CN114707070A

CN114707070A - User behavior prediction method and related equipment thereof

Info

Publication number: CN114707070A
Application number: CN202210379948.7A
Authority: CN
Inventors: 刘卫文; 唐睿明; 张瑞; 傅凌玥; 林江浩; 张伟楠; 俞勇
Original assignee: Huawei Technologies Co Ltd
Current assignee: Huawei Technologies Co Ltd
Priority date: 2022-04-12
Filing date: 2022-04-12
Publication date: 2022-07-05
Also published as: WO2023197910A1

Abstract

The application discloses a user behavior prediction method and related equipment thereof, which can enable the probability of clicking a project obtained by a neural network model by a user to be higher in accuracy and beneficial to accurately recommending the interested project for the user subsequently. The method comprises the following steps: acquiring a first characteristic of a first item and a second characteristic of a second item, wherein the first item and the second item are positioned in different lists or the same list of a target page, and the second item is positioned in front of the first item; acquiring a second feature of the first item based on a first feature of the first item and a second feature of the second item; and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item.

Description

User behavior prediction method and related equipment thereof

Technical Field

The embodiment of the application relates to the technical field of Artificial Intelligence (AI), in particular to a user behavior prediction method and related equipment.

Background

With the rapid development of computer technology, developers tend to show contents in which users are interested on pages more and more in order to meet the internet surfing requirements of users. Based on this, for a certain page, it is often necessary to predict which item or items displayed on the page the user clicks, that is, to predict the behavior of the user for the page, and then to modify the items to be presented on the page, so as to recommend the items of interest to the user.

Generally, the arrangement of items in a certain page is often presented to the user in the form of multiple lists, that is, the page usually includes multiple lists, and each list includes multiple items. In predicting the user's behavior with respect to the page, for any item in the page, a neural network model of AI techniques may be utilized to correlate the probability that the item was clicked on by the user.

However, the neural network model provided by the related art generally only considers the influence on an item caused by the remaining items in the list where the item is located when predicting the probability that the item is clicked by the user. Therefore, the factors considered by the related technology are single, so that the probability that the item is clicked by the user is obtained finally by the model, the accuracy is often low, and the interested item cannot be recommended to the user accurately in the follow-up process.

Disclosure of Invention

The embodiment of the application provides a user behavior prediction method and related equipment thereof, which can enable the probability of clicking a project obtained by a neural network model by a user to be higher in accuracy, and are beneficial to accurately recommending the project which is interested by the user subsequently.

A first aspect of an embodiment of the present application provides a user behavior prediction method, where the method includes:

when user behavior prediction needs to be performed on a target page, that is, when the probability that a first item in the target page is clicked by a user needs to be obtained, a first feature of the first item and a second feature of a second item can be obtained first, and the first feature of the first item and the second feature of the second item are input into a target model. The first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item, that is, the position relationship between the first item and the second item has the following two situations: (1) the first item and the second item may be items in the same list, the second item preceding the first item and the second item being adjacent to the first item. (2) The first item and the second item may be items in different lists, the list in which the second item is located is before the list in which the first item is located, and the second item may be adjacent to the first item or may not be adjacent to the first item.

After the first feature of the first item and the second feature of the second item are input into the target model, the first feature of the first item and the second feature of the second item can be processed through the target model, and therefore the second feature of the first item is obtained. It should be noted that the first feature of the first item may be attribute information of the first item itself, and then the second feature of the first item is information obtained by fusing based on the attribute information of the first item (i.e. the first feature of the first item), and since the obtaining process of the second feature of the second item is the same as the obtaining process of the second feature of the first item, the second feature of the second item is also information obtained by fusing based on the attribute information of the second item (the first feature of the second item).

And finally, processing the second characteristic of the first item through the target model so as to obtain the probability of the first item being clicked by the user.

From the above method, it can be seen that: when it is required to predict the probability that a first item in a target page is clicked by a user, a first feature of the first item and a second feature of a second item may be input to a target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the target model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

In one possible implementation, the method further includes: acquiring a first characteristic of a third item, wherein the first item and the third item are positioned in different lists or the same list of a target page, and the third item is adjacent to the first item; acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item; based on the second feature of the first item, obtaining the probability that the first item is clicked by the user comprises: and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item. In the foregoing implementation manner, the first feature of a third item may also be input into the target model, where the first item and the third item are located in different lists or the same list of the target page, and the third item is adjacent to the first item, that is, the positional relationship between the first item and the third item has the following two cases: (1) the first item and the third item may be items in the same list, the third item being adjacent to the first item. (2) The first item and the third item may be items in different lists, the third item being adjacent to the first item. After the first feature of the third item is input into the target model, the first feature of the first item and the first feature of the third item can be processed through the target model, and therefore the third feature of the first item is obtained. After obtaining the second feature of the first item and the third feature of the first item, the target model may calculate the second feature of the first item and the third feature of the first item, so as to obtain a probability that the first item is clicked by the user. Since the second characteristic of the first item can represent the influence of the second item on the first item, that is, the influence of the items browsed by the user during the actions on the first item when the user browses to the first item by using the sequential browsing action and the skipping action, and the third characteristic of the first item can represent the influence of the third item on the first item, that is, the influence of the items browsed by the user during the actions on the first item when the user browses to the first item by using the comparing action, it can be seen that when the target model predicts the user actions, not only the conventional sequential browsing action but also the browsing actions such as the skipping action and the comparing action are introduced, that is, when the target model considers the user browses to the first item by using the complex and diverse browsing actions, the influence of the items browsed by the user during the actions on the first item, the accuracy of the probability that the first item finally obtained by the target model is clicked by the user can be further improved.

In one possible implementation manner, the obtaining the second feature of the first item based on the first feature of the first item and the second feature of the second item includes: mapping the first characteristic of the first item to obtain a fourth characteristic of the first item; processing the second characteristic of the second item based on a self-attention mechanism to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item to obtain a second feature of the first item. In the foregoing implementation manner, after the first feature of the first item and the second feature of the second item are input into the target model, the target model may map the first feature of the first item on the hidden space to obtain the fourth feature of the first item, and at the same time, the target model may further perform processing based on the self-attention mechanism on the second feature of the second item to obtain the fifth feature of the first item. And obtaining the fourth feature of the first item and the fifth feature of the first item, wherein the target model can utilize the recurrent neural unit to process the fourth feature of the first item and the fifth feature of the first item, so as to accurately obtain the second feature of the first item.

In a possible implementation manner, the first feature of the first item is mapped to obtain a fourth feature of the first item: mapping the first characteristic of the first item, the request of the user for the target page and the probability of the second item being clicked by the user to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item to obtain a fourth feature of the first item. In the foregoing implementation manner, before the fourth feature of the first item is obtained, a request of the user for the target page and a probability of the second item being clicked by the user may be further input into the target model, so that the target model may map the first feature of the first item, the request of the user for the target page, and the probability of the second item being clicked by the user on the hidden space, correspondingly obtain the sixth feature of the first item, the seventh feature of the first item, and the eighth feature of the first item, and then splice the sixth feature of the first item, the seventh feature of the first item, and the eighth feature of the first item, so as to obtain the fourth feature of the first item. Therefore, when the target model analyzes the first item, not only the influence of the attribute information of the first item, but also the influence of external factors such as a request of a user for a target page and the probability of clicking the second item by the user are considered, so that the accuracy of the probability of clicking the first item by the user, which is finally obtained by the target model, is further improved.

In one possible implementation manner, the obtaining, based on the first feature of the first item and the first feature of the third item, the third feature of the first item includes: mapping the first characteristic of the first item and the first characteristic of the third item to obtain a sixth characteristic of the first item and a ninth characteristic of the first item; performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item to obtain a tenth feature of the first item; and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item to obtain a third feature of the first item. In the foregoing implementation manner, after the first feature of the third item is input into the target model, the target model may respectively map the first feature of the first item and the first feature of the third item on the hidden space, so as to correspondingly obtain a sixth feature of the first item and a ninth feature of the first item. Then, the target model may calculate the sixth feature of the first item and the ninth feature of the first item through a comparison function, and perform weighted summation based on the calculation result to obtain the tenth feature of the first item. Finally, the target model can perform an exclusive nor operation on the sixth feature of the first item and the tenth feature of the first item to accurately obtain the third feature of the first item.

In a possible implementation manner, the fourth fusion processing is performed on the sixth feature of the first item and the tenth feature of the first item, and obtaining the third feature of the first item includes: mapping the request of the user for the target page to obtain a seventh characteristic of the first item; and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item to obtain a third feature of the first item. In the foregoing implementation manner, when the third feature of the first item is obtained, the target model may further map the request of the user for the target page in the hidden space to obtain the seventh feature of the first item, and then the target model may perform an exclusive or operation on the sixth feature of the first item, the seventh feature of the first item, and the tenth feature of the first item to obtain the third feature of the first item. Therefore, when the target model analyzes the first item, not only the influence of the attribute information of the first item itself but also the influence of external factors such as a request of a user for a target page are considered, so that the accuracy of the probability that the first item finally obtained by the target model is clicked by the user is further improved.

In one possible implementation, if the first item is the first item in the target page, the second characteristic of the second item is a preset value.

In one possible implementation manner, the target page includes a plurality of lists, a plurality of items in the plurality of lists form a directed acyclic graph, and the plurality of items include a first item, a second item, and a third item.

A second aspect of the embodiments of the present application provides a method for constructing a directed acyclic graph, where the method includes: acquiring eye movement data of a target page browsed by a user; determining browsing behavior of a user on a plurality of items based on the eye movement data, wherein the plurality of items are located in a plurality of lists of the target page; and connecting a plurality of items based on the browsing behavior to obtain the directed acyclic graph.

From the above method, it can be seen that: the browsing behaviors (such as sequential browsing behaviors and skipping behaviors) of the user on the items in the target page can be determined based on eye movement data generated when the user browses the target page, so that the browsing order of the user on the items (such as browsing order of the user in the same list and browsing order of the user between different lists) is often determined, so that the multiple items of the target page are connected according to browsing to obtain a directed acyclic graph for the target page, and the directed acyclic graph can be used for user behavior prediction on the target page subsequently.

In one possible implementation, connecting a plurality of items based on the browsing behavior, and obtaining a directed acyclic graph includes: and connecting the items in the same list browsed by the user in the first sequence according to the first sequence, and connecting the items in different lists browsed by the user in the second sequence according to the second sequence to obtain the directed acyclic graph. In the foregoing implementation manner, the browsing behavior of the user includes two types of browsing behavior. The first browsing behavior refers to that a user browses items in the same list, and includes a first sequential browsing behavior, so that a browsing order of the user in the same list may be referred to as a first order, where the first order includes an order from top to bottom and an order from left to right in the first sequential browsing behavior according to which the user browses all items in the same list. The second browsing behavior refers to a browsing sequence of the user between different lists, and includes a second type of sequential browsing behavior and a comparison behavior, so the browsing sequence of the user between different lists can be called as a second sequence, where the second sequence includes a front-back sequence according to which the user browses adjacent items in two adjacent lists in the second type of sequential browsing behavior, and a jump sequence according to which the user browses two items in two non-adjacent lists in the comparison behavior. Then all the items in the target page may be connected in the first order as well as the second order, resulting in a directed acyclic graph for the target page.

In one possible implementation manner, the eye movement data of the user browsing the target page is acquired: and acquiring eye movement data of a target page browsed by a user through an eye movement instrument.

A third aspect of an embodiment of the present application provides a model training method, including: acquiring a first feature of a first item and a second feature of a second item through a model to be trained, wherein the first item and the second item are positioned in different lists or the same list of a page to be processed, and the second item is positioned in front of the first item; acquiring a second feature of the first item based on a first feature of the first item and a second feature of the second item through the model to be trained, wherein the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusing based on the attribute information of the first item, and the second feature of the second item is information obtained by fusing based on the attribute information of the second item (namely the first feature of the second item); acquiring the probability of the first item clicked by the user based on the second characteristic of the first item through the model to be trained; acquiring a target loss based on the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user, wherein the target loss is used for indicating the difference between the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user; and updating the parameters of the model to be trained based on the target loss until the model training condition is met, so as to obtain the target model.

The target model obtained by the method has the capability of predicting the user behavior of the page. When it is required to predict the probability that a first item in a target page is clicked by a user, a first feature of the first item and a second feature of a second item may be input to a target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the object model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

In one possible implementation, the method further includes: acquiring a first feature of a third item through the model to be trained, wherein the first item and the third item are positioned in different lists or the same list of the page to be processed, and the third item is adjacent to the first item; acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item through the model to be trained; obtaining, by the model to be trained, based on the second feature of the first item, a probability that the first item is clicked by the user includes: and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the model to be trained.

In one possible implementation manner, the obtaining the second feature of the first item based on the first feature of the first item and the second feature of the second item includes: mapping the first characteristic of the first item to obtain a fourth characteristic of the first item; processing the second characteristic of the second item based on a self-attention mechanism to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item to obtain a second feature of the first item.

In a possible implementation manner, the first feature of the first item is subjected to mapping processing, so as to obtain a fourth feature of the first item: mapping the first characteristic of the first item, the request of the user for the page to be processed and the probability of the second item being clicked by the user to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item to obtain a fourth feature of the first item.

In one possible implementation manner, the obtaining, based on the first feature of the first item and the first feature of the third item, the third feature of the first item includes: mapping the first characteristic of the first item and the first characteristic of the third item to obtain a sixth characteristic of the first item and a ninth characteristic of the first item; performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item to obtain a tenth feature of the first item; and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

In a possible implementation manner, the fourth fusion processing is performed on the sixth feature of the first item and the tenth feature of the first item, and obtaining the third feature of the first item includes: mapping the request of the user for the page to be processed to obtain a seventh characteristic of the first item; and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

In one possible implementation, if the first item is the first item in the page to be processed, the second characteristic of the second item is a preset value.

In a possible implementation manner, the page to be processed includes a plurality of lists, a plurality of items in the lists form a directed acyclic graph, and the plurality of items include a first item, a second item, and a third item.

A fourth aspect of an embodiment of the present application provides a user behavior prediction apparatus, including: the first acquisition module is used for acquiring a first characteristic of a first item and a second characteristic of a second item through the target model, wherein the first item and the second item are positioned in different lists or the same list of the target page, and the second item is positioned in front of the first item; a second obtaining module, configured to obtain, by using the target model, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusing the attribute information based on the first item, and the second feature of the second item is information obtained by fusing the attribute information based on the second item (i.e., the first feature of the second item); and the third acquisition module is used for acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item through the target model.

From the above device it can be seen that: when it is required to predict the probability that a first item in a target page is clicked by a user, a first feature of the first item and a second feature of a second item may be input to a target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the target model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

In one possible implementation, the apparatus further includes: the fourth obtaining module is used for obtaining a first feature of a third item through the target model, the first item and the third item are located in different lists or the same list of the target page, and the third item is adjacent to the first item; the fifth obtaining module is used for obtaining a third feature of the first item based on the first feature of the first item and the first feature of the third item through the target model; and the third obtaining module is used for obtaining the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the target model.

In a possible implementation manner, the second obtaining module is configured to: mapping the first characteristics of the first item through the target model to obtain fourth characteristics of the first item; processing the second characteristic of the second item based on a self-attention mechanism through the target model to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item through the target model to obtain a second feature of the first item.

In a possible implementation manner, the second obtaining module is configured to: mapping the first characteristic of the first item, the request of the user for the target page and the probability of the second item being clicked by the user through the target model to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item through the target model to obtain a fourth feature of the first item.

In a possible implementation manner, the fifth obtaining module is configured to: mapping the first feature of the first item and the first feature of the third item through the target model to obtain a sixth feature of the first item and a ninth feature of the first item; performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item through the target model to obtain a tenth feature of the first item; and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item through the target model to obtain a third feature of the first item.

In a possible implementation manner, the fifth obtaining module is configured to: mapping the request of the user for the target page through the target model to obtain a seventh characteristic of the first item; and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item through the target model to obtain a third feature of the first item.

A fifth aspect of an embodiment of the present application provides a directed acyclic graph building apparatus, including: the acquisition module is used for acquiring eye movement data of a target page browsed by a user; the determining module is used for determining browsing behaviors of a user on a plurality of items based on the eye movement data, wherein the plurality of items are positioned in a plurality of lists of the target page; and the connecting module is used for connecting the plurality of items based on the browsing behavior to obtain the directed acyclic graph.

The device can determine browsing behaviors of a user on a plurality of items in the target page based on eye movement data generated when the user browses the target page, and then the browsing behaviors (such as sequential browsing behaviors and skipping behaviors) often determine a browsing sequence of the user on the items (such as a browsing sequence of the user in the same list and a browsing sequence of the user between different lists), so that the plurality of items of the target page are connected according to the browsing to obtain the directed acyclic graph aiming at the target page, and the directed acyclic graph can be used for predicting the user behavior of the target page subsequently.

In a possible implementation manner, the connection module is configured to connect items in the same list browsed by the user in a first order according to the first order, and connect items in different lists browsed by the user in a second order according to the second order, so as to obtain the directed acyclic graph.

In a possible implementation manner, the obtaining module is configured to collect eye movement data of a user browsing a target page through an eye movement instrument.

A sixth aspect of the embodiments of the present application provides a schematic structural diagram of a model training apparatus, including: the first acquisition module is used for acquiring a first feature of a first item and a second feature of a second item through the model to be trained, the first item and the second item are positioned in different lists or the same list of the page to be processed, and the second item is positioned in front of the first item; a second obtaining module, configured to obtain, by using the model to be trained, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusing the attribute information based on the first item, and the second feature of the second item is information obtained by fusing the attribute information based on the second item (i.e., the first feature of the second item); the third acquisition module is used for acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item through the model to be trained; the fourth obtaining module is used for obtaining a target loss based on the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user, wherein the target loss is used for indicating the difference between the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user; and the updating module is used for updating the parameters of the model to be trained based on the target loss until the model training condition is met, so as to obtain the target model.

The target model obtained by the device training has the capability of predicting the user behavior of the page. When it is required to predict the probability that a first item in a target page is clicked by a user, a first feature of the first item and a second feature of a second item may be input to a target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the target model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

In one possible implementation, the apparatus includes: the fifth acquisition module is used for acquiring the first characteristic of a third item through the model to be trained, wherein the first item and the third item are positioned in different lists or the same list of the page to be processed, and the third item is adjacent to the first item; the sixth obtaining module is used for obtaining a third feature of the first item based on the first feature of the first item and the first feature of the third item through the model to be trained; and the third obtaining module is used for obtaining the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the model to be trained.

In a possible implementation manner, the second obtaining module is configured to: mapping the first characteristic of the first item to obtain a fourth characteristic of the first item; processing the second characteristic of the second item based on a self-attention mechanism to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item to obtain a second feature of the first item.

In a possible implementation manner, the second obtaining module is configured to: mapping the first characteristic of the first item, the request of the user for the page to be processed and the probability of the second item being clicked by the user to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item to obtain a fourth feature of the first item.

In a possible implementation manner, the sixth obtaining module is configured to: mapping the first characteristic of the first item and the first characteristic of the third item to obtain a sixth characteristic of the first item and a ninth characteristic of the first item; performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item to obtain a tenth feature of the first item; and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

In a possible implementation manner, the sixth obtaining module is configured to: mapping the request of the user for the page to be processed to obtain a seventh characteristic of the first item; and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

A seventh aspect of an embodiment of the present application provides a user behavior prediction apparatus, including a memory and a processor; the memory stores code and the processor is configured to execute the code, and when executed, the user behavior prediction apparatus performs a method as described in the first aspect or any one of the possible implementations of the first aspect.

An eighth aspect of an embodiment of the present application provides a directed acyclic graph constructing apparatus, including a memory and a processor; the memory stores code and the processor is configured to execute the code, and when the code is executed, the directed acyclic graph building apparatus performs the method according to the second aspect or any one of the possible implementations of the second aspect.

A ninth aspect of an embodiment of the present application provides a model training apparatus, including a memory and a processor; the memory stores code and the processor is configured to execute the code, and when the code is executed, the model training apparatus performs the method according to the third aspect or any one of the possible implementations of the third aspect.

A tenth aspect of embodiments of the present application provides a circuit system, which includes a processing circuit configured to perform the method as described in the first aspect, any one of the possible implementations of the second aspect, or any one of the possible implementations of the third aspect.

An eleventh aspect of an embodiment of the present application provides a chip system, where the chip system includes a processor, configured to invoke a computer program or computer instructions stored in a memory, so as to cause the processor to execute the method according to any one of the first aspect, any one of the possible implementations of the first aspect, the second aspect, or any one of the possible implementations of the third aspect or the third aspect.

In one possible implementation, the processor is coupled to the memory through an interface.

In one possible implementation, the system-on-chip further includes a memory having a computer program or computer instructions stored therein.

A twelfth aspect of embodiments of the present application provides a computer storage medium, which stores a computer program, and when the program is executed by a computer, the computer executes the method according to the first aspect, any one of the possible implementations of the second aspect, or any one of the possible implementations of the third aspect, or the third aspect.

A thirteenth aspect of embodiments of the present application provides a computer program product, which stores instructions that, when executed by a computer, cause the computer to implement the method according to the first aspect, any one of the possible implementations of the second aspect, or any one of the possible implementations of the third aspect, or the third aspect.

In the embodiment of the application, when the probability that the first item in the target page is clicked by the user needs to be predicted, a first feature of the first item and a second feature of a second item may be input into the target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the target model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

Drawings

FIG. 1 is a schematic structural diagram of an artificial intelligence body framework;

fig. 2a is a schematic structural diagram of a user behavior prediction system according to an embodiment of the present application;

fig. 2b is another schematic structural diagram of a user behavior prediction system according to an embodiment of the present application;

fig. 2c is a schematic diagram of a related device for processing a data sequence according to an embodiment of the present application;

FIG. 3 is a diagram illustrating an architecture of the system 100 according to an embodiment of the present application;

fig. 4 is a schematic flowchart of a directed acyclic graph constructing method according to an embodiment of the present application;

FIG. 5 is a schematic diagram of a target page provided by an embodiment of the present application;

fig. 6 is a schematic diagram of an eye tracker according to an embodiment of the present application;

FIG. 7 is a schematic diagram of a directed acyclic graph according to an embodiment of the present application;

fig. 8 is a schematic flowchart of a user behavior prediction method according to an embodiment of the present application;

FIG. 9 is a schematic structural diagram of a target model provided in an embodiment of the present application;

FIG. 10 is a schematic flow chart diagram illustrating a model training method according to an embodiment of the present disclosure;

fig. 11 is a schematic structural diagram of a user behavior prediction apparatus according to an embodiment of the present application;

fig. 12 is a schematic structural diagram of a directed acyclic graph building apparatus according to an embodiment of the present application;

FIG. 13 is a schematic structural diagram of a model training apparatus according to an embodiment of the present application;

fig. 14 is a schematic structural diagram of an execution device according to an embodiment of the present application;

FIG. 15 is a schematic structural diagram of a training apparatus provided in an embodiment of the present application;

fig. 16 is a schematic structural diagram of a chip according to an embodiment of the present application.

Detailed Description

The terms "first," "second," and the like in the description and in the claims of the present application and in the above-described drawings 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 terms so used are interchangeable under appropriate circumstances and are merely descriptive of the various embodiments of the application and how objects of the same nature can be distinguished. 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 elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.

Generally, the arrangement of the items in a certain page is often presented to the user in the form of multiple lists, that is, the page usually includes multiple lists, and each list includes multiple items, and when predicting the behavior of the user for the page, for any item in the page, a neural network model in the AI technical field can be utilized to match the probability that the item is clicked by the user. For example, in a page of an application mall, a plurality of horizontal lists and a plurality of vertical arrangements are displayed, the horizontal lists and the vertical arrangements are staggered, a plurality of applications in the horizontal lists are arranged in a row, and a plurality of applications in the vertical arrangements are arranged in a column, so that the page can display introduction information of various applications for a user in the form of the staggered lists. In order to predict the clicking behavior of the user on the page, the neural network model may be used to analyze each application one by one, so as to obtain the probability that the user clicks each application in the page.

Further, the browsing behavior of the user on the page tends to be complex, for example, when the user browses the items of the current list, the user directly jumps to the items of another list (the other list and the current list are two lists that are not adjacent to each other) to browse. The related art often fails to consider the influence of various complex browsing behaviors, and also reduces the accuracy of the probability that the final item obtained by the model is clicked by the user.

Furthermore, when a certain item is analyzed by the model of the related art, the model often only analyzes related information of the item itself (for example, if the certain item is an application, the related information of the application includes a developer, a type, a size, and the like of the application), and the accuracy of the probability that the item finally obtained by the model is clicked by the user is also reduced without considering the influence of external factors such as the user.

In order to solve the problem, an embodiment of the present application provides a user behavior prediction method, which may be implemented by combining an Artificial Intelligence (AI) technology. AI technology is a technical discipline that simulates, extends and expands the intelligence of a human being using a digital computer or a machine controlled by a digital computer, and obtains the best results by perceiving the environment, acquiring knowledge and using the knowledge. In other words, artificial intelligence technology is a branch 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. Data processing using artificial intelligence is a common application of artificial intelligence.

The general workflow of the artificial intelligence system is described first, please refer to fig. 1, fig. 1 is a schematic structural diagram of an artificial intelligence body framework, and the artificial intelligence body framework is explained below from two dimensions of an "intelligent information chain" (horizontal axis) and an "IT value chain" (vertical axis). Where "intelligent information chain" reflects a list of processes processed from the acquisition of data. For example, the general processes of intelligent information perception, intelligent information representation and formation, intelligent reasoning, intelligent decision making and intelligent execution and output can be realized. In this process, the data undergoes a "data-information-knowledge-wisdom" refinement process. The 'IT value chain' reflects the value of the artificial intelligence to the information technology industry from the bottom infrastructure of the human intelligence, information (realization of providing and processing technology) to the industrial ecological process of the system.

(1) Infrastructure

The infrastructure provides computing power support for the artificial intelligent system, realizes communication with the outside world, and realizes support through a foundation platform. Communicating with the outside through a sensor; the computing power is provided by intelligent chips (hardware acceleration chips such as CPU, NPU, GPU, ASIC, FPGA and the like); the basic platform comprises distributed computing framework, network and other related platform guarantees and supports, and can comprise cloud storage and computing, interconnection and intercommunication networks and the like. For example, sensors and external communications acquire data that is provided to intelligent chips in a distributed computing system provided by the base platform for computation.

(2) Data of

Data at the upper level of the infrastructure is used to represent the data source for the field of artificial intelligence. The data relates to graphs, images, voice and texts, and also relates to the data of the Internet of things of traditional equipment, including service data of the existing system and sensing data such as force, displacement, liquid level, temperature, humidity and the like.

(3) Data processing

Data processing typically includes data training, machine learning, deep learning, searching, reasoning, decision making, and the like.

The machine learning and the deep learning can perform symbolized and formalized intelligent information modeling, extraction, preprocessing, training and the like on data.

Inference refers to the process of simulating human intelligent inference mode in a computer or an intelligent system, using formalized information to think and solve problems of a machine according to an inference control strategy, and the typical function is searching and matching.

The decision-making refers to a process of making a decision after reasoning intelligent information, and generally provides functions of classification, sequencing, prediction and the like.

(4) General capabilities

After the above-mentioned data processing, further general capabilities may be formed based on the results of the data processing, such as algorithms or a general system, for example, translation, analysis of text, computer vision processing, speech recognition, recognition of images, and so on.

(5) Intelligent product and industrial application

The intelligent product and industry application refers to the product and application of an artificial intelligence system in various fields, and is the encapsulation of an artificial intelligence integral solution, the intelligent information decision is commercialized, and the landing application is realized, and the application field mainly comprises: intelligent terminal, intelligent transportation, intelligent medical treatment, autopilot, wisdom city etc..

Several application scenarios of the present application are presented next.

Fig. 2a is a schematic structural diagram of a user behavior prediction system according to an embodiment of the present application, where the user behavior prediction system includes a user device and a data processing device. The user equipment comprises a mobile phone, a personal computer or an intelligent terminal such as an information processing center. The user equipment is an initiator of user behavior prediction for a page, and is an initiator of a user behavior prediction request, and a user usually initiates the request through the user equipment.

The data processing device may be a device or a server having a data processing function, such as a cloud server, a network server, an application server, and a management server. The data processing equipment receives a user behavior prediction request from the intelligent terminal to the array page through the interactive interface, and then performs page processing in the modes of machine learning, deep learning, searching, reasoning, decision making and the like through the memory for storing data and the processor link for processing data. The memory in the data processing device may be a generic term that includes a database that stores locally and stores historical data, either on the data processing device or on other network servers.

In the user behavior prediction system shown in fig. 2a, the user device may receive an instruction of a user, for example, the user device may obtain a page input/selected by the user, and then initiate a request to the data processing device, so that the data processing device executes a user behavior prediction application for the page obtained by the user device, thereby obtaining a processing result for the page. For example, the user equipment may obtain a page input by a user, and then initiate a user behavior prediction request of the page to the data processing equipment, so that the data processing equipment processes features of each item in the page, thereby obtaining a processing result of the page, that is, a probability that each item in the page is clicked by the user.

In fig. 2a, a data processing device may execute the directed acyclic graph constructing method and the user behavior prediction method according to the embodiment of the present application.

Fig. 2b is another schematic structural diagram of the user behavior prediction system according to the embodiment of the present application, in fig. 2b, the user equipment itself may execute the user behavior prediction application, and the user equipment may directly obtain an input from the user and directly perform processing by hardware of the user equipment itself, where a specific process is similar to that in fig. 2a, and reference may be made to the above description, and details are not repeated here.

In the user behavior prediction system shown in fig. 2b, the user equipment may receive an instruction of the user, for example, the user equipment may obtain a page selected by the user in the user equipment, and then the user equipment performs processing on the features of each item in the page, so as to obtain a processing result of the page, that is, a probability that each item in the page is clicked by the user.

In fig. 2b, the user equipment itself may execute the directed acyclic graph constructing method and the user behavior prediction method according to the embodiment of the present application.

Fig. 2c is a schematic diagram of a related device for user behavior prediction processing according to an embodiment of the present application.

The user device in fig. 2a and fig. 2b may specifically be the local device 301 or the local device 302 in fig. 2c, and the data processing device in fig. 2a may specifically be the execution device 210 in fig. 2c, where the data storage system 250 may store data to be processed of the execution device 210, and the data storage system 250 may be integrated on the execution device 210, or may be disposed on a cloud or other network server.

The processor in fig. 2a and 2b may perform data training/machine learning/deep learning through a neural network model or other models (e.g., a model based on a support vector machine), and perform a user behavior prediction application on a page using the model finally trained or learned by the data, so as to obtain a corresponding processing result.

Fig. 3 is a schematic diagram of an architecture of the system 100 according to an embodiment of the present application, in fig. 3, an execution device 110 configures an input/output (I/O) interface 112 for data interaction with an external device, and a user may input data to the I/O interface 112 through a client device 140, where the input data may include: each task to be scheduled, the resources that can be invoked, and other parameters.

During the process that the execution device 110 preprocesses the input data or during the process that the calculation module 111 of the execution device 110 performs the calculation (for example, performs the function implementation of the neural network in the present application), the execution device 110 may call the data, the code, and the like in the data storage system 150 for corresponding processing, and may store the data, the instruction, and the like obtained by corresponding processing into the data storage system 150.

Finally, the I/O interface 112 returns the processing results to the client device 140 for presentation to the user.

It should be noted that the training device 120 may generate corresponding target models/rules based on different training data for different targets or different tasks, and the corresponding target models/rules may be used to achieve the targets or complete the tasks, so as to provide the user with the required results. Wherein the training data may be stored in the database 130 and derived from training samples collected by the data collection device 160.

In the case shown in fig. 3, the user may manually give the input data, which may be operated through an interface provided by the I/O interface 112. Alternatively, the client device 140 may automatically send the input data to the I/O interface 112, and if the client device 140 is required to automatically send the input data to obtain authorization from the user, the user may set the corresponding permissions in the client device 140. The user can view the result output by the execution device 110 at the client device 140, and the specific presentation form can be display, sound, action, and the like. The client device 140 may also serve as a data collection terminal, collecting input data of the input I/O interface 112 and output results of the output I/O interface 112 as new sample data, and storing the new sample data in the database 130. Of course, the input data inputted to the I/O interface 112 and the output result outputted from the I/O interface 112 as shown in the figure may be directly stored in the database 130 as new sample data by the I/O interface 112 without being collected by the client device 140.

It should be noted that fig. 3 is only a schematic diagram of a system architecture provided in an embodiment of the present application, and the position relationship between the devices, modules, and the like shown in the diagram does not constitute any limitation, for example, in fig. 3, the data storage system 150 is an external memory with respect to the execution device 110, and in other cases, the data storage system 150 may also be disposed in the execution device 110. As shown in fig. 3, a neural network may be trained from the training device 120.

The embodiment of the application also provides a chip, which comprises the NPU. The chip may be provided in the execution device 110 as shown in fig. 3 to complete the calculation work of the calculation module 111. The chip may also be disposed in the training apparatus 120 as shown in fig. 3 to complete the training work of the training apparatus 120 and output the target model/rule.

The neural network processor NPU, NPU is mounted as a coprocessor on a main Central Processing Unit (CPU) (host CPU), and tasks are distributed by the main CPU. The core portion of the NPU is an arithmetic circuit, and the controller controls the arithmetic circuit to extract data in a memory (weight memory or input memory) and perform an operation.

In some implementations, the arithmetic circuitry includes a plurality of processing units (PEs) therein. In some implementations, the operational circuit is a two-dimensional systolic array. The arithmetic circuit may also be a one-dimensional systolic array or other electronic circuit capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuitry is a general-purpose matrix processor.

For example, assume that there is an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit fetches the corresponding data of the matrix B from the weight memory and buffers the data on each PE in the arithmetic circuit. The arithmetic circuit takes the matrix A data from the input memory and carries out matrix operation with the matrix B, and partial results or final results of the obtained matrix are stored in an accumulator (accumulator).

The vector calculation unit may further process the output of the arithmetic circuit, such as vector multiplication, vector addition, exponential operation, logarithmic operation, magnitude comparison, and the like. For example, the vector computation unit may be used for network computation of the non-convolution/non-FC layer in the neural network, such as pooling (posing), batch normalization (batch normalization), local response normalization (local response normalization), and the like.

In some implementations, the vector calculation unit can store the processed output vector to a unified buffer. For example, the vector calculation unit may apply a non-linear function to the output of the arithmetic circuit, such as a vector of accumulated values, to generate the activation value. In some implementations, the vector calculation unit generates a normalized value, a combined value, or both. In some implementations, the vector of processed outputs can be used as activation inputs to arithmetic circuitry, e.g., for use in subsequent layers in a neural network.

The unified memory is used for storing input data and output data.

The weight data directly passes through a memory cell access controller (DMAC) to carry input data in the external memory to the input memory and/or the unified memory, store the weight data in the external memory in the weight memory, and store data in the unified memory in the external memory.

And the Bus Interface Unit (BIU) is used for realizing interaction among the main CPU, the DMAC and the instruction fetch memory through a bus.

An instruction fetch buffer (issue fetch buffer) connected to the controller for storing instructions used by the controller;

and the controller is used for calling the instruction cached in the finger memory and realizing the control of the working process of the operation accelerator.

Generally, the unified memory, the input memory, the weight memory, and the instruction fetch memory are On-Chip (On-Chip) memories, the external memory is a memory outside the NPU, and the external memory may be a double data rate synchronous dynamic random access memory (DDR SDRAM), a High Bandwidth Memory (HBM), or other readable and writable memories.

Since the embodiments of the present application relate to the application of a large number of neural networks, for the convenience of understanding, the related terms and related concepts such as neural networks related to the embodiments of the present application will be described below.

(1) Neural network

The neural network may be composed of neural units, and the neural units may refer to operation units with xs and intercept 1 as inputs, and outputs of the operation units may be:

where, s is 1, 2, … … n, n is a natural number greater than 1, Ws is the weight of xs, and b is the bias of the neural unit. f is an activation function (activation functions) of the neural unit for introducing a nonlinear characteristic into the neural network to convert an input signal in the neural unit into an output signal. The output signal of the activation function may be used as an input to the next convolutional layer. The activation function may be a sigmoid function. A neural network is a network formed by a number of the above-mentioned single neural units joined together, i.e. the output of one neural unit may be the input of another neural unit. The input of each neural unit can be connected with the local receiving domain of the previous layer to extract the characteristics of the local receiving domain, and the local receiving domain can be a region composed of a plurality of neural units.

The operation of each layer in a neural network can be described by the mathematical expression y ═ a (Wx + b): from the work of each layer in the physical layer neural network, it can be understood that the transformation of the input space into the output space (i.e. the row space to the column space of the matrix) is accomplished by five operations on the input space (set of input vectors), which include: 1. ascending/descending dimensions; 2. zooming in/out; 3. rotating; 4. translating; 5. "bending". Wherein the operations 1, 2 and 3 are performed by Wx, the operation 4 is performed by + b, and the operation 5 is performed by a (). The expression "space" is used herein because the object being classified is not a single thing, but a class of things, and space refers to the collection of all individuals of such things. Where W is a weight vector, each value in the vector representing a weight value for a neuron in the layer of neural network. The vector W determines the spatial transformation of the input space into the output space described above, i.e. the weight W of each layer controls how the space is transformed. The purpose of training the neural network is to finally obtain the weight matrix (the weight matrix formed by the vectors W of many layers) of all layers of the trained neural network. Therefore, the training process of the neural network is essentially a way of learning the control space transformation, and more specifically, the weight matrix.

Because it is desirable that the output of the neural network is as close as possible to the value actually desired to be predicted, the weight vector of each layer of the neural network can be updated by comparing the predicted value of the current network with the value actually desired to be predicted, and then updating the weight vector according to the difference between the predicted value and the value actually desired (of course, there is usually an initialization process before the first update, that is, the parameters are configured in advance for each layer of the neural network). Therefore, it is necessary to define in advance "how to compare the difference between the predicted value and the target value", which are loss functions (loss functions) or objective functions (objective functions), which are important equations for measuring the difference between the predicted value and the target value. Taking the loss function as an example, if the higher the output value (loss) of the loss function indicates the larger the difference, the training of the neural network becomes a process of reducing the loss as much as possible.

(2) Back propagation algorithm

The neural network can adopt a Back Propagation (BP) algorithm to correct the size of parameters in the initial neural network model in the training process, so that the reconstruction error loss of the neural network model is smaller and smaller. Specifically, the error loss is generated by transmitting the input signal in the forward direction until the output, and the parameters in the initial neural network model are updated by reversely propagating the error loss information, so that the error loss is converged. The back propagation algorithm is a back propagation motion with error loss as a dominant factor, aiming at obtaining the optimal parameters of the neural network model, such as a weight matrix.

The method provided by the present application is described below from the training side of the neural network and the application side of the neural network.

The model training method provided by the embodiment of the application relates to the processing of data sequences, and particularly can be applied to methods such as data training, machine learning and deep learning, and the like, and intelligent information modeling, extraction, preprocessing, training and the like which are symbolized and formalized are carried out on training data (for example, a first feature of a first item of a page to be processed in the application and the like), so that a trained neural network (such as a target model in the application) is finally obtained; in addition, the user behavior prediction method provided in the embodiment of the present application may use the trained neural network to input data (for example, a first feature of a first item of a target page in the present application, and the like) into the trained neural network, so as to obtain output data (for example, in the user behavior prediction method provided in the present application, a probability that the first item is clicked by a user, and the like). It should be noted that the model training method and the user behavior prediction method provided in the embodiments of the present application are inventions based on the same concept, and may also be understood as two parts in a system or two stages of an overall process: such as a model training phase and a model application phase.

It is noted that before the user behavior prediction is performed on the target page, a directed acyclic graph for the target page may be constructed, and the process of constructing the directed acyclic graph is described below. Fig. 4 is a schematic flowchart of a directed acyclic graph building method provided in an embodiment of the present application, and as shown in fig. 4, the method includes:

401. and acquiring eye movement data of a target page browsed by a user.

In this embodiment, when user behavior prediction needs to be performed on a target page, the target page may be obtained first, where the target page includes a plurality of lists, and for any one of the lists, the list includes a plurality of items, and the plurality of items are arranged in a certain order (for example, if the plurality of items are arranged in a row, the list is a horizontal list, and if the plurality of items are arranged in a column, the list is a vertical list). For example, as shown in fig. 5 (fig. 5 is a schematic view of a target page provided in the embodiment of the present application), the target page is a presentation page of an application mall, and the presentation page can present information of multiple applications to a user, so that the user can browse and download applications required by the user on the page. The page contains a vertical list B₁Longitudinal list B₃Longitudinal list B₅List B of horizontal positions₂And a horizontal list B₄The 3 longitudinal columnsThe tables and 2 horizontal lists being staggered, i.e. in vertical list B₁List B of horizontal positions₂Longitudinal list B₃List B of horizontal positions₄Longitudinal list B₅The above sequence is followed by listing. Wherein, list B is horizontal₂And B₄Containing 5 applications, vertical List B₁And B₃Containing 3 applications. In this way, the page may present information for 19 applications to the user.

After the target page is obtained, at least one user can be invited to browse the target page, and eye movement data generated when the users browse the target page is obtained.

Specifically, the eye movement data of the user browsing the target page may be acquired by:

as shown in fig. 6 (fig. 6 is a schematic view of an eye tracker provided in an embodiment of the present application), an eye tracker facing a user may be disposed near a user device for displaying a target page, and the eye tracker may be electrically connected to the user device. In addition, an auxiliary tool may be deployed for stabilizing the head of the user so that the eye tracker can accurately track the line of sight of the user, and in order to better simulate the interaction environment between the user and the user device, a distance between the user and the screen of the user device may be set (e.g., the distance may be set to 30-40cm, etc.), and an inclination angle of the user device (e.g., the inclination angle may be set to 65 ° -70 °, etc.).

When a user starts to browse a target page displayed by user equipment by means of the auxiliary tool, the eye tracker can track and record the sight line position and the sight line moving mode of the user on the target page, eye movement data of the user browsing the target page are generated and sent to the user equipment. Therefore, the user equipment successfully acquires the eye movement data of the target page browsed by the user.

It should be understood that the user equipment here may be the user equipment in the system shown in fig. 2a or fig. 2b, and then, after acquiring the eye movement data of the user browsing the target page, the user equipment may analyze the eye movement data and construct a directed acyclic graph for the target page by itself, or may send the eye movement data to the data processing equipment in the system shown in fig. 2a or fig. 2b, so that the data processing equipment analyzes the eye movement data and constructs a directed acyclic graph for the target page, which is not described in detail later.

402. Based on the eye movement data, browsing behavior of the user on a plurality of items is determined, and the plurality of items are located in a plurality of lists of the target page.

After the eye movement data of the target page browsed by the user is obtained, the eye movement data can be analyzed, so that browsing behaviors of the user on a plurality of items in the target page are obtained, the plurality of items are respectively located in a plurality of lists of the target page, and therefore the plurality of items can be represented as a set i ═_1,1，i_1,2，...，i_o,q]Wherein i is_t,jRepresents the jth entry in the tth list in the target page, t 1.

Specifically, the eye movement data of the target page browsed by the user is obtained, and the following analysis can be performed based on the eye movement data:

(1) three items that are arbitrarily positioned consecutively in the set i are referred to as item a, item B, and item C, respectively. In the destination page, item A precedes and is adjacent to item B (item A and item B can be two items in two adjacent lists, i.e., t)_A＜t_BFor example, item A is the last 1 item of the 1 st list in the destination page, item B is the 1 st item of the 2 nd list in the destination page, and so on. Item A and item B can also be two items in the same list, with item A being positioned further forward, i.e. t_A＝t_BAnd j is_A＜j_BFor example, item a is the 1 st item of the 1 st list in the destination page, item B is the 2 nd item of the 1 st list in the destination page, etc.), item B precedes item C and item B is adjacent to item C.

Based on the eye movement data, the browsing sequence of the user to the item A, the item B and the item C can be counted, and the statistical result is shown in Table 1:

TABLE 1

Browsing sequence	1	2	3	Number of times of browsing	Ratio of occupation of
						Sequence 1	A	B	C	5771	43.03％
Sequence 2	B	A	B	1668	12.44％
						Sequence 3	A	B	A	1627	12.13％
Sequence 4	C	B	A	1320	9.84％
						Sequence 5	B	A	C	858	6.40％
Sequence 6	A	C	B	846	6.31％
						Sequence 7	B	C	A	705	5.26％
Sequence 8	C	A	B	615	4.59％

As can be seen from table 1, the browsing sequence of the item a → the item B → the item C takes the greatest weight, which indicates that when the user browses a plurality of items in the target page, the user mainly browses according to the sequence (sort) of the plurality of items in the target page, and this browsing behavior may be referred to as a sequential browsing behavior (sequential browsing behavior), which includes two main categories, which will be described below:

(1.1) the first-class sequential browsing behavior means that for any list of the target page, if the list is a horizontal list, all items in the list are browsed in the order from left to right, and if the list is a vertical list, all items in the list are browsed in the order from top to bottom. Still as in the example shown in FIG. 5, for List B₂The first type of sequential browsing behavior is: according to i_2,1→i_2,2→i_2,3→i_2,4→i_2,5In order of (2) to browse List B₂In_2,1、i_2,2、i_2,3、i_2,4And i_2,5These 5 items. For list B₁The first type of sequential browsing behavior is: according to i_1,1→i_1,2→i_1,3In order of (2) to browse List B₁In_1,1、i_1,2And i_1,3These 3 items.

(1.2) the second type of sequential browsing behavior refers to that for two adjacent lists in the target page, the adjacent items in the two lists are browsed according to the front-back sequence of the adjacent items. It should be noted that, if the previous list in the two lists is a vertical list, the next list is a horizontal list, and the adjacent items in the two lists include the last item in the vertical list and all items in the horizontal list, that is, all items in the horizontal list can be regarded as items adjacent to the last item in the vertical list. If the previous list is a horizontal list and the next list is a vertical list, the adjacent items in the two lists include the first item in the vertical list and all the items in the horizontal list, that is, all the items in the horizontal list can be regarded as the items adjacent to the first item in the vertical list. Still as in the example shown in FIG. 5, for List B₁And B₂The second type of sequential browsing behavior is: according to i_1,3→i_2,1、i_1,3→i_2,2、i_1,3→i_2,3、i_1,3→i_2,4、i_1,3→i_2,5In order of (2) to browse List B₁And B₂In_1,3、i_2,1、i_2,2、i_2,3、i_2,4And i_2,5These 6 adjacent items. For list B₂And B₃The second type of sequential browsing behavior is: according to i_2,1→i_3,1、i_2,2→i_3,1、i_2,3→i_3,1、i_2,4→i_3,1、i_2,5→i_3,1In order of (2) to browse List B₂And B₃In_2,1、i_2,2、i_2,3、i_2,4、i_2,5And i_3,1These 6 adjacent items.

It will be understood that if both lists are horizontal lists, then for any one item of the previous list, all items in the next list can be considered as items adjacent to that item, and likewise, for any one item of the next list, all items in the previous list can be considered as items adjacent to that item.

(2) When browsing the target page, the user may jump directly from the current list to another list for browsing, and the current list is separated from the other list by at least one list. For convenience of introduction, the current list is referred to as list D, another list is referred to as list E, list D is separated from list E by at least one list, and a list skip length (skip length) l ═ t between list D and list E is defined_D-t_EFor example, when the list D is the 2 nd list in the target page and the list D is the 4 th list in the target page, l is 4-2 is 2. Based on the eye movement data, different list skip lengths can be counted, and the statistical results are shown in table 2:

TABLE 2

List skip length	2	3	4	5	6
						Number of	439	34	17	1	2
Ratio of occupation of	89.05％	6.90％	3.45％	0.20％	0.41％

As can be seen from table 2, the browsing manner with the list skipping length of 2 has the greatest weight, which indicates that the user often sends a behavior of skipping an entire list and directly browsing the next list in addition to the sequential browsing behavior, and this browsing behavior may be referred to as a skipping behavior (block skip). It is noted that, if the target page is a plurality of horizontal lists (also referred to as horizontal (Vertical) blocks) and a plurality of Vertical lists (also referred to as Vertical (Vertical) blocks) staggered pages (i.e., F-type pages), the skip behavior with the list skip length of 2 includes two categories of behavior, the first category of skip behavior refers to skipping from the horizontal list to the horizontal list, and the second category of skip behavior refers to skipping from the Vertical list to the Vertical list, wherein almost all skip behaviors skip from the Vertical list to the Vertical list in a percentage of 94.5%, and the skip behavior from the horizontal list to the horizontal list in a percentage of only 5.5%, indicating that the user is more inclined to skip from the Vertical list to the Vertical list.

Based on this, it can further count the skip action from the vertical list to the vertical list, which mainly occurs on the item of the two lists, i.e. count the starting item and the ending item of the skip action, and the statistical result is shown in table 3:

TABLE 3

Based on table 3, the maximum probability of the starting item of the user's skip action is the last item in a list, and the maximum probability of the ending item of the user's skip action is the first item in another list.

Then, based on the above analysis, the skipping behavior of the user can be summarized as: the user jumps to the first item of the next list to continue browsing in two non-adjacent lists (which are separated by one list) from the last item of the previous list, and the browsing order between the two items can be referred to as a jump order. Still as in the example shown in FIG. 5, the user is browsing i_1,3When directly skipping over B₂Browse i_3，1。

(3) As can be seen from table 1, in addition to the sequential browsing behaviors, the browsing behaviors performed in the order of B → a → B and the browsing behaviors performed in the order of a → B → a, which are most common among the non-sequential browsing behaviors, indicate that the user tends to repeatedly browse two adjacent items to perform the comparison between the items, and such browsing behaviors may be referred to as comparison behaviors (compare behaviors).

Therefore, after the statistical data are analyzed, the browsing behaviors of the user on a plurality of items in the target page can be determined, wherein the browsing behaviors comprise sequential browsing behaviors, skipping behaviors, comparison behaviors and the like.

403. And connecting a plurality of items based on the browsing behavior to obtain the directed acyclic graph.

After the browsing behaviors of the user on the plurality of items in the target page are obtained, the plurality of items in the target page can be connected based on the browsing behaviors, and the directed acyclic graph for the target page is obtained.

In particular, these browsing behaviors include two broad categories of browsing behaviors. The first browsing behavior refers to that the user browses the items in the same list, and includes the aforementioned first-type sequential browsing behavior, so the browsing sequence of the user in the same list may be referred to as a first sequence, where the first sequence includes a sequence from top to bottom and a sequence from left to right in the first-type sequential browsing behavior according to which the user browses all the items in the same list. The second browsing behavior refers to a browsing sequence of the user between different lists, which includes the aforementioned second sequential browsing behavior and the comparison behavior, so the browsing sequence of the user between different lists can be referred to as a second sequence, where the second sequence includes a front-back sequence according to which the user browses adjacent items in two adjacent lists in the second sequential browsing behavior, and a skip sequence according to which the user browses two items in two non-adjacent lists in the comparison behavior. Then, a directed acyclic graph for the target page may be obtained by:

(1) in the target page, the items in the same list browsed by the user in the first order are connected in the first order, that is, for any one list in the target page, if the list is a horizontal list, all the items in the list are connected in the order from left to right, and if the list is a vertical list, all the items in the list are connected in the order from top to bottom, so that the connection inside each list in the target page can be completed. For example, as shown in fig. 7 (fig. 7 is a schematic diagram of a directed acyclic graph provided by an embodiment of the present application, and fig. 7 is drawn on the basis of fig. 5), for list B₁Can be according to i_1,1→i_1,2→i_1,3In order of (2) to connect to the list B₁In_1,1、i_1,2And i_1,3These 3 items. For the columnTABLE B₂Can be according to i_2,1→i_2,2→i_2,3→i_2,4→i_2,5In order of (2) to join list B₂In_2,1、i_2,2、i_2,3、i_2,4And i_2,5These 5 items. For list B₃、B₄And B₅Also, the details are not repeated here, so that the connections within the 5 lists in the target page are completed.

(2) Connecting the items in different lists browsed by the user in the second order according to the second order to obtain the directed acyclic graph, that is, for two adjacent lists in the target page, the adjacent items can be connected according to the front-back order of the adjacent items in the two lists, and for two non-adjacent lists in the target page (the two lists are separated by one list), the last item of the previous list and the first item of the next list in the two lists can be connected according to the jumping order when the user browses the two lists, so that the connection between the lists in the target page can be completed, and the directed acyclic graph can be understood. Still as in the example shown in FIG. 7, for List B₁And B₂Can be according to i_1,3→i_2,1、i_1,3→i_2,2、i_1,3→i_2,3、i_1,3→i_2,4、i_1,3→i_2,5In the order of (a) to (b), i_1,3And i_2,1Is connected with i_1,3And i_2,2Is connected with i_1,3And i_2,3Is connected with i_1,3And i_2,4Is connected with i_1,3And i_2,5Connect, for list B₂And B₃List B₃And B₄List B₄And B₅As such, will not be described in detail herein. Further, for list B₁And B₃According to i_1,3→i_3,1In the order of (a) and (b) are_1,3And i_3,1Connect, for list B₃And B₅And so on, and will not be described in detail herein. In this way, a directed acyclic graph for the target page can be obtained.

In the embodiment of the application, browsing behaviors of a user on a plurality of items in a target page may be determined based on eye movement data generated when the user browses the target page, and then the browsing behaviors (for example, sequential browsing behaviors and skipping behaviors) often determine a browsing sequence of the user on the items (for example, a browsing sequence of the user in the same list and a browsing sequence of the user between different lists), so that the plurality of items of the target page are connected according to the browsing to obtain a directed acyclic graph for the target page, and the directed acyclic graph can be used for user behavior prediction on the target page in the following process.

The above is a detailed description of the directed acyclic graph construction method provided in the embodiment of the present application, and the following describes a user behavior prediction method provided in the embodiment of the present application. Fig. 8 is a schematic flowchart of a user behavior prediction method provided in an embodiment of the present application, and as shown in fig. 8, the method includes:

801. and acquiring a first characteristic of the first item and a second characteristic of the second item through the target model, wherein the first item and the second item are positioned in different lists or the same list of the target page, and the second item is positioned before the first item.

In this embodiment, when the target page is required to perform user behavior prediction, that is, to obtain the probability that each item in the target page is clicked by the user, first features of each item in the target page may be extracted first, it should be noted that, for any item, the first features of the item refer to attribute information of the item itself, for example, when the item is an application in a page of an application mall, the first features of the application may include a developer of the application, a size of the application, a type of the application, an icon of the application, and the like, and when the item is a certain item in a page of a shopping mall, the first features of the item may include a price of the item, a type of the item, a color of the item, and the like.

It should be noted that, since the target page includes a plurality of items, a plurality of rounds of operations may be performed on the target page, and one round of operations may be performed on one item in the target page (i.e., each round may perform steps 801 to 805 once, that is, steps 801 to 805 may be performed on each item). Based on this, the present embodiment schematically introduces one of the items in the destination page, and refers to the item as the first item.

Then, when the probability that the first item is clicked by the user is to be predicted, the first feature of the first item and the second feature of the second item may be input to the target model (trained neural network model). The first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item, that is, the position relationship between the first item and the second item has the following two situations: (1) the first item and the second item may be items in the same list, the second item preceding the first item and the second item being adjacent to the first item. (2) The first item and the second item may be items in different lists, the list in which the second item is located is before the list in which the first item is located, and the second item may be adjacent to the first item or may not be adjacent to the first item.

Since all items in the target page can form the directed acyclic graph, any one of the items can be regarded as a node in the directed acyclic graph, and a unidirectional connection relationship exists between the nodes in the directed acyclic graph. Therefore, the directed acyclic graph has a parent node and a child node, and the connection direction between the parent node and the child node is pointed to the child node by the parent node. Then, the aforementioned first item can be regarded as a child node in the directed acyclic graph for the target page, and the second item is all parent nodes of the child node, as shown in the example of fig. 7, when the first item is i_1,2When the second item is i_1,1. When the first item is i_2,1When the second item is i_1,3. When it comes toAn item is i_2,2When the second item is i_2,1And i_1,3. When the first item is i_3,1When the second item is i_2,1、i_2,2、i_2,3、i_2,4、i_2,5And i_1,3And so on.

It should be understood that the process of acquiring the second feature of the second item may refer to the process of acquiring the second feature of the subsequent first item, which is not described herein again.

It should also be appreciated that when the first item is the first item in the destination page (e.g., i in the example shown in FIG. 7)_1,1) If the second item does not exist, the second feature of the second item can be understood as a preset value (the size of the preset value can be set according to actual needs, and is not limited herein), so that the first feature of the first item and the preset value can also be input into the target model.

802. And acquiring a second feature of the first item based on the first feature of the first item and the second feature of the second item through the target model.

After the first feature of the first item and the second feature of the second item are input into the target model, the first feature of the first item and the second feature of the second item can be processed through the target model, and therefore the second feature of the first item is obtained.

Specifically, the second feature of the first item may be acquired by:

(1) extracting the request of the user for the target page and the probability of the second item being clicked by the user, wherein the request of the user for the target page may contain a keyword or the like which is input by the user on the target page and is used for searching for some items, and the probability of the second item being clicked by the user may be understood as the probability of the item targeted by the previous turn (namely, the previous item of the first item) being clicked by the user.

(2) On the basis of inputting the first feature of the first item and the second feature of the second item into the target model, the request of the user for the target page and the probability of the second item being clicked by the user can be further input into the target model, so that the target model maps the first feature of the first item, the request of the user for the target page and the probability of the second item being clicked by the user on the hidden space (i.e., the mapping processing described above), correspondingly obtains the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item, and then splices the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item (i.e., the second fusion processing described above) to obtain the fourth feature of the first item. Meanwhile, the target model can also perform self attention mechanism (self attention) based processing on the second features of the second item, so as to obtain fifth features of the first item.

For example, as shown in fig. 9 (fig. 9 is a schematic structural diagram of an object model provided in the embodiment of the present application), let the first item be i_t,jThe set of the second item is P_t,jThe k-th second item in the set is i_k(k 1.. n), the first characteristic of the first item is I, and the second characteristic of the kth second item is h_kThe request of the user to the target page is Q, and the probability that the second item is clicked by the user is C.

Combining the first feature I of the first item with the second feature h of the second item₁、...、h_nAfter a request Q of a user for a target page and the probability C of clicking a second item by the user are input into the target model, the target model can map the first feature I of the first item on a hidden space to obtain a sixth feature V of the first item_IMapping the request Q of the user to the target page on the hidden space to obtain a seventh characteristic V of the first item_QMapping the probability C of the second item clicked by the user on the hidden space to obtain an eighth feature V of the first item_C. The object model may then transform the sixth feature V of the first item_ISeventh feature V of the first item_QEighth feature V of the first item_CSplicing to obtain a fourth feature x of the first item_t,j。

At the same time, the target model may also utilize the self-attention mechanism on a second feature h of a second item₁、...、h_nCalculating to obtain the fifth characteristic e of the first item_t,j. Wherein is based onThe attention mechanism is calculated as shown in the following equation:

a_k＝softmax_k(MLP(h_k)),i_k∈P_t,j

(3) the fourth feature of the first item and the fifth feature of the first item are obtained, and the target model may process the fourth feature of the first item and the fifth feature of the first item by using a recurrent neural unit (GRUcell) (i.e., the first fusion process described above), so as to obtain the second feature of the first item.

Still as in the example shown in FIG. 9, the fourth feature x of the first item is obtained_t,jAnd fifth feature e of the first item_t,jThe target model can input the two features into the recurrent neural unit for processing to obtain a second feature h of the first item_t,j. The processing implemented by the recurrent neural unit is shown in the following formula:

h_t,j＝GRUcell(x_t,j,e_t,j) (3)

it will be appreciated that the second characteristic of the first item may be indicative of the effect of the second item on the first item (which may also be understood as a relationship between the second item and the first item), i.e. the effect of the item viewed by the user during the sequential browsing behavior and the skipping behavior on the first item when the user browses to the first item.

It should be understood that, in the process of obtaining the fourth feature of the first feature, the request of the user for the target page and the probability of the second item being clicked by the user may not be input into the target model, so that the target model directly continues to perform the mapping processing on the first feature of the first item, and then the fourth feature of the first feature is obtained.

803. And acquiring a first characteristic of a third item through the target model, wherein the first item and the third item are positioned in different lists or the same list of the target page, and the third item is adjacent to the first item.

In addition, the first feature of a third item can be input into the target model, wherein the first item and the third item are located in different lists or the same list of the target page, and the third item is adjacent to the first item, that is, the position relationship between the first item and the third item has the following two cases: (1) the first item and the third item may be items in the same list, the third item being adjacent to the first item. (2) The first item and the third item may be items in different lists, the third item being adjacent to the first item.

As still shown in the example of FIG. 7, when the first item is i_1,2When the third item is i_1,1. When the first item is i_1,3When the third item is i_1,2And i_2,1. When the first item is i_2,1When the second item is i_1,3、i_2,2And i_3,1And so on.

804. And acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item through the target model.

After the first feature of the third item is input into the target model, the first feature of the first item and the first feature of the third item can be processed through the target model, and therefore the third feature of the first item is obtained.

Specifically, the third feature of the first item may be acquired by:

(1) the target model maps the first feature of the first item, the request of the user to the target page, and the first feature of the third item in the hidden space (i.e., the mapping process described above), so as to obtain the sixth feature of the first item, the seventh feature of the first item, and the ninth feature of the first item, respectively.

As still shown in the example of FIG. 9, let the set of third items be N_t,jThe f-th third item in the set is i_f(f 1.. said, m), the first characteristic of the f-th third item is I_f。

The first characteristic I of the third item₁、...、I_mAfter inputting the target model, the target modelThe first feature I of the first item may be mapped on the hidden space to obtain a sixth feature V of the first item_IMapping the request Q of the user to the target page on the hidden space to obtain a seventh characteristic V of the first item_QThe first characteristic I of the third item₁、...、I_mMapping on the hidden space to obtain a ninth feature V of the first item₁、...、V_m。

(2) Then, the target model may calculate the sixth feature of the first item and the ninth feature of the first item through the comparison function, and perform weighted summation based on the calculation result (calculation of the comparison function and weighted summation calculation, that is, the aforementioned third fusion processing), so as to obtain the tenth feature of the first item.

As in the example shown in FIG. 9, the sixth feature V of the first item is obtained_IAnd ninth feature V of the first item₁、...、V_mThen, the features can be subjected to computation based on the comparison function g and weighted summation computation to obtain a tenth feature d of the first item_t,j. Wherein, the calculation process is shown as the following formula:

γ_f＝soft max_f(g(V_I,V_f)),i_f∈N_t,j

wherein, the comparison function g can be one of the following three functions: inner product function

Function of neural network

Kernel function

(3) Finally, the target model may perform an exclusive nor operation (i.e., the fourth fusion processing) on the sixth feature of the first item, the seventh feature of the first item, and the tenth feature of the first item, so as to obtain the third feature of the first item.

Still as in the example shown in FIG. 9, the tenth feature d of the first item is obtained_t,jThe sixth feature V of the first item_ISeventh feature V of the first item_QAnd tenth feature d of the first item_t,jPerforming an exclusive OR operation to obtain a third feature cp of the first item_t,j. Wherein, the operation process is shown as the following formula:

cp_t，j＝d_t，j⊙V_I⊙V_Q (5)

it is understood that the third feature of the first item may represent an influence of the third item on the first item (which may also be understood as a relationship between the third item and the first item), that is, when the user browses the first item by using the contrast action, the influence of the item browsed by the user in the course of performing the action on the first item.

It should be understood that, when the third feature of the first item is obtained, the user request for the target page may not be input into the target model, so that the target model may perform the fourth fusion process only on the sixth feature of the first item and the tenth feature of the first item to obtain the third feature of the first item.

805. And acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the target model.

After obtaining the second feature of the first item and the third feature of the first item, the target model may calculate the second feature of the first item and the third feature of the first item, so as to obtain a probability that the first item is clicked by the user.

Still as in the example shown in FIG. 9, the second feature h of the first item is obtained_t,jAnd a third feature cp of the first item_t,jThen, the two characteristics can be calculated to obtain the probability C that the first item is clicked by the user_t,j. Wherein, the calculation process is shown as the following formula:

similarly, for the other items except the first item in the target page, the operation similar to the operation performed on the first item may also be performed, so that the probability that all the items in the target page are clicked by the user may be obtained, thereby completing the user behavior prediction on the target page.

It should be understood that, when obtaining the probability that the first item is clicked by the user, step 803 and step 804 may not be executed, so that the target model directly calculates the second feature of the first item, and obtains the probability that the first item is clicked by the user.

In addition, the predicted result of the target model provided by the embodiment of the present application can be compared with the predicted result of the model of the related art, and the comparison result is shown in table 4:

TABLE 4

Based on table 4, it can be seen that the prediction capability exhibited by the target model provided in the embodiment of the present application is significantly improved in both indexes compared with the model provided in the related art.

Furthermore, the target model provided in the embodiment of the present application not only introduces a conventional sequential browsing behavior, but also introduces browsing behaviors such as a jump behavior and a contrast behavior, that is, the target model may consider the influence of the items browsed by the user in the process of performing these behaviors on the first item when the user browses the first item using these complex and diverse browsing behaviors, so as to further improve the accuracy of the probability that the first item finally obtained by the target model is clicked by the user.

Furthermore, when the target model provided in the embodiment of the present application analyzes the first item, not only the influence of the attribute information of the first item itself is considered, but also the influence of external factors such as a request of a user for a target page and the probability of clicking the second item by the user is considered, so as to further improve the accuracy of the probability of clicking the first item by the user, which is finally obtained by the target model.

The foregoing is a detailed description of the user behavior prediction method provided in the embodiment of the present application, and the following describes a model training method provided in the embodiment of the present application. Fig. 10 is a schematic flowchart of a model training method provided in an embodiment of the present application, and as shown in fig. 10, the method includes:

1001. and acquiring a first feature of a first item and a second feature of a second item through the model to be trained, wherein the first item and the second item are positioned in different lists or the same list of the page to be processed, and the second item is positioned before the first item.

In this embodiment, when the model to be trained needs to be trained, a batch of training data may be obtained first, where the batch of training data includes a page to be processed, the page to be processed includes a plurality of lists, and each list includes at least one item. It is noted that in the pending page, the true probability that any one item was clicked on by the user is known.

It should be noted that, regarding the first item, the second item, the first feature of the first item, and the second feature of the second item of the page to be processed, reference may be made to relevant description portions of the first item, the second item, the first feature of the first item, and the second feature of the second item of the target page in step 801 in the embodiment shown in fig. 8, and details are not repeated here.

It is understood that the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusion based on the attribute information of the first item, and the second feature of the second item is information obtained by fusion based on the attribute information of the second item (i.e., the first feature of the second item).

1002. And acquiring a second feature of the first item based on the first feature of the first item and the second feature of the second item through the model to be trained.

After the first characteristic of the first item and the second characteristic of the second item are input into the model to be trained, the first characteristic of the first item and the second characteristic of the second item can be processed through the model to be trained, and therefore the second characteristic of the first item is obtained.

In a possible implementation manner, the first feature of the first item is subjected to mapping processing, so as to obtain a fourth feature of the first item: mapping the first characteristic of the first item, the request of the user for the page to be processed and the probability of the second item being clicked by the user to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item;

and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item to obtain a fourth feature of the first item.

It should be noted that, for the introduction of step 1002, reference may be made to a relevant description part of step 802 in the embodiment shown in fig. 8, and details are not repeated here.

1003. And acquiring a first feature of a third item through the model to be trained, wherein the first item and the third item are positioned in different lists or the same list of the page to be processed, and the third item is adjacent to the first item.

In this embodiment, the first feature of the third item may also be input into the target model, and it should be noted that, regarding the third item of the page to be processed and the first feature of the third item, reference may be made to a relevant description part of the third item of the target page and the first feature of the third item in step 803 in the embodiment shown in fig. 8, which is not repeated herein.

1004. And acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item through the model to be trained.

After the first feature of the third item is input into the model to be trained, the first feature of the first item and the first feature of the third item can be processed through the model to be trained, and therefore the third feature of the first item is obtained.

It should be noted that, for the introduction of step 1004, reference may be made to a relevant description part of step 804 in the embodiment shown in fig. 8, and details are not repeated here.

1005. And acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the model to be trained.

After the second feature of the first item and the third feature of the first item are obtained, the second feature of the first item and the third feature of the first item may be processed through the model to be trained, so as to obtain a probability that the first item is clicked by the user (which may also be referred to as a predicted probability that the first item is clicked by the user).

It should be noted that, for the introduction of step 1005, reference may be made to a related description part of step 805 in the embodiment shown in fig. 8, and details are not repeated here.

1006. And acquiring a target loss based on the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user, wherein the target loss is used for indicating the difference between the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user.

After the prediction probability of the first item clicked by the user is obtained, the prediction probability of the first item clicked by the user and the real probability of the first item clicked by the user are calculated through a preset target loss function, and a target loss is obtained and used for indicating the difference between the prediction probability of the first item clicked by the user and the real probability of the first item clicked by the user.

1007. And updating the parameters of the model to be trained based on the target loss until the model training condition is met, thereby obtaining the target model.

After the target loss is obtained, parameters of the model to be trained may be updated based on the target loss, a next batch of training data is obtained, and the model to be trained after the parameters are updated is continuously trained by using the next batch of training data (i.e., steps 1001 to 1007 are executed again) until the model training condition is satisfied (e.g., the target loss reaches convergence, etc.), so as to obtain the target model in the embodiment shown in fig. 8.

The target model obtained by training in the embodiment of the application has the capability of predicting the user behavior of the page. When it is desired to predict a probability that a first item in the target page is clicked by the user, a first characteristic of the first item and a second characteristic of a second item may be input to the target model, where the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the object model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

The above is a detailed description of the model training method provided in the embodiments of the present application, and the following describes the apparatus and device provided in the embodiments of the present application. Fig. 11 is a schematic structural diagram of a user behavior prediction apparatus according to an embodiment of the present application, and as shown in fig. 11, the apparatus includes:

a first obtaining module 1101, configured to obtain, through a target model, a first feature of a first item and a second feature of a second item, where the first item and the second item are located in different lists or a same list of a target page, and the second item is located before the first item;

a second obtaining module 1102, configured to obtain, by using the target model, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusing based on the attribute information of the first item, and the second feature of the second item is information obtained by fusing based on attribute information of the second item (that is, the first feature of the second item);

a third obtaining module 1103, configured to obtain, by using the target model, a probability that the first item is clicked by the user based on the second feature of the first item.

In one possible implementation, the apparatus further includes: the fourth obtaining module is used for obtaining a first feature of a third item through the target model, the first item and the third item are located in different lists or the same list of the target page, and the third item is adjacent to the first item; the fifth obtaining module is used for obtaining a third feature of the first item based on the first feature of the first item and the first feature of the third item through the target model; a third obtaining module 1103, configured to obtain, by using the target model, a probability that the first item is clicked by the user based on the second feature of the first item and the third feature of the first item.

In a possible implementation manner, the second obtaining module 1102 is configured to: mapping the first characteristics of the first item through the target model to obtain fourth characteristics of the first item; processing the second characteristic of the second item based on a self-attention mechanism through the target model to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item through the target model to obtain a second feature of the first item.

In a possible implementation manner, the second obtaining module 1102 is configured to: mapping the first characteristic of the first item, the request of the user for the target page and the probability of the second item being clicked by the user through the target model to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item through the target model to obtain a fourth feature of the first item.

In a possible implementation manner, the fifth obtaining module is configured to: mapping the first characteristic of the first item and the first characteristic of the third item through the target model to obtain a sixth characteristic of the first item and a ninth characteristic of the first item; performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item through the target model to obtain a tenth feature of the first item; and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item through the target model to obtain a third feature of the first item.

Fig. 12 is a schematic structural diagram of a directed acyclic graph building apparatus according to an embodiment of the present application, and as shown in fig. 12, the apparatus includes:

an obtaining module 1201, configured to obtain eye movement data of a user browsing a target page;

a determining module 1202, configured to determine, based on the eye movement data, browsing behaviors of the user on a plurality of items, where the plurality of items are located in a plurality of lists of the target page;

a connecting module 1203, configured to connect the multiple items based on the browsing behavior, so as to obtain a directed acyclic graph.

In a possible implementation manner, the connecting module 1203 is configured to connect items in the same list browsed by the user in a first order according to the first order, and connect items in different lists browsed by the user in a second order according to the second order, so as to obtain a directed acyclic graph.

In a possible implementation manner, the obtaining module 1201 is configured to collect eye movement data of a user browsing a target page through an eye movement instrument.

Fig. 13 is a schematic structural diagram of a model training apparatus according to an embodiment of the present application, and as shown in fig. 13, the apparatus includes:

a first obtaining module 1301, configured to obtain, through a to-be-trained model, a first feature of a first item and a second feature of a second item, where the first item and the second item are located in different lists or the same list of a to-be-processed page, and the second item is located before the first item;

a second obtaining module 1302, configured to obtain, by using the model to be trained, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature of the first item is attribute information of the first item, the second feature of the first item is information obtained by fusing attribute information based on the first item, and the second feature of the second item is information obtained by fusing attribute information based on the second item (that is, the first feature of the second item);

the third obtaining module 1303 is configured to obtain, through the model to be trained, a probability that the first item is clicked by the user based on the second feature of the first item;

a fourth obtaining module 1304, configured to obtain a target loss based on a probability that the first item is clicked by the user and a true probability that the first item is clicked by the user, where the target loss is used to indicate a difference between the probability that the first item is clicked by the user and the true probability that the first item is clicked by the user;

the updating module 1305 is configured to update parameters of the model to be trained based on the target loss until the model training condition is met, so as to obtain the target model.

The target model obtained by training in the embodiment of the application has the capability of predicting the user behavior of the page. When it is required to predict the probability that a first item in a target page is clicked by a user, a first feature of the first item and a second feature of a second item may be input to a target model, wherein the first item and the second item are located in different lists or the same list of the target page, and the second item is located before the first item. Then, the object model may obtain the second feature of the first item based on the first feature of the first item and the second feature of the second item, and then obtain the probability that the first item is clicked by the user based on the second feature of the first item. In the process, when the target model obtains the probability that the first item is clicked by the user, the influence of a second item located before the first item on the first item is considered, and the second item can be not only the item in the list where the first item is located but also the items in other lists, so that the factors considered by the target model are comprehensive, and the actual situation of the user when the user browses the first item in a target page can be fitted, so that the probability that the first item is clicked by the user, which is finally obtained by the target model, has higher accuracy, and is beneficial to accurately recommending the interested items for the user in the following process.

In one possible implementation, the apparatus includes: the fifth acquisition module is used for acquiring the first characteristic of a third item through the model to be trained, wherein the first item and the third item are positioned in different lists or the same list of the page to be processed, and the third item is adjacent to the first item; the sixth obtaining module is used for obtaining a third feature of the first item based on the first feature of the first item and the first feature of the third item through the model to be trained; the third obtaining module 1303 is configured to obtain, through the model to be trained, a probability that the first item is clicked by the user based on the second feature of the first item and the third feature of the first item.

In a possible implementation manner, the second obtaining module 1302 is configured to: mapping the first characteristic of the first item to obtain a fourth characteristic of the first item; processing the second characteristic of the second item based on a self-attention mechanism to obtain a fifth characteristic of the first item; and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item to obtain a second feature of the first item.

In a possible implementation manner, the second obtaining module 1302 is configured to: mapping the first characteristic of the first item, the request of the user for the page to be processed and the probability of the second item being clicked by the user to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item; and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item to obtain a fourth feature of the first item.

It should be noted that, because the contents of information interaction, execution process, and the like between the modules/units of the apparatus are based on the same concept as the method embodiment of the present application, the technical effect brought by the contents is the same as the method embodiment of the present application, and specific contents may refer to the description in the foregoing method embodiment of the present application, and are not repeated herein.

The embodiment of the present application further relates to an execution device, and fig. 14 is a schematic structural diagram of the execution device provided in the embodiment of the present application. As shown in fig. 14, the execution device 1400 may be embodied as a mobile phone, a tablet, a notebook, a smart wearable device, a server, and the like, which is not limited herein. The execution device 1400 may be deployed with the user behavior prediction apparatus described in the embodiment corresponding to fig. 11 and the directed acyclic graph constructing apparatus described in the embodiment corresponding to fig. 12, and is configured to implement the function of constructing the directed acyclic graph in the embodiment corresponding to fig. 4 and the function of predicting the user behavior in the embodiment corresponding to fig. 8. Specifically, the execution device 1400 includes: a receiver 1401, a transmitter 1402, a processor 1403 and a memory 1404 (wherein the number of processors 1403 in the performing device 1400 may be one or more, for example one processor in fig. 14), wherein the processor 1403 may comprise an application processor 14031 and a communication processor 14032. In some embodiments of the present application, the receiver 1401, the transmitter 1402, the processor 1403, and the memory 1404 may be connected by a bus or other means.

The memory 1404 may include both read-only memory and random access memory and provides instructions and data to the processor 1403. A portion of memory 1404 may also include non-volatile random access memory (NVRAM). The memory 1404 stores a processor and operating instructions, executable modules or data structures, or a subset thereof, or an expanded set thereof, wherein the operating instructions may include various operating instructions for performing various operations.

The processor 1403 controls the operation of the execution apparatus. In a particular application, the various components of the execution device are coupled together by a bus system that may include a power bus, a control bus, a status signal bus, etc., in addition to a data bus. For clarity of illustration, the various buses are referred to in the figures as a bus system.

The method disclosed in the embodiments of the present application may be applied to the processor 1403, or implemented by the processor 1403. The processor 1403 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method can be performed by hardware integrated logic circuits or instructions in software form in the processor 1403. The processor 1403 may be a general-purpose processor, a Digital Signal Processor (DSP), a microprocessor or a microcontroller, and may further include an Application Specific Integrated Circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. The processor 1403 may implement or perform the methods, steps, and logic blocks disclosed in the embodiments of the present application. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in the memory 1404, and the processor 1403 reads the information in the memory 1404 and completes the steps of the above method in combination with the hardware thereof.

The receiver 1401 may be used to receive input numeric or character information and to generate signal inputs related to performing relevant settings and function control of the device. The transmitter 1402 may be used to output numeric or character information through a first interface; the transmitter 1402 may also be configured to send instructions to the disk pack via the first interface to modify data in the disk pack; the transmitter 1402 may also include a display device such as a display screen.

In this embodiment, in one case, the processor 1403 is configured to perform user behavior prediction on a target page through the target model in the embodiment corresponding to fig. 8.

The embodiment of the present application further relates to a training device, and fig. 15 is a schematic structural diagram of the training device provided in the embodiment of the present application. As shown in FIG. 15, the training apparatus 1500 is implemented by one or more servers, where the training apparatus 1500 may vary widely depending on configuration or performance, and may include one or more Central Processing Units (CPUs) 1514 (e.g., one or more processors) and memory 1532, one or more storage media 1530 (e.g., one or more mass storage devices) that store applications 1542 or data 1544. Memory 1532 and storage media 1530 may be, among other things, transient or persistent storage. The program stored on the storage medium 1530 may include one or more modules (not shown), each of which may include a sequence of instructions for operating on the exercise device. Still further, a central processor 1514 may be provided in communication with the storage medium 1530, executing a series of instruction operations in the storage medium 1530 on the exercise device 1500.

Training apparatus 1500 can also include one or more power supplies 1526, one or more wired or wireless network interfaces 1550, one or more input-output interfaces 1558; or, one or more operating systems 1541, such as Windows Server, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

Specifically, the training apparatus may perform the model training method in the embodiment corresponding to fig. 10.

The present embodiment also relates to a computer storage medium, in which a program for signal processing is stored, which, when running on a computer, causes the computer to perform the steps performed by the aforementioned execution apparatus, or causes the computer to perform the steps performed by the aforementioned training apparatus.

Embodiments of the present application also relate to a computer program product having instructions stored thereon, which, when executed by a computer, cause the computer to perform the steps performed by the aforementioned execution apparatus, or cause the computer to perform the steps performed by the aforementioned training apparatus.

The execution device, the training device, or the terminal device provided in the embodiment of the present application may specifically be a chip, where the chip includes: a processing unit, which may be for example a processor, and a communication unit, which may be for example an input/output interface, a pin or a circuit, etc. The processing unit may execute the computer execution instructions stored in the storage unit to enable the chip in the execution device to execute the data processing method described in the above embodiment, or to enable the chip in the training device to execute the data processing method described in the above embodiment. Optionally, the storage unit is a storage unit in the chip, such as a register, a cache, and the like, and the storage unit may also be a storage unit located outside the chip in the wireless access device, such as a read-only memory (ROM) or another type of static storage device that can store static information and instructions, a Random Access Memory (RAM), and the like.

Specifically, referring to fig. 16, fig. 16 is a schematic structural diagram of a chip provided in the embodiment of the present application, where the chip may be represented as a neural network processor NPU 1600, and the NPU 1600 is mounted on a main CPU (Host CPU) as a coprocessor, and the Host CPU allocates tasks. The core part of the NPU is an arithmetic circuit 1603, and the controller 1604 controls the arithmetic circuit 1603 to extract matrix data in the memory and perform multiplication.

In some implementations, the arithmetic circuit 1603 includes a plurality of processing units (PEs) therein. In some implementations, the arithmetic circuitry 1603 is a two-dimensional systolic array. The arithmetic circuit 1603 may also be a one-dimensional systolic array or other electronic circuit capable of performing mathematical operations such as multiplication and addition. In some implementations, the arithmetic circuitry 1603 is a general-purpose matrix processor.

For example, assume that there is an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit fetches the data corresponding to matrix B from the weight memory 1602 and buffers it in each PE in the arithmetic circuit. The arithmetic circuit takes the matrix a data from the input memory 1601 and performs matrix operation with the matrix B, and a partial result or a final result of the obtained matrix is stored in an accumulator (accumulator) 1608.

The unified memory 1606 is used to store input data as well as output data. The weight data directly passes through a Memory Access Controller (DMAC) 1605, and the DMAC is transferred to the weight Memory 1602. The input data is also carried into the unified memory 1606 through the DMAC.

The BIU is a Bus Interface Unit 1613, which is used for interaction between the AXI Bus and the DMAC and an Instruction Fetch Buffer (IFB) 1609.

The Bus Interface Unit 1613(Bus Interface Unit, BIU for short) is configured to obtain an instruction from the external memory by the instruction fetch memory 1609, and also to obtain the original data of the input matrix a or the weight matrix B from the external memory by the storage Unit access controller 1605.

The DMAC is mainly used to transfer input data in the external memory DDR to the unified memory 1606, or to transfer weight data to the weight memory 1602, or to transfer input data to the input memory 1601.

The vector calculation unit 1607 includes a plurality of arithmetic processing units, and further processes the output of the arithmetic circuit 1603 if necessary, such as vector multiplication, vector addition, exponential operation, logarithmic operation, magnitude comparison, and the like. The method is mainly used for non-convolution/full-connection layer network calculation in the neural network, such as Batch Normalization, pixel-level summation, up-sampling of a prediction label plane and the like.

In some implementations, the vector calculation unit 1607 can store the processed output vector to the unified memory 1606. For example, the vector calculation unit 1607 may calculate a linear function; alternatively, a non-linear function is applied to the output of the arithmetic circuit 1603, such as to linearly interpolate the predicted tag planes extracted by the convolutional layers, and then, such as to accumulate a vector of values to generate activation values. In some implementations, the vector calculation unit 1607 generates normalized values, pixel-level summed values, or both. In some implementations, the vector of processed outputs can be used as activation inputs to the arithmetic circuitry 1603, e.g., for use in subsequent layers in a neural network.

An instruction fetch buffer (1609) connected to the controller 1604 for storing instructions used by the controller 1604;

the unified memory 1606, the input memory 1601, the weight memory 1602, and the instruction fetch memory 1609 are all On-Chip memories. The external memory is private to the NPU hardware architecture.

The processor mentioned in any of the above may be a general-purpose central processing unit, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of the program.

It should be noted that the above-described embodiments of the apparatus are merely illustrative, where 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 multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. In addition, in the drawings of the embodiments of the apparatus provided in the present application, the connection relationship between the modules indicates that there is a communication connection therebetween, and may be implemented as one or more communication buses or signal lines.

Through the above description of the embodiments, those skilled in the art will clearly understand that the present application can be implemented by software plus necessary general-purpose hardware, and certainly can also be implemented by special-purpose hardware including special-purpose integrated circuits, special-purpose CPUs, special-purpose memories, special-purpose components and the like. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions may be various, such as analog circuits, digital circuits, or dedicated circuits. However, for the present application, the implementation of a software program is more preferable. Based on such understanding, the technical solutions of the present application may be substantially embodied in the form of a software product, which is stored in a readable storage medium, such as a floppy disk, a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk of a computer, and includes several instructions for enabling a computer device (which may be a personal computer, an exercise device, or a network device) to execute the method according to the embodiments of the present application.

In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product.

The computer program product includes one or more computer instructions. When loaded and executed on a computer, cause the processes or functions described in accordance with the embodiments of the application to occur, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, from one website site, computer, training device, or data center to another website site, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium may be any available medium that a computer can store or a data storage device, such as a training device, data center, etc., that includes one or more available media. The usable medium may be a magnetic medium (e.g., a floppy Disk, a hard Disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a Solid State Disk (SSD)), among others.

Claims

1. A method for predicting user behavior, the method being implemented by an objective model, the method comprising:

acquiring a first characteristic of a first item and a second characteristic of a second item, wherein the first item and the second item are positioned in different lists or the same list of a target page, and the second item is positioned in front of the first item;

acquiring a second feature of the first item based on a first feature of the first item and a second feature of the second item, wherein the first feature is attribute information, and the second feature is information obtained by fusion based on the attribute information;

and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item.

2. The method of claim 1, further comprising:

acquiring a first characteristic of the third item, wherein the first item and the third item are located in different lists or the same list of the target page, and the third item is adjacent to the first item;

acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item;

the obtaining the probability that the first item is clicked by the user based on the second feature of the first item comprises:

and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item.

3. The method of claim 1, wherein the obtaining a second feature of the first item based on a first feature of the first item and a second feature of the second item comprises:

mapping the first characteristic of the first item to obtain a fourth characteristic of the first item;

processing the second characteristic of the second item based on an attention mechanism to obtain a fifth characteristic of the first item;

and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item to obtain a second feature of the first item.

4. The method of claim 3, wherein the mapping the first feature of the first item to obtain the fourth feature of the first item:

mapping the first feature of the first item, the request of the user for the target page and the probability of the second item being clicked by the user to obtain a sixth feature of the first item, a seventh feature of the first item and an eighth feature of the first item;

5. The method of claim 2, wherein the obtaining a third feature of the first item based on the first feature of the first item and the first feature of the third item comprises:

mapping the first feature of the first item and the first feature of the third item to obtain a sixth feature of the first item and a ninth feature of the first item;

performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item to obtain a tenth feature of the first item;

and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

6. The method of claim 5, wherein performing a fourth fusion process on a sixth feature of the first item and a tenth feature of the first item to obtain a third feature of the first item comprises:

mapping the request of the user for the target page to obtain a seventh characteristic of the first project;

and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item to obtain a third feature of the first item.

7. The method according to any one of claims 1 to 6, wherein if the first item is a first item in a target page, the second characteristic of the second item is a preset value.

8. The method according to any one of claims 1 to 7, wherein the destination page comprises a plurality of lists, a plurality of items in the plurality of lists form a directed acyclic graph, and the plurality of items comprise the first item, the second item, and the third item.

9. A directed acyclic graph construction method, the method comprising:

acquiring eye movement data of a target page browsed by a user;

determining browsing behavior of the user on a plurality of items based on the eye movement data, the plurality of items being located in a plurality of lists of the target page;

and connecting the plurality of items based on the browsing behaviors to obtain the directed acyclic graph.

10. The method of claim 9, wherein the connecting the plurality of items based on the browsing behavior to obtain a directed acyclic graph comprises:

and connecting the items in the same list browsed by the user in a first sequence according to the first sequence, and connecting the items in different lists browsed by the user in a second sequence according to the second sequence to obtain the directed acyclic graph.

11. The method according to claim 9 or 10, wherein the obtaining of eye movement data of the user browsing the target page comprises:

and acquiring eye movement data of a target page browsed by a user through an eye movement instrument.

12. A method of model training, the method comprising:

acquiring a first feature of a first item and a second feature of a second item through a model to be trained, wherein the first item and the second item are positioned in different lists or the same list of a page to be processed, and the second item is positioned in front of the first item;

acquiring a second feature of the first project based on a first feature of the first project and a second feature of the second project through the model to be trained, wherein the first feature is attribute information, and the second feature is information obtained by fusion based on the attribute information;

acquiring the probability of the first item clicked by the user based on the second characteristic of the first item through the model to be trained;

acquiring a target loss based on the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user, wherein the target loss is used for indicating the difference between the probability of the first item being clicked by the user and the real probability of the first item being clicked by the user;

and updating the parameters of the model to be trained based on the target loss until model training conditions are met to obtain a target model.

13. The method of claim 12, further comprising:

acquiring a first feature of the third item through the model to be trained, wherein the first item and the third item are located in different lists or the same list of the page to be processed, and the third item is adjacent to the first item;

acquiring a third feature of the first item based on the first feature of the first item and the first feature of the third item through the model to be trained;

the obtaining, by the model to be trained, the probability that the first item is clicked by the user based on the second feature of the first item includes:

and acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item and the third characteristic of the first item through the model to be trained.

14. An apparatus for predicting user behavior, the apparatus comprising:

the first obtaining module is used for obtaining a first feature of a first item and a second feature of a second item through a target model, wherein the first item and the second item are positioned in different lists or the same list of a target page, and the second item is positioned in front of the first item;

a second obtaining module, configured to obtain, by using a target model, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature is attribute information and the second feature is information obtained by fusing based on the attribute information;

and the third acquisition module is used for acquiring the probability of the first item being clicked by the user based on the second characteristic of the first item through the target model.

15. The apparatus of claim 14, further comprising:

a fourth obtaining module, configured to obtain, through a target model, a first feature of the third item, where the first item and the third item are located in different lists or a same list of the target page, and the third item is adjacent to the first item;

a fifth obtaining module, configured to obtain, through a target model, a third feature of the first item based on the first feature of the first item and the first feature of the third item;

the third obtaining module is configured to obtain, through a target model, a probability that the first item is clicked by the user based on the second feature of the first item and the third feature of the first item.

16. The apparatus of claim 14, wherein the second obtaining module is configured to:

mapping the first feature of the first item through a target model to obtain a fourth feature of the first item;

processing the second characteristic of the second item based on a self-attention mechanism through a target model to obtain a fifth characteristic of the first item;

and performing first fusion processing on the fourth feature of the first item and the fifth feature of the first item through a target model to obtain a second feature of the first item.

17. The apparatus of claim 16, wherein the second obtaining module is configured to:

mapping the first characteristic of the first item, the request of a user for the target page and the probability of clicking the second item by the user through a target model to obtain a sixth characteristic of the first item, a seventh characteristic of the first item and an eighth characteristic of the first item;

and performing second fusion processing on the sixth feature of the first item, the seventh feature of the first item and the eighth feature of the first item through an object model to obtain a fourth feature of the first item.

18. The apparatus of claim 17, wherein the fifth obtaining module is configured to:

mapping the first feature of the first item and the first feature of the third item through an object model to obtain a sixth feature of the first item and a ninth feature of the first item;

performing third fusion processing on the sixth feature of the first item and the ninth feature of the first item through a target model to obtain a tenth feature of the first item;

and performing fourth fusion processing on the sixth feature of the first item and the tenth feature of the first item through a target model to obtain a third feature of the first item.

19. The apparatus of claim 18, wherein the fifth obtaining module is configured to:

mapping the request of the user for the target page through a target model to obtain a seventh characteristic of the first item;

and performing fourth fusion processing on the sixth feature of the first item, the seventh feature of the first item and the tenth feature of the first item through an object model to obtain a third feature of the first item.

20. The apparatus according to any one of claims 14 to 19, wherein if the first item is a first item in a destination page, the second characteristic of the second item is a preset value.

21. The apparatus according to any one of claims 14 to 20, wherein the destination page comprises a plurality of lists, a plurality of items in the lists form a directed acyclic graph, and the plurality of items comprise the first item, the second item, and the third item.

22. An apparatus for constructing a directed acyclic graph, the apparatus comprising:

the acquisition module is used for acquiring eye movement data of a target page browsed by a user;

a determining module, configured to determine browsing behaviors of the user on a plurality of items based on the eye movement data, where the plurality of items are located in a plurality of lists of the target page;

and the connecting module is used for connecting the plurality of items based on the browsing behaviors to obtain the directed acyclic graph.

23. The apparatus of claim 22, wherein the connecting module is configured to connect items in the same list browsed by the user in a first order according to the first order, and connect items in different lists browsed by the user in a second order according to the second order, so as to obtain the directed acyclic graph.

24. The apparatus according to claim 22 or 23, wherein the acquiring module is configured to acquire eye movement data of a user browsing a target page through an eye tracker.

25. A model training apparatus, the apparatus comprising:

the system comprises a first acquisition module, a second acquisition module and a processing module, wherein the first acquisition module is used for acquiring a first characteristic of a first item and a second characteristic of a second item through a model to be trained, the first item and the second item are positioned in different lists or the same list of a page to be processed, and the second item is positioned in front of the first item;

a second obtaining module, configured to obtain, by using the model to be trained, a second feature of the first item based on a first feature of the first item and a second feature of the second item, where the first feature is attribute information and the second feature is information obtained by fusing based on the attribute information;

the third obtaining module is used for obtaining the probability that the first item is clicked by the user based on the second characteristic of the first item through the model to be trained;

a fourth obtaining module, configured to obtain a target loss based on the probability that the first item is clicked by the user and the true probability that the first item is clicked by the user, where the target loss is used to indicate a difference between the probability that the first item is clicked by the user and the true probability that the first item is clicked by the user;

and the updating module is used for updating the parameters of the model to be trained based on the target loss until model training conditions are met to obtain the target model.

26. The apparatus of claim 25, wherein the apparatus comprises:

a fifth obtaining module, configured to obtain, through the model to be trained, a first feature of the third item, where the first item and the third item are located in different lists or a same list of the page to be processed, and the third item is adjacent to the first item;

a sixth obtaining module, configured to obtain, by using the model to be trained, a third feature of the first item based on the first feature of the first item and the first feature of the third item;

the third obtaining module is configured to obtain, by the model to be trained, a probability that the first item is clicked by the user based on the second feature of the first item and the third feature of the first item.

27. A user behavior prediction apparatus, characterized in that the apparatus comprises a memory and a processor;

the memory stores code, the processor is configured to execute the code, and when executed, the apparatus performs the method of any of claims 1 to 13.

28. A computer storage medium, characterized in that it stores a computer program which, when executed by a computer, causes the computer to carry out the method of any one of claims 1 to 13.

29. A computer program product having stored instructions which, when executed by a computer, cause the computer to carry out the method of any one of claims 1 to 13.