Information processing method, model training method, equipment, storage medium and program product

Abstract

The embodiment of the invention provides an information processing method, a model training method, equipment, a storage medium and a program product. In the embodiment of the invention, in the process of sensing the interestingness of the user on the information object, not only is the attention duration of the user on the target information object obtained, but also the context information capable of comprehensively representing the user portrait and the interaction environment and the interaction behavior when the user browses the page can be obtained; further, in combination with a duration distribution prediction model trained by a large number of users for the context information of different information objects, the difference of different users for the context information of different information objects is learned, and the model can predict duration distribution information corresponding to the context information by pertinently combining the context information of different users; and furthermore, according to the duration distribution information and the attention duration, the interestingness of the perception user on the information object is more accurately determined.

Description

Technical Field

[0001] This application relates to the field of Internet technologies, and in particular, to an information processing method, a model training method, a device, a storage medium, and a program product. Background Art

[0002] In the information age, users can obtain various types of information objects from Internet applications or platforms, such as texts, pictures, videos, and gifs, etc. For example, users can browse various product entries from e-commerce applications or platforms and can perform operations such as clicking, collecting, or placing orders on products of interest. Another example is that users can view various information such as news and entertainment from social platforms and can perform operations such as viewing and commenting on news and entertainment of interest.

[0003] Taking the example of a user browsing product information, the browsing duration of the user for product information with different degrees of interest often varies, and to a certain extent, the browsing duration can reflect the user's interest in the product information. However, in the case of expressing the same degree of interest, the browsing duration of different users for the same product information may be different, and even the browsing duration of the same user for different product information will also be different. Therefore, how to more accurately use the browsing duration and other duration information expressing the attention to information objects to perceive the user's interest in the information objects is a technical problem that urgently needs to be solved at present. Summary of the Invention

[0004] Embodiments of this application provide an information processing method, a model training method, a device, a storage medium, and a program product, which are used to more accurately use the browsing duration and other duration information expressing the attention to information objects to perceive the user's interest in the information objects.

[0005] Embodiments of this application provide an information processing method, including: responding to an access operation of a target user, presenting a target page, where the target page includes a plurality of information objects displayed on the current screen; for a target information object among the plurality of information objects, obtaining context information of the target information object and the attention duration of the target user for the target information object; inputting the context information into a duration distribution prediction model to predict duration distribution information of the target user's attention to the target information object; and determining the interest degree of the target user in the target information object according to the duration distribution information and the attention duration.

[0006] The embodiment of the present application further provides a model training method, including: obtaining the sample context information of a plurality of sample information objects, where one sample information object corresponds to one sample user, and obtaining the sample attention duration of the sample user for the corresponding sample information object; training an initial prediction model according to the sample context information of the plurality of sample information objects to obtain the sample duration distribution information of each sample user's attention to the corresponding sample information object; calculating a loss function between the sample attention duration and the sample duration distribution information, and updating the model parameters of the initial prediction model according to the loss function until the loss function meets a set termination condition to obtain the duration distribution prediction model.

[0007] The embodiment of the present application further provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor is coupled to the memory and is used to execute the computer program to implement the steps in each method provided by the embodiment of the present application.

[0008] The embodiment of the present application further provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the processor can implement the steps in the above-mentioned method.

[0009] The embodiment of the present application further provides a computer program product, which includes computer programs / instructions. When the computer programs / instructions are executed by a processor, the processor can implement the steps in the above-mentioned method embodiments.

[0010] In the embodiment of the present application, in the process of perceiving the user's interest in an information object, not only the attention duration of the user to the target information object is obtained, but also the context information that can comprehensively represent the user portrait, the interaction environment and interaction behavior when the user browses the page is obtained; further, in combination with the duration distribution prediction model trained by the context information of a large number of users for different information objects, the differences in the context information of different users for different information objects are learned. This model can specifically combine the context information of different users and predict the duration distribution information corresponding to the context information; furthermore, according to this duration distribution information and the attention duration, the user's interest in the information object can be more accurately determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0012] Figure 1 It is a schematic flowchart of an information processing method provided by an exemplary embodiment of the present application;

[0013] Figure 2 This is a schematic diagram of a prediction process using a duration distribution prediction model in another exemplary embodiment of the present application;

[0014] Figure 3 This is a schematic diagram of the structure of a duration distribution prediction model provided in another exemplary embodiment of the present application;

[0015] Figure 4 This is a schematic flowchart of a model training method provided in another exemplary embodiment of the present application;

[0016] Figure 5 This is a schematic diagram of the structure of an electronic device provided in another exemplary embodiment of the present application. Detailed implementation manners

[0017] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions of the present application will be clearly and completely described below in conjunction with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0018] It should be noted that in the case where the embodiments of the present application involve user information, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with the relevant laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse. In addition, various models (including but not limited to language models or large models) involved in the present application comply with relevant laws and standards.

[0019] In addition, it should be noted that in the case where the embodiments of the present application involve user interaction operations or trigger operations, the various access operations or selection operations involved in the embodiments of the present application include but are not limited to: interaction operations in various ways such as touch operations, gesture operations, voice operations, head movement operations, and eye movement operations; among them, touch operations include but are not limited to: click operations, double-click operations, long-press operations, swipe operations, pinch operations, or mouse hover operations, etc. Swipe operations include but are not limited to: straight-line swipes, curved swipes, etc.

[0020] Furthermore, it should be noted that, in the case where the embodiments of the present application involve jumping between the first interface and the second interface, the jumping methods involved in the embodiments of the present application include but are not limited to: jumping directly from the first interface to the second interface, jumping from the first interface to the task interface first and jumping to the second interface after completing the corresponding task operation on the task interface; completing the corresponding task operation on the task interface includes but is not limited to: when the task interface is implemented as a game interface, completing the game operation on the game interface; when the task interface is implemented as an identity authentication interface, completing the identity authentication on the identity authentication interface; when the task interface is implemented as a recharge interface, completing the recharge operation on the recharge interface; and so on.

[0021] In the embodiments of the present application, an information object refers to an object in an Internet application that carries certain information content, such as product information, video information, or audio information, etc. Also, the information expressing the duration of a user's attention to an information object is referred to as "attention duration". The attention duration may be, but is not limited to, display duration, dwell duration, playback duration, browsing duration, and viewing duration, etc., and appropriate duration information may be flexibly selected according to the type of information object.

[0022] Two examples are given below. Example 1 takes the case where the information object is product information and uses the length of time the product information stays on the page as the attention duration to introduce the user's interest in product information. Example 2 takes the case where the information object is video information and uses the length of time the video information is viewed as the attention duration to introduce the user's interest in video information.

[0023] Example 1: Take the example of a user browsing product information. Assuming that multiple product information is displayed on the page in a double-column flow of images and text, the user can browse the product information on the page and update the product information displayed on the screen by sliding up and down; further, during the browsing process, the user can also click on the product information of interest to jump to the product details page corresponding to the product information for browsing. Of course, the user can also collect the product information of interest, or place an order for the product information of interest. In this process, the user's interest in each product information can be measured by the length of time that each product information stays on the screen. The length of stay refers to the time interval from the beginning of the product information being displayed on the screen to the time when it slides out of the screen or to the current moment. Specifically, a time threshold can be set. If the length of time that the product information stays on the screen is greater than the time threshold, it is considered that the user has a high interest in the product information. Otherwise, it is considered that the user has a low interest in the product information.

[0024] Further, in the above process, the dwell time is defined in combination with the exposure situation of the product information on the screen. The dwell time is defined for the product information in the exposure state and with an exposure duration longer than the set duration threshold, while the unexposed or pseudo-exposed product information does not have a dwell time. The product information in the exposure state refers to the product information that is completely exposed on the screen, the unexposed product information refers to the product information that does not fully appear on the screen, and the pseudo-exposed product information refers to the product information that is completely exposed on the screen but with an exposure duration shorter than the set duration threshold. When the product information is unexposed or pseudo-exposed, it means that the user's interest in this product information is relatively low. For the product information in the exposure state and with an exposure duration longer than the set duration threshold, its dwell time on the screen is considered. The longer the dwell time, the higher the user's interest in this product information.

[0025] Example 2: Taking the example of a user watching a video, when the user is watching a video, the playback duration of the video can be obtained. The playback duration refers to the actual playing time length of the video. Different users may perform operations such as pausing, fast-forwarding, or playing at a slow speed when watching the same video, resulting in different playback durations of the video. The playback duration of the video can be used to measure the user's interest in the video. The longer the playback duration, the higher the user's interest in the video. Specifically, the playback duration of the video itself can be segmented, and different playback thresholds can be set at different playback durations to judge the completion degree of the user's video viewing. For example, a playback threshold can be set at 50% of the playback duration of a certain video, and another playback threshold can be set at 90% of the playback duration of this video. The higher the playback threshold, the longer the playback duration, indicating that the user's interest in the video is higher.

[0026] Although the attention durations of users to information objects such as product information and video information with different degrees of interest often vary, comparing the attention durations of information objects can also reflect to a certain extent the interests of different users in different information objects. However, in the case of expressing the same degree of interest, the attention durations of different users to the same information object may be different. For example, user A browses a certain product information for 1 second and then adds the product information to the shopping cart, while user B browses the same product information for 5 seconds and then adds the product information to the shopping cart. In this example, the operation of adding the product to the shopping cart by different users for the same product expresses the same or similar degree of interest. Even for the same user, in the case of expressing the same interest, the attention durations to different information objects will also be different. For example, after browsing the first product information for 3 seconds, the user favorites the first product information, and after browsing the second product information for 10 seconds, the user also favorites the second product information. In this example, the user's favorite operations on different product information express the same or similar degree of interest.

[0027] The essence of the solutions in the above two examples is to judge the user's interest degree by setting a time threshold and paying attention to the relationship between the attention duration and the time threshold. The differences between context information such as different information objects, different terminal devices, and different users are not considered. Even for the same information object, the attention duration of the same user in different context environments may be different. Therefore, simply comparing the attention duration with the corresponding threshold to judge the user's interest degree is inaccurate.

[0028] Regarding the technical problem of how to more accurately use duration information to perceive the user's interest in information objects, in the following embodiments of the present application, in the process of perceiving the user's interest in information objects, not only the attention duration of the user on the target information object is obtained, but also the context information that can comprehensively represent the user portrait, the interaction environment, and the interaction behavior when the user browses the page is obtained; further, in combination with the duration distribution prediction model trained by the context information of a large number of users for different information objects, the differences in the context information of different users for different information objects are learned. This model can specifically combine the context information of different users and predict the duration distribution information corresponding to the context information; furthermore, according to the duration distribution information and the attention duration, the user's interest in the information object is more accurately determined.

[0029] The following will describe in detail the technical solutions provided by the embodiments of the present application with reference to the accompanying drawings.

[0030] Figure 1 It is a schematic flowchart of an information processing method provided by an exemplary embodiment of the present application. As Figure 1 shown, the method includes:

[0031] S101: Respond to the user operation of the target user and display the target page, where the target page includes multiple information objects displayed on the current screen;

[0032] S102: For the target information object among the multiple information objects, obtain the context information of the target information object and the attention duration of the target user on the target information object;

[0033] S103: Input the context information into the duration distribution prediction model to predict the duration distribution information of the target user's attention to the target information object;

[0034] S104: Determine the interest degree of the target user in the target information object according to the duration distribution information and the attention duration.

[0035] In the embodiments of the present application, there is no restriction on the execution entity of the information processing method. Hereinafter, the user's terminal device is used as the execution entity of the information processing method provided in the embodiments of the present application for introduction. The user's terminal device can be implemented as a desktop computer, a laptop computer, a smart phone, or an IOT (Internet of Things) device, etc. By directly running the information processing method locally on the terminal device side, network latency and the processing time consumed on the server device side can be saved, and the real-time response to user operations can be improved. Alternatively, the execution entity of the information processing method provided in the embodiments of the present application can also be the server device corresponding to the above-mentioned user's terminal device. Generally, the server device has stronger information processing capabilities than the user's terminal device and can handle more complex tasks. Among them, the specific implementation of the server device in this embodiment is not limited. For example, it can be a conventional server, a cloud server, or a server array, etc.

[0036] In this embodiment, the target user is not limited. For example, it can be the owner of the terminal device. In this case, the target user can be determined according to the identification information of the terminal device. The identification information of the terminal information can be, but is not limited to, UUID (Universally Unique Identifier). For another example, it can also be an account logged in to a certain application on the terminal device. For example, it can be an account logged in to various applications such as an e-commerce application or a social application. In this case, the target user can be the holder of the account logged in to the application.

[0037] In this embodiment, the terminal device responds to the access operation of the target user on the target application and displays the target page. The target page includes a plurality of information objects displayed on the current screen. Among them, the target application refers to the operation object for which the target user performs the access operation. The target application can be a desktop application or a mobile application, and this is not limited. Hereinafter, a mobile application is used as an example for introduction, but it is not limited thereto. Also, the type of the target application is not limited, including but not limited to: various applications such as e-commerce applications, social applications, search applications, or short video applications. Combining the above examples, "an account logged in to a certain application on the terminal device", "a certain application" can be the target application.

[0038] Among them, different function pages with different functions can be integrated in the target application. Different access operations can trigger the display of pages providing different functions, and these function pages can be some examples of the target page. The target page can be various types of pages embedded in different positions within the target application.

[0039] For example, the target page can be the home page of the target application. The home page is the first page presented to the target user when the access operation is to open the target application. Generally, the home page provides the display function of information objects. In this case, the home page can display multiple product information recommended to the target user, and the product information can be an example of an information object. For another example, the target page can be the search page of the target application. The search page is the page where the target application provides the search function. On this page, the target user can perform an access operation for the search operation. In response to the target user's search operation, a search result page corresponding to the search operation is displayed. The search result page can include multiple search result information, and the search result information can be an example of an information object. For yet another example, the target page can be the shopping cart page provided by the target application. In this case, in response to the access operation to open the shopping cart page, the shopping cart page is displayed, and the shopping cart page includes multiple product information added to the shopping cart by the target user, and the product information can be an example of an information object.

[0040] In this embodiment, the target page includes multiple information objects displayed on the current screen. Optionally, the ways for the target page to display information objects include but are not limited to: waterfall flow, card flow, and list flow. Among them, the waterfall flow is also called the waterfall-style layout, which appears as a multi-column layout with uneven heights on the screen. As the scroll bar of the page continuously scrolls down, new information objects will be continuously loaded and appended to the tail of the current page. The card flow is a layout method that arranges information objects in the form of cards. Each card is an information object, and the design of the card mainly combines pictures and text. The list flow is a text-dominated layout form, and information objects are displayed in the form of a list.

[0041] Among them, the categories of information objects in this embodiment are not limited. The category of an information object refers to the presentation form of the information object, including but not limited to: pictures, texts, videos, audios, and animated pictures (GIFs), etc., at least one of which. As the type of the target application is different, such as an e-commerce application, a social application, or a short video application, etc., the categories of information objects displayed will be different. For example, if the target application is an e-commerce application, the categories of information objects are mainly product information combining picture and text categories. For another example, if the target application is a short video application, the categories of information objects are mainly video information in the video category. It should be understood that this is only for illustration, and the categories of information objects displayed on the target page in this embodiment are not limited. For example, in the case where the target application is an e-commerce application, video information in the video category and information objects of other various categories can also be displayed.

[0042] Among them, the content displayed on the screen of the terminal device can change. The screen that displays the target page is called the current screen. Based on this, when the target page is being displayed, the target page slides to display multiple information objects on the current screen. The target user can also initiate an access operation on the content displayed on the current screen, such as a sliding operation, etc. In response to the target sliding operation, the scroll bar of the page continuously scrolls down, and new information objects will be continuously loaded and appended to the tail of the current screen. In this case, the terminal device will load new information objects on the current screen. It should be understood that regardless of how the information objects on the target page change, the multiple information objects on the target page refer to the information objects displayed on the current screen.

[0043] In this embodiment, the target information object among the multiple information objects is not limited. For example, each of the multiple information objects can be used as the target information object; or, those with a larger exposure area among the multiple information objects on the current screen can be used as the target information object; or, the information objects of the target category among the multiple information objects can be used as the target information object, etc.; or, the information objects of the target type among the multiple information objects can be used as the target information object, etc. Among them, the type of the information object refers to the category to which the information object belongs. For example, if the information object is product information, the corresponding types include clothing, mobile phones, computers, etc.; if the information object is video information, the corresponding types include but are not limited to funny videos, game videos, etc.

[0044] In this embodiment, the work done refers to the degree of investment of the target user's attention in the target information object. The higher the work intensity, the higher the interest of the target user in the target information object; conversely, the lower the interest of the target user in the target information object. Of course, the positive correlation between the work intensity and the interest is just one implementation method, and it can also be a negative correlation or other methods. The interest of the target user in the target information object can be measured by the work intensity.

[0045] The inventor of this case further discovered through creative labor that the work intensity of the target user on the target information object is reflected in two aspects. On the one hand, it is reflected in the duration information of the target user's attention to the target information object. On the other hand, it is also related to the context information of the target information object. The context information of the target information object refers to the information set that can comprehensively represent the user characteristics of the target user and the interaction environment where the target user is located, including the attribute information of the target user, the attribute information of the terminal device, the behavior data generated on the terminal device screen, and the attribute information of the target information object, etc., when the target user browses the target page on the terminal device in a specific scenario.

[0046] Based on this, in order to more accurately perceive the degree of interest of the target user in the target information object, for the target information object, not only the attention duration of the target user on the target information object will be obtained, but also the context information of the target information object will be obtained, and there is no limitation on the order of obtaining the attention duration and the context information of the target information object.

[0047] Among them, the attention duration of the target information object refers to the duration information expressing attention to the target information object. This can include, but is not limited to, the browsing duration, exposure duration, stay duration (also known as display duration), or playback duration, etc. of the target information object. Correspondingly, there is no limitation on the method of obtaining the attention duration of the target user on the information object. For example, obtain the display duration of the target information object on the current screen. The display duration can also be considered as the exposure duration, and use the obtained display duration of the target information object on the current screen as the attention duration. Or, when the target information object is audio-visual information and is being played, obtain the playback duration of the target information object as the attention duration. These methods are simple and fast. Or, more accurately, capture the attention position of the target user's eyes on the current screen through the visual sensor of the terminal device; according to the attention position of the target user on the current screen, obtain the duration during which the attention position falls within the screen range where the target information object is located as the attention duration. The attention duration of the target information object can be obtained more precisely through the visual sensor. Further, in the process of perceiving the attention duration of the target user on the target information object, by collecting the attention information of the target user's eyes, for example, the movement of the eyes can be collected in real time according to the visual sensor. Then, according to the attention information of the eyes, determine the attention duration of the target user on the target information object. For example, it is possible to lock the attention position of the target user's eyes on the current screen through the visual sensor. When the attention position is the target information object, start timing, and end timing when the attention position moves away from the target information object, and use the time interval from the start of timing to the end of timing as the attention duration.

[0048] In this embodiment, a duration distribution prediction model is provided. This model is used to predict the duration distribution information of a target user's attention to a target information object based on the context information of the target information object. The duration distribution information refers to the probability distribution of the target user's attention duration to the target information object in different time intervals under the given context information. By obtaining the probability that the target user's attention duration to the target information object falls into a certain time interval, the interest degree of the target user in the target information object is determined. Each user has independent duration distribution information for each information object, and the duration distribution information is predicted separately through the duration distribution prediction model according to the context information of each information object. Whether it is different users for the same information object or the same user for different information objects, a unified duration threshold is no longer used, but their respective duration distribution information is used to judge the interest degree. The duration distribution prediction model is obtained by training an initial prediction model based on the sample context information of multiple sample information objects. The multiple sample information objects involve multiple sample users. Therefore, the duration distribution prediction model in this case learns the duration distribution under a large amount of data during the training process. Further, during the reasoning process, in the case of a target user for a target information object, the context of the target information object is used to predict the duration distribution information of the target user's attention to the target information object, which can retain the duration distribution of a single user for a single information object, and characterize the interest degree of a single user in a single information object from both the group and individual dimensions, providing conditions for subsequent application stages such as decision-making based on interest degree. For the detailed training process, reference can be made to the subsequent embodiments, which will not be elaborated here.

[0049] Based on the above, the context information is input into the duration distribution prediction model to predict the duration distribution information of the target user's attention to the target information object. As described above, the duration distribution information refers to the probability distribution of the target user's attention duration to the target information object in different time intervals. Furthermore, based on the duration distribution information and the attention duration, the interest degree of the target user in the target information object is determined. For example, the probability of the target user's attention duration to the target information object can be determined by finding the position of the attention duration in the duration distribution information, and this probability can reflect the interest degree of the target user in the target information object. For example, if the probability is higher, it means that the target user's attention duration to the target information object may be longer, and the corresponding interest degree is higher; conversely, if the probability is lower, it means that the target user's attention duration to the target information object may be shorter, and the corresponding interest degree is lower.

[0050] In an embodiment of the present application, in the process of perceiving the user's interest in an information object, not only the attention duration of the user on the target information object is obtained, but also the context information that can comprehensively represent the user profile, the interaction environment and the interaction behavior when the user browses the page is obtained; further, in combination with the duration distribution prediction model trained by the context information of a large number of users for different information objects, the differences in the context information of different users for different information objects are learned, and this model can specifically combine the context information of different users to predict the duration distribution information corresponding to the context information; furthermore, based on this duration distribution information and the attention duration, the user's interest in the information object is more accurately determined.

[0051] In an alternative embodiment, before obtaining the context information of the target information object among multiple information objects, the following operations may also be performed: each of the multiple information objects is used as the target information object, and using each information object as the target information object can facilitate understanding the user's interest in various information objects; or, according to the set selection conditions or the selection operation of the target user, the target information object is determined from the multiple information objects. By the set selection conditions or the selection operation of the target user, the intended target information object is filtered out, which is more targeted. Among them, the set selection conditions in this embodiment are not limited. For example, it may be the one with a larger exposure area among the multiple information objects displayed on the current screen, a certain specific type, a certain specific category, or a certain specific position on the current screen, etc. For example, a certain specific category may be a video category, an audio category, a picture category, etc. A certain specific type may be a mobile phone, a computer, clothes, jewelry, etc. A certain specific position on the current screen may be the top position, the bottom position, etc. Or, another way to determine the target information object is that in the process of determining the target information object according to the selection operation of the target user, if the user performs a selection operation on any one of the multiple information objects, then this information object can be used as the target information object.

[0052] In an optional embodiment, obtaining the context information of the target information object includes: obtaining at least one of the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, the behavior data occurring on the current screen, and the attribute information of the target information object as the context information of the target information object. In this optional embodiment, not only the attention duration of the target user on the target information object is obtained, but also the context information of the target information object is obtained. This context information can be reflected in obtaining the attribute information of the target user, the attribute information of the terminal device, the behavior data generated on the terminal device screen, and the attribute information of the target information object, etc., when browsing the target page. All these information sets can comprehensively represent the user characteristics of the target user, as well as the interaction environment and interaction behaviors of the target user, which helps to accurately predict the interest degree of the target user in the future.

[0053] The following introduces different context information respectively. It should be understood that the types of context information below are only examples, but not limited to this.

[0054] The attribute information of the target user: It refers to the user characteristics abstracted from the information such as the browsing history, search records, social activities, and consumption behaviors of the target user, including but not limited to: the purchasing power of the target user, the hobbies of the target user, the identification information of the target user, etc. Among them, the identification information of the target user can be implemented as the account ID (Identity, identity information) for logging in to the target application. By obtaining the attribute information of the target user, the user portrait of the target user can be constructed more accurately, and the preferences, habits, etc. of the user can be understood, which helps to predict the interest degree of the target user in the target information object.

[0055] The attribute information of the terminal device to which the current screen belongs: It includes the software information and hardware information of the terminal device. The software information refers to the operating environment of the terminal device, including the operating system, the version information of the target application, network information, etc. The hardware information of the terminal device includes the brand, screen length and width, screen resolution, and screen refresh rate, etc. The attribute information of the terminal device, such as the brand and operating system, can reflect the usage habits and preferences of the target user on the one hand. On the other hand, the screen of the terminal device will affect the display effect and display quantity of the information object on the target page. Regarding the terminal device as one type of context information can help to more comprehensively understand the context environment of the target user when browsing the target page.

[0056] Behavior data occurring on the current screen: It refers to the data generated by the interaction behaviors that occur on the target page during the period when the target information object is displayed on the target page of the terminal device until it disappears from the target page. The behaviors occurring on the current screen include, but are not limited to, touch interaction behaviors and browsing interaction behaviors. Touch interaction behaviors include, in turn, swipe operations, click operations, and gesture operations. Among them, swipe operations include backswipe operations and downswipe operations. Among them, the backswipe operation refers to swiping upward. Browsing interaction behaviors include, in turn, following, searching, collecting, adding to the shopping cart, and purchasing, etc. Browsing interaction behaviors can more clearly reflect the intentions of the target user and help to more accurately perceive the target user's interest in the target information object. In this embodiment, browsing interaction behaviors and touch interaction behaviors can be collectively referred to as on-device behaviors. As a result, touch interaction behavior data and browsing interaction behavior data will be generated, and these data can be called on-device behavior data. Among them, the touch interaction behavior data includes the speed of the touch interaction behavior, such as the speed of the swipe operation and the speed of the click operation. The speed in the touch interaction behavior data is one of the important factors for measuring the target user's interest in the target information object.

[0057] Attribute information of the target information object: It refers to the display position of the target information object on the current screen and the category and type to which the target information object belongs. For example, the position of the target information object on the current screen can be in the left half or the right half of the current screen, the top position or the bottom position, etc. The category of the target information object can include at least one of pictures, texts, audios, videos, and animated gifs, etc. The type of the target information object refers to the category of the target information object, which can refer to the foregoing embodiments and will not be elaborated herein. Optionally, the target information object can further include a presentation form, and the presentation form includes, but is not limited to, video cards, picture-text combined cards, and text entries, etc.

[0058] Taking the double-column stream as an example again, assume that each information object has one chance to be exposed to each user. For user A, assume that two information objects are located on the left and right sides of the screen of their terminal device. The exposure duration of these two information objects located on the left and right sides of the screen is the same. If the same duration threshold is used, it will be impossible to distinguish the degree of interest of user A in the two information objects on the left and right. However, by using the method of the embodiment of the present application, the duration distribution information is comprehensively predicted according to the context information of each information object, such as including but not limited to: multi-dimensional information such as the attribute information of the user, the attribute information of the device, the attribute information of the information object itself, and the behavior data occurring on the screen, combined with the duration distribution information of the group of users pre-learned by the duration distribution prediction model. Since these two information objects will also be exposed to other users, when these two information objects are exposed to other users, they may be on the left side or the right side of the screen. Some attribute values of other users may be the same as those of user A. The duration distribution prediction model in this embodiment can learn the similarities and differences among them. For example, the duration threshold for the same item "exposed on the right side" may be slightly smaller than the duration threshold for "exposed on the left side", which will in turn affect the duration distribution information of user A for each of these two information objects. Although the exposure durations of these two information objects are the same, the duration distribution information used to judge the degree of interest is different, so that the degree of interest of user A in these two information objects can be accurately distinguished. For example: Assume that the exposure duration of these two information objects is 5 seconds each. The duration distribution information of user A for one of the information objects includes duration distribution values of 6 seconds and 10 seconds. Since the exposure duration of 5 seconds is less than the duration distribution value of 6 seconds, it is considered that user A's interest in this information object is not high; The duration distribution information of user A for the other information object includes duration distribution values of 3 seconds and 8 seconds. Since the exposure duration of 5 seconds is greater than the duration distribution value of 3 seconds, it is considered that user A's interest in this information object is relatively high. For the concept of the duration distribution value, please refer to the following text.

[0059] In an optional embodiment, the duration distribution information is divided by the duration distribution value. The duration distribution value can be implemented as a quantile or a time threshold, and there is no limitation on this. The duration distribution value refers to the position points obtained by dividing the range of the duration distribution information according to different ratios. Each position point corresponds to a different probability. In this embodiment, the probability corresponding to each position point is defined as the interest level. That is to say, the duration distribution information includes multiple duration distribution values, and different duration distribution values correspond to different interest levels. Furthermore, the attention duration of the target user for the target information object is compared with different duration distribution values to determine the interest level of the target user for the target information object. Among them, whether the duration distribution value and the interest level are positively or negatively correlated is not limited, and it depends on the specific application requirements.

[0060] The following gives an example that can be used to introduce the process of the duration distribution prediction model predicting duration distribution information based on context information.

[0061] In this embodiment, after obtaining the context information of the target information object and the attention duration of the target user to the target information object, as Figure 2 shown, the context information of the target information object is input into the duration distribution prediction model to predict the duration distribution information of the target user's attention to the target information object. Among them, Figure 2 the context information of the target information object includes the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, the behavior data occurring on the current screen, and the attribute information of the target information object as an example, but is not limited to this. Further, as Figure 2 shown, the duration distribution prediction model predicts the corresponding duration distribution information based on the given context information. This duration distribution information includes multiple duration distribution values, as Figure 2 shown, including duration distribution value Q1, duration distribution value Q2, duration distribution value Qn. Among them, n≥2, and n is a natural number. It should be noted here that in subsequent embodiments, "duration distribution value Qn" can be abbreviated as "Qn", for example, duration distribution value Q1 can be abbreviated as Q1.

[0062] Optionally, during the process of the target user browsing the target information object, three stages of perception input, information understanding, and decision-making judgment will be experienced, but are not limited to these three stages. After that, for the target information object, the target user will choose to click or not to click. Among them, each stage corresponds to a different duration distribution value, that is to say, after determining the target duration distribution value, the stage where the target user is located can be determined. In an example, perception input, information understanding, and decision-making judgment correspond to Q1, Q2, and Q3 respectively. Taking Q1 corresponding to 25%, Q2 corresponding to 50%, and Q3 corresponding to 75% as an example, for example, being greater than or equal to Q3 is considered decision-making judgment, being greater than or equal to Q2 is considered information understanding, and being greater than or equal to Q1 is considered perception input.

[0063] Further, in the process of determining the interest degree of the target user in the target information object according to the duration distribution information and the attention duration, the attention duration is compared with multiple duration distribution values respectively to obtain the target duration distribution value corresponding to the attention duration. Among them, the target duration distribution value can be the duration distribution value that is relatively close to the attention duration among multiple duration distribution values. Furthermore, according to the interest degree level characterized by the target duration distribution value, the interest degree of the target user in the target information object is determined.

[0064] Among them, this embodiment does not limit the method of determining the interest degree of the target user in the target information object according to the interest degree level characterized by the target duration distribution value. The following provides two implementation manners, but is not limited thereto. Among them, different duration distribution values correspond to different interest degree levels.

[0065] Based on this, an optional implementation manner is to directly determine, that is, directly select the corresponding interest degree level according to the target duration distribution value, and then the interest degree of the target user in the target information object can be determined according to this interest degree level. In an example, as Figure 2 shown, after inputting the context information corresponding to the target information object into the duration distribution prediction model, the corresponding duration distribution information is obtained. The duration distribution information includes different duration distribution values. Taking the duration distribution information including three duration distribution values as an example, for example, the duration distribution values include three duration distribution values Q1, Q2, and Q3. Different duration distribution values correspond to different interest levels. If the target duration distribution value is Q1, it means that there is a 25% probability that the attention duration of the target user in the target information object is less than or equal to Q1. If the target duration distribution value is Q2, it means that there is a 50% probability that the attention duration of the target user in the target information object is less than or equal to Q2. If the target duration distribution value is Q3, it means that there is a 75% probability that the attention duration of the target user in the target information object is less than or equal to Q3. It can be seen that in this example, different duration distribution values correspond to different interest degree levels, and the interest degree levels are divided into three interest degree levels of 25%, 50%, and 75%. Q1 corresponds to 25%, Q2 corresponds to 50%, and Q3 corresponds to 75%. The corresponding relationship between the duration distribution value and the interest degree level here is only an example. The quantity of the duration distribution information and the specific values of the interest degree levels are also examples and do not constitute a limitation to this embodiment.

[0066] In another implementation manner, in the process of determining the interest degree of the target user in the target information object, it is allowed to adjust the target duration distribution value according to the attribute information of the target user and / or the target information object. Especially when the target user is a new user or a user who uses the target application less frequently; or, the target information object is a new information object that appears as the time line advances. In these cases, adjusting the target duration distribution value helps to reduce the misjudgment of the interest degree of the target user in the target information object. And give the target user a certain time to get familiar with the target application, or give the target information object more appearance frequencies, so as to be more accurate in the subsequent process of judging the interest degree of the target user in the target information object. Among them, whether the duration distribution value and the interest degree level are positively correlated or negatively correlated will affect the way of adjusting the target duration distribution value. The following respectively introduces the adjustment methods of the target duration distribution value in the cases where the duration distribution value and the interest degree level are positively correlated and negatively correlated.

[0067] Optionally, when the duration distribution value is positively correlated with the interest level, that is, the larger the duration distribution value, the higher the interest level. In this case, if the occurrence frequency of the target user is less than the first set threshold or the occurrence frequency of the target information object is less than the second set threshold, this indicates that the target user is a new user or a user who uses the target application less frequently, or the target information object is a new information object launched along the timeline. Among them, the size relationship between the first set threshold and the second set threshold is not limited and depends on the actual situation. When the occurrence frequency of the target user is less than the first set threshold or the occurrence frequency of the target information object is less than the second set threshold, the target duration distribution value is adjusted upward so that the interest level determined based on the adjusted target duration distribution value is not before adjustment.

[0068] When the duration distribution value is negatively correlated with the interest level, that is, the larger the duration distribution value, the lower the interest level. In this case, if the occurrence frequency of the target user is less than the first set threshold or the occurrence frequency of the target information object is less than the second set threshold, this indicates that the target user is a new user or a user who uses the target application less frequently, or the target information object is a new information object launched along the timeline. In this case, the target duration distribution value is adjusted downward so that the interest level determined based on the adjusted target duration distribution value is not before adjustment.

[0069] Further, according to the interest level characterized by the adjusted target duration distribution value, the interest of the target user in the target information object is determined. For the determination method, reference can be made to the foregoing embodiments and will not be elaborated here.

[0070] Optionally, after determining the interest of the target user in the target information object, the following application operations can also be performed based on the interest level.

[0071] Operation 1: According to the interest of the target user in the target information object, display the detailed information of the target information object. For example, when the target user has a high interest in the target information object, a graphic card can be popped up on the target page, and the detailed information of the target information object is introduced in the graphic card. Another example is that in response to the user's sliding operation, the scroll bar of the target page continuously scrolls down, and the detailed information of the target information object can be loaded in the form of a graphic card and appended to the end of the target page. By displaying the detailed information of the target information object with a high interest level to the user, it is convenient for the user to use the target application more immersively and improve the user's stickiness to the target application. When the target application is an e-commerce application and the target information object is product information, the exposure rate of the products that the user is interested in can be increased, thereby improving the conversion rate.

[0072] Operation 2: Highlight the target information object according to the target user's interest in the target information object. For example, if the target user has a high interest in the target information object, the target information object can be highlighted. In this embodiment, the method for highlighting the target information object when the interest level is high is not limited. Any method that can highlight the target information object from among the multiple information objects included in the target page falls within the scope of protection of this embodiment. Among them, for target information objects in different categories, the highlighting methods can be different. For example, for target information objects mainly in the text category, the highlighting methods are not limited to bold font, underlining, italicizing, changing font color, and highlighting. For another example, for target information objects mainly in the picture category, the picture can be enlarged to increase the exposure area, or marketing terms such as "new product" can also be added to the picture. By highlighting the target information object, its display effect can be enhanced, and the attention of the target user to the target information object can be improved.

[0073] Operation 3: Adjust the display order of other information objects included in the target page that are not displayed on the current screen according to the target user's interest in the target information object. For example, if the target user has a high interest in the target information object, in response to the user's swiping operation, as the scroll bar of the target page continuously scrolls down and other information objects are loaded, during this process, the display order of the target information object is adjusted to be before other information objects. If the target user has a low interest in the target information object, the display order of the target information object is adjusted to be after other information objects. By preferentially displaying the target information object with a high interest level of the target user and making accurate inferences for the target user, it is convenient for the target user to browse the target information object of interest first, and filter out the target information objects that the target user is not interested in among the target information objects that the target user has not browsed, improving the user experience and thus increasing the user retention rate.

[0074] Operation 4: Recommend associated information to the target user according to the target user's interest in the target information object. For example. For example, if the target user has a high interest in the target information object, the associated information of the target information object can be recommended to the target user, so that personalized information object recommendations can be provided according to the level of the user's interest, making it easier for users to find the information objects they are interested in, thereby improving the satisfaction of use. If the target user has a high interest in 3C products, the associated information of 3C products can be recommended to the target user, such as including but not limited to mobile phones, computers, and smart wearable devices, etc. Among them, "3C products" is the general term for computer products, communication products, and consumer electronic products.

[0075] In the above embodiments, a duration distribution prediction model is introduced. Through the duration distribution prediction model, the prediction of the duration distribution information of the target user's attention to the target information object can be realized. Among them, the architecture of the duration distribution prediction model is not limited in this embodiment. For example, it can be, but is not limited to, various machine learning models. For example, a deep learning model can be adopted, or a traditional machine learning model other than the deep learning model can be adopted. Among them, the traditional machine learning model refers to a model that uses mathematical methods such as statistics, including but not limited to linear regression, decision tree, support vector machine, etc. These models are all applicable to this solution. When the duration distribution model is implemented using a deep learning model, from the perspective of the model scale, the deep learning model can be a model with a relatively large parameter scale, for example, various large language models (Large Language Model), or a neural network model with a relatively small parameter scale, which is not limited in this regard. Further, for the model architecture of the deep learning model, it can be adopted, but is not limited to: Convolutional Neural Networks (CNN), Recurrent Neural Network (RNN), Deep Neural Networks (DNN), Attention network, and residual network, and is implemented based on architectures such as Encoder-Decoder.

[0076] The following provides a structure of the duration distribution prediction model, but it is not limited to this. As Figure 3 shown, the duration prediction distribution model 30 includes an embedding layer 31, a Recurrent Neural Networks (RNN) 32, a hierarchical attention layer 33, and a multi-task layer 34.

[0077] In this embodiment, the duration distribution prediction model 30 receives the context information of the target information object. In this embodiment, the context information includes, for example, but is not limited to, the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, the behavior data occurring on the current screen, and the attribute information of the target information object.

[0078] Further, the embedding layer 31 converts the input context information into multiple low-dimensional feature vectors for subsequent processing by the recurrent neural network 32. Among them, since the behavior data occurring on the current screen is often time series data, the recurrent neural network 32 is used to process the low-dimensional feature vectors corresponding to the behavior data occurring on the current screen to learn the time dependence of the behavior data, so as to model the low-dimensional feature vectors corresponding to the behavior data as time series feature vectors.

[0079] Further, the hierarchical attention layer 33 processes the temporal feature vectors by using the intra-behavior data sequence attention and the cross-behavior data sequence attention to model the temporal feature vectors as global behavior feature vectors.

[0080] Further, the global behavior feature vectors, along with the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, and the attribute information of the target information object, are input into the multi-task layer 34 to model the global behavior feature vectors as duration distribution information. The duration distribution information includes duration distribution values Q1, Q2, …, Qn; n ≥ 2, and n is a natural number.

[0081] In an embodiment of the present application, a structure of a duration distribution prediction model is further provided. The duration distribution prediction model includes: an embedding layer, a partitioning layer, an extraction layer, an interaction layer, an activation layer, and a multi-layer perceptron (MLP). The embedding layer converts different context information into low-dimensional feature vectors; the partitioning layer splits different types of context information; the extraction layer extracts the key features of the user that can be represented in the context information through a self-attention mechanism; the interaction layer is used to learn the connections between different key features; the activation layer is used to calculate the influence degree of different key features on the interest degree of the target information object, so as to generate a global interest representation; the global interest representation is input into the multi-layer perceptron to obtain duration distribution information.

[0082] In an embodiment of the present application, the duration distribution prediction model is obtained by training an initial prediction model with the goal of optimizing a loss function. The model training process is described below.

[0083] In an optional embodiment, the sample context information of multiple sample information objects is obtained. One sample information object corresponds to one sample user, and one sample user may correspond to one or more sample information objects; the sample attention duration of the sample user for the corresponding sample information object is obtained; the initial prediction model is trained according to the sample context information of the multiple sample information objects to obtain the sample duration distribution information of each sample user's attention to the corresponding sample information object; the loss function between the sample attention duration and the sample duration distribution information is calculated, and the model parameters of the initial prediction model are updated according to the loss function, and the loss function is optimized until the loss function meets the set termination condition, and then the model training is ended to obtain the duration distribution prediction model.

[0084] Further optionally, the sample duration distribution information includes multiple sample duration distribution values, and different sample duration distribution values represent different interest levels; when calculating the loss function between the sample attention duration and the sample duration distribution information, it includes: for any sample duration distribution value, calculating the loss function according to the difference between the sample attention duration and this sample duration distribution value, so that the difference between the sample attention duration and this sample duration distribution value gradually shrinks until the termination condition is met as the goal.

[0085] Further optionally, in the process of calculating the loss function for any sample duration distribution value according to the difference between the sample attention duration and this sample duration distribution value, according to the size relationship between this sample duration distribution value and other sample duration distribution values and the number of multiple sample duration distribution values, where the number of sample duration distribution values is set in advance as needed. Furthermore, calculate the weight of this sample duration distribution value, and then combine this sample duration distribution value, the weight of this sample duration distribution value, and the sample attention duration to calculate the loss function. Among them, in this embodiment, when the size relationship between the sample attention duration and the sample duration distribution value is different, the loss function used will be different. One case is for the first sample attention duration in the sample attention duration that is greater than this sample duration distribution value, calculating the loss function according to the weight and the difference between the first sample attention duration and this sample duration distribution value. Another case is for the second sample attention duration in the sample attention duration that is less than or equal to this sample duration distribution value, calculating the loss function according to the weight and the difference between this sample duration distribution value and the second sample attention duration. In any case, the ultimate goal is to optimize the loss function so that the difference between the sample attention duration and this sample duration distribution value gradually shrinks until the termination condition is met as the goal.

[0086] The following gives an example of the loss function for the above training process, and in combination with this example, the training process is described exemplarily.

[0087]

[0088] In the above loss function L τ (y, t τ ) where y represents the sample attention duration, and t τ represents the sample duration distribution value, which is an unknown variable, and τ i is the weight of each sample duration distribution value, and the weight parameters of different sample duration distribution values are different, and can be calculated according to i / (N + 1) provided above, where i represents the i-th sample duration distribution information, N represents the sample duration distribution value, and " / " represents the division operation. Among them, the above loss function is implemented as a piecewise function with two branches. One branch is in the case of y ≥ t τ , and the loss function is expressed as τ i (y - tτ )。Among them, (y - t τ ) represents obtaining the difference between the sample attention duration and the sample duration distribution value of this sample. In this case, the optimization loss function is used to gradually reduce the difference between y and t τ . In the above loss function, there is another branch. This branch is in other (otherwise) cases except y ≥ t τ . For example, other cases can be but are not limited to y ≤ t τ . In this branch, the loss function is expressed as (1 - τ i )(t τ - y). Among them, the purpose of optimizing the loss function is to gradually reduce the difference between y and t τ until the termination condition is met, at which point the model training ends and a duration distribution prediction model is obtained.

[0089] In an embodiment of the present application, a model training method is further provided. The duration distribution prediction model trained by this method can be applied but is not limited to the information processing method provided in the above embodiment. As Figure 4 shown, this method includes the following steps:

[0090] S401: Obtain the sample context information of multiple sample information objects. One sample information object corresponds to one sample user, and obtain the sample attention duration of the sample user for the corresponding sample information object;

[0091] S402: Train the initial prediction model according to the sample context information of multiple sample information objects to obtain the sample duration distribution information of each sample user's attention to the corresponding sample information object;

[0092] S403: Calculate the loss function between the sample attention duration and the sample duration distribution information, and update the model parameters of the initial prediction model according to the loss function until the loss function meets the set termination condition to obtain a duration distribution prediction model.

[0093] Among them, the model training method provided in this embodiment can be trained with reference to the training process of the model in the above embodiment. The relevant content can be referred to the above embodiment and will not be elaborated here.

[0094] The detailed implementation manners and beneficial effects of each step in the method of this embodiment have been described in detail in the foregoing embodiments and will not be elaborated here.

[0095] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can also be executed by different devices as the execution subject. For example, the execution subject of steps 401 to 403 can be device A; for another example, the execution subject of steps 401 and 402 can be device A, and the execution subject of step 403 can be device B; and so on.

[0096] In addition, in some of the processes described in the above embodiments and the accompanying drawings, a plurality of operations appear in a specific order. However, it should be clearly understood that these operations can be executed not in the order in which they appear in this document or in parallel. The operation numbers such as 401, 402, etc. are only used to distinguish different operations, and the numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and these operations can be executed in sequence or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit that "first" and "second" are of different types.

[0097] Figure 5 It is a schematic structural diagram of an electronic device provided in another exemplary embodiment of the present application. As Figure 5 shown, the electronic device includes: a memory 54 and a processor 55.

[0098] The memory 54 is used to store computer programs and can be configured to store various other data to support operations on the computing platform. Examples of these data include instructions for any application program or method for operating on the computing platform, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0099] The processor 55 is coupled to the memory 54 and is used to execute the computer program in the memory 54 for: in response to an access operation of a target user, presenting a target page, where the target page includes a plurality of information objects displayed on the current screen; for a target information object among the plurality of information objects, obtaining context information of the target information object and the attention duration of the target user for the target information object; inputting the context information into a duration distribution prediction model to predict the duration distribution information of the target user's attention to the target information object; and determining the interest degree of the target user in the target information object according to the duration distribution information and the attention duration.

[0100] In an optional embodiment, when the processor 55 obtains the context information of the target information object, it specifically is used for: obtaining at least one of the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, the behavior data occurring on the current screen, and the attribute information of the target information object as the context information of the target information object.

[0101] In an optional embodiment, when the processor 55 obtains the attention duration of the target user for the information object, it specifically is used for: obtaining the display duration of the target information object on the current screen as the attention duration; or, when the target information object is audio-visual information and is being played, obtaining the playback duration of the target information object as the attention duration; or, according to the attention position of the target user on the current screen, obtaining the duration during which the attention position falls within the screen range where the target information object is located as the attention duration.

[0102] In an optional embodiment, before the processor 55 obtains the context information of the target information object among the multiple information objects, it further is used for: respectively using the multiple information objects as the target information object; or, determining the target information object from the multiple information objects according to set selection conditions or the selection operation of the target user.

[0103] In an optional embodiment, the duration distribution information includes multiple duration distribution values, and different duration distribution values represent different interest level grades; when the processor 55 determines the interest level of the target user for the target information object according to the duration distribution information and the attention duration, it specifically is used for: comparing the attention duration with the multiple duration distribution values to obtain the target duration distribution value corresponding to the attention duration; determining the interest level of the target user for the target information object according to the interest level grade represented by the target duration distribution value.

[0104] In an optional embodiment, when the processor 55 determines the interest level of the target user for the target information object according to the interest level grade represented by the target duration distribution value, it specifically is used for: adjusting the target duration distribution value according to the attribute information of the target user and / or the target information object; determining the interest level of the target user for the target information object according to the interest level grade represented by the adjusted target duration distribution value.

[0105] In an alternative embodiment, when adjusting the target duration distribution value according to the attribute information of the target user and / or the target information object, the processor 55 is specifically configured to: when the duration distribution value is positively correlated with the interest level, if the occurrence frequency of the target user is less than a first set threshold or the occurrence frequency of the target information object is less than a second set threshold, adjust the target duration distribution value upward; or, when the duration distribution value is negatively correlated with the interest level, if the occurrence frequency of the target user is less than a first set threshold or the occurrence frequency of the target information object is less than a second set threshold, adjust the target duration distribution value downward.

[0106] In an alternative embodiment, the processor 55 is further configured to: obtain the sample context information of a plurality of sample information objects, where one sample information object corresponds to one sample user, and obtain the sample attention duration of the sample user for the corresponding sample information object; train an initial prediction model according to the sample context information of the plurality of sample information objects to obtain the sample duration distribution information of each sample user's attention to the corresponding sample information object; calculate a loss function between the sample attention duration and the sample duration distribution information, and update the model parameters of the initial prediction model according to the loss function until the loss function meets a set termination condition, so as to obtain the duration distribution prediction model.

[0107] In an alternative embodiment, the sample duration distribution information includes a plurality of sample duration distribution values, and different sample duration distribution values represent different interest levels; when calculating the loss function between the sample attention duration and the sample duration distribution information, the processor 55 is specifically configured to: for any sample duration distribution value, calculate the loss function according to the difference between the sample attention duration and the any sample duration distribution value.

[0108] In an alternative embodiment, when calculating the loss function according to the difference between the sample attention duration and any sample duration distribution value for any sample duration distribution value, the processor 55 is specifically configured to: calculate the weight of the any sample duration distribution value according to the magnitude relationship between the any sample duration distribution value and other sample duration distribution values and the number of the plurality of sample duration distribution values; for the first sample attention duration greater than the any sample duration distribution value in the sample attention duration, calculate the loss function according to the weight and the difference between the first sample attention duration and the any sample duration distribution value; for the second sample attention duration less than or equal to the any sample duration distribution value in the sample attention duration, calculate the loss function according to the weight and the difference between the any sample duration distribution value and the second sample attention duration.

[0109] In an optional embodiment, the execution subject of the method is the terminal device to which the current screen belongs, and the duration distribution prediction model is deployed on the terminal device.

[0110] In an optional embodiment, after determining the degree of interest of the target user in the target information object, the processor 55 is further configured to perform any of the following operations: display the detailed information of the target information object according to the degree of interest of the target user in the target information object; highlight the target information object according to the degree of interest of the target user in the target information object; adjust the display order of other information objects included in the target page that are not displayed on the current screen according to the degree of interest of the target user in the target information object; recommend associated information to the target user according to the degree of interest of the target user in the target information object.

[0111] Further, as Figure 5 shown, the electronic device further includes: other components such as a communication component 56, a display 57, a power supply component 58, and an audio component 59. Figure 5 Only some components are schematically shown, and it does not mean that the electronic device only includes Figure 5 the components shown. Additionally, Figure 5 the components within the dashed box in Figure 5 are optional components, rather than essential components, and can be determined according to the product form of the electronic device. The electronic device in this embodiment can be implemented as a terminal device such as a desktop computer, a laptop computer, a smart phone, or an IOT device, or can also be a server device such as a conventional server, a cloud server, or a server array. If the electronic device in this embodiment is implemented as a terminal device such as a desktop computer, a laptop computer, or a smart phone, it may include Figure 5 the components within the dashed box in

[0112] This application embodiment further provides an electronic device, and the implementation structure of this electronic device is the same as or similar to the implementation structure of the Figure 5 electronic device shown, and can be implemented with reference to the structure of the Figure 5 electronic device shown. The electronic device provided in this embodiment and Figure 5The differences between the electronic devices in the illustrated embodiments mainly lie in that the functions implemented by the processors in the electronic devices when executing the computer programs stored in the memories are different. For the electronic device provided in this embodiment, when its processor executes the computer program stored in the memory, it can be used to: obtain the sample context information of multiple sample information objects, where one sample information object corresponds to one sample user, and obtain the sample attention duration of the sample user for the corresponding sample information object; train an initial prediction model according to the sample context information of the multiple sample information objects to obtain the sample duration distribution information of each sample user's attention to the corresponding sample information object; calculate the loss function between the sample attention duration and the sample duration distribution information, and update the model parameters of the initial prediction model according to the loss function until the loss function meets the set termination condition to obtain the duration distribution prediction model.

[0113] The above-mentioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0114] The above-mentioned communication component is configured to facilitate communication between the device where the communication component is located and other devices in a wired or wireless manner. The device where the communication component is located can access a wireless network based on a communication standard, such as 2G, 3G, 4G / LTE, 5G and other mobile communication networks, or a combination thereof. In an exemplary embodiment, the communication component receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel.

[0115] The above-mentioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen can be implemented as a touch screen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can not only sense the boundaries of touch or swipe actions, but also detect the duration and pressure associated with the touch or swipe operations.

[0116] The above-mentioned power supply component provides power for various components of the device where the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device where the power supply component is located.

[0117] The above-mentioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC), which is configured to receive external audio signals when the device where the audio component is located is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals can be further stored in a memory or transmitted via a communication component. In some embodiments, the audio component further includes a speaker for outputting audio signals.

[0118] Accordingly, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to be able to implement the steps in the above-mentioned method embodiments. Among them, the computer-readable storage medium can be implemented by volatile or non-volatile or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, Phase-change Random Access Memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), flash memory or other memory technologies, Compact Disc Read Only Memory (CD-ROM), Digital Video Disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium

[0119] Accordingly, an embodiment of the present application further provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is enabled to implement the steps in the above method embodiments. It should be understood that each process or the combination of multiple processes in the above method flow can be implemented by the computer program or instructions. Additionally, these computer programs or instructions can be applied to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices, such that the processors of the general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices can be used as devices to implement the corresponding functions in the above method embodiments.

[0120] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover a non-exclusive inclusion, such that a process, method, commodity or device including a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, commodity or device including the element.

[0121] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. An information processing method, characterized in that: include: In response to an access operation of a target user, a target page is displayed, wherein the target page includes a plurality of information objects displayed on a current screen; For a target information object among the multiple information objects, acquiring context information of the target information object and a duration of attention of the target user to the target information object; Inputting the context information into a duration distribution prediction model to predict the duration distribution information of the target user paying attention to the target information object; The interest degree of the target user in the target information object is determined according to the duration distribution information and the attention duration.

2. The method according to claim 1, characterized in that Acquiring the context information of the target information object includes: Acquire at least one of the attribute information of the target user, the attribute information of the terminal device to which the current screen belongs, the behavior data occurring on the current screen, and the attribute information of the target information object as the context information of the target information object.

3. The method according to claim 1, characterized in that Obtaining the target user's attention duration on the information object includes: Obtaining the display duration of the target information object on the current screen as the attention duration; or When the target information object is audio or video information and is played, obtaining the playing time of the target information object as the attention time; or According to the focus position of the target user on the current screen, the time duration during which the focus position is within the screen range where the target information object is located is acquired as the focus duration.

4. The method according to claim 1, characterized in that: Before acquiring the context information of a target information object among the multiple information objects, the method further includes: taking the multiple information objects as the target information objects respectively; or The target information object is determined from the multiple information objects according to a set selection condition or a selection operation of the target user.

5. The method according to claim 1, characterized in that The duration distribution information includes a plurality of duration distribution values, and different duration distribution values ​​represent different interest levels; Determining the interest of the target user in the target information object according to the duration distribution information and the attention duration includes: Compare the attention duration with a plurality of duration distribution values ​​to obtain a target duration distribution value corresponding to the attention duration; The interest level of the target user in the target information object is determined according to the interest level level represented by the target duration distribution value.

6. The method according to claim 5, characterized in that Determining the interest of the target user in the target information object according to the interest level represented by the target duration distribution value includes: Adjusting the target duration distribution value according to the attribute information of the target user and / or the target information object; The interest level of the target user in the target information object is determined according to the interest level level represented by the adjusted target duration distribution value.

7. The method according to claim 6, characterized in that Adjusting the target duration distribution value according to the attribute information of the target user and / or the target information object includes: In the case where the duration distribution value is positively correlated with the interest level, if the appearance frequency of the target user is less than a first set threshold or the appearance frequency of the target information object is less than a second set threshold, the target duration distribution value is adjusted upward; or In the case where the duration distribution value is negatively correlated with the interest level, if the appearance frequency of the target user is less than a first set threshold or the appearance frequency of the target information object is less than a second set threshold, the target duration distribution value is adjusted downward.

8. The method according to any one of claims 1 to 7, characterized in that: Also includes: Acquire sample context information of multiple sample information objects, where one sample information object corresponds to one sample user, and acquire the sample attention duration of the sample user to the corresponding sample information object; Training the initial prediction model according to the sample context information of the multiple sample information objects to obtain sample duration distribution information of each sample user paying attention to the corresponding sample information object; The loss function between the sample attention duration and the sample duration distribution information is calculated, and the model parameters of the initial prediction model are updated according to the loss function until the loss function meets the set termination condition to obtain the duration distribution prediction model.

9. The method according to claim 8, characterized in that The sample duration distribution information includes a plurality of sample duration distribution values, and different sample duration distribution values ​​represent different interest levels; Calculating the loss function between the sample attention duration and the sample duration distribution information includes: For any sample duration distribution value, the loss function is calculated according to the difference between the sample attention duration and any sample duration distribution value.

10. The method according to claim 9, characterized in that For any sample duration distribution value, the loss function is calculated according to the difference between the sample attention duration and any sample duration distribution value, including: Calculate the weight of any sample duration distribution value according to the magnitude relationship between any sample duration distribution value and other sample duration distribution values ​​and the number of the multiple sample duration distribution values; For a first sample attention duration among the sample attention durations that is greater than any of the sample duration distribution values, calculating the loss function according to the weight and a difference between the first sample attention duration and any of the sample duration distribution values; For a second sample attention duration among the sample attention durations that is less than or equal to any sample duration distribution value, the loss function is calculated according to the weight and a difference between any sample duration distribution value and the second sample attention duration.

11. The method according to any one of claims 1 to 7, characterized in that: The execution subject of the method is a terminal device to which the current screen belongs, and the duration distribution prediction model is deployed on the terminal device.

12. The method according to any one of claims 1 to 7, characterized in that: After determining the interest of the target user in the target information object, the method further includes any of the following operations: Displaying detailed information of the target information object according to the target user's interest in the target information object; According to the interest of the target user in the target information object, highlighting the target information object; According to the interest of the target user in the target information object, adjusting the display order of other information objects included in the target page and not displayed on the current screen; Recommending related information to the target user based on the target user's interest in the target information object.

13. A model training method, characterized in that: include: Acquire sample context information of multiple sample information objects, where one sample information object corresponds to one sample user, and acquire the sample attention duration of the sample user to the corresponding sample information object; Training the initial prediction model according to the sample context information of the multiple sample information objects to obtain sample duration distribution information of each sample user paying attention to the corresponding sample information object; The loss function between the sample attention duration and the sample duration distribution information is calculated, and the model parameters of the initial prediction model are updated according to the loss function until the loss function meets the set termination condition to obtain the duration distribution prediction model.

14. An electronic device, characterized in that: The invention comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is coupled to the memory and is used to execute the computer program to implement the steps in the method according to any one of claims 1 to 12 and claim 13.

15. A computer-readable storage medium storing a computer program / instruction, characterized in that: When the computer program is executed by a processor, the processor is enabled to implement the steps of the method according to any one of claims 1 to 12 and claim 13.

16. A computer program product, characterized in that include: A computer program / instruction, when the computer program / instruction is executed by a processor, enables the processor to implement the steps in any one of claims 1 to 12 and claim 13.

Images