Resource pushing method, model training method and device

Abstract

The invention provides a resource pushing method and device and a model training method and device, and relates to the technical field of large models and artificial intelligence, in particular to the technical field of machine learning and information filtering processing. The specific implementation scheme of the resource pushing method comprises the following steps: in response to a received switching operation of a target object for an initial page, obtaining historical interaction information of the target object for historical pushing resources, a current preference resource, attribute information of the target object and guide information for guiding the target object to browse the initial page; based on the attribute information, the historical interaction information, the current preference resource and the guide information, identifying to obtain the intention of the target object; and in response to the situation that the intention is associated browsing, pushing a first target resource associated with the guide resource through the switched target page.

Description

Technical Field

[0001] The present disclosure relates to the fields of big models and artificial intelligence technology, in particular to the fields of machine learning and information filtering and processing technology, and specifically to a resource push method, a model training method and a device. Background Art

[0002] With the development of big data technology, application platforms will use specific algorithms and strategies to select content that meets user interests and needs from massive amounts of content and display it to users.

[0003] Since the content that users are interested in may be complex and changeable, the degree to which the content pushed by the application platform conforms to the user's interests and needs directly affects the user's experience. Summary of the invention

[0004] The present invention provides a resource pushing method, a model training method and a device.

[0005] According to one aspect of the present disclosure, a resource push method is provided, comprising: in response to receiving a switch operation of a target object for an initial page, obtaining historical interaction information of the target object for historically pushed resources, current preferred resources, attribute information of the target object, and guidance information for guiding browsing of the initial page; based on the attribute information, historical interaction information, current preferred resources, and guidance information, identifying the intention of the target object; and in response to determining that the intention is associated browsing, pushing a first target resource associated with the guidance resource through the switched target page. .

[0006] According to another aspect of the present disclosure, a model training method is provided, which includes: using an intent recognition module included in an initial model to perform intent recognition based on attributes of a sample object, sample historical preference features, content features of sample preference resources, and sample guidance information, and outputting sample intent; wherein the sample historical preference features are obtained based on sample historical interaction information; and the content features are obtained based on sample preference resource extraction; using a resource optimization module included in the initial model to perform resource optimization based on attributes of a sample object, sample historical preference features, content features of sample preference resources, sample guidance information, and content features of sample candidate resources, and outputting sample push resources; based on a first target loss function, calculating a first loss value according to the sample intent, sample push resources, sample intent label, and sample push resource label; and adjusting model parameters of the initial model based on the first loss value to obtain a trained model.

[0007] According to another aspect of the present disclosure, a resource pushing device is provided, including: a first acquisition module, an identification module and a first pushing module.

[0008] The first acquisition module is used to obtain the historical interaction information of the target object for the historical pushed resources, the current preferred resources, the attribute information of the target object and the guidance information for guiding the browsing of the initial page in response to receiving the switching operation of the target object for the initial page.

[0009] The recognition module is used to identify the intention of the target object based on attribute information, historical interaction information, current preferred resources and guidance information.

[0010] The first push module is used for pushing a first target resource associated with the guiding resource through a switched target page in response to determining that the intention is associated browsing.

[0011] According to another aspect of the present disclosure, a model training device is provided, including: an intention recognition module, a resource optimization module, a first loss calculation module and a first adjustment module.

[0012] The initial model includes an intent recognition module, which is used to identify intent based on the attributes of the sample object, the sample historical preference features, the content features of the sample preference resources and the sample guidance information, and output the sample intent; wherein the sample historical preference features are obtained based on the sample historical interaction information; the content features are extracted based on the sample preference resources.

[0013] The initial model includes a resource optimization module, which is used to optimize resources based on the attributes of sample objects, sample historical preference characteristics, content characteristics of sample preference resources, sample guidance information and content characteristics of sample candidate resources, and output sample push resources.

[0014] The first loss calculation module is used to calculate a first loss value based on a first target loss function, according to sample intent, sample push resources, sample intent labels and sample push resource labels.

[0015] The first adjustment module is used to adjust the model parameters of the initial model based on the first loss value to obtain a trained model.

[0016] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described above.

[0017] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute the method described above.

[0018] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method described above is implemented.

[0019] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0021] Figure 1 An exemplary system architecture to which the resource push method and device according to an embodiment of the present disclosure can be applied is schematically shown;

[0022] Figure 2 A flowchart of a resource push method according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 3A A schematic diagram of a page for which the user intends to browse in association according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 3B A schematic diagram of a page in which the user intends to browse non-associated content according to an embodiment of the present disclosure is schematically shown;

[0025] Figure 4 A schematic diagram of intent recognition according to an embodiment of the present disclosure is schematically shown;

[0026] Figure 5 A schematic diagram of performing resource optimization when it is determined that the user intention is associated browsing according to an embodiment of the present disclosure is shown;

[0027] Figure 6 A schematic diagram of performing resource optimization when it is determined that the user intention is associated browsing according to another embodiment of the present disclosure is shown;

[0028] Figure 7 A flowchart of a model training method according to an embodiment of the present disclosure is schematically shown;

[0029] Figure 8 A schematic diagram schematically shows a model training method according to an embodiment of the present disclosure;

[0030] Fig. 9 A schematic diagram schematically shows a model training method according to another embodiment of the present disclosure;

[0031] Fig.10 A block diagram of a resource push device according to an embodiment of the present disclosure is schematically shown;

[0032] Fig.11 A block diagram schematically shows a model training device according to an embodiment of the present disclosure; and

[0033] Fig.12 A block diagram of an electronic device suitable for implementing a resource pushing method or a model training method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0034] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0035] When users use terminal devices such as mobile phones and tablets to browse content pushed by application platforms, the application platform will build a user profile based on the user's historical interaction operations on the platform, such as likes, comments, and shares. The user profile describes the user's long-term interest preferences or interaction habits, so as to push content to the user based on the user profile.

[0036] However, since the content that users are interested in is not static, this method of pushing resources based on users' long-term interactive operations may result in a low match between the pushed content and the user's actual browsing intention. For example, a user quickly scrolls through the pushed content page and does not watch the pushed content completely.

[0037] In addition, the application platform can also identify the user's short-term interests based on the user's real-time interactive operations on the platform, such as clicking or staying to watch specific content, and push content to the user based on the user's short-term interests.

[0038] However, since users' short-term interests are determined based on specific content, and the types of specific content are relatively simple, the content pushed based on short-term interests is also relatively simple, which can easily cause visual fatigue and also lead to a poor match between the pushed content and the user's actual browsing intentions.

[0039] In view of this, the embodiment of the present disclosure combines the historical interactions of the target object with respect to historical push resources, the current preference resources, the attributes of the target object and the guidance information, comprehensively analyzes the user's long-term and short-term preferences, and the relationship between the guidance information and the current preference resources, accurately identifies whether the user has the intention of related browsing, and dynamically adjusts the push strategy based on the identified browsing intention, further improving the matching degree between the push resources and the user's actual browsing intention, further improving the accuracy of high push resources, and thus improving the user experience.

[0040] Figure 1 An exemplary system architecture to which the resource push method and apparatus according to an embodiment of the present disclosure can be applied is schematically shown.

[0041] It should be noted that Figure 1 The examples shown are only examples of system architectures to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. For example, in another embodiment, the exemplary system architecture to which the resource push method and apparatus can be applied may include a terminal device, but the terminal device may implement the resource push method and apparatus provided by the embodiments of the present disclosure without interacting with the server.

[0042] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104 and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0043] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients and / or social platform software, etc. (only as examples).

[0044] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0045] The server 105 may be a server that provides various services, such as a background management server (only an example) that provides support for the content browsed by the user using the terminal devices 101, 102, and 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to the user request) to the terminal device.

[0046] It should be noted that the resource push method provided in the embodiment of the present disclosure can generally be executed by the terminal device 101, 102, or 103. Accordingly, the resource push apparatus provided in the embodiment of the present disclosure can also be set in the terminal device 101, 102, or 103.

[0047] Alternatively, the resource push method provided in the embodiment of the present disclosure may also be generally executed by the server 105. Accordingly, the resource push device provided in the embodiment of the present disclosure may generally be arranged in the server 105. The resource push method provided in the embodiment of the present disclosure may also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the resource push device provided in the embodiment of the present disclosure may also be arranged in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0048] For example, when a user browses a page, the terminal devices 101, 102, 103 may obtain the user's historical interactions with historically pushed resources, current preferred resources, user attributes, and guidance information, and then the terminal devices 101, 102, 103 may send the obtained information to the server 105. The server 105 identifies the user's intention by executing the method of the embodiment of the present disclosure, and feeds back the target resource matching the user's intention to the terminal devices 101, 102, 103. The target resource is displayed on the terminal devices 101, 102, 103 through the switched page.

[0049] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to the implementation requirements.

[0050] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, disclosure and application of user personal information involved comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0051] In the technical solution of the present disclosure, the user's authorization or consent is obtained before obtaining or collecting the user's personal information.

[0052] In the technical solution of the present disclosure, the images used in the examples are all from public image data sources.

[0053] Figure 2 The flowchart of the resource pushing method according to the embodiment of the present disclosure is schematically shown.

[0054] like Figure 2 As shown, the method 200 includes operations S210 to S230.

[0055] In operation S210, in response to receiving a switch operation of the target object for the initial page, historical interaction information of the target object for historical pushed resources, current preferred resources, attribute information of the target object, and guidance information for guiding browsing of the initial page are obtained.

[0056] In operation S220, the intention of the target object is identified based on the attribute information, the historical interaction information, the current preferred resources, and the guidance information.

[0057] In operation S230, in response to determining that the intention is associated browsing, a first target resource associated with the guiding resource is pushed through the switched target page.

[0058] In this embodiment, the acquisition and use of the user's interactive information are known and agreed to by the users, comply with the provisions of relevant laws and regulations, and do not violate public order and good customs.

[0059] According to an embodiment of the present disclosure, the guidance information may be content displayed on the guidance page of the application, and the guidance page may be the homepage of the application. The guidance information may represent a summary of the content displayed on the initial page. The guidance information includes, but is not limited to, text, images, videos, etc. The summary describes the resource type, channel representation, resource duration, resource title length, etc. of the content displayed on the initial page. After the user confirms the guidance information by clicking, selecting, etc., the guidance page jumps to the push page in the immersive scene, and the content corresponding to the guidance information is displayed on the push page.

[0060] For example, the initial page can be a push page in an immersive scene, and the content displayed on the push page can be a video clip of a kitten walking on a rock wall. The guidance information can include "Title: Kitten Walking on Rock Walls", "30s Video Clip", "Channel 1", etc.

[0061] Figure 3A A schematic diagram of a page for associated browsing intended by a user according to an embodiment of the present disclosure is schematically shown.

[0062] like Figure 3A As shown, the guide page 301 may include, from top to bottom: a search box 3011 , a channel bar 3012 , and a guide resource 3013 .

[0063] When the user clicks on the area for displaying the guide resource 3013 in the guide page, the user is redirected to the push page 302, and the content corresponding to the guide resource 3013 is displayed in the push page 302, for example, a video clip of an athlete riding. In the push page 3013, the user can interact with the guide resource by clicking on virtual interactive buttons, such as the favorite button 3021, the comment button 3022, and the like button 3033, so as to dynamically identify the user's browsing intention based on the user's interaction.

[0064] According to an embodiment of the present disclosure, the switching operation for the initial page can be as follows: Figure 3A The sliding operation shown may also be configured with other switching operations according to actual scene requirements, such as gesture switching, somatosensory switching, etc. The embodiment of the present disclosure does not specifically limit the implementation method of the switching operation.

[0065] According to an embodiment of the present disclosure, historical interaction information includes but is not limited to collection, like, comment, share, stop watching, quick switch, etc.

[0066] According to an embodiment of the present disclosure, the current preferred resource may include at least one push resource and at least one guide resource. At least one guide resource is a resource determined by the target object in the guide page during the current period. For example: a resource clicked by the target object in the guide page. At least one push resource is a resource on which the target object has performed a predetermined interactive operation in a push page associated with the guide page.

[0067] For example, the current preferred resources may include the guidance resources clicked by the user within a predetermined time period before the switching operation is performed and the pushed resources that meet the user's browsing intention and are guided by the guidance resources to be browsed. For example, the predetermined time period may be 5 minutes.

[0068] For example, the current preferred resources may also include a predetermined number of guide resources for clicking before performing the switching operation and a predetermined number of push resources that meet the user's browsing intention and are guided by the guide resources to browse. For example, the predetermined number may be 10.

[0069] In some embodiments, when a user collects, likes, shares, or stays to watch a certain push resource for a period exceeding a predetermined viewing threshold, it indicates that the content of the push resource meets the user's browsing intention. When a user quickly switches to a certain push resource, it indicates that the content of the push resource does not meet the user's browsing intention. Therefore, the predetermined interactive operations may include but are not limited to collecting, liking, sharing, or staying to watch for a period exceeding a predetermined viewing threshold.

[0070] According to the embodiments of the present disclosure, the attribute information of the target object can be derived from the user's registration information. Before obtaining the user's attribute information, an authorization request can be sent to the user, and the user's attribute information can be obtained only after obtaining the user's consent. The attribute information includes but is not limited to the user's ID, city, occupation, educational background, etc.

[0071] In some embodiments, a trained intent recognition model can be used to identify the intent of the target object based on attribute information, historical interaction information, current preference resources, and guidance information.

[0072] Since the embodiments of the present disclosure refer to current preference resources and guidance information, in order to improve data processing efficiency and meet the timeliness requirements of real-time analysis and real-time push, the intent recognition model can be a simple neural network model, such as: NN (Neural Network) network.

[0073] According to an embodiment of the present disclosure, the intention of the target object may represent whether the user wants to browse the content associated with the initial page content in the switched page.

[0074] like Figure 3A As shown, when the target object's intention is identified as associated browsing, it means that the user wants to browse the content associated with the initial page content in the switched page. The content displayed in the push page 302 is a picture of an athlete riding.

[0075] The displayed content belongs to the sports competition type. In addition to cycling, the push resources of the sports competition type can also include push resources of various sports such as football, volleyball, shooting, marathon, etc.

[0076] Among the push resources of many projects, comprehensive correlation analysis can be performed based on current preferred resources, historical interactions, and guidance information.

[0077] For example, in historical interactions, the user frequently made positive comments on multiple competition videos of a certain swimmer in the past three months. Before switching to the current page, 8 of the 10 push resources collected by the user were about swimming. The channel to which this guide information belongs is the sports event channel.

[0078] By comprehensively analyzing the user's instant feedback on the pushed resources, it can be determined that the user is more likely to be interested in the pushed resources of the swimming event in the current period. Therefore, when the user switches pages, the picture of the athlete swimming can be displayed in the switched page 303A.

[0079] The disclosed embodiment combines the target object's historical interactions with historical push resources, current preference resources, the target object's attributes and guidance information, comprehensively analyzes the user's long-term and short-term preferences, and the relationship between the guidance information and the current preference resources, accurately identifies whether the user has the intention of associated browsing, and dynamically adjusts the push strategy based on the identified browsing intention, further improving the matching degree between the push resources and the user's actual browsing intention, further improving the accuracy of the push resources, and thus improving the user experience.

[0080] According to an embodiment of the present disclosure, the above method further includes the following operation: in response to determining that the intention is non-associated browsing, pushing the second target resource through the target page.

[0081] According to an embodiment of the present disclosure, the second target resource is determined based on the interactive operation of the target object within a historical period.

[0082] For example, you can build a user profile that describes the user's long-term interests, preferences, and operating habits based on the target object's interactive operations in the historical period, such as viewing time, likes, comments, shares, favorites, etc. Then, based on the user profile, push personalized resources to the target object that are unrelated to the content of the initial page but are in line with the user's long-term interests, preferences, and operating habits.

[0083] Figure 3B A schematic diagram of a page for non-associated browsing intended by a user according to an embodiment of the present disclosure is schematically shown.

[0084] like Figure 3B As shown, the difference between the embodiment 300B and the embodiment 300A is that the identified target object's intention is non-associated browsing, which means that the user wants to browse content unrelated to the initial page content on the switched page.

[0085] For example, the push page 302 displays the cycling pictures of athletes corresponding to the guidance resources. When it is recognized that the user's intention is non-related browsing, based on the user's long-term interest preferences and operating habits, for example, the user often watches images of landscapes, so the user's profile can be marked as liking landscapes. The landscape images can be displayed in the switched page 303B.

[0086] According to the embodiments of the present disclosure, by accurately identifying the user's instant browsing intention and dynamically adjusting the push strategy, the user's instantly changing browsing needs can be dynamically met, further improving the user experience.

[0087] Reference below Figure 4~Figure 6 , combined with specific embodiments Figure 2 The method shown is further explained.

[0088] According to an embodiment of the present disclosure, identifying the intention of a target object based on attribute information, historical interaction information, current preference resources and guidance information may include the following operations: obtaining historical preference features for historical push resources based on historical interaction information; extracting content features of current preference resources; and performing intent identification based on attribute information, historical preference features, content features and guidance information to obtain intent.

[0089] Figure 4 The schematic diagram of intention recognition according to an embodiment of the present disclosure is schematically shown.

[0090] like Figure 4 As shown, in embodiment 400, firstly, statistical calculation is performed based on historical interaction information 401 to obtain historical preference features for historical push resources.

[0091] In some embodiments, historical preference features can be obtained based on the number of historical push resources in the past three days and the historical interaction information of the target objects for the historical push resources in the past three days. The historical preference features include but are not limited to: fast scrolling rate, completion rate, positive comment rate, like rate, collection rate, sharing rate, etc.

[0092] For example, the number of historical resources pushed to users in the past three days is 200. According to statistics, among these historical resources, the number of resources liked by users is 50, the number of resources quickly switched is 80, the number of resources watched in full is 10, the number of resources collected is 20, the number of resources with positive comments is 10, and the number of resources shared with other users is 30. Therefore, the fast scrolling rate is 40%, the completion rate is 5%, the positive comment rate is 5%, the like rate is 25%, the collection rate is 10%, and the sharing rate is 15%.

[0093] Then, the current preferred resource 411 may be input into a content-based collaborative filtering model (CB2CF) to output a content feature 412. The content feature 412 may be used to characterize the similarity and relevance of the user's current preferred resources at the content level.

[0094] Next, the historical preference feature 402 , the content feature 412 , the object attribute 421 , and the guidance information 422 may be concatenated and input into the intention recognition model 420 to obtain the output intention 423 .

[0095] In some embodiments, the object attributes 421 and the guidance information 422 may be vectorized respectively and then concatenated with the historical preference features 402 and the content features 412 , and then input into the intent recognition model 420 .

[0096] According to the embodiments of the present disclosure, historical interaction information and attribute information can reflect the user's long-term interest preferences and operating habits, current preferred resources can reflect the user's short-term interest preferences in the current period, and guidance information can reflect the resource type preferred by the user. Based on the correlation analysis between the resource type to which the user's currently determined guidance information belongs and the long-term / short-term interest preferences, intent recognition is performed, further improving the accuracy of intent recognition.

[0097] In some embodiments, during the current time period, the resources determined by the user in the guidance page can reflect the user's immediate intentions. However, since the number of guidance resources used for determination is limited, when resources are pushed based only on guidance resources, a large number of pushed resources may be repeated, causing visual fatigue to the user.

[0098] Therefore, in the embodiment of the present disclosure, the current preferred resource includes not only at least one push resource, but also at least one guide resource. The at least one guide resource is a resource determined by the target object in the guide page during the current period. The at least one push resource is a resource on which the target object has performed a predetermined interactive operation in the push page associated with the guide page.

[0099] According to an embodiment of the present disclosure, extracting content features of the current preferred resource may include the following operations: extracting the first content feature of the push resource and the second content feature of the guide resource respectively; and concatenating the first content feature with the second content feature to obtain the content feature of the current preferred resource.

[0100] The guided resource is clicked by the user directly on the guided page, and it can be determined that it meets the user's browsing intention. The pushed resource is actively pushed to the user by the application, so the user's browsing intention can be determined based on the user's feedback information on the pushed resource. For example: when the user likes, collects, or watches a pushed resource for a long time, it can be said that the pushed resource meets the user's browsing preference.

[0101] For example, currently preferred resources may include nearly 10 guide resources clicked by the user on the home page (guide page), the 10 guide resources that push the guide page to jump to the push page of the immersive scene, and 10 push resources that the user likes, collects, shares, comments positively, or watches for more than a predetermined time.

[0102] In some embodiments, the 10 guide resources may be input into the CB2CF model to output a first content feature. Then, the 10 push resources may be input into the CB2CF model to output a second content feature. Finally, the first content feature and the second content feature may be concatenated to obtain the content feature of the current preferred resource.

[0103] It should be noted that the number of current preference resources can be configured according to the requirements of actual application scenarios, and the embodiment of the present disclosure does not specifically limit the number of current preference resources.

[0104] According to an embodiment of the present disclosure, the guidance resources are combined with the push resources on which the target object has performed a predetermined interactive operation in a push page associated with the guidance page, so as to expand the number of resources that the user is interested in, so that the intent recognition model can deeply understand the user's real browsing intention in the current period, and further improve the accuracy of intent recognition.

[0105] According to an embodiment of the present disclosure, the above method may further include the following operations: determining a first target resource from a plurality of predetermined candidate resources based on attribute information, historical interaction information, current preferred resources and guidance information; and generating a target page based on the first target resource.

[0106] In some embodiments, the resource type preferred by the user may be determined based on the attribute information, historical interaction information, current preferred resources and guidance information. Then, a candidate resource corresponding to the resource type preferred by the user among a plurality of predetermined candidate resources is determined as the first target resource.

[0107] In some embodiments, based on historical interaction information, resource types that meet the user's long-term interest preferences can be determined from historical push resources. Based on current preferred resources, resource types that meet the user's short-term interest preferences can be determined. Then, the intersection of the resource types of the user's long-term interest preferences and the resource types of the short-term interest preferences can be taken to determine the resource types preferred by the user.

[0108] For example, among 100 historical push resources, the resource types of 40 resources liked by the user and 10 resources watched in full include sports game type and travel type. Therefore, it can be determined that the resource types that the user prefers for a long time are sports game type and travel type. Among the nearly 10 guided resources clicked by the user and the pushed resources collected by the user, 8 resources are sports game type. Therefore, it can be determined that the resource type that the user prefers is sports game type.

[0109] Then, a resource of the sports game type is determined from a plurality of predetermined candidate resources and pushed to the user.

[0110] In some embodiments, the first target resource may be processed according to the page template of the initial page to generate a switched page. Maintaining the same page template before and after the page switch facilitates user interaction with the pushed resource and reduces the probability of affecting the user interaction experience due to the update of the page layout.

[0111] In some embodiments, when there is no intersection between the resource type of the user's long-term interest preference and the resource type of the user's short-term interest preference, the type associated with the resource type indicated by the guidance information may be determined as the resource type preferred by the user.

[0112] For example, when the resource types that the user prefers in the long term are sports games and travel, and the resource types that the user prefers in the short term are music and fashion trends, at this time, the resource type indicated by the guidance information that guides the user to browse the initial page is clothing matching type, and the fashion trend type associated with the clothing matching type can be determined as the resource type preferred by the user.

[0113] According to the embodiments of the present disclosure, when it is determined that the user's browsing intention is associated browsing, the first target resource to be pushed is determined from the predetermined candidate resources based on attribute information, historical interaction information, current preference resources and guidance information. Since the user's long-term preferences and short-term preferences and the resource type indicated by the guidance information are comprehensively considered, the association between the pushed resources and the resource type determined by the user is more accurately defined, the scope of the associated resources to be pushed is further improved, the matching degree between the pushed resources and the user's browsing intention is further improved, and the user experience is improved.

[0114] In some embodiments, determining a first target resource from multiple predetermined candidate resources based on historical interaction information, current preference resources, and guidance information may include the following operations: extracting multiple candidate content features of multiple predetermined candidate resources; performing intent matching based on attribute information, multiple candidate content features, content features of current preference resources, historical preference features, and guidance information to obtain respective matching degrees between each predetermined candidate resource and the intent; and determining a first target resource from multiple predetermined candidate resources based on respective matching degrees.

[0115] Figure 5 The schematic diagram of performing resource optimization when it is determined that the user intention is associated browsing according to an embodiment of the present disclosure is schematically shown.

[0116] like Figure 5 As shown, the difference between this embodiment 500 and the embodiment 400 is that feature extraction is performed on the candidate resource 501 to obtain the candidate content feature 502. After the object attribute 421, the historical preference feature 402, the content feature 412, the guidance information 422 and the candidate content feature 502 are spliced, they are input into the trained resource optimization model 510, and the target resource 503 is output.

[0117] In some embodiments, the resource optimization model 510 may be a neural network model, such as a NN (Neural Network) network model. During the training process, resources that are labeled as satisfactory to users may be used as positive samples, and resources that are labeled as unsatisfactory to users may be used as negative samples to train the initial model and obtain a resource optimization model.

[0118] By using the trained resource optimization model, the resources to be pushed are determined from the candidate resources based on a deep understanding of user preferences, which improves the matching degree between the resources to be pushed and the user's browsing intention, and further improves the user experience.

[0119] After it has been identified that the user's immediate browsing intention is to browse content associated with the content of the initial page, the user does not necessarily want to browse the associated content all the time when browsing pushed resources in an immersive scenario. The user's browsing intention may change at any time. Therefore, before using the resource optimization model to screen predetermined candidate resources, if the number of associated resources that the user wants to browse can be accurately estimated, the efficiency of candidate resource screening can be improved.

[0120] Therefore, according to an embodiment of the present disclosure, the above method may also include the following operations: obtaining network environment information; obtaining preference characteristics of associated resources based on historical interaction information and historical pushed resources; obtaining the target number of associated resources based on attribute information, historical preference characteristics, preference characteristics of associated resources, network environment information, content characteristics of the current guided resources and guidance information; and in response to determining that the intention is associated browsing, pushing the target number of first target resources associated with the guided resources through the target page.

[0121] Figure 6 The schematic diagram of performing resource optimization when it is determined that the user intention is associated browsing according to another embodiment of the present disclosure is schematically shown.

[0122] like Figure 6 As shown, in embodiment 600 , first, preference features 602 of associated resources may be calculated based on historical push resources and historical interaction information 401 .

[0123] In some embodiments, the preference characteristics of the associated resources include, but are not limited to, relevance distribution ratio, relevance duration ratio, relevance fast sliding rate, relevance completion rate, and relevance step length.

[0124] For example: Relevant distribution ratio = the number of historical pushed resources of the same resource type as indicated by the guidance information in the push page in the past three days / the total number of historical pushed resources. Duration ratio = the playback duration of historical pushed resources of the same resource type as indicated by the guidance information in the push page in the past three days / the total playback duration of historical pushed resources.

[0125] According to an embodiment of the present disclosure, the network environment information may be the network environment in which the terminal device used by the user is currently located, including but not limited to the application system type and network traffic of the terminal device.

[0126] Then, historical preference features 402 are obtained based on historical interaction information 401. The statistical method is the same as the statistical method used in the intention recognition process described above, and will not be described in detail here.

[0127] Next, the current guidance resource 603 is input into the CB2CF model, and the content characteristics 604 of the guidance resource are output.

[0128] Then, the preference features 602 of the associated resources, the historical preference features 402, the content features 604 of the guiding resources, the guiding information 422 and the network environment information 605 are concatenated and input into the trained quantity estimation model 610 to output the target quantity 606.

[0129] In some embodiments, the quantity estimation model 610 may also be a NN model, and the training sample label may be an associated quantity, for example, it may be 6.

[0130] Finally, according to the target quantity 606, the resource optimization model described above can be used to determine the target resource 433 from the predetermined candidate resources.

[0131] According to an embodiment of the present disclosure, before using a resource optimization model to screen predetermined candidate resources, the number of associated resources that the user wants to browse can be accurately estimated based on the user's feedback on historically pushed resources, current guided resources, and network environment, thereby further improving the efficiency of candidate resource screening.

[0132] Figure 7 A flowchart of a model training method according to an embodiment of the present disclosure is schematically shown.

[0133] like Figure 7 As shown, the model training method 700 may include: operations S710~S740.

[0134] In operation S710, an intent recognition module is used to perform intent recognition based on attributes of the sample object, sample historical preference features, content features of sample preference resources, and sample guide information, and the sample intent is output.

[0135] In operation S720, a resource optimization module is used to optimize resources based on attributes of sample objects, sample historical preference features, content features of sample preference resources, sample guidance information, and content features of sample candidate resources, and sample push resources are output.

[0136] In operation S730, based on the first target loss function, a first loss value is calculated according to the sample intent, the sample push resource, the sample intent label, and the sample push resource label.

[0137] In operation S740, based on the first loss value, the model parameters of the initial model are adjusted to obtain a trained model.

[0138] Figure 8 A schematic diagram of a model training method according to an embodiment of the present disclosure is schematically shown.

[0139] like Figure 8 As shown, in this embodiment 800, the initial model 810 may include an intent recognition module 810_1 and a resource optimization module 810_2.

[0140] For the intention recognition module 810_1, first, the attributes of the sample object, the sample history preference features, the content features of the sample preference resources and the sample guidance information can be spliced ​​to obtain the first input feature 811. Among them, the sample history preference features are calculated based on the sample history interaction information; the content features are extracted based on the sample preference resources. The statistical method and feature extraction method described above are the same and will not be repeated here.

[0141] Then, the first input feature 811 is input into MLP (Multi Layer Perceptron) for feature recognition, and then after normalization, the associated intent probability is output to obtain the sample intent 812.

[0142] Next, a first target loss function, such as a cross entropy loss function, may be used to calculate the loss based on the sample intent 812 and the sample intent label 813 to obtain a loss value L1814. The sample intent label 813 may include positive sample labels of resources that are annotated with the guiding resource and meet the user's browsing intent, and negative sample labels of resources that are annotated with the guiding resource and do not meet the user's browsing intent.

[0143] For the resource optimization module 810_2, first, the attributes of the sample object, the sample history preference features, the content features of the sample preference resources, the sample guidance information and the content features of the sample candidate resources can be feature spliced ​​to obtain the second input feature 821.

[0144] Then, the second input feature 821 is input into the MLP (Multi Layer Perceptron) for feature recognition, and then classified, and the category probability of each candidate resource is output. The associated intention probability is multiplied by the category probability of each candidate resource, and then input into the MLP (Multi Layer Perceptron) to output the sample recommended resource 822.

[0145] Next, a first target loss function, for example, a cross entropy loss function, may be used to calculate the loss based on the sample push resource 822 and the sample push resource tag 823 to obtain a loss value L2815.

[0146] Finally, the loss value L1814 and the loss value L2815 can be weighted according to the predetermined weight to obtain the total loss L831. The model parameters of the initial model 810 are adjusted based on the total loss L831 until the convergence condition is reached, such as the maximum number of iterations or the loss value converges, to obtain the trained model. The trained model can be used in the resource push method described above to perform the intent recognition task, and when the intent of the target object is identified as associated browsing, the resource optimization task is performed to filter out resources associated with the initial page content from the predetermined candidate resources.

[0147] According to an embodiment of the present disclosure, the intent recognition model and the resource optimization module are jointly trained so that the initial model can classify candidate resources according to user preferences with a deep understanding of the user's browsing intent, thereby improving the efficiency of model training and the accuracy of intent recognition and resource push of the trained model.

[0148] In some embodiments, the quantity estimation model can be trained separately. Since the network structure of the quantity estimation model is the same as that of the intent recognition model and the resource optimization model, an associated quantity determination module can also be added to the initial model.

[0149] According to an embodiment of the present disclosure, the initial model also includes an association quantity determination module. The above-mentioned model training method also includes: using the association quantity determination module to determine the sample quantity of associated resources based on the attributes of the sample object, the sample historical preference characteristics, the preference characteristics of the sample associated resources, the sample network environment information, the content characteristics of the sample guidance resources and the sample guidance information; based on the second target loss function, according to the sample quantity of the associated resources and the label of the sample quantity, calculate the second loss value; and based on the second loss value, in the case of fixing the model parameters of the intent recognition module and the resource optimization module, adjust the model parameters of the association quantity determination module to generate a trained model.

[0150] Fig. 9A schematic diagram of a model training method according to another embodiment of the present disclosure is schematically shown.

[0151] like Fig. 9 As shown, in this embodiment 900, the initial model 910 may include: an intention recognition module 810_1, a resource optimization module 810_2 and an associated quantity determination module 910_1.

[0152] For the association quantity determination module 910_1, first, the attributes of the sample object, the sample historical preference characteristics, the preference characteristics of the sample associated resources, the sample network environment information, the content characteristics of the sample guidance resources and the sample guidance information can be spliced ​​to obtain the third input feature 911.

[0153] Then, the third input feature 911 is input into the association quantity determination module 910_1 , and the sample quantity 912 is output.

[0154] Next, a second objective loss function, such as a MAE loss function (Mean Absolute Error), is used to calculate a loss value L3914 based on the number of samples 912 and the number of sample labels 913 .

[0155] Finally, the model parameters of the association quantity determination module are adjusted based on the loss value L3914 to obtain a trained model. The association quantity determination module of the trained model can estimate the number of associated resources when the target object's intention is identified as associated browsing in the resource push method described above, and then call the resource optimization module based on the number of associated resources to screen out the same number of associated resources from the predetermined candidate resources.

[0156] According to an embodiment of the present disclosure, while fixing the model parameters of the awareness recognition module and the resource optimization module, adjusting the model parameters of the association quantity determination module can further improve the efficiency of model training.

[0157] Fig.10 The block diagram of a resource pushing device according to an embodiment of the present disclosure is schematically shown.

[0158] like Fig.10 As shown, the resource pushing device 1000 may include: a first acquisition module 1010 , an identification module 1020 and a first pushing module 1030 .

[0159] The first acquisition module 1010 is used to obtain the target object's historical interaction information for historical pushed resources, current preferred resources, attribute information of the target object, and guidance information for guiding browsing of the initial page in response to receiving a switch operation of the target object for the initial page.

[0160] The identification module 1020 is used to identify the intention of the target object based on attribute information, historical interaction information, current preferred resources and guidance information.

[0161] The first push module 1030 is configured to push a first target resource associated with the guiding resource through the switched target page in response to determining that the intention is associated browsing.

[0162] According to an embodiment of the present disclosure, the recognition module may include: a calculation submodule, an extraction submodule and a recognition submodule.

[0163] The calculation submodule is used to calculate the historical preference features for the historically pushed resources based on the historical interaction information. The extraction submodule is used to extract the content features of the current preferred resources. The identification submodule is used to identify the intent based on the attribute information, historical preference features, content features and guidance information to obtain the intent.

[0164] According to an embodiment of the present disclosure, the current preference resources include push resources and guide resources; the guide resources are resources determined by the target object in the guide page within the current period; the push resources are resources on which the target object has performed predetermined interactive operations in the push page associated with the guide page.

[0165] According to an embodiment of the present disclosure, the extraction submodule may include: a first extraction unit and a splicing unit. The first extraction unit is used to extract the first content feature of the push resource and the second content feature of the guide resource respectively. The splicing unit is used to splice the first content feature with the second content feature to obtain the content feature of the current preferred resource.

[0166] According to an embodiment of the present disclosure, the first push module may include: a determination submodule and a generation submodule. The determination submodule is used to determine a first target resource from a plurality of predetermined candidate resources based on attribute information, historical interaction information, current preferred resources, and guidance information. The generation submodule is used to generate a target page based on the first target resource.

[0167] According to an embodiment of the present disclosure, the determination submodule includes: a second extraction unit, a matching unit and a determination unit.

[0168] The second extraction unit is used to extract multiple candidate content features of multiple predetermined candidate resources.

[0169] The matching unit is used to match the intent based on attribute information, multiple candidate content features, content features of current preferred resources, historical preference features and guidance information, and obtain the matching degree between each predetermined candidate resource and the intent; wherein the historical preference feature is obtained based on historical interaction information; and the content feature is extracted based on the current preferred resource.

[0170] The determining unit is used to determine a first target resource from a plurality of predetermined candidate resources based on each matching degree.

[0171] According to an embodiment of the present disclosure, the resource pushing device further includes: a second acquisition module, a calculation module, an estimation module and a second pushing module.

[0172] The second acquisition module is used to acquire network environment information.

[0173] The calculation module is used to obtain the preference characteristics of the associated resources based on the historical interaction information and the historical pushed resources.

[0174] The estimation module is used to obtain the target number of associated resources based on attribute information, historical preference characteristics, preference characteristics of associated resources, network environment information, content characteristics and guidance information for current guidance resources.

[0175] The second push module is used for pushing a target number of first target resources associated with the guiding resource through the target page in response to determining that the intention is associated browsing.

[0176] According to an embodiment of the present disclosure, the resource push device also includes: a third push module, which is used to push a second target resource through a target page in response to determining that the intention is non-associated browsing; wherein the second target resource is determined based on the interactive operation of the target object within a historical period.

[0177] Fig.11 A block diagram of a model training device according to an embodiment of the present disclosure is schematically shown.

[0178] like Fig.11 As shown, the model training device 1100 may include: an intention recognition module 1110, a resource optimization module 1120, a first loss calculation module 1130 and a first adjustment module 1140.

[0179] The initial model includes an intent recognition module 1110, which is used to perform intent recognition based on the attributes of the sample object, the sample historical preference characteristics, the content characteristics of the sample preference resources and the sample guidance information, and output the sample intent; wherein the sample historical preference characteristics are obtained based on the sample historical interaction information; the content characteristics are obtained based on the extraction of the sample preference resources.

[0180] The initial model includes a resource optimization module 1120, which is used to optimize resources based on the attributes of sample objects, sample historical preference characteristics, content characteristics of sample preference resources, sample guidance information and content characteristics of sample candidate resources, and output sample push resources.

[0181] The first loss calculation module 1130 is used to calculate a first loss value based on a first target loss function, according to sample intent, sample push resources, sample intent labels and sample push resource labels.

[0182] The first adjustment module 1140 is used to adjust the model parameters of the initial model based on the first loss value to obtain a trained model.

[0183] According to an embodiment of the present disclosure, the model training device also includes: an association quantity determination module, a second loss calculation module and a second adjustment module included in the initial model.

[0184] The initial model includes an associated quantity determination module, which is used to determine the sample quantity of associated resources based on the attributes of the sample object, the sample historical preference characteristics, the preference characteristics of the sample associated resources, the sample network environment information, the content characteristics of the sample guidance resources and the sample guidance information.

[0185] The second loss calculation module is used to calculate a second loss value based on a second target loss function and according to the number of samples of the associated resources and the labels of the number of samples.

[0186] The second adjustment module is used to adjust the model parameters of the association quantity determination module based on the second loss value while fixing the model parameters of the intent recognition module and the resource optimization module to generate a trained model.

[0187] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0188] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described above.

[0189] According to an embodiment of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute the method described above.

[0190] According to an embodiment of the present disclosure, a computer program product includes a computer program, and when the computer program is executed by a processor, the computer program implements the method described above.

[0191] Fig.12A schematic block diagram of an example electronic device 1200 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0192] like Fig.12 As shown, the device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 into a random access memory (RAM) 1203. In the RAM 1203, various programs and data required for the operation of the device 1200 can also be stored. The computing unit 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0193] A number of components in the device 1200 are connected to the I / O interface 1205, including: an input unit 1206, such as a keyboard, a mouse, etc.; an output unit 1207, such as various types of displays, speakers, etc.; a storage unit 1208, such as a disk, an optical disk, etc.; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1209 allows the device 1200 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0194] The computing unit 1201 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1201 performs the various methods and processes described above, such as a resource push method or a model training method. For example, in some embodiments, the resource push method or the model training method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1208. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 1200 via the ROM 1202 and / or the communication unit 1209. When the computer program is loaded into the RAM 1203 and executed by the computing unit 1201, one or more steps of the resource push method or the model training method described above may be performed. Alternatively, in other embodiments, the computing unit 1201 may be configured to execute the resource pushing method or the model training method in any other appropriate manner (eg, by means of firmware).

[0195] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0196] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0197] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0198] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0199] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0200] A computer system may include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises through computer programs running on respective computers and having a client-server relationship to each other. The server may be a cloud server, a server in a distributed system, or a server combined with a blockchain.

[0201] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0202] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A resource push method, comprising: In response to receiving a switch operation of the target object for the initial page, acquiring historical interaction information of the target object for historical pushed resources, current preferred resources, attribute information of the target object, and guidance information for guiding browsing of the initial page; Based on the attribute information, the historical interaction information, the current preferred resource and the guidance information, identifying the intention of the target object; as well as In response to determining that the intention is associated browsing, a first target resource associated with the guiding resource is pushed through the switched target page.

2. The method according to claim 1, wherein: The identifying the intention of the target object based on the attribute information, the historical interaction information, the current preferred resource and the guidance information includes: Based on the historical interaction information, obtaining historical preference features for the historical push resources; extracting content features of the currently preferred resource; and Intent recognition is performed based on the attribute information, the historical preference characteristics, the content characteristics and the guidance information to obtain the intent.

3. The method according to claim 2, wherein: The current preferred resources include push resources and guide resources; the guide resources are resources determined by the target object in the guide page within the current period; The pushed resource is a resource on which the target object has performed a predetermined interactive operation in a push page associated with the guide page; The extracting the content feature of the current preferred resource includes: Respectively extracting a first content feature of the push resource and a second content feature of the guide resource; as well as The first content feature and the second content feature are concatenated to obtain the content feature of the current preferred resource.

4. The method according to any one of claims 1 to 3, wherein The method further comprises: Determine the first target resource from a plurality of predetermined candidate resources based on the attribute information, the historical interaction information, the current preferred resource and the guidance information; The target page is generated based on the first target resource.

5. The method according to claim 4, wherein: The determining the first target resource from a plurality of predetermined candidate resources based on the attribute information, the historical interaction information, the current preferred resource and the guidance information includes: Extracting a plurality of candidate content features of the plurality of predetermined candidate resources; Intent matching is performed based on the attribute information, the plurality of candidate content features, the content features of the current preferred resource, the historical preference features and the guidance information to obtain the matching degrees between each predetermined candidate resource and the intent; wherein the historical preference features are obtained based on the historical interaction information; and the content features are extracted based on the current preferred resource; Based on each of the matching degrees, the first target resource is determined from the plurality of predetermined candidate resources.

6. The method according to any one of claims 1 to 5, wherein: The method further comprises: Obtain network environment information; Based on the historical interaction information and the historical pushed resources, obtaining preference characteristics of associated resources; Based on the attribute information, the historical preference characteristics, the preference characteristics of the associated resources, the network environment information, the content characteristics of the current guided resources and the guidance information, obtaining a target number of associated resources; and In response to determining that the intention is associated browsing, the target number of first target resources associated with the guiding resource are pushed through the target page.

7. The method according to any one of claims 1 to 5, wherein: The method further comprises: In response to determining that the intention is non-associated browsing, a second target resource is pushed through the target page; wherein the second target resource is determined based on the interactive operation of the target object in a historical period.

8. A model training method, the method comprising: Using the intent recognition module included in the initial model, the intent recognition is performed based on the attributes of the sample object, the sample history preference features, the content features of the sample preference resources and the sample guidance information, and the sample intent is output; wherein the sample history preference features are obtained based on the sample history interaction information; and the content features are obtained based on the extraction of the sample preference resources; Utilizing the resource optimization module included in the initial model, optimizing resources based on the attributes of the sample object, the sample historical preference characteristics, the content characteristics of the sample preference resources, the sample guidance information and the content characteristics of the sample candidate resources, and outputting sample push resources; Based on the first objective loss function, calculating a first loss value according to the sample intent, the sample push resource, the sample intent label, and the sample push resource label; Based on the first loss value, the model parameters of the initial model are adjusted to obtain a trained model.

9. The method according to claim 8, wherein: The method further comprises: Using the association quantity determination module included in the initial model, the sample quantity of the associated resources is determined based on the attributes of the sample object, the sample historical preference characteristics, the preference characteristics of the sample associated resources, the sample network environment information, the content characteristics of the sample guidance resources and the sample guidance information; Based on the second objective loss function, calculating a second loss value according to the number of samples of the associated resource and the label of the number of samples; and Based on the second loss value, while fixing the model parameters of the intention recognition module and the resource optimization module, the model parameters of the association quantity determination module are adjusted to generate a trained model.

10. A resource push device, comprising: A first acquisition module, configured to, in response to receiving a switch operation of a target object with respect to an initial page, acquire historical interaction information of the target object with respect to historical pushed resources, current preferred resources, attribute information of the target object, and guidance information for guiding browsing of the initial page; an identification module, configured to identify the intention of the target object based on the attribute information, the historical interaction information, the current preferred resource and the guidance information; as well as The first push module is used to push a first target resource associated with the guiding resource through a switched target page in response to determining that the intention is associated browsing.

11. The device according to claim 10, wherein: The identification module comprises: A calculation submodule, configured to obtain a historical preference feature for the historical push resource based on the historical interaction information; an extraction submodule, configured to extract content features of the currently preferred resource; and The identification submodule is used to perform intent identification based on the attribute information, the historical preference characteristics, the content characteristics and the guidance information to obtain the intent.

12. The device according to claim 11, wherein The current preferred resources include push resources and guide resources; the guide resources are resources determined by the target object in the guide page within the current period; The pushed resource is a resource on which the target object has performed a predetermined interactive operation in a push page associated with the guide page; The extraction submodule comprises: A first extraction unit, used to extract a first content feature of the push resource and a second content feature of the guide resource respectively; as well as The concatenation unit is configured to concatenate the first content feature with the second content feature to obtain the content feature of the current preferred resource.

13. The device according to any one of claims 10 to 12, wherein: The first push module also includes: a determination submodule, configured to determine the first target resource from a plurality of predetermined candidate resources based on the attribute information, the historical interaction information, the current preferred resource and the guidance information; A generating submodule is used to generate the target page based on the first target resource.

14. The device according to claim 13, wherein: The determination submodule comprises: A second extraction unit, configured to extract a plurality of candidate content features of the plurality of predetermined candidate resources; A matching unit, configured to perform intent matching based on the attribute information, the plurality of candidate content features, the content features of the current preferred resource, the historical preference features and the guidance information, and obtain the matching degrees between each predetermined candidate resource and the intent; wherein the historical preference features are obtained based on the historical interaction information; and the content features are extracted based on the current preferred resource; A determining unit is used to determine the first target resource from the multiple predetermined candidate resources based on each of the matching degrees.

15. The device according to any one of claims 10 to 14, wherein: The device also includes: The second acquisition module is used to acquire network environment information; A calculation module, used for obtaining preference characteristics of associated resources based on the historical interaction information and the historical pushed resources; an estimation module, configured to obtain a target number of associated resources based on the attribute information, the historical preference characteristics, the preference characteristics of the associated resources, the network environment information, the content characteristics of the current guided resources, and the guidance information; and The second push module is used for pushing the target number of first target resources associated with the guiding resource through the target page in response to determining that the intention is associated browsing.

16. The method according to any one of claims 10 to 14, wherein: The device also includes: The third push module is used to push a second target resource through the switched target page in response to determining that the intention is non-associated browsing; wherein the second target resource is determined based on the interactive operation of the target object in a historical period.

17. A model training device, comprising: The initial model includes an intention recognition module, which is used to perform intention recognition based on the attributes of the sample object, the sample history preference features, the content features of the sample preference resources and the sample guidance information, and output the sample intention; wherein the sample history preference features are obtained based on the sample history interaction information; the content features are obtained based on the sample preference resource extraction; The initial model includes a resource optimization module, which is used to optimize resources based on the attributes of the sample object, the sample historical preference characteristics, the content characteristics of the sample preference resources, the sample guidance information and the content characteristics of the sample candidate resources, and output sample push resources; A first loss calculation module, configured to calculate a first loss value based on a first target loss function, according to the sample intent, the sample push resource, the sample intent label, and the sample push resource label; and The first adjustment module is used to adjust the model parameters of the initial model based on the first loss value to obtain a trained model.

18. The device according to claim 17, wherein: The device also includes: The initial model includes an association quantity determination module, which is used to determine the sample quantity of the number of associated resources based on the attributes of the sample object, the sample historical preference characteristics, the preference characteristics of the sample associated resources, the sample network environment information, the content characteristics of the sample guidance resources and the sample guidance information; A second loss calculation module, configured to calculate a second loss value based on a second target loss function and according to the number of samples of the associated resource and the label of the number of samples; and The second adjustment module is used to adjust the model parameters of the association quantity determination module based on the second loss value while fixing the model parameters of the intention recognition module and the resource optimization module to generate a trained model.

19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-9.

21. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images