Data processing method and device, computer equipment and medium

Abstract

The embodiment of the invention discloses a data processing method and device, computer equipment and a medium, and the method comprises the steps: obtaining historical interaction information generated by a target user for content in a content set on a target application, and constructing a user behavior track of the target user for the content in the content set based on the historical interaction information; based on the user behavior track of the target user for the content in the content set, determining that the content in the content set generates a scene label corresponding to historical interaction information on the target application; obtaining a target business scene type of a to-be-trained model associated with the target application, and determining a target content set conforming to the target business scene type from the content set based on the scene label of the content in the content set; training the to-be-trained model based on the target content set to obtain a target model; the target model is used for processing a task corresponding to a target business scene type on the target application; the training effect of the model can be improved, so that the task processing effect of the model can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a data processing method, device, computer equipment and medium. Background Art

[0002] With the rapid development of multimedia technology, various business scenarios can exist in one application. The same content may exist in different business scenarios, that is, users can perform related behavioral operations on the same content in different business scenarios, such as browsing, clicking, sharing, etc. In tasks under business scenarios, related models are usually used. Before use, the model needs to be trained using the content in the application. At present, when it is necessary to train a model in a certain business scenario, it is usually trained directly using the content in the application, which has a poor training effect. Therefore, how to improve the training effect of the model has become a current research hotspot. Summary of the invention

[0003] The embodiments of the present application provide a data processing method, apparatus, computer equipment and medium, which can improve the training effect of the model.

[0004] The first aspect of the embodiment of the present application discloses a data processing method, the method comprising:

[0005] Acquire historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on contents in the content collection based on the historical interaction information;

[0006] Based on the user behavior track of the target user with respect to the content in the content set, determining a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application;

[0007] Acquire a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario label of the content in the content set;

[0008] The model to be trained is trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task

[0009] A second aspect of an embodiment of the present application discloses a data processing device, the device comprising:

[0010] a determination unit, configured to obtain historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on contents in the content collection based on the historical interaction information;

[0011] The determining unit is further used to determine, based on the user behavior track of the target user with respect to the content in the content set, a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application;

[0012] The determining unit is further configured to obtain a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario labels of the content in the content set;

[0013] A training unit is used to train the model to be trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task.

[0014] A third aspect of an embodiment of the present application discloses a computer device, comprising a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method of the first aspect above.

[0015] A fourth aspect of an embodiment of the present application discloses a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor executes the method of the first aspect.

[0016] The fifth aspect of the embodiments of the present application discloses a computer program product or a computer program, wherein the computer program product or the computer program comprises program instructions, and when the program instructions are executed by a processor, the method of the first aspect is implemented.

[0017] In an embodiment of the present application, the historical interaction information generated by the target user on the content in the content set on the target application can be obtained, and the user behavior trajectory of the target user on the content in the content set can be constructed based on the historical interaction information; based on the user behavior trajectory of the target user on the content in the content set, the scene label of the content in the content set generating the corresponding historical interaction information on the target application can also be determined, and the scene label is used to indicate the business scenario type on the target application; further, the target business scenario type of the model to be trained associated with the target application can be obtained, and the target content set that matches the target business scenario type can be determined from the content set based on the scene label of the content in the content set; thus, the model to be trained can be trained based on the target content set to obtain the target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain the task processing result for the task. By implementing the above method, the historical interaction information of the user on the content on the target application can be monitored, and the notes of different business scenarios can be summarized and attributed based on the historical interaction information to identify which type of business scenario the content belongs to, and the content of which type of business scenario the historical interaction information belongs to can also be determined to improve the accuracy of the label data of the content. Subsequently, different models can be trained based on the content in different business scenarios. Due to the accuracy of the content label data, the model can accurately consume the historical interaction information of each business scenario during training, thereby improving the accuracy and reliability of model training, and further improving the accuracy of model prediction and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1a It is a flowchart of a data processing solution provided in an embodiment of the present application;

[0020] Figure 1b It is a schematic diagram of a data processing system provided in an embodiment of the present application;

[0021] Figure 2 It is a flowchart of a data processing method provided in an embodiment of the present application;

[0022] Figure 3a This is a schematic diagram of page jump in a recommendation scenario provided by an embodiment of the present application;

[0023] Figure 3bThis is a schematic diagram of page jump in a search scenario provided by an embodiment of the present application;

[0024] Figure 3c This is a schematic diagram of page jump in an attachment scenario provided by an embodiment of the present application;

[0025] Figure 3d This is a schematic diagram of page jump in an author introduction scenario provided by an embodiment of the present application;

[0026] Figure 4 is a structural schematic diagram of another data processing method provided in an embodiment of the present application;

[0027] Figure 5 is a structural schematic diagram of a data processing device provided in an embodiment of the present application;

[0028] Figure 6 It is a structural diagram of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0030] The present application embodiment proposes a data processing solution, for example, see Figure 1a As shown, the principle of the data processing scheme is as follows: the historical interaction information generated by the target user for the content in the content set can be obtained, so as to determine the scenario label corresponding to the content in the content set based on the historical interaction information, and the scenario label can be used to indicate the business scenario type. In one embodiment, the user behavior trajectory of the target user for the content in the content set can be constructed based on the historical interaction information, so as to determine the scenario label of the corresponding user behavior trajectory generated by the content in the content set on the target application based on the user behavior trajectory of the target user.

[0031] When there is a need to train the model to be trained later, the target business scenario type corresponding to the model to be trained can be further obtained, and the target content set that matches the target business scenario type can be determined from the content set based on the scenario label of the content in the content set, so as to train the model to be trained based on the target content set, thereby obtaining the target model, which can be used to process the tasks corresponding to the target business scenario type on the target application. The content in the above-mentioned content set can be the content on the target application, or other applications that have an associated relationship with the target application, that is, in addition to using the historical interaction information of users on the target application for the content to determine the scenario label of the content on the target application, the historical interaction information of users on other applications for the content can also be used to determine the scenario label of the content on the target application.

[0032] Among them, the association relationship is not specifically limited; for example, the association relationship may refer to applications of the same type, such as the target application is multimedia playback application A, and other applications may be multimedia playback application B; for another example, the association relationship may refer to applications under the same enterprise, such as the target application is application 1 of enterprise A, and other applications are application 2 under enterprise A.

[0033] Through the above method, the historical interaction information of users on the content of the target application can be monitored, and the notes of different business scenarios can be summarized and attributed based on the historical interaction information to identify which type of business scenario the content belongs to. At the same time, it can also be determined which type of business scenario the historical interaction information belongs to, so as to improve the accuracy of the content label data. Subsequently, different models can be trained based on the content in different business scenarios. Due to the accuracy of the content label data, the model can accurately consume the historical interaction information of each business scenario during training, thereby improving the accuracy and reliability of model training, and then improving the accuracy of model prediction, and also improving the user experience.

[0034] In a specific implementation, the execution subject of the above-mentioned data processing scheme may be a computer device, which may be a terminal or a server. The terminal mentioned here may be a smart phone, tablet computer, laptop computer, desktop computer and other devices; the server may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, etc.

[0035] For example, when the computer device is a server, the embodiment of the present application provides a data processing system, such as Figure 1b As shown, the data processing system may include at least one terminal and at least one server; the terminal can obtain the historical interaction information generated by the target user on the content set on the target application, and upload the obtained historical interaction information to the server (i.e., computer device), so that the server can obtain the historical interaction information, adopt a data processing solution and implement training of the model to be trained based on the historical interaction information, thereby obtaining the target model.

[0036] Based on the data processing solution provided above, the embodiment of the present application provides a data processing method, which can be executed by the computer device mentioned above. Figure 2 The data processing method includes but is not limited to the following steps:

[0037] S201 , obtaining historical interaction information generated by a target user on contents in a content collection on a target application, and constructing a user behavior track of the target user on the contents in the content collection based on the historical interaction information.

[0038] The target application may refer to an application that can perform content recommendation, such as a browser, a community application, an information flow application, an education application, a multimedia playback application, etc. The target user may refer to a user who uses the target application, and the number of the target users may be one or more. The content on the target application may refer to information published by a publishing user (or author), such as one or more of an image, a video, and a text. The content may be referred to as a note.

[0039] Among them, interaction can refer to actions such as clicking, liking, sharing, following, commenting or blocking, and historical interaction information can refer to the interactive actions performed by users on the content within a historical time period, such as clicking, liking, etc. on notes.

[0040] In one implementation, the log data information generated by the target user on the target application can be obtained first. It should be understood that the application on the terminal device is usually equipped with point-of-sale collection technology, which can refer to a technical means of monitoring, capturing and recording specific user behaviors or events when the user uses the application or visits the website. Based on this, it can be seen that the point-of-sale collection technology can be used to monitor the user's usage in the application or website, such as the user's access to various pages in the application, the time the user visits a page, the length of time the user stays on a page, the user's user operations on a page (such as clicks, sharing, liking, saving, etc.), etc.

[0041] Based on the above description, it can be known that when the target user uses the target application, the various behavior data of the user in the target application can be collected through the data collection points in the target application, and the various behavior data collected can be recorded to obtain log data information. The log data information records a series of user behaviors performed by the target user in the target application. For example, the target user's operation behavior (such as viewing behavior, clicking behavior, sharing behavior, etc.) on the recommended content on the recommendation page (recommendation page) corresponding to the target application; for example, the target user enters a certain search word on the search page (search page) corresponding to the target application, and then the operation behavior of each search result in the search page. Among them, the point collection technology can also obtain the device identifier used by the user and the user identifier of the user. The device identifier can uniquely identify a device, and the user identifier can uniquely identify a user. When the target user uses the target application, the point collection technology in the target application can obtain the user identifier of the target user, so as to use the user identifier of the target user as a keyword to record the log data information of each target user separately.

[0042] After obtaining the log data information, the historical interaction information generated by the target user to the content set on the target application can be further obtained from the log data information, so that the subsequent historical interaction information can construct the user behavior trajectory of the user to the content in the target application. In one embodiment, the historical interaction information of each target user can be summarized based on the log data information to obtain the historical interaction information corresponding to each target user. Here, the user ID of each target user can be used as an independent keyword (key), and the historical interaction information of all target users can be summarized in units of user IDs, that is, one user ID can be associated with the corresponding historical interaction information, so that all subsequent historical interaction information can be attributed based on the key.

[0043] In one implementation, considering that the principle of constructing a user behavior trajectory for any content is similar, the construction of the user behavior trajectory is specifically described by taking a target user and a content in a content set as an example.

[0044] The target event performed by the target user on any content can be first extracted from the historical interaction information; wherein, the target event can be pre-set, such as the target event can refer to one or more of an exposure event, a click event, etc. After obtaining these target events, the user behavior trajectory generated by the target user on any content can be constructed based on the time sequence of the target events and the target events. For example, the target events can be connected according to the time sequence of the target events, and the connected target event sequence is the user behavior trajectory. For example, taking the content as a note, assuming that the target events of the target user for the note are exposure events and click events in time sequence, then the user behavior trajectory of the target user for the note is: exposure event -> click event. It can be seen that the user behavior trajectory here can characterize event operations with a time sequence.

[0045] It can be understood that there is a corresponding page for the target event. For example, the page corresponding to the exposure event of the note can be a search page, a recommendation page, etc., and the page corresponding to the click event is the note details page of the note, etc., that is, a target event can correspond to a page. In addition to using the target event to construct the user behavior trajectory, the page corresponding to the target event can also be used to construct the user behavior trajectory. That is, after determining the target event, the user behavior trajectory can be constructed based on the page corresponding to the target event. Here, the pages are connected according to the chronological order of the pages generated by the target event, and the connected page sequence is the user behavior trajectory. For example, taking the content as a note, assuming that the target event of the target user for the note is an exposure event and a click event, and the pages corresponding to the exposure event and the click event are the recommendation page and the note details page in chronological order, then the user behavior trajectory of the target user for the note is: recommendation page -> note details page. It can be seen that the user behavior trajectory here can characterize the page access with a chronological order.

[0046] For example, see Figure 3aAs shown, a user enters the recommendation page 31 of the target application. There are multiple recommended notes (such as graphic note A, graphic note B, etc.) in the recommendation page 31. Here, the user's target event for graphic note B is an exposure event, and the corresponding page is the recommendation page. The user can perform a trigger operation (such as a click operation) on any recommended note in the recommendation page 31. For example, the user can click on the graphic note B marked by 301 in the recommendation page 31, and the recommendation page 31 can jump to the note details page 32 corresponding to the graphic note B. The user can view the notes in the note details page 32. Here, the user's target event for graphic note B is a click event, and the corresponding page is the note details page. The note details page 32 may contain an author ID 302 for publishing the graphic note B. The user may perform a trigger operation (such as a click operation) on the author ID 302 in the note details page 32, and the note details page 32 may jump to the author details page 33 showing the publishing user corresponding to the author ID 302, that is, the user may view the user information of the publishing user in the author details page 33. Here, the user's target event for the graphic note B is a click event, and the corresponding page is the author details page. In summary, it can be seen that for the above-mentioned user behavior data, the user behavior trajectory constructed based on the user's target event can be exposure time->click event->click event, and the user behavior trajectory constructed based on the page corresponding to the target event can be: recommendation page->note details page->author details page.

[0047] For example, see Figure 3b As shown, a user enters the search page of the target application, and performs a search operation on the search page. After the computer device responds to the search operation, a search result page showing the search results can be displayed, and there are multiple search results in the search result page; the user can perform a trigger operation on a search result in the search result page, and the search result page can jump to the result detail page corresponding to the search result, that is, the user can view the detailed search results in the result detail page. For the above user behavior data, it can be determined that the user behavior trajectory for the user is: search page -> search result page -> result detail page.

[0048] For example, see Figure 3bAs shown, a user enters the search page 34 of the target application. In the search page 34, there are multiple search notes (such as graphic note A, graphic note B, etc.) obtained based on the search term. Here, the user's target event for graphic note B is an exposure event, and the corresponding page is the search page. The user can perform a trigger operation (such as a click operation) on any search note in the search page 34. For example, the user can click on the graphic note B marked by 303 in the search page 34, and the search page 34 can jump to the note details page 35 corresponding to the graphic note B. Here, the user's target event for graphic note B is a click event, and the corresponding page is the note details page. In summary, it can be seen that for the above-mentioned user behavior data, the user behavior trajectory constructed based on the user's target event can be exposure time->click event, and the user behavior trajectory constructed based on the page corresponding to the target event can be: search page->note details page.

[0049] For example, see Figure 3c As shown, a user enters the nearby page 36 of the target application. In the nearby page, there are multiple notes (such as graphic note A, graphic note B, etc.) posted by users whose physical distance from the user is within a preset distance range. Here, the user's target event for graphic note B is an exposure event, and the corresponding page is a nearby page. The user can perform a trigger operation (such as a click operation) on any note in the nearby page 36. For example, the user can click on the graphic note B marked by 304 in the nearby page 36, and the nearby page 36 can jump to the note details page 37 corresponding to the graphic note B. Here, the user's target event for graphic note B is a click event, and the corresponding page is a note details page. In summary, it can be seen that for the above-mentioned user behavior data, the user behavior trajectory constructed based on the user's target event can be exposure time->click event, and the user behavior trajectory constructed based on the page corresponding to the target event can be: nearby page->note details page.

[0050] For example, see Figure 3dAs shown, a user enters the author details page 38 of the target application. The author details page 38 contains information about the author and may also include notes published by the author (such as graphic note B, graphic note G, etc.). Here, the user's target event for graphic note G is an exposure event, and the corresponding page is the author details page. The user can perform a trigger operation (such as a click operation) on any note in the author details page 38. For example, the user can click on the graphic note G marked by 305 in the author details page 38, and the author details page 38 can jump to the note details page 39 corresponding to the graphic note G. Here, the user's target event for graphic note G is a click event, and the corresponding page is the author details page. In summary, it can be seen that for the above-mentioned user behavior data, the user behavior trajectory constructed based on the user's target event can be exposure time->click event, and the user behavior trajectory constructed based on the page corresponding to the target event can be: author details page->note details page.

[0051] In one implementation, historical interaction information may be stored in a target area, and the target area is not specifically limited. For the historical interaction information stored in the target area, a corresponding storage duration may be set, that is, the historical interaction information stored in the target area has an expiration time. That is, the storage duration of the historical interaction information in the target area may be monitored in real time to update the historical interaction information in the target area. While continuously updating, the historical interaction information stored in the target area may be guaranteed to be the latest information, thereby ensuring that the latest historical interaction information may be used to determine the scene label of the content, thereby improving the accuracy of determining the target sample set of the model to be trained based on the scene label, and improving the training effect of the model to be trained, and thus improving the processing effect of the model in performing task processing in the corresponding business scenario type.

[0052] Based on this, it can be known that the storage time of the historical interaction information of the target user in the target area can be monitored in real time. If the storage time of the historical interaction information of the target user reaches the first storage time, it can be further determined whether the target user has historical interaction information within the preset historical time period to determine whether the historical interaction information needs to be deleted. Specifically, if not, the historical interaction information of the target user can be cleared; if so, the first storage time can be extended to the second storage time. Among them, the second storage time is longer than the first storage time, and the first storage time and the second storage time can be pre-set, and their specific values ​​are not limited.

[0053] Among them, the preset historical time period can be pre-set, for example, the preset historical time period can refer to a time period that is before the current time and is separated from the current time by a preset time period (for example, 5 minutes, 10 minutes, etc.), and the current time here can refer to the moment when the storage time of the historical interaction information is detected.

[0054] In one embodiment, the second storage duration may be determined based on the amount of information of the historical interaction information in a preset historical time period. Based on this, it can be known that the amount of information of the historical interaction information in the preset historical time period can be obtained, and the second storage duration can be determined based on the amount of information. For example, the second storage duration may be set according to the principle that the second storage duration is positively correlated with the amount of information, that is, the larger the amount of information, the longer the second storage duration may be, and the smaller the amount of information, the shorter the second storage duration may be.

[0055] In one embodiment, the second storage duration may be determined based on the frequency of generation of historical interaction information within a preset historical time period. The generation frequency may refer to the number of times the historical interaction information is stored in the target area per unit time. Based on this, it can be known that the generation frequency of the historical interaction information within the preset historical time period can be obtained, and the second storage duration can be determined based on the generation frequency. For example, the second storage duration may be set according to the principle that the second storage duration is positively correlated with the generation frequency, that is, the faster the generation frequency, the longer the second storage duration can be, and the slower the generation frequency, the shorter the second storage duration can be.

[0056] In one implementation, the historical interaction information generated by the target user for any content includes the target event executed by the target user for any content. In order to avoid the disorder of various target events stored in the target area, when storing the corresponding target event in the target area, it can be determined whether to store it in the target area based on the execution time of the target event. In a specific implementation, if there is a current event to be saved for any content, it is queried whether there is a target event executed for any content in the target area; if not, the current event can be stored in the target area. If it exists, the first execution time of the target event in the target area can be obtained, and the second execution time of the current event can be obtained; if the second execution time is later than the first execution time, the current event can be stored in the target area; if the second execution time is earlier than the first execution time, the current event can be filtered. Here, when determining whether to store the current event in the target area, the event type of the target event to be judged refers to the same type. For example, assuming that the current event is a click event, the target event to be judged also refers to a click event. If the current event is an exposure event, the target event to be judged is also an exposure event.

[0057] In one implementation, for a user on a target application, there may be no corresponding historical interaction information. For example, the target user may be a new user. Taking these situations into consideration, when obtaining historical interaction information from a target area to construct a user behavior trajectory of a target user for a content, it is possible to query whether there is historical interaction information for the target user from the target area. If the corresponding historical interaction information can be queried from the target area, the queried historical interaction information can be obtained to perform subsequent operations. If the corresponding historical interaction information cannot be queried from the target area, that is, there is no historical interaction information for the target user, the historical interaction information of the target user can be initialized, that is, the historical interaction information of the target user can be initialized. The initialized historical interaction information can be determined based on the historical interaction information of other users on the target application. For example, the historical interaction information of a reference user among other users can be initialized as the historical interaction information of the target user. The reference user can refer to a user who has interactive behavior with the target user. Alternatively, randomly generated historical interaction information can be used as the initialized historical interaction information. Alternatively, the historical interaction information can be initialized in other ways, and comparison is not specifically limited.

[0058] S202 , based on the user behavior trajectory of the target user with respect to the content in the content collection, determine a scenario label of the content in the content collection generating corresponding historical interaction information on the target application.

[0059] Among them, the scene tag can be used to indicate the business scene type. The target application may involve various business scenes. For example, the business scenes involved in the target application may include recommendation, search, advertisement, nearby, attention, etc., and the scene tag for the content here may include a scene tag for indicating the business scene types such as recommendation, search, advertisement, nearby, attention, etc.

[0060] In one implementation, the business scenario type of the content in the content set that generates the corresponding user behavior trajectory on the target application can be determined based on the user behavior trajectory of the target user for the content in the content set, and then a scenario label of the corresponding business scenario type can be added to the content in the content set. The scenario label of the content in the content set can facilitate the subsequent distinction of the business scenario types to which different content belongs.

[0061] In one implementation, the content with scenario tags added this time can be used in the subsequent model training of models of different business scenario types. It can be understood that the models used in different business scenario types play different roles. In order to achieve better training results for the model, when selecting the sample set of the model, the content under the corresponding business scenario type can be selected as the sample set. Based on this, it can be known that corresponding scenario tags can be added to the content of the target application, so that sample sets can be selected for models of different business scenario types based on the scenario tags of the content.

[0062] In one implementation, as described above, the user behavior trajectory of the target user on any content is obtained by processing the target events in the chronological order of the target events executed by the target user on any content. Considering that the principle of determining the scene label of each content is similar, the following is an example of any content in the content set for relevant elaboration. In a specific implementation, for the user behavior trajectory of the target user on any content, the target page generated by each target event in the user behavior trajectory can be determined first, and then the scene label of the corresponding user behavior trajectory generated by any content on the target application can be determined based on the business scenario type corresponding to the target page.

[0063] In one embodiment, it is possible to determine whether the target page generated by the first target event in the user behavior track of the target user for any content is a first-level page of the target application, so as to determine the scene label corresponding to any content based on the determination result. If the target page generated by the first target event in the user behavior track of the target user for any content is a first-level page of the target application, then the scene label corresponding to the business scenario type corresponding to the first-level page can be used as the scene label for the corresponding user behavior track generated by any content on the target application.

[0064] If the page generated by the first target event in the user behavior track of the target user for any content is not a primary page of the target application, then the scenario label of the corresponding user behavior track generated by any content on the target application can be determined based on the logical rule for generating the page corresponding to the first target event. For example, assuming that the page corresponding to the first target event in the user behavior track of a note is the note details page, then based on the logical rule for generating the note details page, it can be determined that the previous page of the note details page can be a recommendation page, a search page, a nearby page, etc., then the business scenario types corresponding to these pages can all be used as the business scenario types of the note, or the business scenario types corresponding to one or more pages can be randomly selected from these pages as the business scenario types of the note.

[0065] Among them, the first-level page can refer to the page of the business scenario corresponding to the business scenario type in the target application. For example, when the business scenario type is recommendation, the first-level page can be understood as a recommendation page, when the business scenario type is search, the first-level page can be understood as a search page, when the business scenario type is nearby, the first-level page can be understood as a nearby page, and when the business scenario type is attention, the first-level page can be understood as a attention page. It can be understood that when a user generates a target event on the target application, there is a corresponding page entry on the target application, that is, the target page corresponding to the target event. The first-level page here can also refer to the first-level page entry of the target application for the generated target event. The goal of the embodiment of the present application is to find which first-level page entry in the target application the content enters from, and then generate subsequent historical interaction information. Then, through the above method, the first-level page entry of each content in the content set can be determined, and the first-level page entry corresponds to the business scenario type, then the business scenario type of each content can be further determined.

[0066] For example, assuming that the target page generated by the first target event in the user behavior track of a content is a recommendation page, the scene tag of the content can be determined to be a tag indicating recommendation. For another example, assuming that the target page generated by the first target event in the user behavior track of a content is a search page, the scene tag of the content can be determined to be a tag indicating search. For another example, assuming that the target page generated by the first target event in the user behavior track of a content is a focus page, the scene tag of the content can be determined to be a tag indicating focus.

[0067] In one implementation, as mentioned above, the user behavior trajectory of the target user for any content is obtained by processing the pages in the chronological order of the target events performed by the target user on any content, and the scene label of the content can be determined directly based on the business scenario type corresponding to the page in the user behavior trajectory. In one embodiment, it can be determined whether the first page in the user behavior trajectory of the target user for any content is a first-level page of the target application, so as to determine the scene label corresponding to any content based on the judgment result. If the first page in the user behavior trajectory of the target user for any content is a first-level page of the target application, the scene label corresponding to the business scenario type corresponding to the first-level page can be used as the scene label for the corresponding user behavior trajectory of any content on the target application.

[0068] If the first page in the user behavior track of the target user for any content is not a primary page of the target application, the scenario label of the corresponding user behavior track generated by any content on the target application can be determined based on the logical rule for generating the first page. For example, assuming that the first page in the user behavior track of a note is the note details page, based on the logical rule for generating the note details page, it can be determined that the previous page of the note details page can be a recommendation page, a search page, a nearby page, etc., then the business scenario types corresponding to these pages can all be used as the business scenario types of the note, or the business scenario types corresponding to one or more pages can be randomly selected from these pages as the business scenario types of the note.

[0069] S203, obtaining a target business scenario type of the to-be-trained model associated with the target application, and determining a target content set that matches the target business scenario type from the content set based on the scenario label of the content in the content set.

[0070] In one implementation, for a target application, a model to be trained for various business scenario types under the target application can be configured for it, that is, one business scenario type is configured with a model to be trained. For example, if the target application includes business scenario types such as recommendation, search, advertising, nearby, and attention, then the target application is associated with models to be trained corresponding to business scenario types such as recommendation, search, advertising, nearby, and attention. Among them, the model architecture of the model to be trained is not specifically limited, and it is only necessary to ensure that the model to be trained has a prediction function.

[0071] Among them, the above-mentioned model to be trained needs to be pre-trained to obtain the corresponding target model, and the target model can be used to process tasks under the corresponding business scenario type. In order to train the model to be trained, a training sample set for training the model to be trained can be obtained first, so that the training sample set can be used to train the model to be trained. It can be understood that the various contents in the above-mentioned content set can involve various business scenario types, such as business scenario types can include recommendation, search, advertising, nearby, attention, etc. Then the content set here can be applied to business scenario types such as recommendation, search, advertising, nearby, attention, etc.

[0072] Based on this, it can be known that in order to realize the training of a certain model to be trained associated with the target application, it is necessary to obtain a target sample set adapted to the model to be trained. In one embodiment, the target business scenario type applicable to the model to be trained can be obtained first. After determining the target business scenario type, the target content set that matches the target business scenario type is determined from the content set based on the scene label of the content in the content set. The target content set is the training sample set. Specifically, the business scenario type indicated by the scene label of each content in the content set can be judged for consistency with the target business scenario type. If a certain content passes the consistency judgment, the content can be added to the target content set. The consistency judgment here refers to determining whether the business scenario type indicated by the scene label is consistent with the target business scenario type. If they are consistent, it is determined that the consistency judgment has passed. If they are inconsistent, it is determined that the consistency judgment has not passed.

[0073] For example, assuming that the target business scenario type is a business scenario type for recommendation, it means that the target model obtained by training the model to be trained needs to be applied to the content recommendation scenario. The target content set required by the model to be trained is the content set in the content recommendation scenario. Then, the content with the business scenario type of recommendation can be selected from the content set, and the selected content can be added to the target content set.

[0074] In summary, by monitoring the user's historical interaction information on the content, the content can be divided twice to determine which type of business scenario the content belongs to. At the same time, the historical interaction information of the content can be divided twice to determine which type of business scenario the content belongs to, and prepare the required accurate label data for the model to avoid the model learning useless information, thereby improving the training effect of the model.

[0075] S204, training the to-be-trained model based on the target content set to obtain a target model; the target model is used to process tasks corresponding to the target business scenario type on the target application to obtain task processing results for the tasks.

[0076] In one implementation, for any target content in the target content set, the data of any target content under the target dimension can be determined based on the user behavior track of any target content, and the tag information of any target content can be determined based on the data under the target dimension. Among them, the target dimension can include one or more of the click dimension, the sharing dimension, the collection dimension, and the like dimension; the tag information can be used to indicate whether any target content has been subjected to a behavioral operation under the target dimension. For example, assuming that the target dimension is the click dimension, the tag information of the target content can be used to indicate whether the target content has been subjected to a behavioral operation under the click dimension. Exemplarily, assuming that the business scenario type of a target content under the corresponding user behavior track is a content recommendation type, if there is a click event in the user behavior track of the target content, then the tag information of the target content can be determined to indicate that the target content has been subjected to a behavioral operation under the click dimension, and if there is no click event in the user behavior track of the target content, then the tag information of the target content can be determined to indicate that the target content has not been subjected to a behavioral operation under the click dimension.

[0077] After label information is associated with each target content in the target content set, the target content associated with label information can be used to train the model to be trained to obtain the target model. In one embodiment, for any target content, any target content can be input into the model to be trained to obtain the training information of any target content, and then the model to be trained can be trained based on the label information and training information of any target content to obtain the target model.

[0078] For example, the model loss value of the model to be trained can be calculated based on the label information and training information of any target content; for example, the model loss function can be used and based on the label information and training information of any target content, the model loss value of the model to be trained can be calculated. The model loss function can be a mean square error (MSE) loss function, a cross entropy loss function or other functions, without limitation. After obtaining the model loss value, the model parameters of the model to be trained can be optimized in the direction of reducing the model loss value to obtain the target model. Among them, model training is an iterative optimization process, the purpose of which is to find parameters that can optimize the model performance. Based on this, in order to ensure that the model has strong generalization ability and avoid overfitting, specified conditions can be set in advance to determine when to stop training to obtain the final target model required.

[0079] The specified condition may be any of the following:

[0080] (1) The number of iterations of the model to be trained during the training process reaches a specified number of iterations; wherein the specified number of iterations can be preset, and its specific value is not limited. Under this condition, it can be understood that when the model to be trained has learned enough information after a certain number of iterations, the training can be stopped.

[0081] (2) The model parameters of the model to be trained change less than a second specified threshold value in M ​​consecutive iterations; wherein the second specified threshold value can be preset, and its specific value is not limited; the M value can be preset, and its specific value is not limited, such as 2, 3, etc. It is understandable that if the model parameters change very little or almost unchanged after several consecutive iterations, it can be considered that the model to be trained has converged, and further training may not significantly improve the performance of the model to be trained, and the training can be stopped.

[0082] (3) During the training process, the model loss value of the model to be trained is less than a first specified threshold; wherein the first specified threshold can be preset, and its specific value is not limited, such as the first specified threshold can be a smaller value. Under this condition, it can be considered that the model to be trained has reached a good performance level, and the training can be stopped.

[0083] (4) When the target performance index (such as prediction accuracy, etc.) of the model to be trained reaches a specified value; the specified value can be preset, and its specific value is not limited, and the specified value can be as high as possible. Under this condition, it can be understood that the performance of the trained model is good enough, and the training can be stopped.

[0084] In summary, the embodiments of the present application can perceive the user behavior trajectory and grasp the user's interests to help analyze and confirm the direction of subsequent model iterations. After obtaining the target model in the above manner, the target model can be used to process the tasks corresponding to the target business scenario type on the target application to obtain the task processing results for the task. For example, taking the content as notes, the target models trained under various business scenario types can be applied to note recommendations for business scenarios such as recommendation, search, advertising, nearby, and attention.

[0085] In one implementation, a set of content to be processed can be obtained, and each content to be processed in the set of content to be processed can be input into the target model to obtain prediction information of each content to be processed under the target business scenario type. The prediction information of any content to be processed can be used to indicate whether any content to be processed has been subjected to a behavioral operation under the target dimension. For example, assuming that the target dimension is a click dimension, the prediction information can indicate whether a click operation has been performed on any content to be processed. Assuming that the target dimension is a sharing operation, the prediction information can indicate whether a sharing operation has been performed on any content to be processed. In summary, in the note recommendation system, the goal of model training is to predict whether the user is willing to interact with the unknown notes recommended in the corresponding business scenario in the future based on the user's historical interaction information for notes in a certain business scenario. If it is predicted that there is a willingness to interact, the note can be used as a recommended note, thereby improving the recommendation effect.

[0086] The above-mentioned content set to be processed may be all the content in the target application, or may be part of the content in the target application, which is not limited.

[0087] As mentioned above, the above target model may be a model for different business scenario types. The following is an elaboration on the specific application of the target model for various business scenario types.

[0088] In one embodiment, taking the target business scenario type as the content recommendation type and the target dimension including the click dimension as an example, the use of the target model in the target application is specifically explained, that is, the target model can be used in the content recommendation scenario of the target application. Specifically, a set of content to be recommended can be obtained, and each content to be recommended in the set of content to be recommended can be input into the target model to obtain prediction information of each content to be recommended; wherein, the prediction information of any content to be recommended can be used to indicate whether any content to be recommended has been subjected to a behavioral operation under the click dimension. Furthermore, the content to be recommended that is the target prediction information in the set of content to be recommended can be determined as recommended content, and content recommendation can be performed based on the recommended content; wherein, the target prediction information is prediction information indicating that a behavioral operation under the click dimension has been performed on the content to be recommended.

[0089] In one embodiment, the content set to be searched here is understood as the above-mentioned content set to be processed. That is, the content set to be recommended can be all the content in the target application, or part of the content in the target application. The part of the content can be determined based on relevant selection rules.

[0090] For example, the selection rule may be a random selection rule, that is, a portion of the content in the target application may be randomly selected as the content set to be recommended. For another example, the selection rule may be based on feature information of the content in the target application, for example, the feature information may include one or more of the number of clicks, exposure, likes, interest features of the user to be recommended, etc.

[0091] In a possible implementation, taking the feature information as the number of clicks, the specific implementation of determining the content set to be recommended may be: the number of clicks of each content in all the content may be counted, and the content set to be recommended may be determined from all the content based on the number of clicks of each content. For example, the content may be sorted in descending order of the number of clicks to obtain the corresponding sorting result; then, the content in the top N positions in the sorting result may be used as the set of content to be recommended. The N value may be preset, and its specific value is not limited.

[0092] In a possible implementation, taking the characteristic information as the number of clicks and exposures, the specific implementation of determining the content set to be recommended may be: the number of clicks and exposures of each content in all the content may be counted, and the content set to be recommended may be determined from all the content based on the number of clicks and exposures of each content. For example, the target quantity of the corresponding content may be determined based on the number of clicks and exposures of each content. For example, for any content, the sum of the number of clicks and exposures of the content may be used as the target quantity of the content, or the number of clicks and exposures may be weighted respectively, and the sum of the weighted results may be used as the target quantity of the content. After determining the target quantity of each content, the contents may be sorted in descending order according to the target quantity to obtain the corresponding sorting result; then, the contents in the top N positions in the sorting result may be used as the data set to be recommended.

[0093] In a possible implementation, taking the characteristic information as the interest characteristic of the user to be recommended, the specific implementation of determining the set of content to be recommended may be: the interest characteristic of the user to be recommended may be obtained, and the content to be recommended may be determined from all the content based on the interest characteristic. The interest characteristic may include the content type of the content of interest to the user to be recommended, for example, the content type may include one or more of the film and television type, music type, entertainment type, science and education type, etc. In a specific implementation, the content type of each content in all the content may be determined, and the content matching the interest characteristic of the user to be recommended may be screened out from all the content based on the content type of each content, and the matched content may be used as the content to be recommended, so as to obtain the set of content to be recommended. The specific implementation of screening out the content matching the interest characteristic of the user to be recommended from all the content based on the content type of each content may be: the content type of each content may be matched with the content type corresponding to the interest characteristic, and if a content passes the consistency match, the content may be used as the content to be recommended.

[0094] Through the above method, the content that the user is more interested in can be selected from all the content of the target application, and then recommendations can be made based on these contents to ensure that the recommended content is likely to be the content that the user is interested in, thereby improving the accuracy and reliability of the recommendation, and then improving the recommendation effect; in addition, different users usually have different interest characteristics, and using the above method to determine the recommended content can also achieve personalized recommendations, thereby ensuring the recommendation effect.

[0095] In one embodiment, taking the target business scenario type as the content search type and the target dimension including the click dimension as an example, the use of the target model in the target application is specifically explained, that is, the target model can be used in the content search scenario of the target application. Specifically, a set of content to be searched can be obtained, and each content to be searched in the set of content to be searched can be input into the target model to obtain prediction information of each content to be searched; wherein, the prediction information of any content to be searched can be used to indicate whether any content to be searched has been subjected to a behavioral operation under the click dimension. Further, the content to be searched that is the target prediction information in the set of content to be searched can be determined as the search content, and the search content can be displayed on the search page; wherein, the target prediction information is prediction information indicating that the content to be searched has been subjected to a behavioral operation under the click dimension. The set of content to be searched here is understood to be the above-mentioned set of content to be processed.

[0096] In summary, the embodiment of the present application can find the original primary page entry of the historical interaction information generated by the content, and determine the business scenario type corresponding to the content based on the primary page entry, and generate a corresponding attribution field for the content, which can be understood as a scenario label added to the content. Based on this, it can be used to generate accurate labels for the content by business, so that the model in the recommendation system can make secondary accurate recommendations.

[0097] To better understand the data processing method proposed in this application, the following combination Figure 4 The method is further explained. In the embodiment of the present application, the model associated with the target application can be modeled to obtain various models suitable for different business scenarios. That is, the data processing method mentioned in the embodiment of the present application can be understood as a modeling method. Here, the content is a note as an example for relevant elaboration. In a specific implementation, the implementation of the modeling method in the embodiment of the present application may involve the following three parts: summarizing the historical interaction information of users on the target application; saving the key target events in the historical interaction information; determining the attribution of the historical interaction information generated by each note, where the attribution is to determine the business scenario type corresponding to the note, or the first-level page entry of the note.

[0098] In one embodiment, for example, see Figure 4 As shown. The log data information generated by different users on the target application can be read, and then the historical interaction information of all users can be summarized based on the log data information to obtain the historical interaction information of each user. Here, the user ID of each user can be used as an independent key, and all subsequent user behavior trajectories are attributed based on this key. After obtaining the historical interaction information of each user on the target application, the notes can be attributed based on the historical interaction information of each user, that is, the scene label of the note can be determined.

[0099] Based on this, it can be known that the modeling process enters the attribution operator, that is, determines the scene label of the note. For a user, first, the historical interaction information of the user can be queried from the target area. If the historical interaction information of the user exists in the target area, the historical interaction information is retrieved. If there is no historical interaction information of the user, the historical interaction information is initialized. Among them, a corresponding storage duration (such as the first storage duration mentioned above) is set for the historical interaction information stored in the target area, that is, a data expiration time needs to be set for the historical interaction information.

[0100] Among them, the historical interaction information includes target events of various event types, such as exposure events, click events, etc. Since users can continuously generate target events on the target application, and the target events need to be stored in the target area, the target events in the target area are continuously updated.

[0101] For example, if the target event is an exposure event, if the latest exposure event needs to be saved, it is necessary to first query whether the exposure event of the same note is stored in the target area. If there is no exposure event of the note, save the exposure event. If the exposure event is stored, in order to avoid disorder, it can be determined whether the event time of the current exposure event of the note is earlier than the event time of the exposure event stored in the target area. If it is earlier, the latest exposure event is stored. If it is not earlier, it is considered that the event is disordered, and the latest exposure event can be filtered.

[0102] For another example, if the target event is a click event, if you need to save the latest click event, you need to first check whether the click event of the same note is stored in the target area. If there is no click event of the note, save the click event. If the click event is stored, in order to avoid disorder, you can determine whether the event time of the current click event of the note is earlier than the event time of the click event stored in the target area. If it is earlier, store the latest click event. If it is not earlier, it is considered that the event is disordered, and the latest click event can be filtered.

[0103] After the necessary target events are saved in the target area, the notes can be attributed based on the target events. In one embodiment, a user behavior trajectory can be constructed based on the user's target events for the notes, so as to attribute the notes based on the target events in the user behavior trajectory. For example, it can be determined whether the target page generated by the first target event in the user behavior trajectory is a first-level page of the target application; if so, the scenario label corresponding to the business scenario type corresponding to the first-level page can be used as the scenario label of the note; if not, the scenario label of the note is determined based on the logical rules of the page that generated the first target event.

[0104] It is understandable that as time goes by, the historical interaction information in the target area may reach the expiration time. Then it can be determined whether there is any historical interaction information in the preset historical time period. If not, the historical status information is cleared. If so, the storage time of the historical status information is extended.

[0105] Through the above method, corresponding scenario labels can be added to the notes on the target application. The scenario labels can be understood as the attribution information of the notes. When the label data of these notes is needed in the future, the data of the same user, the same notes, and the same attribution information can be aggregated to ultimately generate accurate model training data. For example, when it is necessary to train the model to be trained under the target business scenario type, the scenario labels of the notes can be used to determine the notes whose business scenario type is the target business scenario type, and these notes can be used as training data for training the model to be trained.

[0106] In an embodiment of the present application, the user's historical interaction information on the content of the target application can be monitored to aggregate and attribute notes of different business scenarios based on the historical interaction information, identify which type of business scenario the note belongs to, and also determine which type of business scenario the historical interaction information belongs to, so as to improve the accuracy of the content label data. Subsequently, different models can be trained based on the content in different business scenarios. Due to the accuracy of the content label data, the model can accurately consume the historical interaction information of each business scenario during training, thereby improving the accuracy and reliability of model training, and then improving the accuracy of model prediction, and also improving the user experience.

[0107] See also Figure 5 , Figure 5 : is a schematic diagram of the structure of a data processing device provided in an embodiment of the present application. The data processing device described in this embodiment includes:

[0108] A determination unit 501 is configured to obtain historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on the contents in the content collection based on the historical interaction information;

[0109] The determining unit 501 is further configured to determine, based on a user behavior track of the target user with respect to the content in the content set, a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application;

[0110] The determining unit 501 is further configured to obtain a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario labels of the content in the content set;

[0111] The training unit 502 is used to train the model to be trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task.

[0112] In one implementation, the user behavior trajectory of the target user on any content is obtained by processing the target events according to the time sequence of the target events executed by the target user on the any content; the determining unit 501 is specifically used to:

[0113] According to the user behavior track of the target user for any content, determining the target page generated by each target event in the user behavior track;

[0114] Based on the business scenario type corresponding to the target page, it is determined that any one of the contents generates a scenario tag corresponding to the user behavior trajectory on the target application.

[0115] In one implementation, the determining unit 501 is specifically configured to:

[0116] If the target page generated by the first target event in the user behavior track of the target user for any content is a first-level page of the target application, the scenario label corresponding to the business scenario type corresponding to the first-level page is used as the scenario label for the corresponding historical interaction information generated by any content on the target application;

[0117] If the page generated by the first target event in the user behavior trajectory of the target user for any content is not a first-level page of the target application, the scene label of the corresponding historical interaction information generated by any content on the target application is determined based on the logical rules of the page that generates the first target event.

[0118] In one implementation, the training unit 502 is specifically configured to:

[0119] For any target content in the target content set, data of the target content in a target dimension is determined based on historical interaction information of the target content, and tag information of the target content is determined based on the data in the target dimension; the target dimension includes one or more of a click dimension, a share dimension, a favorite dimension, and a like dimension; the tag information is used to indicate whether a behavior operation in the target dimension is performed on the target content;

[0120] Inputting any target content into the model to be trained to obtain training information of any target content;

[0121] The model to be trained is trained based on the label information of any target content and the training information to obtain a target model.

[0122] In one implementation, the determining unit 501 is further configured to:

[0123] Acquire a set of content to be processed, input each content to be processed in the set of content to be processed into the target model, obtain prediction information of each content to be processed under the target business scenario type, and the prediction information of any content to be processed is used to indicate whether any behavior operation under the target dimension is executed on any content to be processed;

[0124] The to-be-processed contents that are concentrated as target prediction information are determined as recommended contents, and content recommendation is performed based on the recommended contents; the target prediction information is prediction information indicating that a behavioral operation under a target dimension is performed on the to-be-processed contents.

[0125] In one implementation, the historical interaction information is stored in the target area; and the determining unit 501 is further configured to:

[0126] If the storage time of the historical interaction information of the target user reaches a first storage time length, determining whether the target user has historical interaction information within a preset historical time period;

[0127] If not, clear the historical interaction information of the target user;

[0128] If so, the first storage duration is extended to a second storage duration; the second storage duration is longer than the first storage duration.

[0129] In one implementation, the historical interaction information generated by the target user on any content includes a target event performed by the target user on the any content; the determining unit 501 is further configured to:

[0130] If there is a current event to be saved for any of the contents, query whether there is a target event executed for any of the contents in the target area;

[0131] If so, obtaining a first execution time of the target event in the target area, and obtaining a second execution time of the current event;

[0132] If the second execution time is later than the first execution time, storing the current event in the target area;

[0133] If the second execution time is earlier than the first execution time, the current event is filtered.

[0134] It is understood that the division of units in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. The functional units in the embodiments of the present application may be integrated into a processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware or in the form of software functional units.

[0135] See also Figure 6 , Figure 6 6 is a schematic diagram of a computer device provided in an embodiment of the present application. The computer device includes: a processor 601 and a memory 602. Optionally, the computer device may also include a network interface 603. The processor 601, the memory 602 and the network interface 603 may exchange data.

[0136] The processor 601 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0137] The memory 602 may include a read-only memory and a random access memory, and provide program instructions and data to the processor 601. A portion of the memory 602 may also include a non-volatile random access memory. When the processor 601 calls the program instructions, it is used to execute:

[0138] Acquire historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on contents in the content collection based on the historical interaction information;

[0139] Based on the user behavior track of the target user with respect to the content in the content set, determining a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application;

[0140] Acquire a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario label of the content in the content set;

[0141] The model to be trained is trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task.

[0142] In one implementation, the user behavior trajectory of the target user on any content is obtained by processing the target events according to the time sequence of the target events executed by the target user on the any content; the processor 601 is specifically configured to:

[0143] According to the user behavior track of the target user for any content, determining the target page generated by each target event in the user behavior track;

[0144] Based on the business scenario type corresponding to the target page, it is determined that any one of the contents generates a scenario tag corresponding to the user behavior trajectory on the target application.

[0145] In one implementation, the processor 601 is specifically configured to:

[0146] If the target page generated by the first target event in the user behavior track of the target user for any content is a first-level page of the target application, the scenario label corresponding to the business scenario type corresponding to the first-level page is used as the scenario label for the corresponding historical interaction information generated by any content on the target application;

[0147] If the page generated by the first target event in the user behavior trajectory of the target user for any content is not a first-level page of the target application, the scene label of the corresponding historical interaction information generated by any content on the target application is determined based on the logical rules of the page that generates the first target event.

[0148] In one implementation, the processor 601 is specifically configured to:

[0149] For any target content in the target content set, data of the target content in a target dimension is determined based on historical interaction information of the target content, and tag information of the target content is determined based on the data in the target dimension; the target dimension includes one or more of a click dimension, a share dimension, a favorite dimension, and a like dimension; the tag information is used to indicate whether a behavior operation in the target dimension is performed on the target content;

[0150] Inputting any target content into the model to be trained to obtain training information of any target content;

[0151] The model to be trained is trained based on the label information of any target content and the training information to obtain a target model.

[0152] In one implementation, the processor 601 is further configured to:

[0153] Acquire a set of content to be processed, input each content to be processed in the set of content to be processed into the target model, obtain prediction information of each content to be processed under the target business scenario type, and the prediction information of any content to be processed is used to indicate whether any behavior operation under the target dimension is executed on any content to be processed;

[0154] The to-be-processed contents that are concentrated as target prediction information are determined as recommended contents, and content recommendation is performed based on the recommended contents; the target prediction information is prediction information indicating that a behavioral operation under a target dimension is performed on the to-be-processed contents.

[0155] In one implementation, the historical interaction information is stored in a target area; and the processor 601 is further configured to:

[0156] If the storage time of the historical interaction information of the target user reaches a first storage time length, determining whether the target user has historical interaction information within a preset historical time period;

[0157] If not, clear the historical interaction information of the target user;

[0158] If so, the first storage duration is extended to a second storage duration; the second storage duration is longer than the first storage duration.

[0159] In one implementation, the historical interaction information generated by the target user on any content includes a target event performed by the target user on the any content; the processor 601 is further configured to:

[0160] If there is a current event to be saved for any of the contents, query whether there is a target event executed for any of the contents in the target area;

[0161] If so, obtaining a first execution time of the target event in the target area, and obtaining a second execution time of the current event;

[0162] If the second execution time is later than the first execution time, storing the current event in the target area;

[0163] If the second execution time is earlier than the first execution time, the current event is filtered.

[0164] The present application also provides a computer storage medium in which program instructions are stored. When the program is executed, it may include: Figure 2 Part or all of the steps of the data processing method in the corresponding embodiment.

[0165] It should be noted that, for the above-mentioned various method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by this application.

[0166] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0167] The embodiment of the present application also provides a computer program product or computer program, which includes program instructions, and when the program instructions are executed by a processor, some or all of the steps in the above method can be implemented. For example, the program instructions are stored in a computer-readable storage medium. The processor of the computer device reads the program instructions from the computer-readable storage medium, and the processor executes the program instructions, so that the computer device executes the steps performed in the embodiments of the above methods.

[0168] The above is a detailed introduction to a data processing method, device, computer equipment and medium provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on contents in the content collection based on the historical interaction information; Based on the user behavior track of the target user with respect to the content in the content set, determining a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application; Acquire a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario label of the content in the content set; The model to be trained is trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task.

2. The method according to claim 1, characterized in that The user behavior track of the target user on any content is obtained by processing the target event in the time sequence of the target event executed by the target user on the any content; the determining, based on the user behavior track of the target user on the content in the content collection, the scene label of the corresponding historical interaction information generated by the content in the content collection on the target application includes: According to the user behavior track of the target user for any content, determining the target page generated by each target event in the user behavior track; Based on the business scenario type corresponding to the target page, it is determined that any one of the contents generates a scenario tag corresponding to the user behavior trajectory on the target application.

3. The method according to claim 2, characterized in that The determining, based on the business scenario type corresponding to the target page, that any content generates a scenario tag corresponding to the historical interaction information on the target application includes: If the target page generated by the first target event in the user behavior track of the target user for any content is a first-level page of the target application, the scenario label corresponding to the business scenario type corresponding to the first-level page is used as the scenario label for the corresponding historical interaction information generated by any content on the target application; If the page generated by the first target event in the user behavior trajectory of the target user for any content is not a first-level page of the target application, the scene label of the corresponding historical interaction information generated by any content on the target application is determined based on the logical rules of the page that generates the first target event.

4. The method according to claim 1, characterized in that: The step of training the model to be trained based on the target content set to obtain a target model includes: For any target content in the target content set, data of the target content in a target dimension is determined based on historical interaction information of the target content, and tag information of the target content is determined based on the data in the target dimension; the target dimension includes one or more of a click dimension, a share dimension, a favorite dimension, and a like dimension; the tag information is used to indicate whether a behavior operation in the target dimension is performed on the target content; Inputting any target content into the model to be trained to obtain training information of any target content; The model to be trained is trained based on the label information of any target content and the training information to obtain a target model.

5. The method according to claim 4, characterized in that Also includes: Acquire a set of content to be processed, input each content to be processed in the set of content to be processed into the target model, obtain prediction information of each content to be processed under the target business scenario type, and the prediction information of any content to be processed is used to indicate whether any behavior operation under the target dimension is executed on any content to be processed; The to-be-processed contents that are concentrated as target prediction information are determined as recommended contents, and content recommendation is performed based on the recommended contents; the target prediction information is prediction information indicating that a behavioral operation under a target dimension is performed on the to-be-processed contents.

6. The method according to claim 1, characterized in that The historical interaction information is stored in the target area; and further comprising: If the storage time of the historical interaction information of the target user reaches a first storage time length, determining whether the target user has historical interaction information within a preset historical time period; If not, clear the historical interaction information of the target user; If so, the first storage duration is extended to a second storage duration; the second storage duration is longer than the first storage duration.

7. The method according to claim 6, characterized in that The historical interaction information generated by the target user on any content includes a target event performed by the target user on the any content; and further includes: If there is a current event to be saved for any of the contents, query whether there is a target event executed for any of the contents in the target area; If so, obtaining the first execution time of the target event in the target area, and obtaining the second execution time of the current event; If the second execution time is later than the first execution time, storing the current event in the target area; If the second execution time is earlier than the first execution time, the current event is filtered.

8. A data processing device, characterized in that: include: a determination unit, configured to obtain historical interaction information generated by a target user on contents in a content collection on a target application, and construct a user behavior track of the target user on contents in the content collection based on the historical interaction information; The determining unit is further used to determine, based on the user behavior track of the target user with respect to the content in the content set, a scenario tag of the content in the content set generating corresponding historical interaction information on the target application, wherein the scenario tag is used to indicate a business scenario type on the target application; The determining unit is further configured to obtain a target business scenario type of the to-be-trained model associated with the target application, and determine a target content set that matches the target business scenario type from the content set based on the scenario labels of the content in the content set; A training unit is used to train the model to be trained based on the target content set to obtain a target model; the target model is used to process the task corresponding to the target business scenario type on the target application to obtain a task processing result for the task.

9. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 7.

Images