steps of any of the method embodiments described above by means of the computer program.

Optionally, in this embodiment, the electronic device may be located in at least one network device of a plurality of network devices of a computer network.

Optionally, in this embodiment, the processor may be configured to execute the following steps by a computer program:

acquiring target account data of a target account in a target time period, wherein the target account data comprises target account information of the target account and target commodity information of a target commodity which is pushed to the target account in advance, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through the first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period;

inputting target account data, a first probability parameter and a current probability parameter into a target neural network model, wherein the target neural network model is a model trained by using sample data, the sample data comprises a second probability parameter of a target account, the first probability parameter, the current account data and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is used for outputting a recognition result, and the recognition result is used for indicating the probability that the target account clicks a target commodity under the condition that the target commodity is pushed to the target account in the target time period;

and determining to push the target commodity to the target account in the target time period or forbidding to push the target commodity to the target account in the target time period according to the identification result.

Alternatively, it can be understood by those skilled in the art that the structure shown in fig. 10 is only an illustration, and the electronic device may also be a terminal device such as a smart phone (e.g., an Android phone, an iOS phone, etc.), a tablet computer, a palmtop computer, a Mobile Internet Device (MID), a PAD, and the like. Fig. 10 is a diagram illustrating a structure of the electronic device. For example, the electronics may also include more or fewer components (e.g., network interfaces, etc.) than shown in FIG. 10, or have a different configuration than shown in FIG. 10.

The memory 1002 may be used to store software programs and modules, such as program instructions/modules corresponding to the product recommendation method and apparatus in the embodiments of the present invention, and the processor 1004 executes various functional applications and data processing by running the software programs and modules stored in the memory 1002, that is, implements the product recommendation method. The memory 1002 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 1002 may further include memory located remotely from the processor 1004, which may be connected to the terminal over a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof. The memory 1002 may be used for storing information, but not limited to. As an example, as shown in fig. 10, the memory 1002 may include, but is not limited to, the first obtaining unit 702, the input unit 704, and the first determining unit 706 in the product recommending apparatus. In addition, the module may further include, but is not limited to, other module units in the aforementioned commodity recommending apparatus, and details are not described in this example.

Optionally, the above-mentioned transmission device 1006 is used for receiving or sending data via a network. Examples of the network may include a wired network and a wireless network. In one example, the transmission device 1006 includes a Network adapter (NIC) that can be connected to a router via a Network cable and other Network devices so as to communicate with the internet or a local area Network. In one example, the transmission device 1006 is a Radio Frequency (RF) module, which is used for communicating with the internet in a wireless manner.

In addition, the electronic device further includes: and a connection bus 1008 for connecting the respective module components in the above-described electronic apparatus.

In other embodiments, the terminal device or the server may be a node in a distributed system, where the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting a plurality of nodes through a network communication. Nodes can form a Peer-To-Peer (P2P, Peer To Peer) network, and any type of computing device, such as a server, a terminal, and other electronic devices, can become a node in the blockchain system by joining the Peer-To-Peer network.

According to a further aspect of an embodiment of the present invention, there is also provided a computer-readable storage medium having a computer program stored thereon, wherein the computer program is arranged to perform the steps of any of the above method embodiments when executed.

Alternatively, in the present embodiment, the above-mentioned computer-readable storage medium may be configured to store a computer program for executing the steps of:

Alternatively, in this embodiment, a person skilled in the art may understand that all or part of the steps in the methods of the foregoing embodiments may be implemented by a program instructing hardware associated with the terminal device, where the program may be stored in a computer-readable storage medium, and the storage medium may include: flash disks, Read-Only memories (ROMs), Random Access Memories (RAMs), magnetic or optical disks, and the like.

The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.

The integrated unit in the above embodiments, if implemented in the form of a software functional unit and sold or used as a separate product, may be stored in the above computer-readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for causing one or more computer devices (which may be personal computers, servers, network devices, etc.) to execute all or part of the steps of the method according to the embodiments of the present invention.

In the above embodiments of the present invention, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.

In the several embodiments provided in the present application, it should be understood that the disclosed client may be implemented in other manners. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one type of division of logical functions, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, units or modules, and may be in an electrical or other form.

The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims

1. A method for recommending an article, comprising:

acquiring target account data of a target account in a target time period, wherein the target account data includes a current probability parameter of the target account in the current time period and a first probability parameter of the target account in a first time period, the target account data includes target behavior characteristics of the target account and target commodity information of a target commodity which is to be pushed to the target account, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through the first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period;

inputting the target account data, the first probability parameter and the current probability parameter into a target neural network model, wherein the target neural network model is a model trained by using sample data, the sample data includes a second probability parameter of the target account, the first probability parameter and the current account data, and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used for indicating a probability that the target account clicks the target commodity when the target commodity is pushed to the target account in the target time period;

and determining to push the target commodity to the target account within the target time period or forbidding to push the target commodity to the target account within the target time period according to the identification result.

2. The method of claim 1, wherein prior to inputting the target account data with the first probability parameter and the current probability parameter into a target neural network model, the method further comprises:

acquiring historical data of the target account;

determining account behavior characteristics of the target account in the historical data in the current time period as current behavior characteristics, and determining information of commodities pushed to the target account in the current time period as current commodity information;

determining the behavior characteristics of the target account in the historical data in the first time period as first behavior characteristics, and determining the information of the commodities pushed to the target account in the first time period as first commodity information;

acquiring the first probability parameter and the second probability parameter;

and training an original neural network model by using the current behavior characteristic, the current commodity information, the first behavior characteristic, the first commodity information, the first probability parameter and the second probability parameter to obtain the target neural network model.

3. The method of claim 2, wherein the obtaining the first probability parameter and the second probability parameter comprises:

acquiring a first number of times that the target account clicks the target commodity pushed to the target account within the first time period and a second number of times that the target account clicks the target commodity pushed to the target account within the second time period

And determining the first probability parameter according to the first times and a first total number of times of pushing the target commodity to the target account in the first time period, and determining the second probability parameter according to the second times and a second total number of times of pushing the target commodity to the target account in the second time period.

4. The method of claim 1, wherein in training an original neural network model to obtain the target neural network model using the sample data, the method further comprises:

training the original neural network model by using the training data, determining a first model parameter of the original neural network model, testing the original neural network model under the first model parameter by using the test data to obtain a test result output by the original neural network model under the first model parameter, and adjusting the first model parameter under the condition that the test result indicates that the recognition accuracy is less than or equal to the first threshold value.

5. The method of claim 1, wherein after acquiring the sample data, the method further comprises:

encoding the first type sample to obtain the encoded first type sample;

determining the first type of sample after encoding and the second type of sample after processing as new sample data.

6. The method according to any one of claims 1 to 5, wherein the determining, according to the identification result, whether to push the target commodity to the target account within the target time period or to prohibit the target commodity from being pushed to the target account within the target time period includes:

acquiring the identification result;

determining to push the target commodity to the target account within the target time period under the condition that the identification result is larger than a target threshold value;

7. The method according to any one of claims 1 to 5, wherein after determining that the target commodity is pushed to the target account within the target time period or is prohibited from being pushed to the target account within the target time period according to the identification result, the method further comprises:

under the condition that the target commodity is pushed to the target account within the target time period, acquiring a target probability parameter of the target account, wherein the target probability parameter is obtained according to the target times of clicking the target commodity by the target account within the target time period and the total times of pushing the target commodity to the target account within the target time period;

predicting a target result of pushing the target commodity to the target account within a third time period after the target time period by using the target probability parameter;

and determining to push the target commodity to the target account within the third time period or forbidding to push the target commodity to the target account within the third time period according to the target result.

8. An article recommendation device, comprising:

a first obtaining unit, configured to obtain target account data of a target account in a target time period, where the current probability parameter of the target account in the current time period and a first probability parameter of the target account in a first time period are included in the target account data, where the target account data includes target behavior characteristics of the target account and target commodity information of a target commodity to be pushed to the target account, the first time period is a time period before the current time period, the target time period is a time period after the current time period, the first probability parameter is obtained through first account data of the target account in the first time period, and the current probability parameter is obtained through the current account data of the target account in the current time period;

an input unit, configured to input the target account data, the first probability parameter, and the current probability parameter into a target neural network model, where the target neural network model is a model trained using sample data, the sample data includes a second probability parameter of the target account, the first probability parameter, the current account data, and the first account data, the second probability parameter is obtained through second account data of the target account in a second time period, the second time period is a time period before the first time period, the target neural network model is configured to output a recognition result, and the recognition result is used to indicate a probability that the target account clicks the target commodity when the target commodity is pushed to the target account in the target time period;

a first determining unit, configured to determine, according to the identification result, to push the target product to the target account within the target time period or prohibit the target product from being pushed to the target account within the target time period.

9. The apparatus of claim 8, further comprising:

a second obtaining unit, configured to obtain historical data of the target account before inputting the target account data, the first probability parameter, and the current probability parameter into a target neural network model;

a second determining unit, configured to determine, in the historical data, account behavior characteristics of the target account in the current time period as current behavior characteristics, and determine information of a commodity pushed to the target account in the current time period as current commodity information;

a third determining unit, configured to determine, in the history data, an account behavior of the target account in the first time period as a first behavior feature, and determine information of a commodity pushed to the target account in the first time period as first commodity information;

a third obtaining unit, configured to obtain the first probability parameter and the second probability parameter;

and the training unit is used for training an original neural network model by using the current behavior characteristic, the current commodity information, the first behavior characteristic, the first commodity information, the first probability parameter and the second probability parameter to obtain the target neural network model.

10. The apparatus of claim 9, wherein the third obtaining unit comprises:

a first obtaining module, configured to obtain a first number of times that the target account clicks the target product pushed to the target account within the first time period in the first time period, and a second number of times that the target account clicks the target product pushed to the target account within the second time period in the second time period

A first determining module, configured to determine the first probability parameter according to the first number and a first total number of times that the target commodity is pushed to the target account within the first time period, and determine the second probability parameter according to the second number and a second total number of times that the target commodity is pushed to the target account within the second time period.

11. The apparatus of claim 8, further comprising:

the first dividing unit is used for dividing the sample data into training data and test data according to a preset proportion in the process of using the sample data to train an original neural network model to obtain the target neural network model;

a first processing unit, configured to perform the following operations on the original neural network model until the recognition accuracy of the original neural network model is greater than a first threshold:

12. The apparatus of claim 8, further comprising:

the second dividing unit is used for dividing the sample data into a first type sample and a second type sample according to a preset condition after the sample data is obtained, wherein the first type sample is a single specific feature, and the second type sample is a continuous feature;

the encoding unit is used for encoding the first type sample to obtain the encoded first type sample;

the second processing unit is used for performing decorrelation processing, normalization processing and feature discretization processing on the second type sample to obtain a processed second type sample;

a fourth determining unit, configured to determine the encoded first type sample and the processed second type sample as new sample data.

13. The apparatus according to any one of claims 8 to 12, wherein the first determining unit comprises:

the second acquisition module is used for acquiring the identification result;

the second determination module is used for determining that the target commodity is pushed to the target account within the target time period under the condition that the identification result is larger than a target threshold value;

and the third determining module is used for determining that the target commodity is prohibited from being pushed to the target account within the target time period when the identification result is smaller than or equal to the target threshold.

14. A computer-readable storage medium comprising a stored program, characterized in that the program when executed performs the method of any of claims 1 to 7.

15. An electronic device comprising a memory and a processor, characterized in that the memory has stored therein a computer program, the processor being arranged to execute the method of any of claims 1 to 7 by means of the computer program.