Content processing method and device, electronic equipment and storage medium

Abstract

The invention relates to a content processing method and device, electronic equipment and a storage medium. The method comprises the steps that to-be-recommended content for a target object is acquired, the to-be-recommended content comprises associated content and target content, and display of the associated content is used for guiding display of the target content; and taking the object information of the target object and the content information of the to-be-recommended content as input, and outputting a corresponding target prediction result by utilizing a target prediction model, so as to recommend the to-be-recommended content based on the target prediction result. According to the method, the adaptability of the target prediction model obtained by training to a specific scene can be improved, so that the content recommendation effect is facilitated. The embodiment of the invention can be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation, intelligent entertainment and the like.

Description

Technical Field

[0001] The present application relates to the field of Internet communication technology, and in particular to a content processing method, device, electronic device and storage medium. Background Art

[0002] With the development of Internet communication technology, various Internet products emerge in an endless stream. Related Internet products can provide users with content recommendation services to meet their information acquisition needs. In related technologies, prediction models are often used to predict the content to be recommended, and then the recommendation logic for the content to be recommended is determined based on the prediction results. However, in the process of training the prediction model, only the difference between positive and negative samples is paid attention to, which affects the adaptability of the prediction model and thus affects the content recommendation effect. Summary of the invention

[0003] In order to solve at least one of the above-mentioned technical problems, the present application provides a content processing method, device, electronic device and storage medium:

[0004] According to a first aspect of the present application, a content processing method is provided, the method comprising:

[0005] Acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and display of the associated content is used to guide display of the target content;

[0006] Taking the object information of the target object and the content information of the content to be recommended as input, outputting a corresponding target prediction result using a target prediction model, so as to recommend the content to be recommended based on the target prediction result;

[0007] Wherein, the target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on a positive set in the sample set that carries a label indicating successful guidance. The first difference is the difference between a first correlation and a second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between a third correlation and a fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

[0008] According to a second aspect of the present application, a content processing device is provided, the device comprising:

[0009] Acquisition module: used to acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and the display of the associated content is used to guide the display of the target content;

[0010] Prediction module: used for taking the object information of the target object and the content information of the content to be recommended as input, outputting the corresponding target prediction result by using the target prediction model, so as to recommend the content to be recommended based on the target prediction result;

[0011] Wherein, the target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on a positive set in the sample set that carries a label indicating successful guidance. The first difference is the difference between a first correlation and a second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between a third correlation and a fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

[0012] According to the third aspect of the present application, 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 at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the at least one processor to implement the content processing method as described in the first aspect.

[0013] According to a fourth aspect of the present application, a computer-readable storage medium is provided, in which at least one instruction or at least one program is stored, and the at least one instruction or at least one program is loaded and executed by a processor to implement the content processing method as described in the first aspect.

[0014] According to a fifth aspect of the present application, a computer program product is provided, which includes at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the content processing method as described in the first aspect.

[0015] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application.

[0016] Implementing this application has the following beneficial effects:

[0017] In the process of training to obtain the target prediction model, the present application combines the characteristics of some scenarios in which direct recommendation content has the need to guide the display of indirect recommendation content (such as the automatic broadcast material as direct recommendation content guides the display of the main film as indirect recommendation content), and constructs two types of differences based on the positive sample to adjust the model parameters to improve the adaptability of the target prediction model obtained by training to specific scenarios, thereby facilitating the content recommendation effect. The two types of differences focus on the difference in the correlation between the object dimension information and the content dimension information in the positive set and the entire sample set (or sample subset), and the difference in the correlation between the first information (sample associated content) and the second information (sample target content) in the positive set and the entire sample set (or sample subset). Taking into account the association between the content dimension information and the object dimension information as a whole in the sample, and the difference in the association in the positive set relative to the entire sample set (or sample subset), it is beneficial to dig out the connection between the object and the content, so as to better recommend associated content and target content for the target object. At the same time, the association between the first information and the second information in the content dimension information in the sample, as well as the difference between the association in the positive concentration and the association in the entire sample set (or sample subset) are taken into consideration, which is conducive to capturing the connection between the associated content and the target content, so as to improve the success rate of displaying the target content.

[0018] Other features and aspects of the present application will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A schematic diagram of an application environment according to an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram showing a flow chart of a content processing method according to an embodiment of the present application;

[0022] Figure 3 A schematic diagram showing a process of training and obtaining a target prediction model according to an embodiment of the present application;

[0023] Figure 4 A schematic diagram of a process for obtaining positive and negative samples according to an embodiment of the present application is shown;

[0024] Figure 5 A schematic diagram showing a process of obtaining candidate associated content according to an embodiment of the present application;

[0025] Figure 6 A schematic diagram showing the architecture of a preset model according to an embodiment of the present application;

[0026] Figure 7-9 A schematic diagram showing an interface of a selected page according to an embodiment of the present application is shown;

[0027] Fig.10 A device block diagram according to an embodiment of the present application is shown;

[0028] Fig.11 A schematic diagram of an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0032] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0033] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0034] In addition, in order to better illustrate the present application, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present application can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present application.

[0035] See also Figure 1 , Figure 1 A schematic diagram of an application environment according to an embodiment of the present application is shown, and the application environment may include a client 10 and a server 20. The client 10 and the server 20 may be directly or indirectly connected via wired or wireless communication. The user as the target object sends a content recommendation request to the server 20 via the client 10. The server 20 obtains the content to be recommended for the target object based on the received content recommendation request, and the content to be recommended includes associated content and target content, and the display of associated content is used to guide the display of the target content; then, the object information of the target object and the content information of the content to be recommended are used as input, and the target prediction model is used to output the corresponding target prediction result, so as to recommend the content to be recommended based on the target prediction result.

[0036] The target prediction model used is obtained by training through the following steps: 1) using the first network to output the object vector corresponding to the sample object information, and using the second network to output the first content vector corresponding to the sample content information; 2) using the second network to output the second content vector corresponding to the first type of information, and multiple first types of information are obtained by performing first modality loss processing on multiple sample content information in the positive set, and the first modality loss processing is used to cover up the second information; 3) using the second network to output the third content vector corresponding to the second type of information, and multiple second types of information are obtained by performing second modality loss processing on multiple sample content information in the positive set, and the second modality loss processing is used to cover up the first information; 4) based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set, the first difference is obtained; 5) based on the third content vector corresponding to each second type of information in the positive set and the second content vector corresponding to each first type of information in the sample set, the second difference is obtained; 6) based on the first difference and the second difference, the parameters of the preset model are adjusted to obtain the target prediction model. It should be noted that Figure 1 Just an example.

[0037] The client 10 may be a physical device such as a smart phone, a computer (such as a desktop computer, a tablet computer, a laptop computer), an augmented reality (AR) / virtual reality (VR) device, a digital assistant, an intelligent voice interaction device (such as an intelligent speaker), an intelligent wearable device, an intelligent home appliance, a vehicle-mounted terminal, etc., or may be software running in a physical device, such as a computer program. The operating system of the client 10 may be an Android system (Android system), an iOS system (a mobile operating system developed by Apple), a Linux system (an operating system), a Microsoft Windows system (Microsoft Windows operating system), etc.

[0038] The server end 20 can 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, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server can include a network communication unit, a processor, a memory, etc. The server end can provide background services for the corresponding client.

[0039] In practical applications, the server end that obtains the target prediction model through the training steps provided in the embodiment of the present application may be server end A, and the server end that processes the content to be recommended through the content processing method provided in the embodiment of the present application may be server end B. Server end A and server end B may indicate the same server end or two different server ends. In addition, the client may also process the content to be recommended using the locally stored target prediction model.

[0040] It should be noted that for sample object information, target object information, etc. that are associated with user information, when the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0041] Figure 2 A flow chart of a content processing method according to an embodiment of the present application is shown as follows: Figure 2 As shown, the method includes:

[0042] S201: Acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and the display of the associated content is used to guide the display of the target content;

[0043] In an embodiment of the present application, the server or client obtains the content to be recommended for the target object. The target object may be a target user, a target account, etc. The content to be recommended may be content randomly extracted from the content pool, or may be content extracted from the content pool according to preset rules. The preset rules are intended to better meet the needs of the target object. The content to be recommended includes associated content and target content, and the display of associated content is used to guide the display of target content. The expected display timing of associated content is earlier than the expected display timing of target content. It can be understood that the substantial content of content recommendation to the target object is the target content, but the associated content is recommended to the target object before the target content, and the associated content will play a guiding role, so that the target object performs a preset operation to trigger the target content to be displayed. The preset operation occurs under the premise that the associated content is displayed, and the preset operation may be an operation with positive intentions, which indicates that the target object is interested in the associated content and its related content, and the operation with positive intentions may be clicking, long pressing, etc. in the page area where the associated content is displayed and / or in the adjacent page area of ​​the page area. On the contrary, reverse intent indicates that the target object is not interested in the associated content and its relevance. Operations with reverse intent can include closing the page displaying the associated content, or swiping up or down to move away from the page area displaying the associated content.

[0044] For example, the associated content is the auto-play material, and the target content is the main film. The auto-play material is displayed on the first page presented to the user, and the user can watch the auto-play material. The user triggers the relevant interactive controls on the first page, and the page jumps to the second page, on which the main film is displayed, and the user can watch the main film. Of course, the page may not jump, and the user may trigger the relevant interactive controls on the first page, and the main film is displayed in the relevant area of ​​the first page, and the user can watch the main film. The relevant area may be the page area that originally displayed the auto-play material, or it may be other areas, such as the adjacent page area that originally displayed the auto-play material.

[0045] Automatic playback material is a form of related content. Automatic playback material is related content that is automatically played through a preset window. In the content recommendation scenario, automatic playback material is automatically played for display. Automatic playback material can be a video with a certain length. For example, the target content is a feature film, which is the full version of a certain movie. The automatic playback material can be the trailer of the movie; the feature film is the full version of a certain episode in a certain TV series, and the automatic playback material can be the highlights of the episode. Figure 7-9 As shown, the automatically played material can be presented on the featured pages of related Internet products.

[0046] The associated content is related to the target content. The associated content is generally generated based on the target content or the theme of the target content. The same target content can correspond to at least one associated content. One target content and the first associated content constitute a content to be recommended. The content form of the associated content can be text, image, graphic, audio, video (including short video), etc. The content form of the target content can be text, image, graphic, audio, video (including short video), etc.

[0047] S202: taking the object information of the target object and the content information of the content to be recommended as input, using a target prediction model to output a corresponding target prediction result, so as to recommend the content to be recommended based on the target prediction result;

[0048] In an embodiment of the present application, the object information of the target object and the content information of the content to be recommended are used as input, and the server or client uses the target prediction model to output the corresponding target prediction result to recommend the content to be recommended based on the target prediction result. The corresponding target prediction result can be obtained through the target prediction model, and the target prediction result reflects the prediction result of the target object's object information under at least one business goal and the characterization of the prediction result of the content information of the content to be recommended under at least one business goal. In this way, with the help of the adaptability and reliability of the target prediction model, the efficiency and accuracy of the obtained target prediction results can be improved. Subsequently, the content to be recommended can be recommended based on the target prediction result. For example, whether to recommend the content to be recommended to the target object is determined based on the target prediction result and a preset threshold.

[0049] Regarding "recommending the content to be recommended based on the target prediction result", the process may include the following steps: when it is determined based on the target prediction result that the content to be recommended meets the preset recommendation requirements, configuring a first display mode for the associated content and a second display mode for the target content, the first display mode indicating that the associated content is automatically played through a preset window, and the second display mode indicating that the playback of the target content is triggered based on a preset operation for the associated content; recommending the content to be recommended to the target object based on the first display mode and the second display mode.

[0050] The target prediction result can be used as a basis for judging whether to recommend the content to be recommended to the target object. If the content to be recommended needs to be recommended to the target object, the display mode is configured for the associated content and the target content that constitute the content to be recommended. Different display modes fit the different roles of associated content and target content in content recommendation scenarios. As the automatically played associated content of direct recommended content, it can increase the amount of information delivered to users and help attract user attention. As the target content of indirect recommended content, it plays in response to preset operations for associated content, which provides a guarantee for retaining the original playback environment of the target content (such as the target content is played under a specific website link), which is conducive to maintaining a high-quality and orderly target content production environment.

[0051] The target prediction model is described in detail below. The target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on the positive set of labels indicating successful guidance in the sample set. The first difference is the difference between the first correlation and the second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between the third correlation and the fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

[0052] Whether the display of the sample associated content successfully guides the display of the sample target content. The sample associated content and the sample target content can refer to the records of the associated content and the target content in the aforementioned step S201, and will not be repeated here. Correspondingly, successful guidance indicates that the above preset operation has occurred, and unsuccessful guidance indicates that the above preset operation has not occurred.

[0053] like Figure 3 As shown, the preset model includes a first network and a second network; the target prediction model is obtained by training through the following steps:

[0054] S301: using the first network to output an object vector corresponding to the sample object information, and using the second network to output a first content vector corresponding to the sample content information;

[0055] S302: using the second network to output a second content vector corresponding to the first type of information, where the plurality of the first types of information are obtained by performing first modality deletion processing on the plurality of sample content information in the positive set, respectively, and the first modality deletion processing is used to cover up the second information;

[0056] S303: using the second network to output a third content vector corresponding to the second type of information, where the plurality of the second types of information are obtained by performing a second modality deletion process on the plurality of the sample content information in the positive set, respectively, and the second modality deletion process is used to cover up the first information;

[0057] S304: Obtaining the first difference based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set;

[0058] S305: Obtain the second difference based on the third content vector corresponding to each piece of the second-category information in the positive set and the second content vector corresponding to each piece of the first-category information in the sample set;

[0059] S306: Based on the first difference and the second difference, adjust parameters of the preset model to obtain the target prediction model.

[0060] Exemplarily, the sample set includes N samples, sample i is the i-th sample among the N samples, and the value range of i is 1-N. Sample i is composed of sample object information oi and sample content information ci. Sample content information i includes first information ci1 of sample associated content i and second information ci2 of sample target content i. Sample i has a label for describing whether the display of sample associated content i successfully guides the display of sample target content i. If it is successfully guided, label label=1. If it is not successfully guided, label label=0.

[0061] The sample object information oi is input into the first network, and the first network outputs the corresponding object vector oi. The sample content information ci is input into the second network, and the second network outputs the corresponding first content vector ci. To obtain the first content vector ci, the data input into the second network is the first information ci1 and the second information ci2.

[0062] The M samples with label label=1 in the sample set constitute the positive set, where M is less than N. Sample j is the jth sample among the M samples, and the value range of j is 1-M. Correspondingly, sample j is composed of sample object information oj and sample content information cj. Sample content information j includes first information cj1 of sample associated content j and second information cj2 of sample target content j. Compared with the input data of the above-mentioned second network: the first information ci1 and the second information ci2, the second information cj2 is subjected to modal missing processing. In this way, the data input to the second network is the first information cj1, and the second network outputs the corresponding second content vector cj. Similarly, compared with the input data of the above-mentioned second network: the first information ci1 and the second information ci2, the first information cj1 is subjected to modal missing processing. In this way, the data input to the second network is the second information cj2, and the second network outputs the corresponding third content vector cj.

[0063] Since the first difference is the difference between the first correlation and the second correlation, the first correlation represents the correlation between the object dimension information and the content dimension information in the positive set, and the second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set provide the data source for obtaining the first difference.

[0064] Since the second difference is the difference between the third correlation and the fourth correlation, the third correlation represents the correlation between the first information and the second information in the positive set, and the fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set. The third content vector corresponding to each second type of information in the positive set and the second content vector corresponding to each first type of information in the sample set provide the data source for obtaining the second difference.

[0065] When the first difference and the second difference are used to adjust the parameters of the preset model, a target loss function can be constructed based on the first difference and the second difference, and then the target loss function is used to adjust the parameters of the preset model. In the constructed target loss function, different weight coefficients can be configured for the two types of differences according to historical feedback, etc.

[0066] Model training is ongoing, and the target prediction model can be the result of the previous training or the basis for the next training. The target prediction model is a model with high generalization ability obtained through training. Using the target prediction model for prediction can improve the adaptability and reliability of the prediction. During training, compared with learning to bring the same sample closer and push different samples further away, learning to mine the connection between objects and content and capture the connection between related content and target content based on positive samples is more in line with the needs of some scenarios where direct recommendation content has the need to guide the display of indirect recommendation content (such as the automatic broadcast material as direct recommendation content guides the display of the main film as indirect recommendation content), which is beneficial to the content recommendation effect.

[0067] For the four correlations mentioned above, the two parts of the correlation to be calculated can be regarded as two wholes, and the inner product calculation, summation, or splicing of the two wholes can be performed to obtain the corresponding correlation. Taking the calculation of the second correlation as an example, the object dimension information in the positive set is regarded as a whole, and the whole can be obtained based on the object vector corresponding to each sample object information in the positive set. The content dimension information in the sample set is regarded as another whole, and the whole can be obtained based on the first content vector corresponding to each sample content information in the sample set. The two parts of the correlation to be calculated can also be subjected to fine-grained calculation of the sample dimension, and then these fine-grained calculation results are integrated to obtain the corresponding correlation. Taking the calculation of the first correlation as an example, for each sample in the positive set, the object vector corresponding to the sample object information of the sample and the first content vector corresponding to the sample content information are determined to obtain the fine-grained calculation of the sample dimension. The fine-grained calculations of multiple sample dimensions are integrated to obtain the corresponding correlation.

[0068] In one embodiment, obtaining the first difference based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set may include the following steps: first, based on the first content vector corresponding to each sample content information in the sample set, determining the first content vector corresponding to each sample content information in a first candidate set and the first content vector corresponding to each sample content information in the positive set, the first candidate set being composed of a portion extracted from the sample set; then, based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information, obtaining the first relevance; further, based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the first candidate set, obtaining the second relevance; finally, determining the first difference between the first relevance and the second relevance.

[0069] In obtaining the second relevance, the association between the content dimension information and the object dimension information in the sample as a whole, as well as the difference between the association in the positive set and the association in the sample subset (corresponding to the first candidate set mentioned above) are taken into account. Compared with the contribution of the content dimension information of the entire sample set, the contribution of the content dimension information of the sample subset is more conducive to the mining of the connection between objects and content; on the other hand, the reduction in the amount of data to be calculated can save computing resources, increase the speed of obtaining the first difference, and reduce the time cost of trial and error for selecting a more appropriate method for calculating the first difference.

[0070] In one embodiment, obtaining the second difference based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information in the sample set may include the following steps: first, based on the second content vector corresponding to each first-category information in the sample set, determining the second content vector corresponding to each first-category information in a second candidate set and the second content vector corresponding to each first-category information in the positive set, the second candidate set being composed of a part extracted from the sample set; then, based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information, obtaining the third correlation; further, based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information in the second candidate set, obtaining the fourth correlation; finally, determining the second difference between the third correlation and the fourth correlation.

[0071] In obtaining the fourth relevance, the association between the first information and the second information in the content dimension information in the sample and the difference between the association in the positive set and the association in the sample subset (corresponding to the second candidate set mentioned above) are taken into account. Compared with the first information contribution of the entire sample set, the first information contribution of the sample subset is more conducive to capturing the connection between the associated content and the target content; on the other hand, the reduction in the amount of data to be calculated can save computing resources, increase the speed of obtaining the second difference, and reduce the time cost of trial and error for selecting a more appropriate method for calculating the second difference.

[0072] In practical applications, the first candidate set may be composed of some samples randomly selected from the sample set. The second candidate set may be composed of some samples randomly selected from the sample set. It should be noted that, in addition to the above “based on the first content vector corresponding to each sample content information in the sample set, determine the first content vector corresponding to each sample content information in the first candidate set, and the first content vector corresponding to each sample content information in the positive set”, the sample content information of the relevant samples may be input into the second network on the basis of the known first candidate set and the positive set, so as to obtain the first content vector corresponding to each sample content information in the first candidate set, and the first content vector corresponding to each sample content information in the positive set. In addition to the above “based on the second content vector corresponding to each first category information in the sample set, determine the second content vector corresponding to each first category information in the second candidate set, and the second content vector corresponding to each first category information in the positive set”, the sample content information of the relevant samples may be input into the second network on the basis of the known second candidate set and the positive set, so as to obtain the second content vector corresponding to each first category information in the second candidate set, and the second content vector corresponding to each first category information in the positive set.

[0073] In addition, the differences involved in parameter adjustment may include a third difference in addition to the first difference and the second difference. The third difference indicates the difference in prediction results corresponding to two types of samples with different labels. Accordingly, the parameter adjustment of the preset model based on the first difference and the second difference to obtain the target prediction model may include the following steps: based on the first difference, the second difference and the third difference, the parameter adjustment of the preset model to obtain the target prediction model.

[0074] The third difference focuses on the output difference between the positive set and the negative set in the sample set after the preset model. Compared with the first difference and the second difference, which focus on the differences brought by the internal information of the sample, the third difference focuses on the differences brought by different types of samples. The application of the three types of differences is conducive to improving the accuracy and adaptability of the target prediction model obtained through training. The target loss function can be constructed based on the first difference, the second difference and the third difference, and then the target loss function is used to adjust the parameters of the preset model. In the constructed target loss function, different weight coefficients can be configured for the two types of differences based on historical feedback, etc.

[0075] like Figure 6 As shown, the preset model has a first network and three second networks, and the four networks are four towers. Compared with the double-tower model, the preset model is better in terms of expression enhancement and alignment. The first network corresponds to Figure 6 The leftmost deep neural network (DNN) in the figure, the three second networks correspond to Figure 6The three deep neural networks (DNNs) on the right side of the figure. In the figure, user_output represents the output result of sample object information; positive_item_output represents the output result of sample content information of positive samples; sampling_item_output represents the output result of sample content information of sampling samples, and the sampling samples correspond to the samples in the first candidate set and the second candidate set mentioned above; positive_main_output represents the output result of the second type of information of positive samples, and the second type of information is the result of masking the first information in the sample content information here; positive_mid_output represents the output result of the first type of information of positive samples, and the first type of information is the result of masking the second information in the sample content information here; sampling_mid_output represents the output result of the first type of information of sampling samples.

[0076] For example, there are 100 samples, of which 60 are positive samples and 40 are negative samples. Randomly sample k samples from the 100 samples, and take k as 30. In the calculation of the first difference, the inner product of user_output and positive_item_output can be calculated as the positive sample logit (corresponding to the first correlation above); then, the inner product of user_output and sampling_item_output is calculated as the negative sample logits (corresponding to the second correlation above); and then softmax_cross_entropy_with_logits is performed on the positive sample logit and the negative sample logits:

[0077] -logexp(inner product of positive samples) / ∑(inner product of positive and negative samples),

[0078] In this way, the first difference can be obtained. The user_output used in both places here indicates the output results of the sample object information of 60 positive samples.

[0079] In the calculation of the second difference, the inner product can be calculated by positive_main_output and positive_mid_output as the positive sample logit (corresponding to the third correlation above); then, the inner product is calculated by positive_main_output and sampling_mid_output as the negative sample logits (corresponding to the fourth correlation above); and then softmax_cross_entropy_with_logits is performed on the positive sample logit and the negative sample logits:

[0080] -logexp(inner product of positive samples) / ∑(inner product of positive and negative samples)

[0081] In this way, the second difference can be obtained. The difference used for parameter adjustment = first difference * weight coefficient 1 + second difference * weight coefficient 2 + third difference.

[0082] If the associated content is automatic play material, the target content is the main film. In the video application, there may be functions that support automatic play material playback and main film playback, and the core goal of the video application is the main film playback, so the design of the automatic play material is also to facilitate the main film playback. The above-mentioned first difference can be used to strengthen the user's interest in the main film, and the second difference can be used to strengthen the association of the material that facilitates the conversion to the main film with its main film. The target prediction model provided by the embodiment of the present application can enhance the relationship between the automatic play material that facilitates the playback of the main film and its corresponding main film through the first difference and the second difference, so as to strengthen the information mining of the model material conversion to the main film, and then enhance the recommendation effect of the rough sorting and recall models.

[0083] The content processing solution provided in the embodiment of the present application can be applied to the featured pages of related Internet products. The related Internet products can be video products, including but not limited to app products, PC products, and OTT (over the top, which refers to providing various application services to users through the Internet) products. The content processing solution provided in the embodiment of the present application is applied to the featured pages of app products, PC products, and OTT products, respectively. Figure 7-9 When a user enters the video product, he will automatically jump to the featured page. Compared with the original solution that does not apply the content processing solution provided by the embodiment of the present application, the content processing solution provided by the embodiment of the present application can achieve significant improvements in indicators such as average per capita feature film playback time and average per capita feature film playback times.

[0084] In one embodiment, Figure 4 As shown, the method also includes:

[0085] S401: for the candidate associated content delivered to the candidate object, taking the time point when the candidate associated content is displayed as the monitoring time starting point, monitoring whether the candidate object performs a preset operation within a preset time length, wherein the preset operation is used to trigger the candidate target content to be displayed;

[0086] S402: in the case where it is monitored that the candidate object performs the preset operation, determining the object information of the candidate object as the sample object information, determining the content information of the candidate associated content and the content information of the candidate target content as the sample content information, and obtaining a positive sample based on the sample object information and the sample content information;

[0087] S403: In the case that the candidate object is not detected to perform the preset operation, determining that the object information of the candidate object is the sample object information, determining that the content information of the candidate associated content and the content information of the candidate target content are the sample content information, and obtaining a negative sample based on the sample object information and the sample content information.

[0088] The target prediction model is obtained using a sample set, which consists of multiple positive samples and multiple negative samples. For multiple positive samples and multiple negative samples, a method for obtaining positive and negative samples is provided here. Both positive and negative samples involve object dimension information and content dimension information. By marking a sample as a positive sample or a negative sample, it is possible to focus on whether the display of the candidate associated content delivered successfully guides the display of the candidate target content. The starting point of the monitoring time is the time point when the candidate associated content delivered is displayed. On this basis, positive and negative samples are labeled according to whether a preset operation occurs within a preset time length. This method of obtaining positive and negative samples is convenient and efficient, which is conducive to improving the efficiency of training to obtain the target prediction model.

[0089] Furthermore, 1) both positive and negative samples involve object dimension information and content dimension information. Object dimension information comes from object information of candidate objects. Object information is used to characterize the characteristics of candidate objects. Object information may be object identifier, object attribute, object type, etc. Content dimension information comes from content information of candidate associated content and content information of candidate target content. Content information of candidate associated content is used to characterize the characteristics of candidate associated content. Content information of candidate target content is used to characterize the characteristics of candidate target content. Specifically, determining the content information of the candidate associated content and the content information of the candidate target content as the sample content information includes: 1) obtaining a first identifier of the candidate associated content and first content description information for the candidate associated content to obtain the first information; 2) obtaining a second identifier of the candidate target content and second content description information for the candidate target content to obtain the second information; 3) obtaining the sample content information based on the first information and the second information.

[0090] The first identifier reflects the id feature of the candidate associated content, and the first content description information reflects the Sideinfo feature of the candidate associated content. The second identifier reflects the id feature of the candidate target content, and the first content description information reflects the Sideinfo feature of the candidate target content. The candidate associated content or the candidate target content can be regarded as an item. Sideinfo features can be other features of the item except the id feature. Taking the content as a video as an example, the Sideinfo features can include the release time, publisher information, creator information (such as director information), cast information, video type, video rating, etc. of the video.

[0091] A granular method for constructing sample content information is provided here, which is conducive to ensuring the richness of content dimension information, and further conducive to the accuracy of the target prediction model obtained through training.

[0092] 2) If Figure 5 As shown, for the candidate associated content delivered to the candidate object, the time point when the candidate associated content is displayed is used as the monitoring time starting point, and before monitoring whether the candidate object performs a preset operation within a preset time length, the method also includes:

[0093] S501: When the candidate target content is a target video, extract at least one candidate video segment from the target video;

[0094] S502: Performing preset processing on the at least one candidate video segment to obtain the candidate associated content; wherein the preset processing includes at least one of the following: splicing, adding visual materials, and editing audio tracks.

[0095] The associated content is related to the target content. The target content can provide a data source for obtaining the associated content. In the case where the target content is in the form of a video, the data source provided by the target content can be at least one video clip. The associated content is derived from processing of at least one video clip.

[0096] If at least one candidate video segment is at least two candidate video segments, the at least two candidate video segments can be spliced ​​to obtain related content. Each candidate video segment can indicate a progress bar time interval of the target video, and the progress bar time intervals corresponding to different candidate video segments can be different. At least two candidate video segments can be spliced ​​according to the order of appearance of the progress bar time intervals. At least two candidate video segments can also be spliced ​​according to the order of the subject popularity of the candidate video segments. Visual material addition is used to add visual material to at least one candidate video segment. The added visual material can be a watermark, a filter, a video frame that plays a connecting role (such as an opening role), etc. Audio track editing can update (such as delete, add, replace, etc.) the original audio track of at least one candidate video segment.

[0097] Here, a method of obtaining related content based on the target content is provided. The obtained related content has a more flexible and rich presentation form, which is conducive to increasing the user's interest in the display of the target content. At the same time, the obtained related content is consistent with the theme of the target content, which can avoid affecting the user's viewing experience due to the large difference between the target content and the related content.

[0098] It can be seen from the technical solution provided by the above embodiment of the present application that in the process of training to obtain the target prediction model, the embodiment of the present application combines the characteristics of some direct recommended content with the need to guide the display of indirect recommended content (such as the automatic broadcast material as the direct recommended content guides the display of the feature film as the indirect recommended content), and constructs two types of differences based on the positive sample to adjust the model parameters to improve the adaptability of the target prediction model obtained by training to specific scenarios, thereby facilitating the content recommendation effect. The two types of differences focus on the difference in the correlation between the object dimension information and the content dimension information in the positive set and the sample set (or sample subset), and the difference in the correlation between the first information (sample associated content) and the second information (sample target content) in the positive set and the sample set (or sample subset). Taking into account the association between the content dimension information and the object dimension information as a whole in the sample, and the difference in the association in the positive set relative to the sample set (or sample subset), it is beneficial to dig out the connection between the object and the content, so as to better recommend the associated content and the target content for the target object. At the same time, the association between the first information and the second information in the content dimension information in the sample, as well as the difference between the association in the positive concentration and the association in the entire sample set (or sample subset) are taken into consideration, which is conducive to capturing the connection between the associated content and the target content, so as to improve the success rate of displaying the target content.

[0099] The present application also provides a content processing device, such as Fig.10 As shown, the content processing device 100 includes:

[0100] Acquisition module 1001: used to acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and the display of the associated content is used to guide the display of the target content;

[0101] Prediction module 1002: used to take the object information of the target object and the content information of the content to be recommended as input, output a corresponding target prediction result using a target prediction model, and recommend the content to be recommended based on the target prediction result;

[0102] Wherein, the target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on a positive set in the sample set that carries a label indicating successful guidance. The first difference is the difference between a first correlation and a second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between a third correlation and a fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

[0103] In one embodiment, the preset model includes a first network and a second network; the target prediction model is obtained by training through the following steps: using the first network to output an object vector corresponding to the sample object information, and using the second network to output a first content vector corresponding to the sample content information; using the second network to output a second content vector corresponding to the first type of information, wherein the plurality of the first types of information are obtained by performing a first modality deletion process on the plurality of the sample content information in the positive set, respectively, and the first modality deletion process is used to cover the second information; using the second network to output a third content vector corresponding to the second type of information, wherein the plurality of the second types of information are obtained by performing a second modality deletion process on the plurality of the sample content information in the positive set, respectively, and the second modality deletion process is used to cover the first information; based on the object vector corresponding to each of the sample object information in the positive set and the first content vector corresponding to each of the sample content information in the sample set, the first difference is obtained; based on the third content vector corresponding to each of the second type of information in the positive set and the second content vector corresponding to each of the first type of information in the sample set, the second difference is obtained; based on the first difference and the second difference, the preset model is parameter adjusted to obtain the target prediction model.

[0104] In one embodiment, obtaining the first difference based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set includes: determining the first content vector corresponding to each sample content information in a first candidate set and the first content vector corresponding to each sample content information in the positive set based on the first content vector corresponding to each sample content information in the sample set, the first candidate set being composed of a portion extracted from the sample set; obtaining the first relevance based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information; obtaining the second relevance based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the first candidate set; and determining the first difference between the first relevance and the second relevance.

[0105] In one embodiment, the second difference is obtained based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information in the sample set, including: determining the second content vector corresponding to each first-category information in a second candidate set and the second content vector corresponding to each first-category information in the positive set based on the second content vector corresponding to each first-category information in the sample set, the second candidate set being composed of a part extracted from the sample set; obtaining the third correlation based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information; obtaining the fourth correlation based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information in the second candidate set; and determining the second difference between the third correlation and the fourth correlation.

[0106] In one embodiment, the third difference indicates the difference in prediction results corresponding to two types of samples carrying different labels, and adjusting the parameters of the preset model based on the first difference and the second difference to obtain the target prediction model includes: adjusting the parameters of the preset model based on the first difference, the second difference and the third difference to obtain the target prediction model.

[0107] In one embodiment, the method further includes: for the candidate associated content delivered to the candidate object, taking the time point when the candidate associated content is displayed as the monitoring time starting point, monitoring whether the candidate object performs a preset operation within a preset time length, and the preset operation is used to trigger the candidate target content to be displayed; in the case where the candidate object is monitored to perform the preset operation, determining that the object information of the candidate object is the sample object information, determining that the content information of the candidate associated content and the content information of the candidate target content are the sample content information, and obtaining a positive sample based on the sample object information and the sample content information; in the case where the candidate object is not monitored to perform the preset operation, determining that the object information of the candidate object is the sample object information, determining that the content information of the candidate associated content and the content information of the candidate target content are the sample content information, and obtaining a negative sample based on the sample object information and the sample content information.

[0108] In one embodiment, the determining that the content information of the candidate associated content and the content information of the candidate target content are the sample content information includes: obtaining a first identifier of the candidate associated content and first content description information for the candidate associated content to obtain the first information; obtaining a second identifier of the candidate target content and second content description information for the candidate target content to obtain the second information; and obtaining the sample content information based on the first information and the second information.

[0109] In one embodiment, for the candidate associated content delivered to the candidate object, the time point when the candidate associated content is displayed is used as the monitoring time starting point, and before monitoring whether the candidate object performs a preset operation within a preset time length, the method also includes: when the candidate target content is a target video, extracting at least one candidate video segment from the target video; performing preset processing on the at least one candidate video segment to obtain the candidate associated content; wherein the preset processing includes at least one of the following: splicing, adding visual materials, and audio track editing.

[0110] It should be noted that the device in the device embodiment and the method embodiment are based on the same inventive concept.

[0111] In some embodiments, the functions or modules included in the device provided in the embodiments of the present application can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments. For the sake of brevity, it will not be repeated here.

[0112] The embodiment of the present application also provides a computer-readable storage medium, wherein at least one instruction or at least one program is stored in the computer-readable storage medium, and the at least one instruction or at least one program is loaded and executed by a processor to implement the above method. The computer-readable storage medium may be a non-volatile computer-readable storage medium.

[0113] An embodiment of the present application also provides an electronic device, which includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the at least one processor to implement the above method.

[0114] The electronic device may be provided as a terminal, a server, or a device in other forms.

[0115] Fig.11 1 shows a block diagram of an electronic device according to an embodiment of the present application. For example, the electronic device 1900 can be provided as a server. Fig.11 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0116] The electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output (I / O) interface 1958. The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™ or the like.

[0117] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to perform the above method.

[0118] The present application may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present application.

[0119] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: 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), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0120] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0121] The computer program instructions for performing the operation of the present application can be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C+, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer, partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet). In some embodiments, by using the state information of computer-readable program instructions to personalize electronic circuits, such as programmable logic circuits, field programmable gate arrays (FPGAs) or programmable logic arrays (PLAs), the electronic circuits can execute computer-readable program instructions, thereby realizing various aspects of the present application.

[0122] Various aspects of the present application are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0123] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0124] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0125] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present application. In this regard, each square frame in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the above-mentioned module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the standard function in the square frame can also occur in a sequence different from the standard in the accompanying drawings. For example, two continuous square frames can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square frame in the block diagram and / or flow chart, and the combination of the square frames in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0126] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A content processing method, characterized in that: The method comprises: Acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and the display of the associated content is used to guide the display of the target content; Taking the object information of the target object and the content information of the content to be recommended as input, outputting a corresponding target prediction result using a target prediction model, so as to recommend the content to be recommended based on the target prediction result; Wherein, the target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on a positive set in the sample set that carries a label indicating successful guidance. The first difference is the difference between a first correlation and a second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between a third correlation and a fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

2. The method according to claim 1, characterized in that The preset model includes a first network and a second network; the target prediction model is obtained by training through the following steps: Outputting an object vector corresponding to the sample object information using the first network, and outputting a first content vector corresponding to the sample content information using the second network; Outputting a second content vector corresponding to the first type of information using the second network, wherein the plurality of the first type of information is obtained by performing first modality deletion processing on the plurality of sample content information in the positive set, respectively, and the first modality deletion processing is used to cover up the second information; Outputting a third content vector corresponding to the second type of information using the second network, wherein the plurality of the second types of information are obtained by performing second modality deletion processing on the plurality of the sample content information in the positive set, respectively, and the second modality deletion processing is used to cover the first information; Obtaining the first difference based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set; Obtaining the second difference based on the third content vector corresponding to each piece of the second-category information in the positive set and the second content vector corresponding to each piece of the first-category information in the sample set; Based on the first difference and the second difference, parameters of the preset model are adjusted to obtain the target prediction model.

3. The method according to claim 2, characterized in that The obtaining the first difference based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the sample set includes: Based on the first content vector corresponding to each sample content information in the sample set, determine the first content vector corresponding to each sample content information in a first candidate set and the first content vector corresponding to each sample content information in the positive set, wherein the first candidate set is composed of a portion extracted from the sample set; Obtaining the first relevance based on the object vector corresponding to each piece of sample object information and the first content vector corresponding to each piece of sample content information in the positive set; Obtaining the second relevance based on the object vector corresponding to each sample object information in the positive set and the first content vector corresponding to each sample content information in the first candidate set; The first difference between the first correlation and the second correlation is determined.

4. The method according to claim 2, characterized in that: The obtaining the second difference based on the third content vector corresponding to each second-category information in the positive set and the second content vector corresponding to each first-category information in the sample set includes: Based on the second content vector corresponding to each piece of the first category of information in the sample set, determine the second content vector corresponding to each piece of the first category of information in a second candidate set and the second content vector corresponding to each piece of the first category of information in the positive set, wherein the second candidate set is composed of a portion extracted from the sample set; Obtaining the third relevance based on the third content vector corresponding to each piece of the second-category information in the positive set and the second content vector corresponding to each piece of the first-category information; Obtain the fourth relevance based on the third content vector corresponding to each piece of the second-category information in the positive set and the second content vector corresponding to each piece of the first-category information in the second candidate set; The second difference between the third correlation and the fourth correlation is determined.

5. The method according to claim 2, characterized in that: The third difference indicates a difference in prediction results corresponding to two types of samples with different labels, and the parameter adjustment of the preset model based on the first difference and the second difference to obtain the target prediction model includes: Based on the first difference, the second difference and the third difference, parameters of the preset model are adjusted to obtain the target prediction model.

6. The method according to claim 2, characterized in that The method further comprises: For the candidate associated content delivered to the candidate object, the time point when the candidate associated content is displayed is used as the monitoring time starting point, and whether the candidate object performs a preset operation within a preset time length is monitored, and the preset operation is used to trigger the candidate target content to be displayed; In the case where it is monitored that the candidate object performs the preset operation, determining the object information of the candidate object as the sample object information, determining the content information of the candidate associated content and the content information of the candidate target content as the sample content information, and obtaining a positive sample based on the sample object information and the sample content information; In the case that the candidate object is not detected to perform the preset operation, the object information of the candidate object is determined to be the sample object information, the content information of the candidate associated content and the content information of the candidate target content are determined to be the sample content information, and a negative sample is obtained based on the sample object information and the sample content information.

7. The method according to claim 6, characterized in that The step of determining the content information of the candidate associated content and the content information of the candidate target content as the sample content information includes: Acquire a first identifier of the candidate associated content and first content description information of the candidate associated content to obtain the first information; Acquire a second identifier of the candidate target content and second content description information of the candidate target content to obtain the second information; The sample content information is obtained based on the first information and the second information.

8. The method according to claim 6, characterized in that For the candidate associated content delivered to the candidate object, taking the time point when the candidate associated content is displayed as the monitoring time starting point, monitoring whether the candidate object performs a preset operation within a preset time length, the method also includes: In a case where the candidate target content is a target video, extracting at least one candidate video segment from the target video; Performing preset processing on the at least one candidate video segment to obtain the candidate associated content; The preset processing includes at least one of the following: splicing, adding visual materials, and editing audio tracks.

9. The method according to claim 1, characterized in that: The recommending the content to be recommended based on the target prediction result includes: In a case where it is determined based on the target prediction result that the content to be recommended meets the preset recommendation requirement, configuring a first display mode for the associated content and a second display mode for the target content, wherein the first display mode indicates that the associated content is automatically played through a preset window, and the second display mode indicates that the playing of the target content is triggered based on a preset operation for the associated content; The content to be recommended is recommended to the target object based on the first presentation mode and the second presentation model.

10. A content processing device, characterized in that: The device comprises: Acquisition module: used to acquire content to be recommended for a target object, wherein the content to be recommended includes associated content and target content, and the display of the associated content is used to guide the display of the target content; Prediction module: used for taking the object information of the target object and the content information of the content to be recommended as input, outputting the corresponding target prediction result by using the target prediction model, so as to recommend the content to be recommended based on the target prediction result; Wherein, the target prediction model is obtained by training based on a sample set and adjusting the parameters of a preset model based on a first difference and a second difference during training. Each sample in the sample set is composed of sample object information and sample content information. The sample content information includes first information of sample-related content and second information of sample target content. The sample carries a label for describing whether the display of the sample-related content successfully guides the display of the sample target content. The first difference and the second difference are based on a positive set in the sample set that carries a label indicating successful guidance. The first difference is the difference between a first correlation and a second correlation. The first correlation represents the correlation between the object dimension information and the content dimension information in the positive set. The second correlation represents the correlation between the object dimension information in the positive set and the content dimension information in the sample set. The second difference is the difference between a third correlation and a fourth correlation. The third correlation represents the correlation between the first information and the second information in the positive set. The fourth correlation represents the correlation between the second information in the positive set and the first information in the sample set.

11. An electronic device, characterized in that: The electronic device includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the at least one processor to implement the content processing method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement the content processing method according to any one of claims 1 to 9.

Images