Account searching method and related device

Abstract

The invention relates to the technical field of computers, and provides an account number searching method and a related device, which are used for improving the account number searching precision, and the method comprises the following steps: when an account number searching instruction aiming at a target search word is received, obtaining an account number label set corresponding to each candidate account number, each account tag set is determined according to the content information of each published content under the corresponding candidate account; and respectively matching each account tag contained in each acquired account tag set with the target keyword, and screening out a target account matched with the target search word from each candidate account according to a matching result. The embodiment of the invention can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic, auxiliary driving and the like. And the function of searching the account in the search scene by the content is realized by utilizing various published contents published by the account.

Description

Technical Field

[0001] The present application relates to the field of computer technology and provides an account search method and related devices. Background Art

[0002] Account search is an important part of the search system. How to quickly and accurately retrieve the target account based on the search term is a very important ability in account search.

[0003] In related technologies, when performing an account search, the search term is usually matched with the account name and account profile of each candidate account, so that based on the matching results, a target account that matches the search term is selected from each candidate account.

[0004] However, the account name and account profile are often unrelated to the content of the account itself. If the search terms are keywords related to the content of the post, it is difficult to accurately retrieve the corresponding target account based on the account name and account profile. Summary of the Invention

[0005] The embodiments of the present application provide an account search method and related devices to enhance the account recall capability and improve the account search accuracy.

[0006] In a first aspect, an embodiment of the present application provides an account search method, comprising:

[0007] When an account search instruction for a target search term is received, an account tag set corresponding to each candidate account is obtained; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account;

[0008] Each account tag contained in each of the obtained account tag sets is matched with the target keyword respectively, and based on the matching results, a target account matching the target search term is screened out from the candidate accounts.

[0009] In a second aspect, an embodiment of the present application provides an account search device, comprising:

[0010] A tag acquisition unit, configured to, upon receiving an account search instruction for a target search term, acquire an account tag set corresponding to each candidate account; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account;

[0011] The account matching unit is used to match each account tag contained in each acquired account tag set with the target keyword, and screen out the target account matching the target search term from the candidate accounts based on the matching results.

[0012] As a possible implementation, the tag acquisition unit is further configured to obtain an account tag set corresponding to a candidate account in the following manner:

[0013] Extracting content from each of the candidate accounts to obtain corresponding content information, each piece of content information including: a set of content tags corresponding to the corresponding content in at least one evaluation dimension;

[0014] Based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition is screened out from the content tags as the account tag in the account tag set.

[0015] As a possible implementation, based on the obtained tag confidences of the content tags corresponding to the respective published contents, the tag acquisition unit selects at least one content tag that meets a set confidence condition from the content tags and uses it as the account tag in the account tag set. Specifically, the tag acquisition unit is configured to:

[0016] For the at least one evaluation dimension, perform the following operations:

[0017] Aggregating content tags corresponding to each of the published content under an evaluation dimension to obtain candidate tags corresponding to the candidate account under the evaluation dimension;

[0018] Obtaining a label confidence corresponding to each candidate label based on the label confidence of each content label corresponding to each published content under the one evaluation dimension;

[0019] Based on the tag confidences corresponding to the candidate tags, at least one candidate tag that meets a set confidence condition is screened out from the candidate tags as the account tag in the account tag set.

[0020] As a possible implementation, based on the tag confidence corresponding to each candidate tag, at least one candidate tag that meets a set confidence condition is screened out from the candidate tags and used as the account tag in the account tag set. The tag acquisition unit is specifically configured to:

[0021] Based on the label confidences corresponding to the candidate labels, a total label confidence corresponding to the candidate account in the evaluation dimension is obtained, and based on the total label confidence, a confidence ratio corresponding to each candidate label is determined;

[0022] If there is at least one candidate tag among the candidate tags whose corresponding tag confidence exceeds the set confidence threshold and whose confidence ratio exceeds the set ratio threshold, the at least one candidate tag is used as the account tag in the account tag set.

[0023] As a possible implementation, when aggregating the content tags corresponding to the respective published contents under one evaluation dimension to obtain the candidate tags corresponding to the candidate account under the one evaluation dimension, the tag acquisition unit is specifically configured to:

[0024] Aggregating content tags corresponding to each of the published content under the one evaluation dimension to obtain initial tags corresponding to the candidate account under the one evaluation dimension;

[0025] Based on the label confidences corresponding to the respective initial labels, initial labels whose corresponding label confidences exceed a set label threshold are screened out from the initial labels as the candidate labels corresponding to the candidate account in the evaluation dimension.

[0026] As a possible implementation, when extracting content from a published content under the candidate account and obtaining corresponding content information, the tag acquisition unit is specifically configured to perform at least one of the following operations:

[0027] Based on the content modality type of the published content, category extraction is performed on the published content to obtain a content tag set corresponding to the category evaluation dimension of the published content;

[0028] Based on the published content and its corresponding historical feedback information, entity extraction is performed on the published content to obtain a content tag set corresponding to the published content in an entity evaluation dimension.

[0029] As a possible implementation, each account tag contained in each acquired account tag set is matched with the target keyword, and based on the matching results, a target account matching the target search term is screened out from the candidate accounts. The account matching unit is specifically configured to:

[0030] Matching the basic account information of each candidate account with the target search term to obtain corresponding initial matching results;

[0031] Matching each account tag in the account tag set corresponding to each candidate account with the target search term to obtain a corresponding reference matching result;

[0032] Based on the obtained initial matching results and reference matching results, a target account that matches the target search term is screened out from the candidate accounts.

[0033] As a possible implementation manner, the label acquisition unit is further configured to:

[0034] When one of the candidate accounts updates its content, the account tag set of the candidate account is updated according to the new content; or

[0035] The account tag set of each candidate account is updated according to a set update cycle.

[0036] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above method.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to perform the steps of any of the above methods.

[0038] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium. The processor of an electronic device reads and executes the computer program from the computer-readable storage medium, so that the electronic device performs the steps of any of the above methods.

[0039] In an embodiment of the present application, when performing an account search for a target search term, each candidate account is matched with the target search term using the account tag set corresponding to each candidate account. Since the account tag set is determined based on the content information of each published content under the corresponding candidate account, when performing an account search, the account can be matched based on the content published by the candidate account. In this way, when keywords related to the content are obtained, accounts that have published content of the corresponding type can be searched based on the account tags, thereby enhancing the account call-back capability and improving the account search accuracy.

[0040] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purposes and other advantages of the present application can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0042] Figure 1 A schematic diagram of an application scenario provided in an embodiment of the present application;

[0043] Figure 2 A flowchart of an account search method provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of a candidate account provided in an embodiment of the present application;

[0045] Figure 4 A flowchart of a method for extracting an account tag set provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of a content tag provided in an embodiment of the present application;

[0047] Figure 6 A logical diagram of a candidate tag screening process provided in an embodiment of the present application;

[0048] Figure 7 A logical diagram of a tag confidence determination process provided in an embodiment of the present application;

[0049] Figure 8 A logical diagram of an account tag screening process provided in an embodiment of the present application;

[0050] Figure 9 A logical diagram of an account matching process provided in an embodiment of the present application;

[0051] Figure 10 A logical diagram of an account index construction process provided in an embodiment of the present application;

[0052] Figure 11 This is a schematic diagram of the structure of an account search device provided in an embodiment of the present application;

[0053] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of the technical solutions of this application, but not all of them. Based on the embodiments described in this application document, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the technical solutions of this application.

[0055] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0056] It is understandable that when the various embodiments of this application are applied to specific products or technologies, relevant licenses or consents need to be obtained, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0057] In related technologies, when performing an account search, the search term is usually matched with the account name and account profile of each candidate account, so that based on the matching results, a target account that matches the search term is selected from each candidate account.

[0058] However, the account name and account profile are often unrelated to the content of the account itself. If the search terms are keywords related to the content of the post, it is difficult to accurately retrieve the corresponding target account based on the account name and account profile.

[0059] In an embodiment of the present application, when performing an account search for a target search term, the account tag set corresponding to each candidate account is used to match each candidate account with the target search term. Since the account tag set is determined based on the content information of each published content under the corresponding candidate account, when performing an account search, the account can be matched based on the content published by the candidate account. Therefore, when keywords related to the content are obtained, accounts that have published content of the corresponding type can be searched based on the account tags, thereby enhancing the account call-back capability and improving the account search accuracy.

[0060] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that seeks to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to possess the capabilities of perception, reasoning, and decision-making.

[0061] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, pre-trained models, operating / interaction systems, and mechatronics. Pre-trained models, also known as large models or basic models, can be fine-tuned and widely applied to downstream tasks across various AI disciplines. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0062] Computer vision (CV) is the science of making machines "see." Specifically, it refers to the use of cameras and computers to replace the human eye in identifying, monitoring, and measuring objects, and further image processing to transform the computer into images more suitable for human observation or transmission to instruments. As a scientific discipline, computer vision studies related theories and technologies, aiming to build artificial intelligence systems that can extract information from images or multidimensional data. Large model technology has brought significant changes to the development of computer vision technology. Pre-trained models in the field of vision, such as the Swin Transformer, ViT, V-MOE, and MAE, can be fine-tuned to quickly and widely apply to specific downstream tasks. Computer vision technology generally includes image processing, image recognition, image semantic understanding, image retrieval, optical character recognition (OCR), video processing, video semantic understanding, video content / action recognition, 3D object reconstruction, 3D technology, virtual reality, augmented reality, simultaneous localization and mapping, and other technologies. It also includes common biometric recognition technologies such as facial recognition and fingerprint recognition.

[0063] Natural language processing (NLP) is an important field in the fields of computer science and artificial intelligence. It studies various theories and methods that enable effective communication between humans and computers using natural language. Natural language processing involves natural language, the language people use in daily life, and is closely related to linguistics research; it also involves computer science and mathematics. Pre-training models, an important technology for model training in the field of artificial intelligence, are developed from large language models (LLMs) in the field of NLP. After fine-tuning, large language models can be widely used in downstream tasks. Natural language processing technologies generally include text processing, semantic understanding, machine translation, robot question answering, knowledge graphs, and other technologies.

[0064] Machine learning (ML) is a multidisciplinary field that encompasses probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications span all areas of AI. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning. Pretrained models are the latest development in deep learning, integrating these techniques.

[0065] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, autonomous driving, drones, digital twins, virtual humans, robots, artificial intelligence generated content (AIGC), conversational interaction, smart medical care, smart customer service, game AI, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0066] The solution provided in the embodiments of this application primarily involves artificial intelligence machine learning technology. Specifically, machine learning technology is used to extract the type of candidate accounts using a type extraction model and to extract the entities of candidate accounts using an entity extraction model. In addition, a style extraction model can also be used to extract the style of candidate accounts.

[0067] The following briefly introduces the application scenarios to which the technical solutions of the embodiments of the present application can be applied. It should be noted that the application scenarios described below are only used to illustrate the embodiments of the present application and are not limiting. In the specific implementation process, the technical solutions provided by the embodiments of the present application can be flexibly applied according to actual needs.

[0068] The solution provided in the embodiments of this application can be applied to various account searches, such as searches for various accounts in browsers, social platforms, video platforms, etc. This solution can be applied as a basic technology in various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving, etc.

[0069] See Figure 1 , which is a schematic diagram of an application scenario provided in an embodiment of the present application. The application scenario includes a terminal device 110 and a server 120. The number of terminal devices 110 can be one or more. The number of servers 120 can also be one or more. This application does not specifically limit the number of terminal devices 110 and servers 120.

[0070] In the embodiments of the present application, terminal device 110 may be, but is not limited to, a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, IoT device, intelligent voice interaction device, smart home appliance, vehicle-mounted terminal, aircraft, etc. Terminal device 110 may include a client that performs the corresponding account search function. The client may be in the form of, but is not limited to, an application, mini-program, or webpage.

[0071] Server 120 is the backend server corresponding to the client. Server 120 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0072] The terminal device 110 and the server 120 may be connected directly or indirectly via wired or wireless communication, which is not limited in this application.

[0073] The account search method mentioned in the embodiments of the present application can be executed by a server or a terminal device, or can be executed jointly by a server and a terminal device, without limitation.

[0074] As one possible implementation, the terminal device obtains the target search term input by the target object in response to a search term input operation on the target object. If the account search method is applied to the terminal device, the terminal device can determine that an account search instruction for the target search term has been received after the target object triggers the search term input operation, and then obtain the account tag set corresponding to each candidate account. The terminal device matches each account tag contained in each obtained account tag set with the target keyword, and based on the matching results, selects the target account that matches the target search term from each candidate account. If the account search method is applied to a server, the terminal device can send an account search request carrying the target search term to the server after the target object triggers the search term input operation. In this way, after receiving the account search request, the server determines that an account search instruction for the target search term has been received, and then obtains the account tag set corresponding to each candidate account. The terminal device matches each account tag contained in each obtained account tag set with the target keyword, and based on the matching results, selects the target account that matches the target search term from each candidate account.

[0075] See Figure 2 As shown, it is a flow chart of an account search method provided in an embodiment of the present application. The method can be applied to a terminal device or a server. The following is only described using the application to the server as an example. The specific process is as follows:

[0076] S201. When an account search instruction for a target search term is received, an account tag set corresponding to each candidate account is obtained; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account.

[0077] In the embodiment of the present application, the target search term can be a text composed of one or more keywords. For example, the target search term is "National Day Travel", where "National Day" is part of the target account's account name and "Travel" is the relevant content posted by the target account.

[0078] The account tag set corresponding to a candidate account includes: one or more account tags determined based on the content information of each published content under the candidate account.

[0079] Published content refers to the content published by the corresponding candidate account. The types of published content include, but are not limited to, videos, graphics, text, and images. The types of each published content can be the same or different.

[0080] Taking the video scenario as an example, any registered account in the video platform that is allowed to be searched can be a candidate account. The content published by the candidate account is usually a video. Assume that account A is an account in the video platform that is allowed to be searched, that is, account A is a candidate account. Figure 3As shown, the account name of Account A is "xxx Eleven", and the account profile is "*This is an example profile*". Account A has published 10 videos such as Video 1, Video 2, and Video 3. Then, the account tag set corresponding to Account A is determined based on the content information of the 10 videos.

[0081] In some embodiments, the account tag set of a candidate account can be pre-extracted before account search or can be extracted after receiving an account search instruction, and this is not limited. Since the extraction methods of the account tag sets of each candidate account are the same, below, only taking candidate account x as an example, the process of extracting the account tag set will be described. Candidate account x can be any one of the candidate accounts.

[0082] To improve the accuracy of account tags, in the embodiments of the present application, the content extraction can be separately performed on each piece of published content under a candidate account to obtain the content tag set corresponding to the corresponding published content in at least one evaluation dimension. Through the content tags of one or more evaluation dimensions for subsequent screening, compared with the basic account information such as the account name and account profile, the tags related to the content can be effectively supplemented, so as to obtain the semantic class tags in the account dimension; further, based on the tag confidence levels of the respective content tags corresponding to each piece of published content obtained, the account tags are screened from the content tags. Since the tag confidence level can reflect the relevance between the published content and the content tag to a certain extent, the screened account tags can more accurately represent the posted content of the account, thereby further improving the account search accuracy. In this way, even when using content-oriented target search terms, the corresponding account can be accurately retrieved, thus improving the user search experience.

[0083] Specifically, refer to Figure 4 As shown, it is a schematic flowchart of a process for extracting an account tag set provided in the embodiments of the present application. The specific process is as follows:

[0084] S401: Respectively perform content extraction on each piece of published content under candidate account x to obtain the corresponding content information, and each content information includes: the content tag set corresponding to the corresponding published content in N evaluation dimensions.

[0085] In an embodiment of the present application, label mining can be performed from one or more evaluation dimensions, that is, the value of N is a positive integer. The evaluation dimensions include, but are not limited to: the category to which the published content belongs, the entities contained in the published content, the content style of the published content, etc. For example, the categories include, but are not limited to, television, movies, variety shows, or a specific game, movie, TV series, etc., the entities include, but are not limited to, people, one or a certain type of animal, one or a certain type of building, etc., and the content style includes, but is not limited to, funny, warm, cute pets, etc., without limitation to this. The specific categories, entities, and content styles can be set according to the actual application scenario.

[0086] Assume that the content published by candidate account x (i.e., published content) includes: content 1, content 2, ..., content m, and the value of m is a positive integer. For content 1, content 2, ..., content m, content 1, content 2, ..., content m, respectively, obtain the content information corresponding to content 1, content 2, ..., content m content 1, content 2, ..., content m.

[0087] Taking content i as an example, content i can be any of content 1, content 2, ..., content m. The content information of content i includes the content tag sets corresponding to content i in N evaluation dimensions. For example, the content information of content i includes the content tag set of content i in the category evaluation dimension and the content tag set of content i in the entity evaluation dimension.

[0088] Taking the two evaluation dimensions of category and entity as an example, see Figure 5 As shown, content extraction is performed on each published content under candidate account x, including content 1, content 2, ..., content m. For each published content, a content label set corresponding to the category evaluation dimension and a content label set corresponding to the entity evaluation dimension can be obtained. Taking each content label set as an example, where each content label set contains three content labels, the content label set corresponding to content 1 in the category evaluation dimension includes labels c1, c2, and c3, and the content label set corresponding to content 1 in the entity evaluation dimension includes labels e1, e2, and e3; the content label set corresponding to content 2 in the category evaluation dimension includes labels c1, c3, and c4, and the content label set corresponding to content 2 in the entity evaluation dimension includes labels e1, e2, and e3; the content label set corresponding to content 3 in the category evaluation dimension includes labels c3, c4, and c6, and the content label set corresponding to content 3 in the entity evaluation dimension includes labels e1, e2, and e3. Similarly, the content label sets of the remaining published content are not repeated.

[0089] Below, the process of extracting content tag sets is explained using the two evaluation dimensions of category and entity as examples.

[0090] (1) Category evaluation dimensions

[0091] For each published content under the candidate account x, the categories of the published content are extracted respectively to obtain the content label set corresponding to the category evaluation dimension of the published content.

[0092] In one possible implementation, for each published content under the candidate account x, category extraction is performed on the published content based on the content modality type of the published content, and a content tag set corresponding to the published content in the category evaluation dimension is obtained.

[0093] Taking content i as an example, a category extraction model can be used to extract categories from content i and obtain the label confidence of each target tag in the category evaluation dimension. Then, based on the label confidence corresponding to each target tag, a number of category tags are selected from each target tag to serve as the content label for content i in the category evaluation dimension. The label confidence represents the probability that the published content is associated with the corresponding label.

[0094] As an example, the top k category labels with the highest label confidence can be filtered out as the content labels of content i in the category evaluation dimension. Category labels whose label confidence reaches a set threshold can also be filtered out as the content labels of content i in the category evaluation dimension, but it is not limited to this.

[0095] In addition, in the content label set corresponding to the category evaluation dimension of content i, in addition to one or more content labels corresponding to the content i in the category evaluation dimension, the label confidence corresponding to each content label can also be included. Assume that CategoryList is used to represent the content label set corresponding to the category evaluation dimension of content i, CategoryList = [(category1, s c1 ), (cate2, s c2 ), (cate3, s c3 )], where cate1, cate2, and cate3 are three content tags, s c1 is the label confidence of cate1, s c2 is the label confidence of cate2, s c3 is the label confidence of cate3.

[0096] Taking account A as an example, assume that the category tags include travel, photography, home, pets, food preparation, dish sharing, food appreciation, game commentary, game recording, and other target tags. Account A publishes 10 videos, including video 1, video 2, and video 3. For video 1, the category extraction model is used to extract the category of video 1 and obtain the label confidence of video 1 under various target tags. Then, based on the obtained label confidence, the top 3 category tags are selected from various target tags as the content tag set corresponding to video 1 in the category evaluation dimension (referred to as tag set 1). Tag set 1 = [(travel, 0. 9), (food making, 0.8), (dish sharing, 0.7)]; for video 2, the category extraction model is used to extract the category of video 2, and the label confidence of video 2 under each target label is obtained. Then, according to the obtained label confidence, the top 3 category labels are screened out from each target label as the content label set corresponding to video 2 in the category evaluation dimension (referred to as label set 2), label set 2 = [(travel, 0.9), (pets, 0.85), (dish sharing, 0.7)]; similarly, for the remaining 8 videos released by account A, the content label set corresponding to the category evaluation dimension of the corresponding video can be obtained.

[0097] In some implementations, different category extraction models can be set for different types of published content. Thus, when performing type extraction, based on the content modality type of the published content, the category extraction model corresponding to the content modality type is used to extract the category of the published content, thereby obtaining a set of content tags corresponding to the category evaluation dimension of the published content. The content modality type refers to the content type of the published content, such as video, graphic, text, image, etc.

[0098] For example, for the published content of video, picture, text, and image types, the corresponding category extraction models are trained respectively. If the published content is video, then the category extraction model corresponding to the video is used for type extraction. If the published content is picture and text, then the category extraction model corresponding to the picture and text is used for type extraction. If the published content is text, then the category extraction model corresponding to the text is used for type extraction. If the published content is image, then the category extraction model corresponding to the image is used for type extraction.

[0099] (2) Entity Assessment Dimension

[0100] For each published content under the candidate account x, perform entity extraction on the published content and obtain the content label set corresponding to the entity evaluation dimension of the published content.

[0101] Taking content i as an example, the entity extraction model can be used to extract entities from content i and obtain the label confidence of each entity label of content i in the entity evaluation dimension. Then, according to the label confidence corresponding to each entity label, several entity labels are screened out from each entity label as the content label of content i in the entity evaluation dimension.

[0102] As an example, the top k entity tags with the highest tag confidence can be screened out as the content tags of content i in the entity evaluation dimension. Entity tags whose tag confidence reaches a set threshold can also be screened out as the content tags of content i in the entity evaluation dimension, but this is not limited to this.

[0103] In addition, the content label set corresponding to the entity evaluation dimension of content i may include, in addition to one or more content labels corresponding to the content i in the entity evaluation dimension, the label confidence corresponding to each content label. Assume that EntityList is used to represent the content label set corresponding to the entity evaluation dimension of content i, EntityList = [(e1, s e1 ),(e2,s e2 ),(e3,s e3 )], where e1, e2, and e3 are three content tags, s e1 is the label confidence of e1, s e2 is the label confidence of e2, s e3 is the label confidence of e3.

[0104] Taking the candidate account as account A as an example, assuming that the entity tags include: film and television character name 1, film and television character name 2, film and television character name 3, game character name 1, game character name 2, dog, cat, etc., for video 1, the entity extraction model is used to extract entities from video 1 and obtain the label confidence of video 1 under each entity label. Then, based on the obtained label confidence, the top 3 entity labels are selected from each entity label as the content label set corresponding to video 1 in the entity evaluation dimension (referred to as label set 1). Label set 1 = [(RV, 0.9), (hot pot, 0.88) , (snow, 0.87)]; for video 2, the entity extraction model is used to extract entities from video 2 and obtain the label confidence of video 2 under each entity label. Then, according to the obtained label confidence, the top 3 entity labels are screened out from each entity label as the content label set corresponding to video 2 in the entity evaluation dimension (referred to as label set 2 for short), label set 2 = [(RV, 0.9), (dog, 0.85), (xx lake, 0.7)]; similarly, for the remaining 8 videos released by account A, the content label set corresponding to the corresponding video in the entity evaluation dimension can be obtained.

[0105] In one possible implementation, entity extraction is performed on each post under candidate account x based on the post's content and its corresponding historical feedback information, obtaining a set of content labels corresponding to the entity evaluation dimension. This historical feedback information includes, but is not limited to, comments and comments. In other words, historical feedback information, such as comments and comments, is further combined with the post's content to assist in entity extraction, thereby improving the accuracy of entity extraction.

[0106] In some implementations, different entity extraction models can be set for different types of published content. In this way, when performing type extraction, based on the content modality type of the published content, the entity extraction model corresponding to the content modality type is used to perform entity extraction on the published content to obtain the content label set corresponding to the entity evaluation dimension of the published content.

[0107] In an embodiment of the present application, two types of content tags, category and entity, are aggregated and screened to obtain the final account tag. Since the category can reflect the overall type bias of the account's published content, and the entity can reflect the specific content bias of the account's publication, this can effectively help the account expand its tags as needed, thereby enhancing the account class recall capability and improving the retrieval accuracy.

[0108] S402: Based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition is selected from the content tags as the account tag in the account tag set.

[0109] Specifically, when executing S402, as a first screening method, the following operations may be performed for each of the N evaluation dimensions:

[0110] First, aggregate the content labels corresponding to each published content under an evaluation dimension to obtain the candidate labels corresponding to the candidate account x in an evaluation dimension;

[0111] Secondly, based on the label confidence of each content label corresponding to each published content under an evaluation dimension, the label confidence corresponding to each candidate label is obtained;

[0112] Finally, based on the tag confidence corresponding to each candidate tag, at least one candidate tag that meets the set confidence condition is screened out from the candidate tags as the account tag in the account tag set.

[0113] That is to say, in an embodiment of the present application, after obtaining the content tags corresponding to each published content, for each of the N evaluation dimensions, the content tags under the corresponding evaluation dimension are screened to obtain the account tag set of the candidate account x. In this way, the account tags can be expanded from the N evaluation dimensions. The refined tag design can effectively improve the accuracy of the account tags, thereby improving the retrieval accuracy to a certain extent.

[0114] As a possible implementation method, the content tags corresponding to each published content under an evaluation dimension are aggregated to obtain the candidate tags corresponding to the candidate account x in an evaluation dimension. The following two possible implementation methods may be used, but are not limited to:

[0115] The first method is to aggregate the content labels corresponding to each published content under an evaluation dimension to obtain the initial labels corresponding to the candidate account x in the evaluation dimension, and directly use the obtained initial labels corresponding to the candidate account x in the evaluation dimension as the candidate labels corresponding to the candidate account x in the evaluation dimension.

[0116] Among them, when aggregating the content tags corresponding to each published content under an evaluation dimension, considering that the published content may have repeated tags, the content tags can be deduplicated during aggregation.

[0117] See Figure 6 As shown, taking the category evaluation dimension as an example, the content labels corresponding to content 1, content 2, ..., content m under the category evaluation dimension are aggregated to obtain the initial labels corresponding to candidate account x in the category evaluation dimension. Each initial label is the candidate label for account A under the category evaluation dimension, and each initial label includes: label c1, label c2, label c3, label c4, label c5, label c6, etc. Similarly, the content labels corresponding to content 1, content 2, ..., content m under the entity evaluation dimension are aggregated to obtain the initial labels corresponding to candidate account x in the entity evaluation dimension. Each initial label is the candidate label for account A under the entity evaluation dimension, and each initial label includes: label e1, label e2, label e3, label e4, label e5, label e6, etc.

[0118] Taking account A as an example, among the 10 videos released by account A, taking video 1, video 2 and video 3 as examples, the content tags corresponding to the category evaluation dimension of video 1 include: travel, food preparation, and dish sharing; the content tags corresponding to the category evaluation dimension of video 2 include: travel, pets, and dish sharing; the content tags corresponding to video 3 include: travel, game commentary, and pets. Then, aggregate the content tags under the category evaluation dimension of each of the 10 videos to obtain the initial tags corresponding to account A in the category evaluation dimension. The initial tags are the candidate tags of account A under the category evaluation dimension. The initial tags corresponding to the category evaluation dimension include: travel, food preparation, dish sharing, pets, game commentary, etc.

[0119] Similarly, the content labels corresponding to the entity evaluation dimension of video 1 include: RV, hot pot, snow; the content labels corresponding to video 2 include: RV, dog, xx Lake; and the content labels corresponding to video 3 include: RV, cute pet, hot pot. The content labels under the entity evaluation dimension of each of the 10 videos are aggregated to obtain the initial labels corresponding to account A in the category evaluation dimension. The initial labels are the candidate labels of account A under the entity evaluation dimension. The initial labels corresponding to the category evaluation dimension include: RV, cute pet, dog, hot pot, snow, xx Lake, etc.

[0120] The second method is to first aggregate the content labels corresponding to each published content under an evaluation dimension to obtain the initial labels corresponding to the candidate account x in the evaluation dimension. Then, based on the label confidence corresponding to each initial label, the initial labels whose corresponding label confidence exceeds the set label threshold are screened out from the initial labels as the candidate labels corresponding to the candidate account x in the evaluation dimension.

[0121] Among them, when aggregating the content tags corresponding to each published content under an evaluation dimension, considering that the published content may have repeated tags, the content tags can be deduplicated during aggregation.

[0122] In some implementations, the tag threshold may be set for some of the N evaluation dimensions, or may be set for all N evaluation dimensions, and there is no limitation on this.

[0123] In some implementations, the sum of the label confidences of the initial label corresponding to the published content associated with the initial label is used as the label confidence of the initial label under the candidate account x. For example, see Figure 7As shown in the figure, among the content 1, content 2, ..., content m published by candidate account x, content 1, content 2, and content 3 are associated with label c3. Then, the label confidence of label c3 of content 1, the label confidence of label c3 of content 2, and the label confidence of label c3 of content 3 are taken as the label confidence of label c3 under candidate account x. Of course, in actual applications, other indicators such as the average, variance, and standard deviation of signature confidence can also be used, but are not limited to these.

[0124] Taking account A releasing videos 1, 2, and 3 as an example, for the category evaluation dimension, the content tags corresponding to video 1 in the category evaluation dimension include: travel (tag confidence 0.9), food preparation (tag confidence 0.8), and dish sharing (tag confidence 0.7); the content tags corresponding to video 2 in the category evaluation dimension include: travel (tag confidence 0.9), pets (tag confidence 0.85), and dish sharing (tag confidence 0.7); the content tags corresponding to video 3 in the category evaluation dimension include: travel (tag confidence 0.8), game commentary (tag confidence 0.7), and pets (tag confidence 0.5). Aggregate the content tags for each of the three videos under the category evaluation dimension to obtain the initial tags corresponding to account A in the category evaluation dimension. The initial tags corresponding to account A in the category evaluation dimension include: travel, food preparation, dish sharing, pets, and game commentary. The tag confidence corresponding to the initial tag (travel) is 0.9+0.9+0.8=2.6, the tag confidence corresponding to the initial tag (food preparation) is 0.8, the tag confidence corresponding to the initial tag (dish sharing) is 0.8+0.7=1.5, the tag confidence corresponding to the initial tag (pets) is 0.85+0.5=1.35, and the tag confidence corresponding to the initial tag (game commentary) is 0.7. Assuming that the set tag threshold for the category evaluation dimension is 1, then based on the tag confidence corresponding to each initial tag, select the initial tags with a corresponding tag confidence greater than 1 from the initial tags as the candidate tags corresponding to account A in the category evaluation dimension. The candidate tags include: travel, dish sharing, and pets. Similarly, for the entity evaluation dimension, the second implementation method can also be used to obtain the candidate labels corresponding to account A in the entity evaluation dimension. Since the process is similar, it will not be repeated here.

[0125] In the second method, content tags with low confidence are filtered out, thereby preventing content tags with low confidence from becoming account tags of candidate accounts, thereby improving the accuracy of account tags and further improving the accuracy of subsequent account searches.

[0126] As a possible implementation method, based on the label confidence of each content label corresponding to each published content under an evaluation dimension, the label confidence corresponding to each candidate label is obtained, including:

[0127] For each candidate tag in each candidate tag, the sum of the tag confidences corresponding to each published content associated with the candidate tag can be used as the tag confidence of the candidate tag under candidate account x. For example, among content 1, content 2, ..., content m published by candidate account x, content 1, content 2, and content 3 are associated with tag c3. Then, the tag confidence of tag c3 of content 1, the tag confidence of tag c3 of content 2, and the tag confidence of tag c3 of content 3 are summed as the tag confidence of tag c3 under candidate account x. Of course, in actual applications, other indicators such as the average, variance, and standard deviation of the signature confidence can also be used, but are not limited to this.

[0128] In some implementations, a corresponding label threshold may be set for each evaluation dimension. As another possible implementation, the same label threshold may also be set for each evaluation dimension.

[0129] x_score is used to represent the label threshold set for the category evaluation dimension, and y_score is used to represent the label threshold set for the entity evaluation dimension. If the label confidence of an initial label under the category evaluation dimension is less than x_score, then the confidence of the candidate label is considered too low and it is not counted as a candidate for the account dimension label. That is, the initial label is filtered out and is not considered a candidate label. Similarly, if the label confidence of an initial label under the entity evaluation dimension is less than y_score, then the confidence of the candidate label is considered too low and it is not counted as a candidate for the account dimension label. That is, the initial label is filtered out and is not considered a candidate label.

[0130] As a possible implementation, based on the tag confidence corresponding to each candidate tag, at least one candidate tag that meets the set confidence condition is screened out from each candidate tag as the account tag in the account tag set, including at least one of the following operations:

[0131] Operation 1: If there is at least one candidate tag among the candidate tags whose corresponding tag confidence exceeds a set confidence threshold, the at least one candidate tag is used as the account tag in the account tag set.

[0132] Operation 2: Based on the label confidence corresponding to each candidate label, obtain the total label confidence corresponding to a candidate account in an evaluation dimension, and based on the total label confidence, determine the confidence ratio corresponding to each candidate label; if among the candidate labels, there is at least one candidate label whose corresponding label confidence exceeds the set confidence threshold and whose confidence ratio exceeds the set ratio threshold, then add at least one candidate label to the account label set.

[0133] See Figure 8 As shown, for label c1, based on the label confidence and confidence ratio of label c1, when the label confidence of label c1 exceeds the set confidence threshold and the confidence ratio exceeds the set ratio threshold, label c1 is used as the account label; for label c2, based on the label confidence and confidence ratio of label c2, assuming that the label confidence of label c2 does not exceed the set confidence threshold, then label c2 is not used as an account label; for label c3, based on the label confidence and confidence ratio of label c3, assuming that the label confidence of label c3 exceeds the set confidence threshold and the confidence ratio exceeds the set ratio threshold, label c3 is used as the account label. Similarly, candidate labels such as label c5 whose label confidence exceeds the set confidence threshold and whose confidence ratio exceeds the set ratio threshold can be used as account labels.

[0134] In an embodiment of the present application, after obtaining the candidate tags, the candidate tags are further screened to select account tags that can reflect the overall content of the account, thereby improving the accuracy of the account tags.

[0135] As an example, the sum of the label confidences corresponding to each candidate label can be used as the total confidence, and then the ratio of the label confidence corresponding to each candidate label to the total label confidence can be used as the corresponding confidence ratio. For example, if score is used to represent the total confidence, then the confidence ratio of a candidate label can be expressed as Ri = ci / score, where Ri represents the confidence ratio of the i-th candidate label and ci represents the label confidence of the i-th candidate label.

[0136] In some implementations, a corresponding confidence threshold and a proportion threshold may be set for each evaluation dimension. Of course, a unified confidence threshold and a unified proportion threshold may also be used for each evaluation dimension.

[0137] For example, for the category evaluation dimension, the candidate labels of a candidate account in the category evaluation dimension include: label 1, label 2, label 3 and label 4. First, the label confidence corresponding to label 1, the label confidence corresponding to label 2, the label confidence corresponding to label 3 and the label confidence corresponding to label 4 are summed as the total label confidence. Then, based on the total label confidence, the confidence ratio corresponding to label 1, label 2, label 3 and label 4 is determined, where the confidence ratio corresponding to label 1 is R1=c1 / score, and the confidence ratio corresponding to label 2 is R2=c1 / score. The corresponding confidence ratio R2 = c2 / score, the confidence ratio R3 = c3 / score corresponding to label 3, and the confidence ratio R4 = c4 / score corresponding to label 4. Assume that the ratio threshold corresponding to the category evaluation dimension is rate_x, and the confidence threshold corresponding to the category evaluation dimension is cate_x. If c_i> = cate_x and rate_i> = rate_x, c_i refers to the i-th candidate label among the candidate labels, then the i-th candidate label is used as the account label of the candidate account.

[0138] For the category evaluation dimension, the candidate labels of the category evaluation dimension include: label 1', label 2', label 3' and label 4'. First, the label confidence corresponding to label 1', the label confidence corresponding to label 2', the label confidence corresponding to label 3' and the label confidence corresponding to label 4' are summed as the total label confidence. Then, based on the total label confidence, the confidence ratios corresponding to label 1', label 2', label 3' and label 4' are determined. Among them, the confidence ratio corresponding to label 1' is R1=c1 / score, and the confidence ratio corresponding to label 4' is R2=c1 / score. The confidence ratio corresponding to label 2' is R2 = c2 / score, the confidence ratio corresponding to label 3' is R3 = c3 / score, and the confidence ratio corresponding to label 4' is R4 = c4 / score. Assume that the ratio threshold corresponding to the category evaluation dimension is rate_e, and the confidence threshold corresponding to the category evaluation dimension is e_x. If c_i> = e_x and rate_i> = rate_e, c_i refers to the i-th candidate label among the candidate labels, then the i-th candidate label is used as the account label of the candidate account.

[0139] Through the above implementation method, based on the comprehensive score and proportion of candidate tags in the full account dimension, the account tag of the account dimension is finally obtained, which can enable the account tag to better reflect the content published by the account as a whole, thereby improving the accuracy of the account tag and thus improving the subsequent recall accuracy.

[0140] As a second screening method, in an embodiment of the present application, the content labels corresponding to each of the published contents under N evaluation dimensions can also be aggregated to obtain candidate accounts x candidate labels; then, based on the label confidence of each content label corresponding to each of the published contents, the label confidence corresponding to each candidate label is obtained; finally, based on the label confidence corresponding to each candidate label, at least one candidate label that meets the set confidence condition is screened from each candidate label as the account label in the account label set. In other words, the content labels of each of the published contents under N evaluation dimensions can be directly aggregated and screened. Since the process of aggregating and screening the content labels of each of the published contents under N evaluation dimensions is similar to the process of aggregating and screening the content labels of each of the published contents under one evaluation dimension, it will not be repeated here.

[0141] S202: Match each account tag contained in each acquired account tag set with the target keyword, and select a target account matching the target search term from each candidate account based on the matching result.

[0142] Specifically, in the embodiment of the present application, the following matching methods may be used but are not limited to:

[0143] First, the basic account information of each candidate account is matched with the target search term to obtain the corresponding initial matching results;

[0144] Secondly, each account label in the account label set corresponding to each candidate account is matched with the target search term to obtain the corresponding reference matching results;

[0145] Finally, based on the obtained initial matching results and reference matching results, target accounts matching the target search terms are screened out from the candidate accounts.

[0146] The basic account information includes, but is not limited to, one or more items of the account name and account profile of the corresponding candidate account.

[0147] In one possible implementation, the initial matching result between a candidate account's basic account information and the target keyword may include the basic information matching degree between the candidate account's basic account information and the target search term. For example, for account A, whose basic account information includes the account name and account profile, the initial matching result includes the basic information matching degree between account A's account name and the target search term, as well as the basic information matching degree between account A's account profile and the target search term.

[0148] In one possible implementation, the basic information matching degree between the basic account information and the target search term can be expressed as text similarity. It is understood that if the target search term is entered in text form, the basic information matching degree between the basic account information and the target search term can be directly calculated. If the target search term is entered in voice form, speech recognition technology can be used to convert the input voice data into the target search term in text form, thereby calculating the basic information matching degree between the basic account information and the target search term.

[0149] In one possible implementation, when matching each account tag in the account tag set corresponding to a candidate account with a target search term, each account tag in the account tag set can be matched with the target search term respectively, thereby determining the account tag matching degree between each account tag and the target search term.

[0150] In one possible implementation, the basic information matching degree between the account basic information and the target search term can be represented by text similarity.

[0151] In one possible implementation, based on the obtained initial matching results and the reference matching results, a target account that matches the target search term is screened out from the candidate accounts, including:

[0152] For each candidate account, based on the corresponding initial matching result and reference matching result, combined with the corresponding weights of the initial matching result and the reference matching result, the corresponding account evaluation value is obtained; then, according to the corresponding account evaluation value of each candidate account, the candidate account that meets the set search conditions is screened out from each candidate account as the target account.

[0153] It should be noted that in the embodiment of the present application, there can be one or more target accounts matching the target search term, and there is no limitation on this.

[0154] The account evaluation value represents the degree of match between the candidate account and the target search term. The account evaluation value can be expressed as a numerical table or a graded table. This article uses numerical values ​​as an example. The higher the account evaluation value, the closer the match between the candidate account and the target search term.

[0155] Among them, the set search conditions include but are not limited to at least one of the following: the account evaluation value of the candidate account exceeds the set evaluation value threshold; the account evaluation value of the candidate account is within a certain range; the account evaluation value of the candidate account is in the top K.

[0156] For example, when screening candidate accounts that meet the set search criteria from the candidate accounts based on the account evaluation values ​​corresponding to each candidate account, the candidate accounts can be sorted according to their corresponding account evaluation values, and based on the sorting results, the candidate accounts that meet the set search criteria can be screened from the candidate accounts as target accounts. The candidate accounts can be sorted from largest to smallest based on the account evaluation values. In actual applications, the order can also be sorted from smallest to largest based on the account evaluation values, and this is not limited to this.

[0157] For example, see Figure 9 As shown, the target search term "National Day Travel" is matched against candidate accounts such as Account A, Account B, and Account C. Taking the matching process between the target search term and Account A as an example, the target search term is string-matched with Account A's account name and profile to obtain the basic information matching degree corresponding to the account basic information. Account A's account tags include: travel, cute pets, and food. The target search term is string-matched with each of Account A's account tags to obtain the account label matching degree between each account tag and the target search term. The account label matching degree with the highest value is used as the final account label matching degree in the account label dimension. Then, based on the basic information matching degree and the account label matching degree and their corresponding weights, the account evaluation value corresponding to Account A is calculated. Similarly, the account evaluation values ​​corresponding to the candidate accounts such as Account A, Account B, and Account C can be calculated. Then, from the candidate accounts such as Account A, Account B, and Account C, the top three candidate accounts with the highest values ​​are selected as the target accounts.

[0158] Since users usually use account names to search, they will further combine content with keywords to search when the account name cannot be fully used as a search term. Therefore, in the above matching method, based on basic account information such as account name and account profile, combined with account tags to search, it can effectively improve search efficiency while ensuring search accuracy.

[0159] In some embodiments, considering that new content may be published or existing content may be deleted in an account over time, in order to maintain the accuracy of the account label of the candidate account and ensure that the target account can be retrieved in a timely and accurate manner during subsequent account searches, in an embodiment of the present application, the account label of the candidate account can be updated.

[0160] In one possible implementation, to ensure timely updates of account tags, when one of the candidate accounts updates content, the account tag set for that candidate account is updated based on the new content. Content updates include, but are not limited to, publishing new content or deleting existing content. New content refers to content currently published by a candidate account after the update.

[0161] Since the process of updating the account tag set based on the new content is the same as the process of extracting the account tag set mentioned above, it will not be described here in detail.

[0162] For example, when account A deletes video 1, the account tag set of account A is updated based on videos 2 to 10. For another example, when account A releases a new video, the account tag set of account A is updated based on the existing videos 1 to 10 and the newly released video 11.

[0163] In one possible implementation, considering the limited computing resources and to reduce computing pressure, in an embodiment of the present application, the account tag set of each candidate account may also be updated according to a set update cycle. The update cycle can be set according to actual needs, for example, based on the usage of resources (including but not limited to CPU, memory, etc.). When the resource usage rate is high, the update cycle can be daily, and when the resource usage rate is low, the update cycle can be hourly, but the present invention is not limited thereto.

[0164] In some embodiments, content tags can be represented by more fine-grained tags, for example, by using secondary categories of categories, secondary categories of entities, etc. For example, the secondary categories of categories can be game 1-movies, game 1-derivative videos, etc. Through finer-grained tag representation, they are subsequently aggregated into finer-grained candidate tags, and then filtered according to the finer-grained candidate tags to obtain finer-grained account tags, thereby improving the accuracy of account tags and further improving the account search accuracy.

[0165] Below, this application is described with reference to a specific embodiment.

[0166] See Figure 10 As shown, taking account 1 as an example, it is assumed that the content published by account 1 includes video a, picture and text b, and article c. Figure 10In the figure, pentagons are used to represent content labels corresponding to the category evaluation dimension, and triangles are used to represent content labels corresponding to the style evaluation dimension. Specifically, the pentagon labeled 1 represents category label 1 (Game 1-Game Explanation), the pentagon labeled 2 represents category label 2 (Game 1-Live Recording), the pentagon labeled 3 represents category label 3 (Game 1-Game Strategy), the pentagon labeled 4 represents category label 4 (Game 2-Game Strategy), the pentagon labeled 5 represents category label 5 (Game 3-Game Strategy), and the pentagon labeled 6 represents category label 6 (Game 2-Movie Commentary). The triangle labeled 1 represents style label 1 (Funny), the triangle labeled 2 represents style label 2 (Warm), the triangle labeled 3 represents style label 3 (Newbie), and the triangle labeled 4 represents style label 4 (Daily).

[0167] Among them, the content labels corresponding to video a in the category evaluation dimension include: category label 1, category label 2 and category label 3; the content labels corresponding to picture and text b in the category evaluation dimension include: category label 1, category label 3 and category label 4; the content labels corresponding to article c in the category evaluation dimension include: category label 2, category label 5 and category label 6.

[0168] The content labels corresponding to video a in the style evaluation dimension include: style label 1, style label 2, and style label 3; the content labels corresponding to image and text b in the style evaluation dimension include: style label 2, style label 3, and style label 4; the content labels corresponding to article c in the style evaluation dimension include: style label 1, style label 2, and style label 3.

[0169] For the category evaluation dimension, we aggregate the corresponding content tags for video a, image b, and article c, obtaining candidate tags for account 1 in the category evaluation dimension: Category Tag 1, Category Tag 2, Category Tag 3, Category Tag 4, and Category Tag 5. Note that because the tag confidence for Category Tag 6 is below the set threshold, Category Tag 6 was filtered out during the aggregation process.

[0170] For the style evaluation dimension, the content labels corresponding to video a, image b, and article c in the style evaluation dimension are aggregated to obtain the candidate labels of account 1 in the style evaluation dimension: style label 1, style label 2, style label 3, and style label 4.

[0171] Next, for the category evaluation dimension, based on the label confidence corresponding to each candidate label, the candidate labels that meet the set confidence conditions are screened out from category label 1, category label 2, category label 3, category label 4 and category label 5 as account labels. The account labels include: category label 1, category label 2, category label 3 and category label 4.

[0172] For the style evaluation dimension, based on the label confidence corresponding to each candidate label, candidate labels that meet the set confidence conditions are screened out from style label 1, style label 2, style label 3 and style label 4 as account labels. The account labels include: style label 1, style label 2 and style label 3.

[0173] In summary, Account 1's account tags include: Category Tag 1, Category Tag 2, Category Tag 3, Category Tag 4, Style Tag 1, Style Tag 2, and Style Tag 3. When the target search term is "xxx game 1 commentary," Account 1 can be accurately retrieved based on its account name "xxxyyy," account profile, and account tags.

[0174] Based on the same inventive concept, the embodiment of the present application provides an account search device. Figure 11 As shown, it is a schematic diagram of the structure of the account search device 1100, which may include:

[0175] The tag acquisition unit 1101 is configured to, upon receiving an account search instruction for a target search term, acquire an account tag set corresponding to each candidate account; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account;

[0176] The account matching unit 1102 is configured to match each account tag contained in each acquired account tag set with the target keyword, and select a target account matching the target search term from the candidate accounts based on the matching results.

[0177] As a possible implementation, the tag acquisition unit 1101 is further configured to obtain an account tag set corresponding to a candidate account in the following manner:

[0178] Extracting content from each of the candidate accounts to obtain corresponding content information, each piece of content information including: a set of content tags corresponding to the corresponding content in at least one evaluation dimension;

[0179] Based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition is screened out from the content tags as the account tag in the account tag set.

[0180] As a possible implementation, based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition is screened out from the content tags and used as the account tag in the account tag set. The tag acquisition unit 1101 is specifically configured to:

[0181] For the at least one evaluation dimension, perform the following operations:

[0182] Aggregating content tags corresponding to each of the published content under an evaluation dimension to obtain candidate tags corresponding to the candidate account under the evaluation dimension;

[0183] Obtaining a label confidence corresponding to each candidate label based on the label confidence of each content label corresponding to each published content under the one evaluation dimension;

[0184] Based on the tag confidences corresponding to the candidate tags, at least one candidate tag that meets a set confidence condition is screened out from the candidate tags as the account tag in the account tag set.

[0185] As a possible implementation, based on the tag confidence corresponding to each candidate tag, at least one candidate tag that meets a set confidence condition is screened out from the candidate tags and used as the account tag in the account tag set. The tag acquisition unit 1101 is specifically configured to:

[0186] Based on the label confidences corresponding to the candidate labels, a total label confidence corresponding to the candidate account in the evaluation dimension is obtained, and based on the total label confidence, a confidence ratio corresponding to each candidate label is determined;

[0187] If there is at least one candidate tag among the candidate tags whose corresponding tag confidence exceeds the set confidence threshold and whose confidence ratio exceeds the set ratio threshold, the at least one candidate tag is used as the account tag in the account tag set.

[0188] As a possible implementation, when aggregating the content tags corresponding to the respective published contents under one evaluation dimension to obtain the candidate tags corresponding to the candidate account under the one evaluation dimension, the tag acquisition unit 1101 is specifically configured to:

[0189] Aggregating content tags corresponding to each of the published content under the one evaluation dimension to obtain initial tags corresponding to the candidate account under the one evaluation dimension;

[0190] Based on the label confidences corresponding to the respective initial labels, initial labels whose corresponding label confidences exceed a set label threshold are screened out from the initial labels as the candidate labels corresponding to the candidate account in the evaluation dimension.

[0191] As a possible implementation, when extracting content from a published content under the candidate account and obtaining corresponding content information, the tag acquisition unit 1101 is specifically configured to perform at least one of the following operations:

[0192] Based on the content modality type of the published content, category extraction is performed on the published content to obtain a content tag set corresponding to the category evaluation dimension of the published content;

[0193] Based on the published content and its corresponding historical feedback information, entity extraction is performed on the published content to obtain a content tag set corresponding to the published content in an entity evaluation dimension.

[0194] As a possible implementation, each account tag contained in each acquired account tag set is matched with the target keyword, and based on the matching results, a target account matching the target search term is selected from the candidate accounts. The account matching unit 1102 is specifically configured to:

[0195] Matching the basic account information of each candidate account with the target search term to obtain corresponding initial matching results;

[0196] Matching each account tag in the account tag set corresponding to each candidate account with the target search term to obtain a corresponding reference matching result;

[0197] Based on the obtained initial matching results and reference matching results, a target account that matches the target search term is screened out from the candidate accounts.

[0198] As a possible implementation, the label acquisition unit 1101 is further configured to:

[0199] When one of the candidate accounts updates its content, the account tag set of the candidate account is updated according to the new content; or

[0200] The account tag set of each candidate account is updated according to a set update cycle.

[0201] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0202] Regarding the apparatus in the above embodiment, the specific manner in which each unit executes the request has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0203] Those skilled in the art will appreciate that various aspects of the present application can be implemented as systems, methods, or program products. Therefore, various aspects of the present application can be specifically implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation that combines hardware and software aspects, which may be collectively referred to herein as a "circuit," "module," or "system."

[0204] Based on the same inventive concept, an embodiment of the present application further provides an electronic device. In one embodiment, the electronic device may be a server or a terminal device. Figure 12 , which is a schematic structural diagram of a possible electronic device provided in an embodiment of the present application, Figure 12 In the embodiment, the electronic device 1200 includes: a processor 1210 and a memory 1220.

[0205] The memory 1220 stores a computer program that can be executed by the processor 1210 . The processor 1210 can execute the steps of the above-mentioned account search method by executing the instructions stored in the memory 1220 .

[0206] Memory 1220 may be a volatile memory, such as random-access memory (RAM); a non-volatile memory, such as read-only memory (ROM), flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1220 may also be a combination of the above memories.

[0207] The processor 1210 may include one or more central processing units (CPUs) or digital processing units, etc. The processor 1210 is configured to implement the above-mentioned account search method when executing the computer program stored in the memory 1220 .

[0208] In some embodiments, the processor 1210 and the memory 1220 may be implemented on the same chip. In some embodiments, they may also be implemented on separate chips.

[0209] In the embodiment of the present application, the specific connection medium between the processor 1210 and the memory 1220 is not limited. In the embodiment of the present application, the processor 1210 and the memory 1220 are connected via a bus as an example. Figure 12 The connections between the other components are shown in bold lines for illustration only and are not intended to be limiting. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Figure 12 The diagram shows a single thick line, but this does not indicate that there is only one bus or one type of bus.

[0210] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to perform the steps of the above-mentioned account search method. In some possible implementations, various aspects of the account search method provided by the present application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to perform the steps in the above-mentioned account search method. For example, the electronic device can perform the following steps: Figure 2 Follow the steps shown in .

[0211] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, RAM, ROM, an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0212] The program product of the embodiments of the present application may be a CD-ROM and include a computer program, and may be run on an electronic device. However, the program product of the present application is not limited thereto. In this document, a readable storage medium may be any tangible medium that contains or stores a computer program that can be used by or in conjunction with a command execution system, apparatus, or device.

[0213] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a readable computer program. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a computer program for use by or in conjunction with a command execution system, apparatus, or device.

[0214] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0215] Obviously, those skilled in the art may make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application is intended to include these modifications and variations.

Claims

1. An account search method, characterized in that: include: When an account search instruction for a target search term is received, an account tag set corresponding to each candidate account is obtained; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account; Each account tag contained in each of the obtained account tag sets is matched with the target keyword respectively, and based on the matching results, a target account matching the target search term is screened out from the candidate accounts.

2. The method according to claim 1, wherein The account tag set corresponding to a candidate account is obtained in the following way: Extracting content from each of the candidate accounts to obtain corresponding content information, each piece of content information including: a set of content tags corresponding to the corresponding content in at least one evaluation dimension; Based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition is screened out from the content tags as the account tag in the account tag set.

3. The method according to claim 2, wherein The step of selecting, based on the obtained tag confidences of the content tags corresponding to the respective published contents, at least one content tag that meets a set confidence condition from the content tags as the account tag in the account tag set includes: For the at least one evaluation dimension, perform the following operations: Aggregating the content tags corresponding to the respective published contents under an evaluation dimension to obtain the candidate tags corresponding to the candidate account under the evaluation dimension; Obtaining a label confidence corresponding to each candidate label based on the label confidence of each content label corresponding to each published content under the one evaluation dimension; Based on the tag confidences corresponding to the candidate tags, at least one candidate tag that meets a set confidence condition is screened out from the candidate tags as the account tag in the account tag set.

4. The method according to claim 3, wherein The step of selecting at least one candidate tag that meets a set confidence condition from the candidate tags based on the tag confidences corresponding to the candidate tags as the account tag in the account tag set includes: Based on the label confidences corresponding to the candidate labels, a total label confidence corresponding to the candidate account in the evaluation dimension is obtained, and based on the total label confidence, a confidence ratio corresponding to each candidate label is determined; If there is at least one candidate tag among the candidate tags whose corresponding tag confidence exceeds the set confidence threshold and whose confidence ratio exceeds the set ratio threshold, the at least one candidate tag is used as the account tag in the account tag set.

5. The method according to claim 3, wherein Aggregating the content tags corresponding to the respective published contents under one evaluation dimension to obtain the candidate tags corresponding to the candidate account under the one evaluation dimension includes: Aggregating content tags corresponding to each of the published content under the one evaluation dimension to obtain initial tags corresponding to the candidate account under the one evaluation dimension; Based on the label confidences corresponding to the respective initial labels, initial labels whose corresponding label confidences exceed a set label threshold are screened out from the initial labels as the candidate labels corresponding to the candidate account in the evaluation dimension.

6. The method according to any one of claims 2 to 5, wherein: Extracting content from a published content under the candidate account to obtain corresponding content information includes at least one of the following operations: Based on the content modality type of the published content, category extraction is performed on the published content to obtain a content tag set corresponding to the category evaluation dimension of the published content; Based on the published content and its corresponding historical feedback information, entity extraction is performed on the published content to obtain a content tag set corresponding to the published content in an entity evaluation dimension.

7. The method according to any one of claims 1 to 5, wherein Match each account tag contained in each acquired account tag set with the target keyword, and select target accounts matching the target search term from the candidate accounts based on the matching results, including: Matching the basic account information of each candidate account with the target search term to obtain corresponding initial matching results; Matching each account tag in the account tag set corresponding to each candidate account with the target search term to obtain a corresponding reference matching result; Based on the obtained initial matching results and reference matching results, a target account that matches the target search term is screened out from the candidate accounts.

8. The method according to any one of claims 1 to 5, wherein Also includes: When one of the candidate accounts updates its content, updating the account tag set of the candidate account according to the new content; or, The account tag set of each candidate account is updated according to a set update cycle.

9. An account search device, characterized in that: include: a tag acquisition unit configured to, upon receiving an account search instruction for a target search term, acquire an account tag set corresponding to each candidate account; wherein each account tag set is determined based on content information of each published content under the corresponding candidate account; The account matching unit is used to match each account tag contained in each acquired account tag set with the target keyword respectively, and screen out the target account matching the target search term from the candidate accounts based on the matching results.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to perform the steps of any one of the methods of claims 1 to 8.

11. A computer-readable storage medium, characterized in that It includes a computer program, which performs the steps of any one of the methods described in 1 to 8.

12. A computer program product, characterized in that It includes a computer program, which is stored in a computer-readable storage medium. A processor of an electronic device reads and executes the computer program from the computer-readable storage medium, so that the electronic device performs the steps of any one of the methods of claims 1 to 8.

Images