A data processing, model training method, device, equipment and storage medium

Abstract

The present disclosure relates to a data processing and model training method, apparatus, device, and storage medium, and relates to the field of computer technology. The method can accurately determine the correlation between a topic tag and a short video. The data processing method includes: determining first feature data, second feature data, and third feature data; the first feature data is used to represent the matching degree between a multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent consumption data of the multimedia resource; and the third feature data is used to represent user profile data of the account that published the multimedia resource; and determining the correlation between the multimedia resource and the topic tag based on the first, second, and third feature data.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and particularly to a data processing, model training method, apparatus, device and storage medium. Background Art

[0002] The rise of short videos has gradually become one of the main means for people to record their lives and one of the main forms of people's daily consumption and entertainment. Topic tags (Hashtags) are the definitions of short videos by the publishing accounts, which can help viewing accounts quickly understand the basic information of the videos. A video can have multiple hashtags.

[0003] However, the publishing accounts of short videos may randomly edit hashtags due to exposure, which will have a relatively serious impact on the search and recommendation services provided by short video platforms. For example, the publishing account of a short video publishes a video of running in the evening, but in order to gain more exposure, it tags the video with "#Is Brand A mobile phone good?". At this time, when other viewing accounts search for Brand A mobile phones, they will get this video, and this video is an irrelevant result to Brand A mobile phones.

[0004] Therefore, how to determine and process hashtags irrelevant to short videos is a technical problem that urgently needs to be solved at present. Summary of the Invention

[0005] The present disclosure provides a data processing, model training method, apparatus, device and storage medium, which can accurately determine the relevance between topic tags and short videos.

[0006] The technical solution of the embodiment of the present disclosure is as follows:

[0007] According to the first aspect of the embodiment of the present disclosure, a data processing method is provided, and this method can be applied to an electronic device. This method may include: determining first feature data, second feature data and third feature data; the first feature data is used to represent the matching degree between the multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent the consumption data of the multimedia resource; the third feature data is used to represent the user profile data of the account that publishes the multimedia resource; and determining the relevance between the multimedia resource and the topic tag according to the first feature data, the second feature data and the third feature data.

[0008] Optionally, the method for determining the relevance between the multimedia resource and the topic tag according to the first feature data, the second feature data and the third feature data specifically includes: inputting the first feature data, the second feature data and the third feature data into a relevance detection model to obtain a relevance; the relevance detection model is a model that has been pre-trained to a convergence state and is used to detect the relevance between the multimedia resource and the topic tag.

[0009] Optionally, the method for determining the first feature data specifically includes: obtaining associated data of the multimedia resource; extracting the text feature vector and image feature vector of the multimedia resource from the associated data; determining the first matching data between the topic tag and the text feature vector; obtaining the image feature vector of the associated multimedia resource related to the topic tag; determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource; and determining the first matching data and the second matching data as the first feature data.

[0010] Optionally, the associated data includes at least one of audio data, cover data, image frame data, category label, account information and comment information of the multimedia resource.

[0011] Optionally, the method of extracting text feature vectors and image feature vectors of multimedia resources from associated data specifically includes: extracting credible text and uncredible text from the associated data; credible text is text data whose degree of association with the multimedia resource is a first degree of association; uncredible text is text data whose degree of association with the multimedia resource is a second degree of association; the first degree of association is greater than the second degree of association; converting the credible text into a first text feature vector, and converting the uncredible text into a second text feature vector, and determining that the text feature vector includes the first text feature vector and the second text feature vector; using a pre-trained image feature extraction model, extracting the first image feature vector of the cover data and the second image feature vector of the image frame data, and determining that the image feature vector includes the first image feature vector and the second image feature vector.

[0012] Optionally, the trusted text includes: text in the classification label, text in the account information, at least one of the music name text and singer text extracted from the audio data using speech recognition technology; the untrusted text includes: text in the comment information, lyrics text extracted from the audio data using speech recognition technology, at least one of the text recognized from the image frame data using text recognition technology.

[0013] Optionally, the method for determining the first matching data between a topic tag and a text feature vector specifically includes: obtaining at least one associated text feature vector corresponding to the topic tag; determining the literal matching degree and semantic matching degree of each associated text feature vector in the at least one associated text feature vector with the text feature vector; and determining the first matching data based on the literal matching degree and the semantic matching degree.

[0014] Optionally, the method for obtaining the image feature vector of the associated multimedia resource related to the topic tag specifically includes: obtaining the cover data and image frame data of the associated multimedia resource: using the image feature extraction model to extract the third image feature vector of the cover data of the associated multimedia resource and the fourth image feature vector of the image frame data of the associated multimedia resource; determining that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector.

[0015] Optionally, the method for determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource specifically includes: determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector.

[0016] Optionally, the data processing method further includes: when the correlation between the multimedia resource and the topic tag is less than a preset correlation threshold, releasing the correlation between the multimedia resource and the topic tag.

[0017] According to a second aspect of an embodiment of the present disclosure, a model training method is provided, which can be applied to an electronic device. The model training method includes: obtaining training data of multimedia resources within a preset time period; the training data includes training sample input data and training sample label data; the training sample input data includes: the matching degree between the multimedia resource and the multimedia resource's topic tag, multimedia resource consumption data, and user profile data of the account that publishes the multimedia resource; the training sample label data includes: the correlation between the multimedia resource and the multimedia resource's topic tag; based on the training data, training a correlation detection model; the correlation detection model is used to detect the correlation between the multimedia resource and the topic tag.

[0018] According to a third aspect of an embodiment of the present disclosure, a data processing device is provided, which can be applied to an electronic device. The device may include: a determining unit; the determining unit is configured to determine first feature data, second feature data, and third feature data; the first feature data is configured to indicate a matching degree between a multimedia resource and a topic tag associated with the multimedia resource; the second feature data is configured to indicate consumption data of the multimedia resource; and the third feature data is configured to indicate user profile data of an account that publishes the multimedia resource; and the determining unit is further configured to determine a correlation between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data.

[0019] Optionally, the determination unit is specifically used to: input the first feature data, the second feature data and the third feature data into a correlation detection model to obtain the correlation; the correlation detection model is a model that is pre-trained to a convergence state and is used to detect the correlation between multimedia resources and topic tags.

[0020] Optionally, the determination unit is specifically used to: obtain associated data of multimedia resources; extract text feature vectors and image feature vectors of multimedia resources from the associated data; determine first matching data between topic tags and text feature vectors; obtain image feature vectors of associated multimedia resources related to topic tags; determine second matching data between the image feature vectors of associated multimedia resources and the image feature vectors of multimedia resources; and determine that the first feature data includes first matching data and second matching data.

[0021] Optionally, the associated data includes at least one of audio data, cover data, image frame data, category label, account information and comment information of the multimedia resource.

[0022] Optionally, the determination unit is specifically used to: extract credible text and uncredible text from associated data; credible text is text data whose association degree with multimedia resources is a first association degree; uncredible text is text data whose association degree with multimedia resources is a second association degree; when the first association degree is greater than the second association degree, the credible text is converted into a first text feature vector, and the uncredible text is converted into a second text feature vector, and it is determined that the text feature vector includes the first text feature vector and the second text feature vector; using a pre-trained image feature extraction model, the first image feature vector of the cover data and the second image feature vector of the image frame data are extracted, and it is determined that the image feature vector includes the first image feature vector and the second image feature vector.

[0023] Optionally, the trusted text includes: text in the classification label, text in the account information, at least one of the music name text and singer text extracted from the audio data using speech recognition technology; the untrusted text includes: text in the comment information, lyrics text extracted from the audio data using speech recognition technology, at least one of the text recognized from the image frame data using text recognition technology.

[0024] Optionally, the determination unit is specifically used to: obtain at least one associated text feature vector corresponding to the topic tag; determine the literal matching degree and semantic matching degree of each associated text feature vector in the at least one associated text feature vector with the text feature vector; and determine the first matching data based on the literal matching degree and the semantic matching degree.

[0025] Optionally, the determination unit is specifically used to: obtain the cover data and image frame data of the associated multimedia resource: use the image feature extraction model to extract the third image feature vector of the cover data of the associated multimedia resource and the fourth image feature vector of the image frame data of the associated multimedia resource; determine that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector.

[0026] Optionally, the determining unit is specifically configured to determine second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector.

[0027] Optionally, the determining unit is further configured to disassociate the multimedia resource from the topic tag when the association between the multimedia resource and the topic tag is less than a preset association threshold.

[0028] According to the fourth aspect of the embodiment of the present disclosure, a model training device is provided, including: an acquisition unit and a training unit; the acquisition unit is used to acquire training data of multimedia resources within a preset time period; the training data includes training sample input data and training sample label data; the training sample input data includes: the matching degree between the multimedia resource and the topic tag of the multimedia resource, the consumption data of the multimedia resource, and the user portrait data of the account that publishes the multimedia resource; the training sample label data includes: the correlation between the multimedia resource and the topic tag of the multimedia resource; the training unit is used to train a correlation detection model based on the training data; the correlation detection model is used to detect the correlation between the multimedia resource and the topic tag.

[0029] According to a fifth aspect of an embodiment of the present disclosure, an electronic device is provided, which may include: a processor and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement any one of the optional data processing methods in the first aspect above, or to implement the model training method in the second aspect.

[0030] According to the sixth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided, on which instructions are stored. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute any one of the optional data processing methods in the first aspect above, or implement the model training method in the second aspect.

[0031] According to the seventh aspect of the embodiments of the present disclosure, a computer program product is provided, which includes computer instructions. When the computer instructions are run on an electronic device, the electronic device executes the data processing method as described in any optional implementation method of the first aspect, or implements the model training method of the second aspect.

[0032] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

[0033] The technical solutions provided by the embodiments of the present disclosure bring at least the following beneficial effects:

[0034] Based on any of the above aspects, in the present disclosure, after determining the first feature data, the second feature data and the third feature data, since the first feature data is used to represent the matching degree between the multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent the consumption data of the multimedia resource; the third feature data is used to represent the user portrait data of the account that publishes the multimedia resource, therefore, according to the first feature data, the second feature data and the third feature data, the correlation between the multimedia resource and the topic tag can be obtained quickly and accurately, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource, and improving the user experience of the multimedia resource in the search recommendation scenario. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The accompanying drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute an improper limitation of the present disclosure.

[0036] Figure 1 A flow chart of a data processing method provided by an embodiment of the present disclosure is shown;

[0037] Figure 2 A flow chart showing another data processing method provided by an embodiment of the present disclosure is shown;

[0038] Figure 3 A flow chart showing another data processing method provided by an embodiment of the present disclosure is shown;

[0039] Figure 4 A flow chart showing another data processing method provided by an embodiment of the present disclosure is shown;

[0040] Figure 5 A flow chart showing another data processing method provided by an embodiment of the present disclosure is shown;

[0041] Figure 6 A flow chart showing another data processing method provided by an embodiment of the present disclosure is shown;

[0042] Figure 7 A flow chart of a model training method provided by an embodiment of the present disclosure is shown;

[0043] Figure 8 A schematic structural diagram of another data processing device provided by an embodiment of the present disclosure is shown;

[0044] Figure 9 A schematic structural diagram of a model training device provided by an embodiment of the present disclosure is shown;

[0045] Figure 10A schematic diagram of the structure of a terminal provided by an embodiment of the present disclosure is shown;

[0046] Figure 11 A schematic structural diagram of a server provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0047] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0048] It should be noted that the terms "first," "second," and the like in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the numbers used in this manner are interchangeable where appropriate so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure as detailed in the appended claims.

[0049] It will also be understood that the term “comprising” indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements and / or components.

[0050] The data involved in this disclosure may be data authorized by the user or fully authorized by all parties.

[0051] As described in the background, short video publishers may arbitrarily edit hashtags to maximize exposure, significantly impacting the search and recommendation services provided by short video platforms. For example, a short video publisher might post a video of an evening run, but to maximize exposure, they tag it with the hashtag "#BrandAMobilephoneIsItGood" (#BrandAMobilephone). When other viewers search for BrandAMobilephone, they will see this video, which is irrelevant to BrandA. Therefore, identifying and handling hashtags that are irrelevant to the short video is a pressing technical issue.

[0052] Based on this, the embodiment of the present disclosure provides a data processing method. After determining the first feature data, the second feature data and the third feature data, since the first feature data is used to represent the matching degree between the multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent the consumption data of the multimedia resource; and the third feature data is used to represent the user portrait data of the account that publishes the multimedia resource, therefore, according to the first feature data, the second feature data and the third feature data, the correlation between the multimedia resource and the topic tag can be obtained quickly and accurately, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource, and improving the user experience of the multimedia resource in the search recommendation scenario.

[0053] The following is an exemplary description of the data processing method provided in the embodiment of the present disclosure:

[0054] The data processing method provided by the present disclosure can be applied to electronic devices.

[0055] In some embodiments, the electronic device may be a server, a terminal, or other electronic device for performing data processing, which is not limited in this disclosure.

[0056] The server may be a single server, or a server cluster consisting of multiple servers. In some implementations, the server cluster may also be a distributed cluster. The present disclosure does not limit the specific implementation of the server.

[0057] The terminal can be a mobile phone, tablet computer, desktop computer, laptop computer, handheld computer, notebook computer, ultra-mobile personal computer (UMPC), netbook computer, cellular phone, personal digital assistant (PDA), augmented reality (AR) or virtual reality (VR) device, etc., which can install and use content community applications (such as Kuaishou). The present disclosure does not impose any special restrictions on the specific form of the electronic device. It can interact with the user through one or more methods such as keyboard, touchpad, touch screen, remote control, voice interaction or handwriting device.

[0058] The data processing method provided in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0059] like Figure 1 As shown, when the data processing method is applied to an electronic device, the data processing method may include:

[0060] S101: The electronic device determines first feature data, second feature data, and third feature data.

[0061] Specifically, when determining the relevance between the multimedia resource and the topic tag, the electronic device may first determine the first feature data, the second feature data, and the third feature data.

[0062] The first feature data is used to indicate the matching degree between the multimedia resource and the topic tag of the multimedia resource.

[0063] Optionally, the first feature data may be matching data between the topic tag and text data in the associated data of the multimedia resource, or matching data between the image features of the multimedia resource associated with the topic tag and the image features of the multimedia resource, or other feature data used to represent the degree of matching between the multimedia resource and the topic tag of the multimedia resource. This disclosure does not limit this.

[0064] The second characteristic data is used to represent consumption data of the multimedia resource.

[0065] Optionally, the consumption data includes the number of interactive operations performed on the multimedia resource and the content of the interactive operations performed.

[0066] Exemplarily, the number of interactive operations performed includes: playback volume, click volume, comment volume, like volume, share volume, etc. The content of the interactive operations performed includes: comment content, search query content that triggers clicks, etc.

[0067] The third feature data is used to represent user portrait data of the account that publishes multimedia resources.

[0068] Optionally, the user portrait data includes the account information of the user who publishes the multimedia resources and the user's behavior data.

[0069] For example, account information includes: account name information, friend information, etc. Behavior data includes: search term data, category label data, etc.

[0070] S102: The electronic device determines the relevance between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data.

[0071] Optionally, when the electronic device determines the correlation between multimedia resources and topic tags based on the first feature data, the second feature data and the third feature data, the first feature data, the second feature data and the third feature data can be input into a correlation detection model to obtain the correlation between the multimedia resources and the topic tags.

[0072] The relevance detection model is a model that is pre-trained to a convergence state and is used to detect the relevance between multimedia resources and topic tags.

[0073] Specifically, after determining the first feature data, the second feature data, and the third feature data, the first feature data, the second feature data, and the third feature data may be input into a relevance detection model to obtain the relevance between the multimedia resource and the topic tag.

[0074] Optionally, the association detection model may be a tree model (XGBoost), or a model in an end-to-end model for determining the association, or other models, which is not limited in the present disclosure.

[0075] Alternatively, the relevance detection model can learn the association between input feature data through feature combination and output a score. By empirically classifying the score, the relevance between the multimedia resource and the topic tag is output.

[0076] Exemplarily, the relevance between multimedia resources and topic tags includes three levels: A (corresponding to a model output score of 80-100 points), B (corresponding to a model output score of 60-80 points), and C (corresponding to a model output score of 0-60 points). After determining the first feature data, the second feature data, and the third feature number, the first feature data, the second feature data, and the third feature data can be input into the relevance detection model to obtain a score of 85 for the relevance between multimedia resources and topic tags. In this case, the level of relevance between multimedia resources and topic tags is determined to be A.

[0077] Optionally, when the electronic device determines the correlation between multimedia resources and topic tags based on the first feature data, the second feature data and the third feature data, it can also quantify the first feature data, the second feature data and the third feature data, assign weights to the quantized feature data respectively, and determine the correlation between multimedia resources and topic tags based on the similarity algorithm.

[0078] The similarity algorithm may be a Euclidean distance algorithm, a Pearson correlation coefficient algorithm, a cosine similarity algorithm, or the like.

[0079] Optionally, the electronic device may also determine the relevance between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data through other existing relevance detection algorithms, which is not limited in the present disclosure.

[0080] The technical solution provided by the above embodiment brings at least the following beneficial effects: It can be seen from S101-S102 that after determining the first feature data, the second feature data and the third feature data, since the first feature data is used to represent the matching degree between the multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent the consumption data of the multimedia resource; the third feature data is used to represent the user portrait data of the account that publishes the multimedia resource, therefore, according to the first feature data, the second feature data and the third feature data, the correlation between the multimedia resource and the topic tag can be obtained quickly and accurately, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource, and improving the user experience of the multimedia resource in the search recommendation scenario.

[0081] In one embodiment, the data processing method further includes:

[0082] When the association degree between the multimedia resource and the topic tag is less than a preset association degree threshold, the electronic device releases the association relationship between the multimedia resource and the topic tag.

[0083] Specifically, after inputting the first feature data, the second feature data, and the third feature data into the association detection model to obtain the association between the multimedia resource and the topic tag, it can be determined whether the association between the multimedia resource and the topic tag is less than a preset association threshold. If the association between the multimedia resource and the topic tag is less than the preset association threshold, it indicates that the topic tag is not a topic tag related to the multimedia resource, and therefore, the electronic device disassociates the multimedia resource from the topic tag.

[0084] Correspondingly, when the relevance between the multimedia resource and the topic tag is greater than or equal to the preset relevance threshold, it indicates that the topic tag is a topic tag related to the multimedia resource, and therefore, the electronic device does not need to perform any processing.

[0085] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from the above, after the first feature data, the second feature data, and the third feature data are input into the correlation detection model to obtain the correlation between the multimedia resource and the topic tag, it can be determined whether the correlation between the multimedia resource and the topic tag is less than the preset correlation threshold. When the correlation between the multimedia resource and the topic tag is less than the preset correlation threshold, it means that the topic tag is not a topic tag related to the multimedia resource. Therefore, the electronic device disassociates the multimedia resource from the topic tag, avoids irrelevant topic tags affecting the search service of the multimedia resource, and improves the user experience of the multimedia resource in the search recommendation scenario.

[0086] In one embodiment, combining Figure 1 ,like Figure 2 As shown, in the above S101, the method for the electronic device to determine the first feature data specifically includes:

[0087] S201: The electronic device obtains associated data of multimedia resources.

[0088] Specifically, when determining the first characteristic data, the electronic device may obtain associated data of the multimedia resource.

[0089] Optionally, the associated data includes at least one of audio data, cover data, image frame data, category label, account information and comment information of the multimedia resource.

[0090] Since the above-mentioned association data is detailed data associated with the multimedia resource, the electronic device can subsequently accurately determine the first feature data based on the above-mentioned association data, and then accurately determine the association between the target multimedia resource and the topic tag based on the first feature data.

[0091] For example, when the multimedia resource is a short video, the audio data may be the background music of the short video. The cover data may be the cover image of the short video. The image frame data may be the image of each frame in the short video. The classification label may be the machine classification label of the short video obtained by a classifier (for example, the classification labels of a basketball-themed short video include "basketball", "sports", etc.). The account information may include the account name of the account that posted the short video, the name of the friend, etc. The comment information may be the comments in the short video.

[0092] S202: The electronic device extracts text feature vectors and image feature vectors of multimedia resources from the associated data.

[0093] Specifically, after obtaining the associated data of the multimedia resource, since the associated data includes at least one of the multimedia resource's audio data, cover data, image frame data, classification label, account information, and comment information, the electronic device can extract the text feature vector and image feature vector of the multimedia resource from the associated data.

[0094] Optionally, when the electronic device extracts the text feature vector of the multimedia resource from the associated data, it may use a pre-trained text feature vector extraction model to extract the text feature vector of the multimedia resource from the associated data.

[0095] Optionally, when the electronic device extracts the image feature vector of the multimedia resource from the associated data, it may use a pre-trained image feature vector extraction model to extract the image feature vector of the multimedia resource from the associated data.

[0096] S203: The electronic device determines first matching data between the topic tag and the text feature vector.

[0097] Specifically, after extracting the text feature vector and the image feature vector of the multimedia resource from the associated data, the electronic device may determine first matching data between the topic tag and the text feature vector.

[0098] Optionally, when the electronic device determines the first matching data between the topic tag and the text feature vector, it may use a pre-trained text semantic matching model to determine the first matching data between the topic tag and the text feature vector.

[0099] S204: The electronic device obtains an image feature vector of an associated multimedia resource related to the topic tag.

[0100] Specifically, after extracting the text feature vector and the image feature vector of the multimedia resource from the associated data, the electronic device may further obtain the image feature vector of the associated multimedia resource related to the topic tag.

[0101] Among them, the associated multimedia resources related to the topic tag can be all multimedia resources obtained by searching the topic tag, or multimedia resources whose partial correlation meets the preset conditions, or other associated multimedia resources related to the topic tag, which is not limited in this disclosure.

[0102] For example, the topic tag is "Brand A mobile phone". The electronic device can mine the top five videos with the best consumption effects when "Brand A mobile phone" is used as the search query based on the search log as the associated multimedia resources related to the topic tag "Brand A mobile phone".

[0103] S205: The electronic device determines second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource.

[0104] Specifically, after obtaining the image feature vector of the associated multimedia resource related to the topic tag, the electronic device may determine second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource.

[0105] Specifically, when determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource, the image feature vector of the associated multimedia resource may be obtained.

[0106] Specifically, when obtaining the image feature vector of the associated multimedia resource, the cover image and image frame of the associated multimedia resource may be obtained, and the cover image and image frame of the associated multimedia resource may be converted into the image feature vector of the associated multimedia resource.

[0107] When determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource, the image feature vector of the multimedia resource may also be obtained.

[0108] Specifically, when obtaining the image feature vector of the multimedia resource, the cover image and image frame of the multimedia resource may be obtained, and the cover image and image frame of the multimedia resource may be converted into the image feature vector of the multimedia resource.

[0109] After obtaining the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource, second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource is determined based on the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource.

[0110] It should be noted that the order in which the electronic device obtains the image feature vector of the associated multimedia resource and obtains the image feature vector of the multimedia resource is not limited.

[0111] Optionally, the electronic device may first obtain the image feature vector of the associated multimedia resource and then obtain the image feature vector of the multimedia resource; it may also first obtain the image feature vector of the multimedia resource and then obtain the image feature vector of the associated multimedia resource; it may also simultaneously obtain the image feature vector of the associated multimedia resource and obtain the image feature vector of the multimedia resource.

[0112] S206: The electronic device determines the first matching data and the second matching data as first feature data.

[0113] After determining the first matching data and the second matching data, the electronic device determines the first matching data and the second matching data to be included as first feature data.

[0114] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from S201-S206, after obtaining the associated data of the multimedia resource, the electronic device can extract the text feature vector and image feature vector of the multimedia resource from the associated data, and determine the first matching data between the topic tag and the text feature vector. The electronic device can also obtain the image feature vector of the associated multimedia resource related to the topic tag, and determine the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource. Subsequently, the electronic device determines the first matching data and the second matching data as the first feature data, thereby providing a specific implementation method for determining the first feature data.

[0115] Since the first feature data includes first matching data and second matching data, the first matching data is the matching data between the topic tag and the text feature vector, and the second matching data is the matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource. Therefore, through the first feature data, the electronic device can accurately determine the correlation between the multimedia resource and the topic tag, avoid irrelevant topic tags affecting the search service of multimedia resources, and improve the user experience of multimedia resources in the search recommendation scenario.

[0116] In one embodiment, combining Figure 2 ,like Figure 3 As shown, in the above S202, the method for the electronic device to extract the text feature vector and the image feature vector of the multimedia resource from the associated data specifically includes:

[0117] S301. The electronic device extracts credible text and uncredible text from associated data.

[0118] Specifically, when the electronic device extracts the text feature vector of the multimedia resource from the associated data, it can extract the credible text and the uncredible text from the associated data.

[0119] Among them, the credible text is text data with a first degree of association with the multimedia resource; the uncredible text is text data with a second degree of association with the multimedia resource; and the first degree of association is greater than the second degree of association.

[0120] Optionally, since the credible text is used to indicate that the degree of association between the text and the multimedia resource is highly credible, the credible text includes: text in the classification label, text in the account information, at least one of the music name text and singer text extracted from the audio data using speech recognition technology.

[0121] Optionally, since the credibility of the untrustworthy text used to represent the degree of association between the text and the multimedia resource is low, the untrustworthy text includes at least one of: text in the comment information, lyrics text extracted from audio data using speech recognition technology, and text recognized from image frame data using text recognition technology.

[0122] Since the above-mentioned trusted text and untrusted text are both text data in the associated data of the multimedia resources, the electronic device can subsequently accurately determine the first feature data based on the above-mentioned associated data, and then accurately determine the correlation between the target multimedia resource and the topic tag based on the first feature data.

[0123] S302: The electronic device converts the credible text into a first text feature vector, and converts the uncredible text into a second text feature vector, and determines that the text feature vector includes the first text feature vector and the second text feature vector.

[0124] Specifically, after extracting credible text and uncredible text from the associated data, the electronic device may convert the credible text into a first text feature vector and the uncredible text into a second text feature vector, and determine that the text feature vector includes the first text feature vector and the second text feature vector.

[0125] Optionally, when the electronic device converts the trusted text into the first text feature vector and converts the untrusted text into the second text feature vector, it can use a feature vector conversion algorithm to convert the trusted text into the first text feature vector and convert the untrusted text into the second text feature vector.

[0126] S303. The electronic device uses a pre-trained image feature extraction model to extract a first image feature vector of the cover data and a second image feature vector of the image frame data, and determines that the image feature vector includes the first image feature vector and the second image feature vector.

[0127] Specifically, when the electronic device extracts the image feature vector of the multimedia resource from the associated data, it can use a pre-trained image feature extraction model to extract the first image feature vector of the cover data and the second image feature vector of the image frame data, and determine that the image feature vector includes the first image feature vector and the second image feature vector.

[0128] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from S301-S303, when the electronic device extracts the text feature vector of the multimedia resource from the associated data, it can first extract the credible text and the uncredible text from the associated data. Then, the electronic device can convert the credible text into a first text feature vector, and convert the uncredible text into a second text feature vector, and determine that the text feature vector includes the first text feature vector and the second text feature vector. When the electronic device extracts the image feature vector of the multimedia resource from the associated data, it can use a pre-trained image feature extraction model to extract the first image feature vector of the cover data and the second image feature vector of the image frame data, and determine that the image feature vector includes the first image feature vector and the second image feature vector, thereby providing a specific implementation method for obtaining the text feature vector and image feature vector of the multimedia resource.

[0129] Since the text feature vector includes a first text feature vector (text feature vector of credible text) and a second text feature vector (text feature vector of uncredible text), and the image feature vector includes a first image feature vector (image feature vector of cover data) and a second image feature vector (image feature vector of image frame data), the text feature vector and image feature vector of multimedia resources can be used to accurately determine the correlation between multimedia resources and topic tags, thereby avoiding the impact of irrelevant topic tags on the search service of multimedia resources and improving the user experience of multimedia resources in search recommendation scenarios.

[0130] In one embodiment, combining Figure 3 ,like Figure 4 As shown, in the above S203, the method for the electronic device to determine the first matching data between the topic tag and the text feature vector specifically includes:

[0131] S401: The electronic device obtains at least one associated text feature vector corresponding to a topic tag.

[0132] Specifically, when determining the first matching data between the topic tag and the text feature vector, the electronic device may obtain at least one associated text feature vector corresponding to the topic tag.

[0133] Optionally, when the electronic device obtains at least one associated text feature vector corresponding to a topic tag, it can first rewrite the topic tag and perform entity association to obtain a reliable transformation form, and then perform feature vector processing on the transformed topic tag to obtain at least one associated text feature vector.

[0134] For example, the topic tag is: "How is Brand A mobile phone?" The electronic device can rewrite and transform the topic tag "How is Brand A mobile phone?" and associate entities to obtain reliable transformed forms, including: "Brand A mobile phone," "Brand A," "mobile phone," etc.

[0135] S402: The electronic device determines a literal matching degree and a semantic matching degree of each associated text feature vector in at least one associated text feature vector with a text feature vector.

[0136] Specifically, after obtaining at least one associated text feature vector corresponding to the topic tag, the electronic device determines the literal matching degree and semantic matching degree of each associated text feature vector in the at least one associated text feature vector with the text feature vector.

[0137] Optionally, when the electronic device determines the literal matching degree between each associated text feature vector in at least one associated text feature vector and the text feature vector, it can first obtain the text weight in the text feature vector, and then determine the literal matching degree between each associated text feature vector in at least one associated text feature vector and the text feature vector based on the text weight in the text feature vector.

[0138] Optionally, when the electronic device determines the semantic matching degree between each associated text feature vector in at least one associated text feature vector and the text feature vector, it can determine the semantic matching degree between each associated text feature vector in at least one associated text feature vector and the text feature vector based on a pre-trained text semantic matching model.

[0139] S403: The electronic device determines first matching data based on the literal matching degree and the semantic matching degree.

[0140] Specifically, after obtaining the literal matching degree and the semantic matching degree, the electronic device determines the first matching data based on the literal matching degree and the semantic matching degree.

[0141] For example, for a credible text, the electronic device may calculate a literal matching score between each associated text of at least one associated text and each text in the credible text, and take an average or maximum value as the literal matching degree between the associated text and the credible text.

[0142] Accordingly, the electronic device may calculate the semantic matching score between each associated text of at least one associated text and each text in the credible text, and take the average or maximum value as the semantic matching degree between the associated text and the credible text.

[0143] Optionally, the electronic device may further select associated texts having a score greater than a preset credible text score from calculating the literal matching score and semantic matching score of each associated text of at least one associated text and each text in the credible text, and obtain the number of associated texts having a score greater than the preset credible text score.

[0144] For the untrusted text, the electronic device may calculate a literal matching score between each associated text of at least one associated text and each text in the untrusted text, and take an average or maximum value as the literal matching degree between the associated text and the untrusted text.

[0145] Accordingly, the electronic device may calculate the semantic matching score between each associated text of at least one associated text and each text in the untrustworthy text, and take the average or maximum value as the semantic matching degree between the associated text and the untrustworthy text.

[0146] Optionally, the electronic device may also select associated texts with a score greater than a preset untrustworthy text score from calculating the literal matching score and semantic matching score of each associated text of at least one associated text and each text in the untrustworthy text, and obtain the number of associated texts with a score greater than the preset untrustworthy text score.

[0147] Subsequently, the electronic device determines that the first matching data includes: the literal matching degree between the associated text and the trusted text, the semantic matching degree between the associated text and the trusted text, the literal matching degree between the associated text and the untrusted text, the semantic matching degree between the associated text and the untrusted text, the number of associated texts greater than the preset trusted text score, and the number of associated texts greater than the preset untrusted text score.

[0148] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from S401-S403, when the electronic device determines the first matching data between the topic tag and the text feature vector, it can obtain at least one associated text feature vector corresponding to the topic tag and determine the literal matching degree and semantic matching degree of each associated text feature vector with the text feature vector. Subsequently, the electronic device determines the first matching data based on the literal matching degree and the semantic matching degree. A specific implementation method for obtaining the first matching data is provided.

[0149] Since the first matching data includes: each associated text feature vector in at least one associated text feature vector, the literal matching degree and semantic matching degree with the text feature vector, the electronic device can accurately determine the correlation between the multimedia resource and the topic tag through the first matching data, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource and improving the user experience of the multimedia resource in the search recommendation scenario.

[0150] In one embodiment, combining Figure 4 ,like Figure 5 As shown, in the above S204, the method for the electronic device to obtain the image feature vector of the associated multimedia resource related to the topic tag specifically includes:

[0151] S501: The electronic device obtains cover data and image frame data associated with multimedia resources.

[0152] Specifically, when the electronic device obtains the image feature vector of the associated multimedia resource related to the topic tag, it can obtain the cover data and image frame data of the associated multimedia resource.

[0153] Exemplarily, when the associated multimedia resource is a short video, the electronic device may obtain the video cover of the short video and each image frame in the video.

[0154] S502: The electronic device uses an image feature extraction model to extract a third image feature vector of the cover data associated with the multimedia resource and a fourth image feature vector of the image frame data associated with the multimedia resource.

[0155] Specifically, after obtaining the cover data and image frame data of the associated multimedia resource, the electronic device uses an image feature extraction model to extract a third image feature vector of the cover data of the associated multimedia resource and a fourth image feature vector of the image frame data of the associated multimedia resource.

[0156] S503: The electronic device determines that the image feature vector associated with the multimedia resource includes a third image feature vector and a fourth image feature vector.

[0157] Specifically, after using the image feature extraction model to extract the third image feature vector of the cover data of the associated multimedia resource and the fourth image feature vector of the image frame data of the associated multimedia resource, the electronic device determines that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector.

[0158] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from S501-S503, when the electronic device obtains the image feature vector of the associated multimedia resource related to the topic tag, it can obtain the cover data and image frame data of the associated multimedia resource. Subsequently, the electronic device can use the image feature extraction model to extract the third image feature vector of the cover data of the associated multimedia resource and the fourth image feature vector of the image frame data of the associated multimedia resource, and determine that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector. This provides a specific implementation method for obtaining the image feature vector of the associated multimedia resource related to the topic tag.

[0159] Since the image feature vector of the associated multimedia resource includes the third image feature vector (the image feature vector of the cover data of the associated multimedia resource) and the fourth image feature vector (the image feature vector of the image frame data of the associated multimedia resource), the electronic device can accurately determine the correlation between the multimedia resource and the topic tag by associating the image feature vector of the multimedia resource, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource and improving the user experience of the multimedia resource in the search recommendation scenario.

[0160] In one embodiment, combining Figure 5 ,like Figure 6 As shown, in the above S205, the method for the electronic device to determine the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource specifically includes:

[0161] S601: The electronic device determines second matching data between an image feature vector of an associated multimedia resource and an image feature vector of a multimedia resource based on a first image feature vector, a second image feature vector, a third image feature vector, and a fourth image feature vector.

[0162] Specifically, after obtaining the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, the electronic device can determine the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector.

[0163] Optionally, the electronic device may obtain the similarity between the first image feature vector and the third image feature vector, and obtain the similarity between the second image feature vector and the fourth image feature vector, and determine the obtained similarity as the second matching data. The electronic device may also obtain the similarity between the first image feature vector and the fourth image feature vector, and obtain the similarity between the second image feature vector and the third image feature vector, and determine the obtained similarity as the second matching data. This disclosure is not limited to this.

[0164] Alternatively, when the number of the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector is multiple, the electronic device may further obtain the average value or maximum value of the first image feature vector, the average value or maximum value of the second image feature vector, the average value or maximum value of the third image feature vector, and the average value or maximum value of the fourth image feature vector, and then obtain the similarity between the average value or maximum value of the first image feature vector and the average value or maximum value of the fourth image feature vector, and obtain the similarity between the average value or maximum value of the second image feature vector and the average value or maximum value of the third image feature vector, and determine the obtained similarity as the second matching data. This disclosure is not limited to this.

[0165] The technical solution provided by the above embodiment brings at least the following beneficial effects: It can be seen from S601 that after obtaining the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, the electronic device can determine the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, and provides a specific implementation method for determining the second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource.

[0166] Since the second matching data is determined based on the first image feature vector (feature vector of the cover data of the multimedia resource), the second image feature vector (feature vector of the image frame data of the multimedia resource), the third image feature vector (feature vector of the cover data of the associated multimedia resource), and the fourth image feature vector (feature vector of the image frame data of the associated multimedia resource), the electronic device can accurately determine the correlation between the multimedia resource and the topic tag through the second matching data, thereby avoiding the influence of irrelevant topic tags on the search service of the multimedia resource and improving the user experience of the multimedia resource in the search recommendation scenario.

[0167] In one embodiment, if Figure 7 As shown, the embodiment of the present application also provides a model training method, including:

[0168] S701: The electronic device obtains training data of multimedia resources within a preset time period.

[0169] Among them, the training data includes training sample input data and training sample label data; the training sample input data includes: the matching degree between multimedia resources and the topic tags of multimedia resources, the consumption data of multimedia resources, and the user portrait data of the account that publishes multimedia resources; the training sample label data includes: the correlation between multimedia resources and the topic tags of multimedia resources.

[0170] Specifically, before inputting the first feature data, the second feature data, and the third feature data into the association detection model to obtain the association between the multimedia resource and the topic tag, the electronic device may also obtain training data of the multimedia resource within a preset time period to train the association detection model. S702: The electronic device trains the association detection model based on the training data.

[0171] Specifically, after obtaining training data of multimedia resources within a preset time period, the electronic device trains a correlation detection model based on the training data.

[0172] Among them, the relevance detection model is used to detect the relevance between multimedia resources and topic tags.

[0173] The technical solution provided by the above embodiment brings at least the following beneficial effects: As can be seen from the above, the electronic device can also obtain the training data of the multimedia resources within a preset time period, and based on the training data, train a correlation detection model, so that subsequent electronic devices can quickly and accurately obtain the correlation between multimedia resources and topic tags based on the trained correlation detection model, avoiding irrelevant topic tags affecting the search service of multimedia resources, and improving the user experience of multimedia resources in the search recommendation scenario.

[0174] It is understandable that, in actual implementation, the terminal / server described in the embodiment of the present disclosure may include one or more hardware structures and / or software modules for implementing the aforementioned corresponding data processing methods, and these execution hardware structures and / or software modules may constitute an electronic device. It should be readily apparent to those skilled in the art that, in combination with the algorithmic steps of each example described in the embodiments disclosed herein, the present disclosure can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present disclosure.

[0175] Based on this understanding, the embodiments of the present disclosure also provide a data processing device that can be applied to electronic devices. Figure 8 FIG. 1 shows a schematic diagram of the structure of a data processing device provided by an embodiment of the present disclosure. Figure 8 As shown, the data processing device may include: a determining unit 801;

[0176] Determining unit 801 is used to determine first feature data, second feature data, and third feature data; the first feature data is used to represent the matching degree between the multimedia resource and the topic tag of the multimedia resource; the second feature data is used to represent the consumption data of the multimedia resource; and the third feature data is used to represent the user profile data of the account that publishes the multimedia resource;

[0177] The determining unit 802 is further configured to determine the relevance between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data.

[0178] Optionally, the determination unit 802 is specifically used to: input the first feature data, the second feature data and the third feature data into the association detection model to obtain a model for detecting the association between multimedia resources and topic tags, which is pre-trained to a convergence state and is used to detect the association between multimedia resources and topic tags.

[0179] Optionally, the determining unit 801 is specifically configured to:

[0180] Get the associated data of multimedia resources;

[0181] Extracting text feature vectors and image feature vectors of multimedia resources from linked data;

[0182] Determining first matching data between the topic tag and the text feature vector;

[0183] Obtaining image feature vectors of associated multimedia resources related to the topic tag;

[0184] determining second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource;

[0185] Determining the first feature data includes first matching data and second matching data.

[0186] Optionally, the associated data includes at least one of the multimedia resource's audio data, cover data, image frame data, classification label, account information, and comment information.

[0187] Optionally, the determining unit 801 is specifically configured to:

[0188] Extracting credible text and uncredible text from the associated data; the credible text is text data having a first degree of association with the multimedia resource; the uncredible text is text data having a second degree of association with the multimedia resource; the first degree of association being greater than the second degree of association;

[0189] Converting the credible text into a first text feature vector, and converting the uncredible text into a second text feature vector, and determining that the text feature vector includes the first text feature vector and the second text feature vector;

[0190] A pre-trained image feature extraction model is used to extract a first image feature vector of the cover data and a second image feature vector of the image frame data, and it is determined that the image feature vector includes the first image feature vector and the second image feature vector.

[0191] Optionally, the credible text includes: text in the classification label, text in the account information, at least one of the music name text and singer text extracted from the audio data using speech recognition technology; the uncredible text includes: text in the comment information, lyrics text extracted from the audio data using speech recognition technology, at least one of the text recognized from the image frame data using text recognition technology.

[0192] Optionally, the determining unit 801 is specifically configured to:

[0193] Obtaining at least one associated text feature vector corresponding to the topic tag;

[0194] Determining a literal match and a semantic match between each associated text feature vector in at least one associated text feature vector and the text feature vector;

[0195] Based on the literal matching degree and the semantic matching degree, first matching data is determined.

[0196] Optionally, the determining unit 801 is specifically configured to:

[0197] Get the cover data and image frame data of the associated multimedia resource:

[0198] Using an image feature extraction model, extracting a third image feature vector of cover data associated with the multimedia resource and a fourth image feature vector of image frame data associated with the multimedia resource;

[0199] The image feature vector associated with the multimedia resource is determined to include a third image feature vector and a fourth image feature vector.

[0200] Optionally, the determining unit 801 is specifically configured to:

[0201] Based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource is determined.

[0202] Optionally, the determining unit 801 is further configured to disassociate the multimedia resource from the topic tag when the association between the multimedia resource and the topic tag is less than a preset association threshold.

[0203] The embodiment of the present disclosure also provides a model training device that can be applied to electronic devices. Figure 9 FIG. 1 shows a schematic diagram of the structure of the model training device provided by the embodiment of the present disclosure. Figure 9 As shown, the model training device may include: an acquisition unit 901 and a training unit 902;

[0204] The acquisition unit 901 is configured to acquire training data of multimedia resources within a preset time period; the training data includes training sample input data and training sample label data; the training sample input data includes: the matching degree between the multimedia resource and the multimedia resource's topic label, the multimedia resource's consumption data, and the user profile data of the account that published the multimedia resource; the training sample label data includes: the correlation degree between the multimedia resource and the multimedia resource's topic label;

[0205] The training unit 902 is used to train a relevance detection model based on the training data; the relevance detection model is used to detect the relevance between multimedia resources and topic tags.

[0206] As described above, the embodiments of the present disclosure can divide the electronic device into functional modules according to the above method examples. Among them, the above-mentioned integrated modules can be implemented in the form of hardware or in the form of software functional modules. In addition, it should be noted that the division of modules in the embodiments of the present disclosure is schematic and is only a logical functional division. There may be other division methods in actual implementation. For example, each functional module can be divided according to each function, or two or more functions can be integrated into one processing module.

[0207] Regarding the data processing device in the above embodiment, the specific manner in which each module performs operations and the beneficial effects thereof have been described in detail in the aforementioned method embodiment and will not be repeated here.

[0208] The embodiment of the present disclosure further provides a terminal, which may be a user terminal such as a mobile phone or a computer. Figure 10 FIG. 1 is a schematic diagram showing the structure of a terminal provided by an embodiment of the present disclosure. The terminal may be a data processing device including at least one processor 61 , a communication bus 62 , a memory 63 and at least one communication interface 64 .

[0209] The processor 61 may be a processor (central processing unit, CPU), microprocessor unit, ASIC, or one or more integrated circuits for controlling the execution of the program of the disclosed solution. As an example, in combination with Figure 8 The function implemented by the determination unit 801 in the electronic device is similar to Figure 10 The functions implemented by the processor 61 in are the same.

[0210] The communication bus 62 may include a pathway for transmitting information between the aforementioned components.

[0211] The communication interface 64 uses any transceiver or other device for communicating with other devices or communication networks, such as servers, Ethernet, radio access networks (RAN), wireless local area networks (WLAN), etc. As an example,

[0212] The memory 63 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory may exist independently and be connected to the processing unit via a bus. The memory may also be integrated with the processing unit.

[0213] The memory 63 is used to store application code for executing the solution of the present disclosure, and the execution is controlled by the processor 61. The processor 61 is used to execute the application code stored in the memory 63, thereby realizing the functions of the method of the present disclosure.

[0214] In a specific implementation, as an embodiment, the processor 61 may include one or more CPUs, such as Figure 10 CPU0 and CPU1 in.

[0215] In a specific implementation, as an embodiment, the terminal may include multiple processors, such as Figure 10 61 and processor 65 in FIG. Each of these processors can be a single-CPU processor or a multi-CPU processor. A processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0216] In a specific implementation, as an embodiment, the terminal may further include an input device 66 and an output device 67. The input device 66 communicates with the output device 67 and can accept user input in various ways. For example, the input device 66 can be a mouse, keyboard, touch screen device, or sensor device. The output device 67 communicates with the processor 61 and can display information in various ways. For example, the output device 61 can be a liquid crystal display (LCD), a light emitting diode (LED) display device, etc.

[0217] Those skilled in the art will understand that Figure 10 The structure shown in the figure does not constitute a limitation on the terminal, and may include more or fewer components than shown in the figure, or combine certain components, or adopt a different component arrangement.

[0218] The embodiment of the present disclosure also provides a server. Figure 11 The following is a schematic diagram of the structure of the server provided by an embodiment of the present disclosure. The server can be a data processing device. The server may have relatively large differences due to different configurations or performance, and may include one or more processors 71 and one or more memories 72. The memory 72 stores at least one instruction, and the at least one instruction is loaded and executed by the processor 71 to implement the data processing methods provided by the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output. The server may also include other components for implementing device functions, which will not be described in detail here.

[0219] The present disclosure also provides a computer-readable storage medium including instructions, on which instructions are stored. When the instructions in the computer-readable storage medium are executed by a processor of a computer device, the computer is enabled to execute the data processing method provided in the above-mentioned embodiment. For example, the computer-readable storage medium may be a memory 63 including instructions, and the above-mentioned instructions may be executed by the processor 61 of the terminal to complete the above-mentioned method. For another example, the computer-readable storage medium may be a memory 72 including instructions, and the above-mentioned instructions may be executed by the processor 71 of the server to complete the above-mentioned method. Optionally, the computer-readable storage medium may be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium may be a ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0220] The present disclosure also provides a computer program product, which includes computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes the above Figure 1-Figure 7 The data processing method shown in any of the accompanying drawings, or Figure 8 The model training method shown.

[0221] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the following claims.

[0222] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A data processing method, characterized in that: include: Determining first feature data, second feature data, and third feature data; the first feature data is used to represent the matching degree between a multimedia resource and a topic tag of the multimedia resource; the second feature data is used to represent consumption data of the multimedia resource; the third feature data is used to represent user profile data of an account that publishes the multimedia resource, the user profile data including search term data and classification tag data of the user who published the multimedia resource; the multimedia resource is obtained based on a search of the topic tag; Determining the relevance between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data; When the association between the multimedia resource and the topic tag is less than a preset association threshold, the association between the multimedia resource and the topic tag is released.

2. The data processing method according to claim 1, wherein: The determining, based on the first feature data, the second feature data, and the third feature data, of the relevance between the multimedia resource and the topic tag includes: The first feature data, the second feature data and the third feature data are input into an association detection model to obtain the association; the association detection model is a model that is pre-trained to a convergence state and is used to detect the association between the multimedia resource and the topic tag.

3. The data processing method according to claim 1, wherein: The determining of the first characteristic data includes: Acquiring the associated data of the multimedia resource; Extracting the text feature vector and the image feature vector of the multimedia resource from the associated data; Determining first matching data between the topic tag and the text feature vector; Obtaining an image feature vector of an associated multimedia resource related to the topic tag; determining second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource; The first matching data and the second matching data are determined as first feature data.

4. The data processing method according to claim 3, wherein: The associated data includes at least one of the audio data, cover data, image frame data, classification label, account information and comment information of the multimedia resource.

5. The data processing method according to claim 4, characterized in that: The step of extracting the text feature vector and the image feature vector of the multimedia resource from the associated data includes: Extracting credible text and untrustworthy text from the associated data; the credible text is text data having a first degree of association with the multimedia resource; the untrustworthy text is text data having a second degree of association with the multimedia resource; the first degree of association being greater than the second degree of association; Converting the credible text into a first text feature vector, and converting the uncredible text into a second text feature vector, and determining that the text feature vector includes the first text feature vector and the second text feature vector; A pre-trained image feature extraction model is used to extract a first image feature vector of the cover data and a second image feature vector of the image frame data, and it is determined that the image feature vector includes the first image feature vector and the second image feature vector.

6. The data processing method according to claim 5, characterized in that: The credible text includes: the text in the classification label, the text in the account information, and at least one of the music name text and singer text extracted from the audio data using speech recognition technology; the untrustworthy text includes: the text in the comment information, the lyrics text extracted from the audio data using the speech recognition technology, and at least one of the text recognized from the image frame data using text recognition technology.

7. The data processing method according to claim 5, characterized in that: The determining of first matching data between the topic tag and the text feature vector includes: Obtaining at least one associated text feature vector corresponding to the topic tag; Determining a literal matching degree and a semantic matching degree of each associated text feature vector in the at least one associated text feature vector with the text feature vector; The first matching data is determined based on the literal matching degree and the semantic matching degree.

8. The data processing method according to claim 5, characterized in that: The obtaining of an image feature vector of an associated multimedia resource related to the topic tag includes: Obtain the cover data and image frame data of the associated multimedia resource: Using the image feature extraction model, extracting a third image feature vector of the cover data of the associated multimedia resource and a fourth image feature vector of the image frame data of the associated multimedia resource; Determine that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector.

9. The data processing method according to claim 8, characterized in that: The determining of second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource includes: Based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource is determined.

10. A model training method, characterized in that: include: Obtaining training data of multimedia resources within a preset time period; The training data includes training sample input data and training sample label data; The training sample input data includes: the matching degree between the multimedia resource and the topic tag of the multimedia resource, the consumption data of the multimedia resource, and the user profile data of the account that published the multimedia resource, wherein the user profile data includes the search term data and classification label data of the user who published the multimedia resource; the training sample label data includes: the correlation degree between the multimedia resource and the topic tag of the multimedia resource; Based on the training data, a relevance detection model is trained; the relevance detection model is used to detect the relevance between the multimedia resource and the topic tag of the multimedia resource.

11. A data processing device, characterized in that: include: Identify the unit; The determining unit is configured to determine first feature data, second feature data, and third feature data; the first feature data is configured to indicate a degree of matching between a multimedia resource and a topic tag of the multimedia resource; the second feature data is configured to indicate consumption data of the multimedia resource; the third feature data is configured to indicate user profile data of an account that publishes the multimedia resource; and the multimedia resource is obtained based on a search of the topic tag. The determining unit is further configured to determine the relevance between the multimedia resource and the topic tag based on the first feature data, the second feature data, and the third feature data; The determining unit is further configured to disassociate the multimedia resource from the topic tag when the association between the multimedia resource and the topic tag is less than a preset association threshold.

12. The data processing device according to claim 11, characterized in that The determining unit is specifically configured to: The first feature data, the second feature data and the third feature data are input into an association detection model to obtain the association; the association detection model is a model that is pre-trained to a convergence state and is used to detect the association between the multimedia resource and the topic tag.

13. The data processing device according to claim 11, characterized in that The determining unit is specifically configured to: Acquiring the associated data of the multimedia resource; Extracting the text feature vector and the image feature vector of the multimedia resource from the associated data; Determining first matching data between the topic tag and the text feature vector; Obtaining an image feature vector of an associated multimedia resource related to the topic tag; determining second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource; The first matching data and the second matching data are determined as first feature data.

14. The data processing device according to claim 13, characterized in that The associated data includes at least one of the audio data, cover data, image frame data, classification label, account information and comment information of the multimedia resource.

15. The data processing device according to claim 14, characterized in that The determining unit is specifically configured to: Extracting credible text and untrustworthy text from the associated data; the credible text is text data having a first degree of association with the multimedia resource; the untrustworthy text is text data having a second degree of association with the multimedia resource; the first degree of association being greater than the second degree of association; Converting the credible text into a first text feature vector, and converting the uncredible text into a second text feature vector, and determining that the text feature vector includes the first text feature vector and the second text feature vector; A pre-trained image feature extraction model is used to extract a first image feature vector of the cover data and a second image feature vector of the image frame data, and it is determined that the image feature vector includes the first image feature vector and the second image feature vector.

16. The data processing device according to claim 15, characterized in that The credible text includes: the text in the classification label, the text in the account information, and at least one of the music name text and singer text extracted from the audio data using speech recognition technology; the untrustworthy text includes: the text in the comment information, the lyrics text extracted from the audio data using the speech recognition technology, and at least one of the text recognized from the image frame data using text recognition technology.

17. The data processing device according to claim 15, characterized in that The determining unit is specifically configured to: Obtaining at least one associated text feature vector corresponding to the topic tag; Determining a literal matching degree and a semantic matching degree of each associated text feature vector in the at least one associated text feature vector with the text feature vector; The first matching data is determined based on the literal matching degree and the semantic matching degree.

18. The data processing device according to claim 17, characterized in that The determining unit is specifically configured to: Obtain the cover data and image frame data of the associated multimedia resource: Using the image feature extraction model, extracting a third image feature vector of the cover data of the associated multimedia resource and a fourth image feature vector of the image frame data of the associated multimedia resource; Determine that the image feature vector of the associated multimedia resource includes the third image feature vector and the fourth image feature vector.

19. The data processing device according to claim 18, characterized in that The determining unit is specifically configured to: Based on the first image feature vector, the second image feature vector, the third image feature vector, and the fourth image feature vector, second matching data between the image feature vector of the associated multimedia resource and the image feature vector of the multimedia resource is determined.

20. A model training device, characterized in that: include: Acquisition units and training units; The acquisition unit is used to acquire the training data of the multimedia resource within a preset time period; The training data includes training sample input data and training sample label data; The training sample input data includes: the matching degree between the multimedia resource and the topic tag of the multimedia resource, the consumption data of the multimedia resource, and the user profile data of the account that published the multimedia resource, wherein the user profile data includes the search term data and classification label data of the user who published the multimedia resource; the training sample label data includes: the correlation degree between the multimedia resource and the topic tag of the multimedia resource; The training unit is used to train a relevance detection model based on the training data; the relevance detection model is used to detect the relevance between the multimedia resource and the topic tag of the multimedia resource.

21. An electronic device, characterized in that: The electronic device comprises: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the data processing method according to any one of claims 1 to 9, or the model training method according to claim 10.

22. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the data processing method as described in any one of claims 1 to 9, or the model training method as described in claim 10.

23. A computer program product comprising instructions, characterized in that When the instruction is executed on an electronic device, the electronic device executes the data processing method described in any one of claims 1 to 9, or the model training method described in claim 10.

Images