Training method of data processing model, data processing method and device

Abstract

The present disclosure provides a data processing model training method and device, and a data processing method and device, relates to the technical field of data processing, and in particular to large model technology and content recommendation technology. The implementation scheme is as follows: determining a training sample set, wherein each training sample in the training sample set comprises spliced sample content and a review text for the sample content; determining at least one matching sample matched with the training sample; encoding the training sample by using a data processing model having current parameters to obtain a representation vector of the training sample; encoding the at least one matching sample by using the data processing model having the current parameters to obtain a representation vector of the matching sample; and adjusting the current parameters of the data processing model to minimize the loss between the representation vector of the training sample and the representation vector of the matching sample.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of data processing, in particular to large model technology and content recommendation technology, and specifically to a data processing model training method and device, a data processing method and device, an electronic device, a computer readable storage medium, and a computer program product. BACKGROUND

[0002] Artificial intelligence is a discipline that studies enabling computers to simulate some thinking processes and intelligent behaviors of humans (such as learning, reasoning, thinking, planning, etc.), and includes both hardware and software technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, and big data processing. Artificial intelligence software technologies mainly include computer vision technology, speech recognition technology, natural language processing technology, and machine learning / deep learning, big data processing technology, knowledge graph technology, and several other major directions.

[0003] Content representation vector construction aims to convert the content to be recommended into a vector that can be used for a recommendation algorithm, and is an important link in a recommendation system.

[0004] The methods described in this section are not necessarily the methods that have been previously conceived or employed. Unless otherwise indicated, it should not be assumed that any of the methods described in this section are considered to be prior art merely because of their inclusion in this section. Similarly, issues mentioned in this section should not be assumed to have been admitted to be prior art in any jurisdiction unless otherwise indicated. SUMMARY

[0005] The present disclosure provides a data processing model training method, a data processing method, a device, an electronic device, a computer readable storage medium, and a computer program product.

[0006] According to an aspect of the present disclosure, a data processing model training method is provided, including: determining a training sample set, wherein each training sample in the training sample set includes spliced sample content and review text for the sample content; determining at least one matching sample matching the training sample; encoding the training sample using a data processing model having current parameters to obtain a representation vector of the training sample; encoding the at least one matching sample using the data processing model having the current parameters to obtain a representation vector of the matching sample; and adjusting the current parameters of the data processing model to minimize the loss between the representation vector of the training sample and the representation vector of the matching sample.

[0007] According to another aspect of the present disclosure, a data processing method is provided, including: determining to-be-processed content and comment text for the to-be-processed content; splicing the to-be-processed content and the comment text to obtain to-be-encoded content; and encoding the to-be-encoded content by using a data processing model to obtain a representation vector for the to-be-processed content, wherein the data processing model is trained by using the training method as described above.

[0008] According to another aspect of the present disclosure, a training apparatus of a data processing model is provided, including: a sample determination unit configured to determine a training sample set, wherein each training sample in the training sample set includes spliced sample content and comment text for the sample content; a matching sample determination unit configured to determine at least one matching sample matched with the training sample; an encoding unit configured to encode the training sample by using a data processing model with current parameters to obtain a representation vector of the training sample, and encode the at least one matching sample by using the data processing model with the current parameters to obtain a representation vector of the matching sample; and a parameter adjustment unit configured to adjust the current parameters of the data processing model to minimize the loss between the representation vector of the training sample and the representation vector of the matching sample.

[0009] According to another aspect of the present disclosure, a data processing apparatus is provided, including: an input unit configured to determine to-be-processed content and comment text for the to-be-processed content; a splicing unit configured to splice the to-be-processed content and the comment text to obtain to-be-encoded content; and an encoding unit configured to encode the to-be-encoded content by using a data processing model to obtain a representation vector for the to-be-processed content, wherein the data processing model is trained by using the training method as described above.

[0010] According to another aspect of the present disclosure, an electronic device is also provided, including: at least one processor; and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the embodiments of the present disclosure.

[0011] According to another aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is also provided, and the computer instructions are used to make the computer perform the method according to the embodiments of the present disclosure.

[0012] According to another aspect of the present disclosure, a computer program product is also provided, including a computer program, wherein the computer program, when executed by a processor, implements the method according to the embodiments of the present disclosure.

[0013] According to another aspect of the present disclosure, there is also provided a content recommendation system, comprising: at least one processor; and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the embodiments of the present disclosure.

[0014] According to one or more embodiments of the present disclosure, the model can be enabled to encode both the content itself and the comment text for the content together when generating the representation vector of the content, so that the comment text for the content is taken as supplementary information when generating the representation vector of the content, and the accuracy and comprehensiveness of the content representation vector are improved.

[0015] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments and together with the description serve to explain exemplary implementations of the application. The illustrated embodiments are merely examples and do not limit the scope of the claims. In all the drawings, like reference numerals refer to like elements throughout the accompanying drawings.

[0017] Figure 1 A schematic diagram of an exemplary system in which the various methods described herein can be implemented according to embodiments of the present disclosure is shown;

[0018] Figure 2 An exemplary process of a training method of a data processing model according to embodiments of the present disclosure is shown;

[0019] Figure 3 An exemplary flowchart of a data processing method according to embodiments of the present disclosure is shown;

[0020] Figure 4 A use scenario of a content recommendation method according to embodiments of the present disclosure is shown;

[0021] Figure 5 An exemplary block diagram of a training device of a data processing model according to embodiments of the present disclosure is shown;

[0022] Figure 6 An exemplary block diagram of a data processing device according to embodiments of the present disclosure is shown;

[0023] Figure 7 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0024] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of the embodiments of the present disclosure to assist in understanding them. These should be considered in their context only. Thus, those of ordinary skill in the art will recognize the various changes and modifications of the embodiments described herein, which do not depart from the scope of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for the sake of clarity and conciseness.

[0025] In the present disclosure, the terms "first", "second", and the like are used to describe various elements only and do not intend to limit the positional relationship, the chronological relationship, or the importance of the elements, and such terms are used only to distinguish one element from another. In some examples, the first element and the second element can refer to the same instance of the element, and in some cases, based on the context of the description, they can also refer to different instances.

[0026] The terms used in the description of various described examples in the present disclosure are only for the purpose of describing particular examples and are not intended to be limiting. Unless the context clearly indicates otherwise, the term can be one or more if the number of elements is not specifically limited. In addition, the term "and / or" used in the present disclosure encompasses any one of the listed items and all possible combinations thereof.

[0027] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0028] Figure 1 A schematic diagram of an example system 100 in which various methods and apparatus described herein can be implemented according to embodiments of the present disclosure is shown. Referring to Figure 1 , the system 100 includes one or more client devices 101, 102, 103, 104, 105, and 106, a server 120, and one or more communication networks 110 coupling the one or more client devices to the server 120. The client devices 101, 102, 103, 104, 105, and 106 can be configured to execute one or more application programs.

[0029] In embodiments of the present disclosure, the server 120 can run one or more services or software applications that enable the execution of methods according to embodiments of the present disclosure.

[0030] In certain embodiments, the server 120 can also provide other services or software applications, which can include non-virtual and virtual environments. In certain embodiments, these services can be provided as web-based services or cloud services, for example, under a software as a service (SaaS) model to users of the client devices 101, 102, 103, 104, 105, and / or 106.

[0031] In Figure 1 In the illustrated configuration, the server 120 can include one or more components that implement the functionality performed by the server 120. These components can include software components that are executable by one or more processors, hardware components, or combinations thereof. Users operating the client devices 101, 102, 103, 104, 105, and / or 106 can in turn utilize one or more client applications to interact with the server 120 to utilize the services provided by the components. It should be understood that a wide variety of system configurations are possible, which can differ from system 100. Therefore, Figure 1 is one example of a system for implementing the various methods described herein and is not intended to be limiting.

[0032] The client devices 101, 102, 103, 104, 105, and / or 106 can be used by users to take user inputs and provide output information to the users. The client devices can provide interfaces that enable users of the client devices to interact with the client devices. The client devices can also output information to the users via the interfaces. Although Figure 1 Only six client devices are depicted, but those skilled in the art will understand that the present disclosure can support any number of client devices.

[0033] Client devices 101, 102, 103, 104, 105, and / or 106 can include various types of computer devices, such as portable handheld devices, general purpose computers (such as personal computers and laptop computers), workstation computers, wearable devices, smart screen devices, self-service kiosk devices, service robots, gaming systems, thin clients, various messaging devices, sensors or other sensing devices, and the like. These computer devices can run various types and versions of software applications and operating systems, such as MICROSOFT Windows, APPLE iOS, UNIX-like operating systems, Linux or Linux-like operating systems (such as GOOGLE Chrome OS); or including various mobile operating systems, such as MICROSOFT Windows Mobile OS, iOS, Windows Phone, Android. Portable handheld devices can include cellular telephones, smartphones, tablet computers, personal digital assistants (PDAs), and the like. Wearable devices can include head-mounted displays (such as smart glasses) and other devices. Gaming systems can include various handheld gaming devices, Internet-enabled gaming devices, and the like. Client devices are capable of executing a variety of different applications, such as various Internet-related applications, communication applications (such as email applications), short message service (SMS) applications, and can use various communication protocols.

[0034] Network 110 can be any type of network familiar to those skilled in the art that can support data communications using any of a variety of available protocols, including without limitation TCP / IP, SNA, IPX, etc. As examples, one or more of networks 110 can be a LAN, an Ethernet network, a Token Ring network, a WAN, the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, a local area network (LAN), a wide area network (WAN), a wireless network, a public switched telephone network (PSTN), an infrared network, a wireless network (e.g., a Bluetooth network), and / or any combination of these and / or other networks.

[0035] Server 120 can include one or more general purpose computers, special purpose server computers (e.g., PC (personal computer) servers, UNIX servers, mid-range servers), blade servers, large mainframe computers, server clusters, or any other appropriate arrangement and / or combination. Server 120 can include one or more virtual machines running virtual operating systems, or other computing architectures involving virtualization (e.g., one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices for servers). In various embodiments, server 120 can run one or more services or software applications that provide the functionality described below.

[0036] The computing units in the server 120 can run one or more operating systems including any of the operating systems described above, as well as any commercially available server operating systems. Server 120 can also run any of a variety of additional server applications and / or mid-tier applications, including HTTP servers, FTP servers, CGI servers, JAVA servers, database servers, etc.

[0037] In some embodiments, the server 120 can include one or more applications to analyze and consolidate data feeds and / or event updates from users of the client devices 101, 102, 103, 104, 105, and / or 106. The server 120 can also include one or more applications to display the data feeds and / or real-time events via one or more display devices of the client devices 101, 102, 103, 104, 105, and / or 106.

[0038] In some embodiments, the server 120 can be a server of a distributed system, or a server combined with a blockchain. The server 120 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. The cloud server is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and virtual private server (VPS, Virtual Private Server) services.

[0039] The system 100 can also include one or more databases 130. In certain embodiments, these databases can be used to store data and other information. For example, one or more of the databases 130 can be used to store information such as audio files and video files. The databases 130 can reside in a variety of locations. For example, databases used by the server 120 can reside locally to the server 120, or can be remote from the server 120 and can communicate with the server 120 via a network- or application-specific connection. The databases 130 can be of different types. In certain embodiments, databases used by the server 120 can be, for example, relational databases. One or more of these databases can store, update, and retrieve data to and from the databases in response to commands.

[0040] In certain embodiments, one or more of the databases 130 can also be used by applications to store application data. Databases used by applications can be different types of databases, such as key-value stores, object stores, or regular stores backed by file systems.

[0041] Figure 1The system 100 can be configured and operated in various ways to enable the application of various methods and apparatuses described in accordance with the present disclosure.

[0042] In a content recommendation system, the construction of content representation vectors aims to convert items (such as articles, videos, etc.) into vectors that can be used in recommendation algorithms. The construction of content representation vectors is a crucial link in recommendation systems. The construction of content representation vectors mainly relies on the text description, image information of the content itself, or the historical interaction behavior between the user and the content.

[0043] However, these traditional representation methods largely ignore the importance of interactive comments, a key piece of information. In fact, user comments not only reflect the user's intuitive feedback and initial impression of the pushed content, but also reveal the depth and breadth of the user's interest. User comments are not just a collection of words; they contain the user's deep understanding and multi-dimensional views of the pushed content, which can go beyond the expression information of the content itself, further reflecting personal preferences, emotional tendencies, values, and even social and cultural backgrounds. Therefore, user comments are a valuable and effective supplement to content representation vectors. The information of user comments can provide a more rich, three-dimensional, and user-need-and-emotion-experience-oriented description for content vectors.

[0044] Therefore, by including user comments in the consideration of content representation, not only can the personalization level of the recommendation system be improved, making it more accurately match the user's personalized needs and interest points, but also the user's recognition and satisfaction with the recommended content can be enhanced, thereby building a more positive and healthy interactive relationship between the user and the recommendation system.

[0045] Although user comments can be an effective supplement to the representation of content vectors, due to the unstructured, diverse, and complex nature of comment information, it is difficult to effectively extract, understand, and represent key information and user emotions in the comments, which in turn affects the accuracy and effectiveness of content representation vectors.

[0046] In view of this, the present disclosure proposes a new model training method and data processing method for generating content representation vectors. The method according to the embodiments of the present disclosure aims to comprehensively consider the text information of the content itself (such as title, abstract, body, etc.), the content interaction behavior of the user (such as clicking, browsing time, liking, sharing, etc.), and the user comments in the content comment area, and by utilizing the powerful capabilities of pre-trained data processing models such as large language models, these information is deeply mined and integrated, thereby generating more accurate and comprehensive content representation vectors. This integration strategy not only enhances the expression ability of content representation, but also helps the recommendation system to better understand user preferences, thereby improving the personalization and relevance of recommendations, optimizing user experience, and ultimately achieving significant improvement in recommendation effectiveness.

[0047] Figure 2 An exemplary process of a training method of a data processing model according to an embodiment of the present disclosure is shown.

[0048] As shown in step S202, a training sample set is determined, wherein each training sample in the training sample set comprises spliced sample content and review text for the sample content. Figure 2

[0049] In step S204, at least one matching sample matching the training sample is determined.

[0050] In step S206, the training sample is encoded by the data processing model with the current parameters to obtain a representation vector of the training sample. The at least one matching sample is encoded by the data processing model with the current parameters to obtain a representation vector of the matching sample.

[0051] In step S208, the current parameters of the data processing model are adjusted to minimize the loss between the representation vector of the training sample and the representation vector of the matching sample.

[0052] The training method of the data processing model provided by the embodiments of the present disclosure can enable the model to encode the content itself and the review text for the content together when generating the representation vector of the content, so that the review text for the content is taken as supplementary information when generating the representation vector of the content, and the accuracy and comprehensiveness of the content representation vector are improved.

[0053] The principles of the present disclosure will be described in detail below.

[0054] In step S202, a training sample set is determined, wherein each training sample in the training sample set comprises spliced sample content and review text for the sample content.

[0055] The sample content can include any combination of text, image, audio, video, etc. The specific form of the sample content is not limited in the present disclosure.

[0056] In step S204, at least one matching sample matching the training sample is determined.

[0057] In some embodiments, step S204 can include determining the first sample as the matching sample of the training sample in response to determining that the target user of the first sample and the target user of the training sample are matched. The target user of the first sample is determined according to the user interaction associated with the first sample, and the target user of the training sample is determined according to the user interaction associated with the training sample.

[0058] ​The matching target user mentioned here refers to the same target user between two samples being higher than a predetermined threshold, such as a predetermined user quantity threshold or a user proportion threshold. In the above process, it is not necessary to consider whether the actual content of the two training samples is the same or similar. When the content of the two samples has the same or similar target user, it can be considered that the content of the two samples is substantially similar in the sense of content recommendation. By determining the samples with matching target users as matching samples in the training process, the model can be trained so that the content representation vectors generated by the model have similar representations for content with similar user interaction situations.

[0059] In an example, the user interaction can include at least one of the following: browsing, viewing duration, collection, comment, sharing, purchase, rating, etc. Those skilled in the art can determine the specific content of the user interaction according to the actual application scenario. It can be understood that other user interaction modes can also be specified according to actual conditions without departing from the principles of the present disclosure.

[0060] In some embodiments, a user who has user interaction with the sample content can be determined as a target user. In other embodiments, the users who interact with the sample content can be further screened to determine that the users who have a higher interest in the sample content are target users, thereby improving the accuracy of determining matching samples based on user interaction.

[0061] In an example, the interaction history of at least one candidate user who has user interaction with the training sample can be counted to obtain the interest level of each candidate user for the training sample. Then, the target user can be determined from the at least one candidate user, wherein the target user has an interest level higher than a predetermined interest threshold.

[0062] For example, taking the user interaction as viewing duration as an example, the viewing duration of the user can be taken as the interest level of the user for the sample content, and the users of the sample content can be sorted in descending order of viewing duration. The exemplary interest threshold can be a viewing duration threshold, or a sorting order threshold of all users of the sample content. For example, the viewing duration threshold can be determined as 10 minutes. The sorting order threshold can be determined as the top 20%.

[0063] When the user interactions include multiple different types, the different types of user interactions can be normalized based on predetermined rules, and the level of interest of the user in the sample content can be determined by weighted combination of the normalized scores of the different types of user interactions. For example, the user interactions can include the number of views, the viewing duration, and the sharing. For the number of views, the number of views below 5 times can be determined as low frequency, and the normalized score is 1. The number of views not less than 5 times and less than 10 times can be determined as medium frequency, and the normalized score is 3. The number of views not less than 10 times can be determined as high frequency, and the normalized score is 5. For the viewing duration, the normalized score of the viewing duration within 10 minutes can be determined as 1, the normalized score of the viewing duration from 10 to 30 minutes can be determined as 3, and the normalized score of the viewing duration higher than 30 minutes can be determined as 5. For the sharing, the normalized score of the sharing behavior can be determined as 5, and the normalized score of the non-sharing behavior can be determined as 0. The normalized scores of the above different user interaction behaviors can be weighted and summed as the interest score of the user in the sample content. Then, the users of the sample content can be sorted according to the interest score, and the users whose interest scores are higher than a predetermined score threshold or are ranked higher than a ranking order threshold among all users can be determined as the target users of the training sample.

[0064] It can be understood that the above rules for determining the interest score of the user are only exemplary descriptions. Those skilled in the art can set any suitable rules for determining the interest level according to the user interactions according to actual conditions.

[0065] In some embodiments, step S204 can include determining a text similarity between the text information of the training sample and the text information of the second sample. In response to the text similarity being higher than a predetermined similarity threshold, the second sample is determined as a matching sample of the training sample. The matching sample determined by the text similarity can enable the content representation vector generated by the model to have similar representations for semantically similar contents.

[0066] For sample content in the form of text, the text information can be the text itself of the sample content. For sample content in the form of image, video, audio, etc., the text information can be the text obtained by character recognition from the image, the text recognized by voice recognition from the audio, the text in the subtitle information of the video, etc.

[0067] In some examples, the second sample can be content of other training samples. The text with similarity to the text information of the training sample higher than a predetermined similarity threshold can be determined as the second text by any suitable semantic similarity determination manner. In some other examples, the second sample can be generated by scrambling the text information of the training sample. In examples, part of the text in the text content can be replaced by text in other content. For example, the text information of the training text can be separated by sentence, and a certain proportion of the sentences can be randomly selected and replaced by sentences in other content. The above-mentioned proportion can be determined according to the predetermined similarity threshold. For example, 70% of the sentences can be randomly selected and replaced by sentences in other content. The proportion of the selected sentences for replacement when performing text replacement can be between 50% and 90%, more preferably between 60% and 80%. Too high a replacement proportion will result in the text obtained after replacement completely losing similarity, while too low a replacement proportion will result in overfitting of the results output by the data processing model.

[0068] In some embodiments, the first sample or a combination of the first sample and the second sample can be used as the matching sample of the training sample. In the process of constructing the content representation vector, in addition to the requirement that the representation vectors of the matching samples with the same or highly similar content should be as close as possible to reflect their semantic equivalence or high correlation, the importance of user behavior patterns and interaction habits in recommendation also needs to be considered. The representation vectors of those contents with similar user consumption can also maintain high similarity, thereby generating a representation vector that can reflect the similarity of user interaction habits.

[0069] In examples, the number of second samples can be less than the number of first samples. In examples, the proportion of first samples can be 70%, and the proportion of second samples can be 30%. Those skilled in the art can adjust the number proportion of first samples and second samples according to actual conditions. By adjusting the number proportion of first samples and second samples, it can be controlled whether the model generates content representation vectors more considering the similarity of target users between contents or the semantic similarity between contents.

[0070] In step S206, the data processing model with the current parameters is used to encode the training sample to obtain the representation vector of the training sample. Step S206 also includes using the data processing model with the current parameters to encode at least one matching sample to obtain the representation vector of the matching sample.

[0071] As described previously, the training target of the data processing model is to make the representation vectors of the training sample and its matching sample as similar as possible.

[0072] The data processing model used in the method of the embodiments of the present disclosure can be a large model, such as a large language model (LLM) or a visual large model (CV), etc. Without departing from the principles of the present disclosure, any suitable data processing model can be used to implement the method provided by the present disclosure.

[0073] In some embodiments, in the case where the content of the training sample is text, the data processing model with the current parameters can be used to encode the text information of the training sample, wherein the text information of the training sample includes the spliced content text and the review text. By using the above method, the content of the review text can be used to effectively supplement the content representation vector. In some implementations, the content text can include the title, category, interest point, abstract, etc. of the content. Similarly, the data processing model can be used to encode the text information of the matching sample, wherein the text information of the matching sample includes the content text corresponding to the matching sample and the result of splicing the review text for the content text of the matching sample.

[0074] In other embodiments, when the content of the training sample contains information such as images, audio, etc. in addition to text, the vector representation of the multi-modal information of the content and the embedding vector of the review text can be spliced, and the data processing model with the current parameters can be used to encode the result obtained by splicing the information of the sample content and the review text to obtain the representation vector of the training sample. Various suitable methods can be used to vectorize the multi-modal information in the sample content, and the specific vectorization method is not limited herein.

[0075] In some implementations, the text length of the review text can be less than or equal to the text length of the content text. As mentioned earlier, the information of the review text can supplement the information contained in the content representation vector when encoding, but since there is a certain difference between the content of the review text and the sample content, the length of the review text needs to be controlled to avoid the final obtained content representation vector being unable to correctly express the information of the content itself.

[0076] In an example, the review text for the sample content can be obtained by the following method: for each piece of review information for the sample content, determining a review quality score of the review information; determining at least one piece of review information with the highest score from the multiple pieces of review information in descending order of the review quality score as the review text, wherein the sum of the text lengths of the determined at least one piece of review information with the highest score is less than or equal to the text length of the content text.

[0077] In order to introduce the content vector representation of the user comments, the user comments need to be cleaned. First, even for the same content, the quality of each comment is uneven, so an effective screening mechanism needs to be established to eliminate irrelevant, repetitive or low-quality comments from the massive comments, and to ensure that the effective comments with reference value and reflecting the user's demand and opinion are finally retained. Secondly, due to the differences in the popularity and topic nature of different contents, the number of comments attracted by different contents will also be different. Therefore, when obtaining the comment text, the number of comments added needs to be reasonably set according to the information amount of the content, so as to avoid the overload of comment information on the basis of ensuring that the comment information effectively supplements the content representation.

[0078] For example, the comment quality score can be obtained by linearly adding the like amount, the dislike amount and the reply amount according to the comment posterior index, and the high-score comment is a high-quality comment. Then, all the comments under the content are sorted in descending order according to the score, and the comment with the highest score is obtained from high to low according to the score until the ratio of the total number of comments to the number of text of the content itself is close to 1. In some other implementation modes, one or more comments with the highest score can be selected from all the comments according to the score without sorting until the ratio of the total number of selected comments to the number of text of the content itself is close to 1.

[0079] In step S208, the current parameters of the data processing model are adjusted to minimize the loss between the representation vectors of the training samples and the representation vectors of the matching samples. The above method can minimize the difference between the representation vectors of the two matching samples.

[0080] In some embodiments, the loss between the representation vectors of the training samples and the representation vectors of the matching samples minimizes the similarity between the negative samples while maximizing the similarity between the matching samples. Using the above method, similar data samples can be pulled closer and dissimilar data samples can be pushed farther apart. In an example, the above loss can be an InfoNCE loss, and the mathematical expression is:

[0081]

[0082] where z i and z j are a pair of matching positive samples, N is the number of samples, sim is a similarity function, and γ is a predetermined parameter for controlling the convergence speed. z i and z k are a pair of non-matching negative samples. In an embodiment of the present disclosure, for the training sample z i , any other training sample in the training data set that is not its matching sample can be randomly selected as a negative sample.

[0083] By using the method provided by the embodiment of the present disclosure, a content representation method is provided, which fuses content itself information, user behavior and user comments by using a data processing model. The content representation vector generated by the method is closer to user preference and performs excellently in stimulating user interaction.

[0084] Figure 3 An exemplary flowchart of a data processing method according to an embodiment of the present disclosure is shown.

[0085] As shown in Figure 3 In step S302, the content to be processed and the comment text for the content to be processed can be determined.

[0086] In step S304, the content to be processed and the comment text can be spliced to obtain the content to be encoded.

[0087] In step S306, the data processing model can be used to encode the content to be encoded to obtain the representation vector for the content to be processed. The data processing model is trained by using the training method described in combination Figure 2 with the present disclosure.

[0088] In some embodiments, the data processing method 300 can further include generating recommended content for a user according to the similarity between the representation vector of the content to be processed and the representation vector of the user's interest. The content representation vector obtained by using the method of the embodiment of the present disclosure can be used for content recommendation. In an example, the representation vector of the user's interest can include the content representation vector of the content consumed by the user. In the content recommendation method, the existing content can be input into the trained model one by one, and their respective content representation vectors are generated. Next, the similarity between these content representation vectors and the content representation vector of the content consumed by the user is calculated, and a similarity score with the content consumed by the user is generated for each content. Finally, based on these similarity scores, the content with higher similarity scores, i.e., the content most similar to or most consistent with the user's interest, can be recommended to the user.

[0089] In some embodiments, similar to the way of screening the comment text in the training method, the text length of the comment text is less than or equal to the text length of the content text of the content to be processed. The comment text for the content text can be obtained by: determining a comment quality score of each piece of comment information in the plurality of pieces of comment information for the content to be processed; sorting the plurality of pieces of comment information based on the comment quality score; and determining at least one piece of comment information with the highest ranking in the sorted plurality of pieces of comment information as the comment text, wherein the sum of the text lengths of the at least one piece of comment information with the highest ranking is less than or equal to the text length of the content text.

[0090] Figure 4The following illustrates a use case of the content recommendation method according to an embodiment of the present disclosure.

[0091] like Figure 4 As shown, in recommendation scenarios 401, such as news and social media recommendations with comments, user interaction content 402 containing comments can be obtained. This can be achieved by combining... Figure 2 , Figure 3 The described method involves filtering high-quality comments 406 and target users 407 for each piece of content based on user interaction content 402, and constructing a training sample set 403 for model tuning. The training sample set 403 can be used to fine-tune the large language model, enabling the fine-tuned large language model 404 to consider comment content as supplementary information when generating content representation vectors 405, and to identify similarities between content with similar target users. The content representation vectors 405 generated by the fine-tuned large model can then be used for content recommendation in recommendation scenario 401.

[0092] Figure 5 An exemplary block diagram of a training apparatus for a data processing model according to an embodiment of the present disclosure is shown. It can be utilized... Figure 5 The described training device achieves the combination Figure 2 The training method described.

[0093] like Figure 5 As shown, the training device 500 may include a sample determination unit 510, a matching sample determination unit 520, an encoding unit 530, and a parameter adjustment unit 540.

[0094] The sample determination unit 510 can be configured to determine a training sample set, wherein each training sample in the training sample set includes spliced ​​sample content and comment text for the sample content.

[0095] The matching sample determination unit 520 can be configured to determine at least one matching sample that matches the training sample.

[0096] The encoding unit 530 can be configured to encode training samples using a data processing model with current parameters to obtain a representation vector of the training samples, and to encode at least one matching sample using a data processing model with current parameters to obtain a representation vector of the matching sample.

[0097] The parameter tuning unit 540 can be configured to tune the current parameters of the data processing model to minimize the loss between the representation vectors of the training samples and the representation vectors of the matched samples.

[0098] In some embodiments, encoding the training sample by using the data processing model with the current parameters comprises: encoding text information of the training sample by using the data processing model with the current parameters, wherein the text information of the training sample comprises the concatenated content text and the review text.

[0099] In some embodiments, the text length of the review text is less than or equal to the text length of the content text.

[0100] In some embodiments, the review text is obtained by: for each piece of review information in the plurality of pieces of review information for the sample content, determining a review quality score of the review information; sorting the plurality of pieces of review information based on the review quality scores; and determining at least one piece of review information with the highest ranking in the sorted plurality of pieces of review information as the review text, wherein the sum of the text lengths of the at least one piece of review information with the highest ranking is less than or equal to the text length of the content text.

[0101] In some embodiments, determining at least one matching sample matching the training sample comprises: in response to determining that the target user of the first sample and the target user of the training sample are matched, determining the first sample as the matching sample of the training sample, wherein the target user of the first sample is determined according to user interactions associated with the first sample, and the target user of the training sample is determined according to user interactions associated with the training sample.

[0102] In some embodiments, the user interactions comprise at least one of the following: browsing, viewing duration, collecting, commenting, sharing, purchasing, rating.

[0103] In some embodiments, the target user of the training sample is determined by: counting interaction histories of at least one candidate user interacting with the training sample to obtain an interest level of each candidate user for the training sample; and determining the target user from the at least one candidate user, wherein the target user has an interest level higher than a predetermined interest threshold.

[0104] In some embodiments, determining at least one matching sample matching the training sample further comprises: determining a text similarity between the text information of the training sample and the text information of the second sample, and in response to the text similarity being higher than a predetermined similarity threshold, determining the second sample as the matching sample of the training sample.

[0105] In some embodiments, the second sample is generated by scrambling the text information of the training sample.

[0106] In some embodiments, the number of the second samples is less than the number of the first samples.

[0107] In some embodiments, the loss between the representation vectors of the training samples and the representation vectors of the matching samples minimizes the similarity between the negative samples while maximizing the similarity between the matching samples.

[0108] In some embodiments, the loss is an InfoNCE loss.

[0109] In some embodiments, the data processing model is a large language model.

[0110] It should be understood that Figure 5 The various modules or units of the apparatus 500 shown in FIG. 5 can correspond to the various steps in the method 200 described with reference to Figure 2 The operations, features and advantages described above for the method 200 apply equally to the apparatus 500 and the modules and units included therein. For the sake of brevity, certain operations, features and advantages are not described again here.

[0111] Figure 6 An exemplary block diagram of a data processing apparatus according to embodiments of the disclosure is shown. The data processing apparatus shown in Figure 6 may be utilized to implement the data processing method described in conjunction with Figure 3 .

[0112] As shown in Figure 6 , the data processing apparatus 600 can include an input unit 610, a concatenating unit 620, and an encoding unit 630.

[0113] The input unit 610 can be configured to determine content to be processed and comment text for the content to be processed.

[0114] The concatenating unit 620 can be configured to concatenate the content to be processed and the comment text to obtain content to be encoded.

[0115] The encoding unit 630 can be configured to encode the content to be encoded using a data processing model to obtain a representation vector for the content to be processed, wherein the data processing model is trained using the training method described in conjunction with Figure 2 .

[0116] In some embodiments, the data processing apparatus 600 further includes a recommendation unit configured to generate recommended content for a user according to the similarity between the representation vector of the content to be processed and a representation vector of interest of the user.

[0117] In some embodiments, the text length of the comment text is less than or equal to the text length of the content text of the content to be processed.

[0118] In some embodiments, the comment text is obtained by: for each piece of comment information in the plurality of pieces of comment information for the content to be processed, determining a comment quality score of the comment information; sorting the plurality of pieces of comment information based on the comment quality scores; and determining at least one piece of comment information with the highest ranking in the sorted plurality of pieces of comment information as the comment text, wherein a sum of text lengths of the at least one piece of comment information with the highest ranking is less than or equal to the text length of the content text.

[0119] It should be understood that Figure 6 The various modules or units of the apparatus 600 shown in FIG. 6 can correspond to the various steps in the method 300 described with reference to FIG. 3. Thus, the operations, features and advantages described above for the method 300 apply equally to the apparatus 600 and the modules and units included therein. For the sake of brevity, certain operations, features and advantages are not described again here. Figure 3

[0120] Although specific functions are discussed above with reference to particular modules, it should be noted that the functions of the various units discussed herein can be divided among more units, and / or at least some of the functions of multiple units can be combined into a single unit.

[0121] It should also be understood that the techniques described herein can be described in the general context of software hardware elements or program modules. The various units described above with respect to Figure 5 , Figure 6 may be implemented in hardware or in hardware combined with software and / or firmware. For example, the units can be implemented as computer program code / instructions configured to be executed in one or more processors and stored in a computer-readable storage medium. Alternatively, the units can be implemented as hardware logic / circuitry. For example, in some embodiments, one or more of the units 510-540, 610-630 can be implemented together in a System on Chip (SoC). The SoC can include an integrated circuit chip (which includes one or more of a processor (e.g., a Central Processing Unit (CPU), a microcontroller, a microprocessor, a Digital Signal Processor (DSP), etc.), a memory, one or more communication interfaces, and / or other circuitry), and can optionally execute received program code and / or include embedded firmware to perform functions.

[0122] ​According to another aspect of the present disclosure, an electronic device is also provided, including: at least one processor; and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the embodiments of the present disclosure.

[0123] According to another aspect of the present disclosure, a non-transitory computer readable storage medium storing computer instructions is also provided, and the computer instructions are used to make the computer perform the method according to the embodiments of the present disclosure.

[0124] According to another aspect of the present disclosure, a computer program product is also provided, including a computer program, and the computer program, when executed by a processor, implements the method according to the embodiments of the present disclosure.

[0125] According to another aspect of the present disclosure, a content recommendation system is also provided, including: at least one processor; and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to the embodiments of the present disclosure.

[0126] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in the technical solution all comply with the relevant legal regulations and do not violate public order and good customs.

[0127] According to the embodiments of the present disclosure, an electronic device, a readable storage medium and a computer program product are also provided.

[0128] Reference Figure 7 A block diagram of the structure of an electronic device 700 that can be a server or a client of the present disclosure will now be described, which is an example of a hardware device that can be applied to various aspects of the present disclosure. The electronic device is intended to represent a variety of forms of digital electronic computing devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections, and relationships, and their functions, are meant only as examples, and are not meant to limit implementations of the present disclosure described and / or claimed in this document.

[0129] As Figure 7As shown, the electronic device 700 includes a computing unit 701 that can perform various appropriate actions and processes in accordance with a computer program stored in a read only memory (ROM) 702 or a computer program loaded into a random access memory (RAM) 703 from a storage unit 708. Various programs and data required for the operation of the electronic device 700 can also be stored in the RAM 703. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0130] A plurality of components in the electronic device 700 are connected to the I / O interface 705, including an input unit 706, an output unit 707, a storage unit 708, and a communication unit 709. The input unit 706 can be any type of device that can input information to the electronic device 700, can receive inputted digital or character information, and generate key signal inputs related to user settings and / or function controls of the electronic device, and can include, but is not limited to, a mouse, a keyboard, a touch screen, a track pad, a track ball, a joystick, a microphone, and / or a remote controller. The output unit 707 can be any type of device that can present information, and can include, but is not limited to, a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 708 can include, but is not limited to, a magnetic disk, an optical disk. The communication unit 709 allows the electronic device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include, but is not limited to, a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, an 802.11 device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0131] The computing unit 701 can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The computing unit 701 performs various methods and processes described above, such as the methods 200, 300. For example, in some embodiments, the methods 200, 300 can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded onto the RAM 703 and executed by the computing unit 701, one or more steps of the methods 200, 300 described above can be performed. Alternatively, in other embodiments, the computing unit 1001 can be configured to perform the methods 200, 300 by other any appropriate means, such as by means of firmware.

[0132] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a complex programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0133] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces a means for implementing the functions / acts specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a stand-alone software package, partially on a machine and partially on a remote machine or entirely on a remote machine or server.

[0134] In the context of this disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0135] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0136] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0137] The computer system can include clients and servers. This relationship can be remote or on-site. The servers can be cloud servers, servers of a distributed system, or servers combined with a blockchain.

[0138] It should be understood that the various forms of flow shown above can be used to reorder, add, or remove steps. For example, the steps described in the present disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, which are not limited herein.

[0139] Although embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-described methods, systems, and devices are merely exemplary embodiments or examples, and the scope of the present disclosure is not limited by these embodiments or examples, but only by the granted claims and their equivalent scope. Various elements in the embodiments or examples can be omitted or replaced by equivalent elements. In addition, each step can be performed in an order different from that described in the present disclosure. Further, various elements in the embodiments or examples can be combined in various ways. It is important that many of the elements described herein can be replaced by equivalent elements that appear after the present disclosure as technology evolves.

Claims

1. A method of training a data processing model, wherein, The data processing model is configured to encode the to-be-processed content and the comment text for the to-be-processed content together to obtain a representation vector of the to-be-processed content, and the training method comprises: determining a training sample set, wherein each training sample in the training sample set comprises spliced sample content and comment text for the sample content; determining at least one matching sample matched with the training sample, wherein a first target user interacting with sample content of the matching sample and a second target user interacting with sample content of the training sample have a same target user higher than a predetermined threshold; encoding the training sample by using the data processing model with the current parameters to obtain a representation vector of the training sample; encoding the at least one matching sample by using the data processing model with the current parameters to obtain a representation vector of the matching sample; and adjusting the current parameters of the data processing model to minimize the loss between the representation vector of the training sample and the representation vector of the matching sample.

2. The training method of claim 1, wherein, The encoding of the training sample by using the data processing model with the current parameters comprises: encoding the text information of the training sample by using the data processing model with the current parameters, wherein the text information of the training sample comprises spliced content text and the comment text.

3. The training method of claim 2, wherein, The text length of the comment text is less than or equal to the text length of the content text.

4. The training method of claim 3, wherein, The comment text is obtained by: for each piece of comment information in a plurality of pieces of comment information for the sample content, determining a comment quality score of the comment information; determining at least one piece of comment information with the highest score from the plurality of pieces of comment information in descending order of comment quality score as the comment text, wherein the sum of the text lengths of the at least one piece of comment information with the highest score is less than or equal to the text length of the content text.

5. The training method of claim 2, wherein, The determination of the at least one matching sample matched with the training sample comprises: in response to determining that the target user of the first sample and the target user of the training sample are matched, determining the first sample as the matching sample of the training sample, wherein the target user of the first sample is determined according to user interactions associated with the first sample, and the target user of the training sample is determined according to user interactions associated with the training sample.

6. The training method of claim 5, wherein, The user interactions comprise at least one of the following: browsing, viewing duration, collecting, commenting, sharing, purchasing, rating.

7. The training method of claim 5 or 6, wherein, The target user of the training sample is determined by: counting interaction histories of at least one candidate user interacting with the training sample to obtain an interest level of each candidate user for the training sample; determining the target user from the at least one candidate user, wherein the target user has an interest level higher than a predetermined interest threshold.

8. The training method of claim 5, wherein, The determination of the at least one matching sample matched with the training sample further comprises: determining a text similarity between the text information of the training sample and the text information of the second sample; determine the second sample as a matching sample of the training sample in response to the text similarity being higher than a predetermined similarity threshold.

9. The training method of claim 8, wherein, the second sample is generated by scrambling text information of the training sample.

10. The training method of claim 9, wherein, a number of the second samples is less than a number of the first samples.

11. The training method of claim 1, wherein, a loss between the representation vector of the training sample and the representation vector of the matching sample minimizes similarity between negative samples while maximizing similarity between matching samples.

12. The training method of claim 11, wherein, the loss is an InfoNCE loss.

13. The training method of claim 1, wherein, the data processing model is a large language model.

14. A data processing method, comprising: determining a content to be processed and a comment text for the content to be processed; concatenating the content to be processed and the comment text to obtain a content to be encoded; and encoding the content to be encoded using a data processing model to obtain a representation vector for the content to be processed, wherein the data processing model is trained using a training method according to any one of claims 1-13.

15. The data processing method of claim 14, further comprising: generating recommended content for a user according to similarity between the representation vector of the content to be processed and a representation vector of interest of the user.

16. The data processing method of claim 14, wherein, a text length of the comment text is less than or equal to a text length of content text of the content to be processed.

17. The data processing method of claim 16, wherein, the comment text is obtained by: for each piece of comment information in a plurality of pieces of comment information for the content to be processed, determining a comment quality score of the comment information; determining at least one piece of comment information with the highest score from the plurality of pieces of comment information in descending order of comment quality score as the comment text, wherein a sum of text lengths of the at least one piece of comment information with the highest score is less than or equal to the text length of the content text.

18. An apparatus for training a data processing model, wherein, the data processing model is used to encode the content to be processed and the comment text for the content to be processed together to obtain a representation vector of the content to be processed, and the training device comprises: a sample determination unit configured to determine a training sample set, wherein each training sample in the training sample set comprises concatenated sample content and a comment text for the sample content; a matching sample determination unit configured to determine at least one matching sample matching the training sample, wherein a first target user interacting with the sample content of the matching sample and a second target user interacting with the sample content of the training sample have a same target user higher than a predetermined threshold; an encoding unit configured to encode the training sample using a data processing model with current parameters to obtain a representation vector of the training sample, and encode the at least one matching sample using the data processing model with the current parameters to obtain a representation vector of the matching sample; and a parameter adjustment unit configured to adjust the current parameters of the data processing model to minimize a loss between the representation vector of the training sample and the representation vector of the matching sample.

19. The exercise device of claim 18, wherein, the encoding the training sample using the data processing model with the current parameters comprises: encoding the text information of the training sample by using the data processing model with the current parameters, wherein the text information of the training sample comprises spliced content text and the comment text.

20. The exercise device of claim 19, wherein, The text length of the comment text is less than or equal to the text length of the content text.

21. The exercise device of claim 20, wherein, The comment text is obtained by: For each piece of comment information in the plurality of pieces of comment information for the sample content, determining a comment quality score of the comment information; determining at least one piece of comment information with the highest score from the plurality of pieces of comment information in descending order of comment quality score as the comment text, wherein the sum of the text lengths of the at least one piece of comment information with the highest score is less than or equal to the text length of the content text.

22. The exercise device of claim 21, wherein, The determining of the at least one matching sample matched with the training sample comprises: in response to determining that the target user of the first sample and the target user of the training sample are matched, determining the first sample as a matching sample of the training sample, wherein the target user of the first sample is determined according to user interaction associated with the first sample, and the target user of the training sample is determined according to user interaction associated with the training sample.

23. The exercise device of claim 22, wherein, The user interaction comprises at least one of the following: browsing, viewing duration, collecting, commenting, sharing, purchasing, and scoring.

24. The training device of claim 22 or 23, wherein, The target user of the training sample is determined by: counting the interaction history of at least one candidate user who interacts with the training sample to obtain the interest level of each candidate user for the training sample; determining the target user from the at least one candidate user, wherein the target user has an interest level higher than a predetermined interest threshold.

25. The exercise device of claim 22, wherein, The determining of the at least one matching sample matched with the training sample further comprises: determining a text similarity between the text information of the training sample and the text information of the second sample, in response to the text similarity being higher than a predetermined similarity threshold, determining the second sample as a matching sample of the training sample.

26. The exercise device of claim 25, wherein, The second sample is generated by scrambling the text information of the training sample.

27. The exercise device of claim 26, wherein, The number of the second samples is less than the number of the first samples.

28. The exercise device of claim 18, wherein, The loss between the representation vector of the training sample and the representation vector of the matching sample minimizes the similarity between negative samples while maximizing the similarity between matching samples.

29. The exercise device of claim 28, wherein, The loss is an InfoNCE loss.

30. The exercise device of claim 18, wherein, The data processing model is a large language model.

31. A data processing apparatus, comprising: an input unit configured to determine content to be processed and comment text for the content to be processed; a splicing unit configured to splice the content to be processed and the comment text to obtain content to be encoded; and an encoding unit configured to encode the content to be encoded by using a data processing model to obtain a representation vector for the content to be processed, wherein the data processing model is trained by using the training method according to any one of claims 1-13.

32. The data processing apparatus of claim 31, further comprising a recommendation unit configured to: The recommendation content for the user is generated according to the similarity between the feature vector of the to-be-processed content and the feature vector of the user's interest.

33. The data processing apparatus of claim 31, wherein, The text length of the comment text is less than or equal to the text length of the content text of the to-be-processed content.

34. The data processing apparatus of claim 33, wherein, The comment text is obtained by: For each piece of comment information in the plurality of pieces of comment information for the to-be-processed content, a comment quality score of the comment information is determined; At least one piece of comment information with the highest score is determined from the plurality of pieces of comment information in descending order of comment quality score, as the comment text, wherein the sum of the text lengths of the at least one piece of comment information with the highest score is less than or equal to the text length of the content text.

35. An electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-17.

36. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, The computer instructions are used to enable the computer to perform the method according to any one of claims 1-17.

37. A computer program product comprising a computer program, wherein, The computer program, when executed by a processor, implements the method of any one of claims 1-17.

38. A content recommendation system, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 14-17.

Images