Label recognition method and device, storage medium and electronic device

Abstract

The present invention discloses a method and device for label recognition, a storage medium, and an electronic device. Among them, the method includes: obtaining a set of target description information, where there is a one-to-one correspondence between the set of target description information and a set of media resources, each piece of target description information includes the description information in the original description information of a media resource except for the target label, and the original description information of each media resource includes the target label; determining a set of content vectors according to the words in the set of target description information, where there is a one-to-one correspondence between the set of content vectors and the set of target description information; performing clustering processing on the set of content vectors to obtain a clustering result; and recognizing the target label according to the clustering result to obtain a label recognition result. The present invention solves the technical problem of low efficiency in recognizing topic labels.

Description

Technical Field

[0001] The present invention relates to the field of computers, and in particular, to a method and apparatus for label recognition, a storage medium, and an electronic device. Background Art

[0002] Subject labels are used to group the content of short-content media information (for example, Weibo, video accounts, etc.), and can be applied to theme condensation and search. They have increasingly attracted the attention of video account owners in terms of operation. In the prior art, assigning subject labels to a video account is completely determined by the video account owner independently, and the platform operator will not interfere. Some video account owners, in order to gain popularity, will stack some subject labels unrelated to the video account in the video account. For example, many video accounts will be labeled with "Popular" and "Video Account".

[0003] For the platform governance side, it is necessary to identify those meaningless subject labels through a certain method to facilitate the extraction of various subject label keywords in the future. In the prior art, generally, a subject label word list is collected manually according to experience, and the manual processing method has low efficiency, cannot be updated in time, and is affected by the subjectivity of the processing personnel. There are many disadvantages that rely on manual experience and have no scientific standards.

[0004] In view of the problem of low efficiency in subject label recognition in the related art, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of the present invention provide a method and apparatus for label recognition, a storage medium, and an electronic device to at least solve the technical problem of low efficiency in subject label recognition.

[0006] According to one aspect of the embodiments of the present invention, a method for label recognition is provided, including: obtaining a set of target description information, where the set of target description information has a one-to-one correspondence with a set of media resources, and each piece of the target description information includes the description information in the original description information of a media resource except for the target label, and the original description information of each media resource includes the target label; determining a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; performing clustering processing on the set of content vectors to obtain a clustering result; and recognizing the target label according to the clustering result to obtain a label recognition result.

[0007] According to another aspect of the embodiments of the present invention, there is also provided a tag recognition device, including: an acquisition module, configured to acquire a set of target description information, where the set of target description information has a one-to-one correspondence with a set of media resources, and each piece of the target description information includes the description information in the original description information of a media resource except for the target tag, and the original description information of each media resource includes the target tag; a determination module, configured to determine a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; a processing module, configured to perform clustering processing on the set of content vectors to obtain a clustering result; and an identification module, configured to identify the target tag according to the clustering result to obtain a tag recognition result.

[0008] According to still another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium storing a computer program, where the computer program is configured to execute the above-mentioned tag recognition method when running.

[0009] According to still another aspect of the embodiments of the present invention, there is also provided an electronic device including a memory and a processor, where the memory stores a computer program, and the processor is configured to execute the above-mentioned tag recognition method through the computer program.

[0010] In the embodiments of the present invention, by acquiring a set of target description information that has a one-to-one correspondence with a set of media resources, where each piece of target description information includes the description information in the original description information of a media resource except for the target tag, and the original description information of each media resource includes the target tag; determining a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; performing clustering processing on the set of content vectors to obtain a clustering result; and identifying the target tag according to the clustering result to obtain a tag recognition result, the purpose of automatically recognizing the topic tags in the media resources is achieved, thereby realizing the technical effect of improving the efficiency of topic tag recognition, and further solving the technical problem of low efficiency of topic tag recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of this application. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation to the present invention. In the drawings:

[0012] Figure 1 is a schematic diagram of an application environment of an optional tag recognition method according to an embodiment of the present invention;

[0013] Figure 2It is a flowchart of an optional tag recognition method according to an embodiment of the present invention;

[0014] Figure 3 It is a schematic diagram of an optional media resource according to an embodiment of the present invention Figure 1 ;

[0015] Figure 4 It is a schematic diagram of an optional media resource according to an embodiment of the present invention Figure 2 ;

[0016] Figure 5 It is a flowchart of an optional content vector processing according to an embodiment of the present invention;

[0017] Figure 6 It is a schematic diagram of an optional clustering cluster according to an embodiment of the present invention;

[0018] Figure 7 It is a schematic diagram of an optional word2vec model structure according to an embodiment of the present invention;

[0019] Figure 8 It is a schematic diagram of the structure of an optional tag recognition device according to an embodiment of the present invention;

[0020] Figure 9 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present invention. Detailed implementation manners

[0021] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those clearly listed steps or units, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0023] According to one aspect of an embodiment of the present invention, a tag recognition method is provided. Optionally, as an alternative implementation, the above-mentioned tag recognition method can be applied, but is not limited to, a system environment as shown in Figure 1 Figure 1, where the system includes a user device 102, a network 110, and a server 112.

[0024] Optionally, in this embodiment, the above-mentioned user device may be a terminal device, and may include, but is not limited to, at least one of the following: mobile phone (such as Android mobile phone, iOS mobile phone, etc.), laptop computer, tablet computer, handheld computer, MID (Mobile Internet Devices, mobile Internet device), PAD, desktop computer, smart TV, etc. The user device may be configured with a target client, and the target client may be a video client, an instant messaging client, a browser client, an education client, a shopping client, etc. In this embodiment, the user device may include, but is not limited to: a memory 104, a processor 106, and a display 108. The memory 104 may be used to store data, for example, may be used to store the above-mentioned media resources and target description information. The processor may be used to process the target description information. The display 108 may be used to display the tag recognition result.

[0025] Optionally, the above-mentioned network 110 may include, but is not limited to: a wired network, a wireless network, where the wired network includes: local area network, metropolitan area network, and wide area network, and the wireless network includes: Bluetooth, WIFI, and other networks that implement wireless communication.

[0026] Optionally, the above-mentioned server 112 may be a single server, or may be a server cluster composed of multiple servers, or a cloud server. The server 112 may include, but is not limited to: a database 114 and a processing engine 116. The above-mentioned database 114 may be used to store data, for example, may be used to store the above-mentioned media resources and target description information. The processing engine is used to process the target description information. The above is only an example, and this embodiment does not make any limitations thereto.

[0027] Optionally, as an alternative implementation, as shown in Figure 2 Figure 2, the above-mentioned tag recognition method includes:

[0028] Step S202, obtaining a set of target description information, where the set of target description information has a one-to-one correspondence with a set of media resources, and each piece of the target description information includes the description information other than the target tag in the original description information of a media resource, and the original description information of each media resource includes the target tag;

[0029] Step S204: Determine a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information;

[0030] Step S206: Perform clustering processing on the set of content vectors to obtain a clustering result;

[0031] Step S208: Identify the target label according to the clustering result to obtain a label identification result.

[0032] Through the above steps, by obtaining a set of target description information, the set of target description information has a one-to-one correspondence with a set of media resources, each piece of target description information includes the description information other than the target label in the original description information of a media resource, and the original description information of each media resource includes the target label; determine a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; perform clustering processing on the set of content vectors to obtain a clustering result; identify the target label according to the clustering result to obtain a label identification result, achieving the purpose of automatically identifying the topic labels in the media resources, thus realizing the technical effect of improving the efficiency of topic label identification, and further solving the technical problem of low efficiency of topic label identification.

[0033] As an optional implementation manner, the above media resources may be media resources published on a multimedia platform, such as Weibo or short videos. A set of media resources may include multiple media resources, and the multiple media resources may come from the same multimedia platform. For example, they may be multiple short videos published by the same user on the same platform, or multiple short videos published by different users on the same platform. The multiple media resources may also be media resources published by different users across platforms. For example, the multiple media resources may include the media resources published by user A on platform A and the media resources published by user B on platform B.

[0034] As an optional implementation manner, the media resources published by users on the media platform may be added with description information. The description information usually describes the media resources in text, such as Figure 3 shown is the schematic diagram of the media resource according to the optional embodiment of the present invention Figure 1 assuming Figure 3Shown is short video media resources. In the figure, "Popular", "Appearance of Youth", and "Be a Happy Girl" are the description information of the short video. Usually, the description information of a video should match the content in the short video. However, since the current description information is added by users themselves, some users add some description information that does not match the content of the short video in order to gain popularity. In this embodiment, the tags in the description information can be identified through a clustering algorithm to determine whether the tags match the content of the short video.

[0035] As an alternative implementation, assume that a group of media resources are short videos that all include the "Popular" tag, such as Figure 4 is a schematic diagram of media resources according to an alternative embodiment of the present invention Figure 2 In the figure, multiple short videos shown all include the "Popular" tag. For Figure 4 the short videos with the same "Popular" tag shown, semantic aggregation is performed. The other description information in the short video except the "Popular" tag is converted into content vectors, and the content vectors are clustered. According to the clustering result, it is determined whether the "Popular" is a meaningless tag that does not match the video content.

[0036] Optionally, clustering the group of content vectors to obtain a clustering result includes: clustering the group of content vectors according to the distance between each two content vectors in the group of content vectors to obtain the clustering result, where the group of content vectors includes at least two content vectors.

[0037] As an alternative implementation, the distance between different samples in the clustering is measured as a whole. The distance between content vectors can represent similarity. Multiple content vectors that satisfy a preset value can form a clustering cluster. For tags with a relatively clear theme, their clustering clusters are relatively concentrated, while for tags with a relatively broad theme, their clustering clusters are more. Therefore, for those meaningless tags, the degree of dispersion is greater and the number of clustering clusters is more. Regarding a group of content vectors as a cluster, each content vector as a point in the cluster. For each point in the cluster, if the number of its adjacent points within a given radius range exceeds a preset threshold, then the density of the cluster is significantly higher than the density of noise points, and it is considered that the tag is a meaningful tag; otherwise, the tag is a meaningless tag. In this embodiment, the content vectors are clustered through the distance between content vectors, and then the tags can be identified according to the clustering result, achieving the purpose of automatically identifying tags and improving the efficiency of tag identification.

[0038] Optionally, clustering the set of content vectors according to the distances between every two content vectors in the set of content vectors to obtain the clustering result, including: repeatedly performing the following steps until all the content vectors in the set of content vectors are processed: determining a current core content vector from the unprocessed content vectors in the set of content vectors according to the distances between every two content vectors in the set of content vectors; determining a first content vector set from the unprocessed content vectors in the set of content vectors according to the distances between every two content vectors in the set of content vectors, where the relationship between the content vectors in the first content vector set and the current core content vector is density-reachable, and the current core content vector and the first content vector set form a clustering cluster obtained by clustering the set of content vectors.

[0039] As an optional implementation manner, assume that a sample set composed of a set of content vectors is D = (p1, p2,..., p n ), for the content vector p i , where i is greater than 1 and less than or equal to n, its neighborhood includes a sub-sample set in the sample set D whose distance from p i is not greater than a preset threshold e, and the preset threshold e can be determined according to actual situations, for example, it can be 0.01, 0.1, etc.

[0040] As an optional implementation manner, if the number of content vectors in the neighborhood of a given content vector p i is greater than or equal to a preset threshold t, then the content vector p i is called a core content vector, and the preset threshold t can be the minimum number of content vectors required for the center point period of the content vector cluster set in advance, and can be set according to actual situations, for example, it can be 3, 4, 5, 10, etc.

[0041] As an optional implementation manner, if the content vector p i is located in the neighborhood of the content vector p j , and p j is a core content vector, then p i is density-reachable from p j . For the content vectors included in the sample set, a string of content vector points p1, p2... p n is given. Assume p = p1, q = p n . Suppose the object p i is directly density-reachable from p i-1 , where i is greater than 1 and less than or equal to n, then q is density-reachable from p, and density reachability satisfies transitivity. At this time, p1, p2... p n-1 in the sequence are all core content vectors.

[0042] As an optional implementation, for content vector p i and content vector p j , if there exists a core content vector p k such that both p i and p j are density-reachable from p k , then p i and p j are density-connected. In this embodiment, density-reachability is the transitive closure of direct density-reachability, and this relationship is asymmetric, while density-connectivity is a symmetric relationship.

[0043] As an optional implementation, as Figure 5 shown is the content vector processing flowchart according to an optional embodiment of the present invention. For each content vector in a set of content vectors D = (p1, p2,..., p n ), the following steps are performed:

[0044] Step S51: Extract an unprocessed content vector from the set of content vectors;

[0045] Step S52: If the extracted content vector is a core content vector, find all reachable content vectors from the current core content vector, and cluster the current core vector and all reachable content vectors to obtain a clustering cluster;

[0046] Step S53: If the extracted content vector is a marginal content vector (non-core content vector), break out of the current loop and extract other unprocessed content vectors from the set of content vectors;

[0047] Step S54: Until all content vectors in the set of content vectors are processed.

[0048] Optionally, determining the current core content vector among the unprocessed content vectors in the set of content vectors according to the distance between every two content vectors in the set of content vectors includes: selecting the current content vector to be processed among the unprocessed content vectors in the set of content vectors; when there exists a second set of content vectors among the unprocessed content vectors in the set of content vectors and the number of content vectors in the second set of content vectors is greater than or equal to a first preset threshold, determining the current content vector as the current core content vector, where the distance between the content vectors in the second set of content vectors and the current content vector is less than or equal to a second preset threshold.

[0049] As an optional implementation, if the number of content vectors within the neighborhood of a given content vector p i is greater than or equal to a preset threshold t, then this content vector p iCalled the core content vector, the preset threshold t can be the minimum number of content vectors required for the center point period of the preset content vector cluster, which can be set according to the actual situation. For example, it can be 3, 4, 5, 10, etc. Specifically, in this embodiment, a set of content vectors D = (p1, p2,..., p n ) can be used to extract unprocessed content vectors. If the number of sample points within the Εps neighborhood is greater than or equal to MinPts, then this content vector is called a core content vector. For a content vector p i , at least MinPts content vectors are included within its Εps neighborhood, and the content vectors p i within the neighborhood and other content vectors have a distance less than or equal to the preset threshold, which can be determined according to the actual situation. For example, it can be 0.1 cm, 0.01 cm, that is, |N eps (p i )|≥MinPts, and p i is a core content vector. In this embodiment, by determining the core content vector according to the number of content vectors within the neighborhood, the purpose of clustering the content vectors can be achieved, and whether the label is a meaningless label can be accurately identified according to the clustering result, improving the label recognition efficiency.

[0050] Optionally, determining the first content vector set from the unprocessed content vectors in the set of content vectors according to the distance between every two content vectors in the set of content vectors includes: determining a core content vector set from the unprocessed content vectors in the set of content vectors, where the core content vector set includes the current core content vector, and each content vector in the core content vector set is a core content vector, and the distance between each content vector in the core content vector set and at least one content vector in the core content vector set is less than or equal to a second preset threshold; determining the first content vector set from the unprocessed content vectors in the set of content vectors, where the distance between the content vectors in the first content vector set and at least one content vector in the core content vector set is less than or equal to the second preset threshold.

[0051] As an alternative implementation, as Figure 6 shows, it is a schematic diagram of a clustering cluster according to an optional embodiment of the present invention. The origin in the figure is used to represent the content vector, the midpoint within the solid circle is the core content vector, and the content vectors connected by the double-headed arrow are used to represent reachable content vectors. All reachable content vectors of each core content vector form a clustering cluster. Specifically, for a set of content vectors D = (p1, p2,..., p n ), the core content vectors are determined from the unprocessed content vectors to obtain the core vector set, as Figure 6The midpoints of all solid circles are the core content vectors, and the set formed is the core content vector set. The content vectors within the solid circles and the core content vectors form the first content vector set. The distances between the content vectors in this content vector set and the core content vector satisfy being less than or equal to a second preset threshold, and the second preset threshold can be determined according to the actual situation. For example, it can be 0.1 cm, 0.01 cm, etc. In this embodiment, the core content vector is determined through the distances between the content vectors. Based on the core content vector, a set of content vectors can be clustered to obtain a clustering cluster, and based on the clustering cluster, the label can be recognized, achieving the technical effect of improving the accuracy of label recognition.

[0052] Optionally, determining a set of content vectors according to the words in the set of target description information includes: when the set of target description information includes N pieces of target description information, respectively performing word segmentation on each piece of target description information to obtain N word sets, where N is a natural number greater than 1; respectively converting the N word sets into content vectors to obtain N content vectors, where the set of content vectors includes the N content vectors.

[0053] As an optional implementation manner, after removing the target label from the original description information in each media resource, the remaining description information, that is, the target description information, can be processed as a complete text as follows: The text is segmented using a word segmenter to obtain a word set M, and the word set can be converted into a content vector based on word2vec. As Figure 7 shown is a schematic diagram of the word2vec model structure according to an optional embodiment of the present invention. This model can be a shallow and double-layer neural network used to train and reconstruct the linguistic word text. The network is represented by words. Under the assumption of the bag-of-words model in word2vec, the order of words is not important. After training, the word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words. This vector is the hidden layer of the neural network. The input of the word2vec model is a large amount of segmented text, and the output is to represent each word with a dense vector. The important significance of word vectors is to convert natural language into vectors that can be understood by a computer. Compared with models such as the bag-of-words model and TF-IDF, word vectors can capture the context and semantics of words and measure the similarity between words, playing an important role in many natural language processing fields such as text classification and sentiment analysis. In this embodiment, through the word segmentation and vector conversion processing of the media resource description information, vectors that can be recognized by a computer can be obtained, achieving the purpose of automatically recognizing labels by a computer and improving the accuracy of label recognition.

[0054] Optionally, identifying the target label according to the clustering result to obtain a label identification result includes: determining the clustering dispersion degree of the target label according to the clustering result; and identifying the target label according to the clustering dispersion degree of the target label to obtain the label identification result.

[0055] As an optional implementation manner, the dispersion score of the target label can be determined according to the clustering result of a group of content vectors. Generally, the larger the dispersion degree, the more likely the label is a meaningless label. In this embodiment, a threshold can be preset, and a label with a dispersion degree greater than the threshold is considered a meaningless label. The threshold can be determined according to the actual situation, such as 0.5, 0.6, 0.8, etc. In this embodiment, by setting a threshold K according to the dispersion score of the label, a label with a dispersion degree greater than K is considered a meaningless label, and labels with a lower dispersion degree can be identified as accurate labels that match the video content, which can provide a factor for calculating the importance of the label or extracting key labels and improve the accuracy of label identification.

[0056] Optionally, determining the clustering dispersion degree of the target label according to the clustering result includes: determining the clustering dispersion degree of the target label according to the number of clustering clusters obtained by clustering the group of content vectors, where the clustering result includes the number of clustering clusters obtained by clustering the group of content vectors.

[0057] As an optional implementation, based on clustering, the clustering number distribution of a set of media resources related to the target label can be obtained. Because what is identified are those meaningless labels, rather than labels with a relatively large dispersion degree but significantly smaller. Assuming that the normal mean of the clustering number distribution is generally in the middle, the dispersion degree of these clustering numbers is solved to find a set of media resources with the same label that have a very large dispersion degree and a significantly larger number of clustering. The dispersion analysis method is a method for measuring the dispersion degree of a set of data. Generally speaking, the dispersion degree reflects the degree to which a set of data is far from its central value, so it is also called the tendency away from the center. In this embodiment, whether the label is a meaningless label is identified through the clustering dispersion degree of the label. If the dispersion degree is greater than the preset threshold, the label is considered a meaningless label. Through the dispersion degree, the purpose of automatically identifying labels by a computer can be achieved, avoiding the problems of low accuracy and efficiency of manual identification, and achieving the technical effect of improving the accuracy and efficiency of label identification.

[0058] Optionally, determining the clustering dispersion degree of the target label according to the number of clustering clusters obtained by clustering the group of content vectors includes: determining the clustering dispersion degree S of the target label through the following formula:

[0059] S = D × tag_idf × Cst / FC

[0060] Among them, Cst represents the number of clustering clusters obtained by clustering the set of content vectors, FC represents the number of media resources in the set of media resources, D represents the variance of the number of clustering clusters relative to the target mean, each number of clustering clusters is the number of clusters obtained by clustering the content vectors of a media resource set, and the original description information of the media resources in each media resource set includes the same one tag, and tag_idf represents the distinguishability of the target tag in the set of media resources.

[0061] As an optional implementation manner, the dispersion S = D×tag_idf×Cst / FC, where Cs represents the number of clusters obtained after classifying all media resources containing the target tag, and FC represents the number of all media resources containing the target tag. And D represents the variance of the number of media resources in all media resource clustering clusters under the target tag relative to the mean of the number of media resources in all media resource clustering clusters under the target tag. Generally, the larger the variance, the farther the target is from the mean, that is, either the target tag is more accurate or it is too broad to be a meaningless tag. Then whether it is a meaningless tag or not can be determined by tag_idf in this formula. tag_idf is used to represent the distinguishability of the target tag in the set of media resources. In this embodiment, the dispersion of the target tag is calculated by the above formula, and whether the tag is a meaningless tag is determined according to the dispersion, thereby improving the accuracy of tag recognition.

[0062] Optionally, the method further includes: determining the distinguishability tag_idf through the following formula:

[0063]

[0064] Among them, SFC represents the total number of media resources on the preset platform, and the set of media resources is the media resources on the preset platform.

[0065] As an optional implementation manner, the distinguishability of the target tag can be obtained by dividing the number of all media resources in the platform by the number of media resources containing the target tag and then taking the logarithm of the obtained quotient:

[0066]

[0067] Among them, SFC represents the total number of all media resources on the platform, FC represents the number of media resources in a set of media resources, and the media resources in a set of media resources are the media resources on this platform.

[0068] Optionally, the method further includes: determining the variance through the following formula:

[0069]

[0070]

[0071] Among them, m is used to represent the number of the clustering clusters, m is a natural number greater than 1, and x j is used to represent the number of content vectors in the j-th clustering cluster.

[0072] As an optional implementation manner, variance is mainly used to measure the dispersion degree of numerical data. Variance refers to the average of the squares of the deviations of each value in a set of data from its mean. The larger the variance value, the higher the dispersion degree of the data. On the contrary, the lower the dispersion degree of the data. For the number distribution data of the clustering clusters of media resources under the target label, assume that the number of clustering clusters is m, and the number of content vectors included in each clustering cluster are x1, x2,..., x m , then the variance is:

[0073]

[0074]

[0075] The dispersion score of the target label calculated by the above formula, set a threshold K, and for the label with a dispersion greater than K, it is considered a meaningless label. This embodiment proposes a method for identifying meaningless labels based on the calculation of semantic clustering dispersion, which can make more use of the clustering information between media resources under the same label to determine whether the content represented by the current label is sufficiently focused in the clustering cluster. It is considered that the less focused the clustering is, the more likely the label is meaningless, so as to achieve the technical effect of improving the label recognition efficiency.

[0076] Optionally, the identifying the target label according to the clustering dispersion of the target label to obtain the label recognition result includes: when the clustering dispersion is greater than a third preset threshold, identifying the target label to obtain a first recognition result, where the label recognition result includes the first recognition result, and the first recognition result is used to represent that the target label is an invalid label; when the clustering dispersion is less than the third preset threshold, identifying the target label to obtain a second recognition result, where the label recognition result includes the second recognition result, and the second recognition result is used to represent that the target label is a valid label.

[0077] In this embodiment, a method for identifying meaningless tags based on semantic clustering dispersion calculation is proposed. It can make more use of the clustering information among media resources under the same tag to determine whether the content represented by the current tag is sufficiently focused in terms of clustering. It is considered that the less focused the clustering is, the more likely the tag is a meaningless tag. Specifically, if the dispersion is greater than a preset threshold, it is determined that the content represented by the current tag is not sufficiently focused in terms of clustering and is a meaningless tag, and it is recognized that this tag does not match the video content and is an invalid tag. Further operations such as deletion can be performed on this tag. For a dispersion less than the preset threshold, it is determined that the content represented by the current tag is sufficiently focused in terms of clustering and is an accurate tag, and it is recognized that this tag matches the video content and is a valid tag. Further, the video can be classified according to the tag. In this embodiment, tags are identified through dispersion, and based on the identification results of the tags, the tags and the media resources identified by the tags can be further processed, achieving the technical effect of improving the tag identification efficiency.

[0078] It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0079] According to another aspect of the embodiments of the present invention, there is also provided a tag recognition device for implementing the above-mentioned tag recognition method. As Figure 8 shown, the device includes: an acquisition module 802, configured to acquire a set of target description information, where the set of target description information has a one-to-one correspondence with a set of media resources, and each piece of the target description information includes the description information other than the target tag in the original description information of a media resource, and the original description information of each media resource includes the target tag; a determination module 804, configured to determine a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; a processing module 806, configured to perform clustering processing on the set of content vectors to obtain a clustering result; and an identification module 808, configured to identify the target tag according to the clustering result to obtain a tag identification result.

[0080] Optionally, the above device is used to perform clustering processing on the set of content vectors according to the distance between every two content vectors in the set of content vectors to obtain the clustering result, where the set of content vectors includes at least two content vectors.

[0081] Optionally, the above-mentioned device is used to repeatedly execute the following steps until all the content vectors in the set of content vectors are processed: determine a current core content vector from the unprocessed content vectors in the set of content vectors according to the distance between every two content vectors in the set of content vectors; determine a first set of content vectors from the unprocessed content vectors in the set of content vectors according to the distance between every two content vectors in the set of content vectors, wherein the relationship between the content vectors in the first set of content vectors and the current core content vector is density-reachable, and the current core content vector and the first set of content vectors form a clustering cluster obtained by clustering the set of content vectors.

[0082] Optionally, the above-mentioned device is used to select a current content vector to be processed from the unprocessed content vectors in the set of content vectors; when there is a second set of content vectors in the unprocessed content vectors in the set of content vectors and the number of content vectors in the second set of content vectors is greater than or equal to a first preset threshold, determine the current content vector as the current core content vector, wherein the distance between the content vectors in the second set of content vectors and the current content vector is less than or equal to a second preset threshold.

[0083] Optionally, the above-mentioned device is used to determine a set of core content vectors from the unprocessed content vectors in the set of content vectors, wherein the set of core content vectors includes the current core content vector, each content vector in the set of core content vectors is a core content vector, and the distance between each content vector in the set of core content vectors and at least one content vector in the set of core content vectors is less than or equal to the second preset threshold; determine the first set of content vectors from the unprocessed content vectors in the set of content vectors, wherein the distance between the content vectors in the first set of content vectors and at least one content vector in the set of core content vectors is less than or equal to the second preset threshold.

[0084] Optionally, when the set of target description information includes N pieces of target description information, each piece of the target description information is tokenized respectively to obtain N sets of words, where N is a natural number greater than 1; the N sets of words are respectively converted into content vectors to obtain N content vectors, wherein the set of content vectors includes the N content vectors.

[0085] Optionally, the above-mentioned device is used to determine the clustering dispersion degree of the target label according to the clustering result; identify the target label according to the clustering dispersion degree of the target label to obtain the label identification result.

[0086] Optionally, the above-mentioned device is used to determine the clustering dispersion of the target label according to the number of clustering clusters obtained by clustering the set of content vectors, where the clustering result includes the number of clustering clusters obtained by clustering the set of content vectors.

[0087] Optionally, the above-mentioned device is used to determine the clustering dispersion S of the target label through the following formula:

[0088] S = D × tag_idf × Cst / FC

[0089] Wherein, Cst represents the number of clustering clusters obtained by clustering the set of content vectors, FC represents the number of media resources in the set of media resources, D represents the variance of the number of clustering clusters relative to the target mean, each number of clustering clusters is the number of clusters obtained by clustering the content vectors of a media resource set, the original description information of the media resources in each media resource set includes the same one label, and tag_idf represents the discrimination degree of the target label in the set of media resources.

[0090] Optionally, the above-mentioned device is used to determine the discrimination degree tag_idf through the following formula:

[0091]

[0092] Wherein, SFC represents the total number of media resources on the preset platform, and the set of media resources is the media resources on the preset platform.

[0093] Optionally, the above-mentioned device is used to determine the variance through the following formula:

[0094]

[0095]

[0096] Wherein, m is used to represent the number of clustering clusters, m is a natural number greater than 1, x j represents the number of content vectors in the j-th clustering cluster among the m clustering clusters obtained by clustering the set of content vectors.

[0097] Optionally, the above-mentioned device is used to identify the target label when the clustering dispersion is greater than the third preset threshold to obtain a first identification result, where the label identification result includes the first identification result, and the first identification result is used to indicate that the target label is an invalid label; when the clustering dispersion is less than the third preset threshold, the target label is identified to obtain a second identification result, where the label identification result includes the second identification result, and the second identification result is used to indicate that the target label is a valid label.

[0098] According to another aspect of the embodiments of the present invention, there is also provided an electronic device for implementing the above-mentioned label recognition method, and this electronic device can be Figure 1 the terminal device or server shown in the figure. In this embodiment, the electronic device is taken as an example of the terminal device for illustration. As Figure 9 shown in the figure, the electronic device includes a memory 902 and a processor 904. A computer program is stored in the memory 902, and the processor 904 is configured to execute the steps in any one of the above method embodiments through the computer program.

[0099] Optionally, in this embodiment, the above-mentioned electronic device may be at least one of multiple network devices in a computer network.

[0100] Optionally, in this embodiment, the above-mentioned processor may be configured to execute the following steps through a computer program:

[0101] S1. Obtain a set of target description information, where there is a one-to-one correspondence between the set of target description information and a set of media resources, and each piece of the target description information includes the description information in the original description information of a media resource except for the target label, and the original description information of each media resource includes the target label;

[0102] S2. Determine a set of content vectors according to the words in the set of target description information, where there is a one-to-one correspondence between the set of content vectors and the set of target description information;

[0103] S3. Perform clustering processing on the set of content vectors to obtain a clustering result;

[0104] S4. Identify the target label according to the clustering result to obtain a label recognition result.

[0105] Optionally, those of ordinary skill in the art can understand that Figure 9 the structure shown in the figure is only schematic, and the electronic device can also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, and a mobile Internet device (Mobile Internet Devices, MID), a PAD and other terminal devices. Figure 9 It does not limit the structure of the above-mentioned electronic device. For example, the electronic device may further include more or fewer components (such as a network interface, etc.) than those shown in Figure 9 , or have a different configuration from that shown in Figure 9 .

[0106] Among them, the memory 902 can be used to store software programs and modules, such as the program instructions / modules corresponding to the label recognition method and device in the embodiments of the present invention. The processor 904 executes various functional applications and data processing by running the software programs and modules stored in the memory 902, that is, implements the above-mentioned label recognition method. The memory 902 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memories. In some instances, the memory 902 may further include a memory remotely disposed relative to the processor 904, and these remote memories may be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranet, local area network, mobile communication network, and combinations thereof. Among them, the memory 902 may specifically but not limitedly be used to store information such as sample features of items and target virtual resource accounts. As an example, as Figure 9 shown, the above memory 902 may but not limitedly include the acquisition module 802, determination module 804, processing module 806, and recognition module 808 in the above label recognition device. In addition, it may also include but not limited to other module units in the above label recognition device, which will not be elaborated in this example.

[0107] Optionally, the above transmission device 906 is used to receive or send data via a network. Specific examples of the above network may include a wired network and a wireless network. In one instance, the transmission device 906 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers through a network cable so as to communicate with the Internet or a local area network. In one instance, the transmission device 906 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0108] In addition, the above electronic device further includes: a display 908 for displaying the above order information to be processed; and a connection bus 910 for connecting each module component in the above electronic device.

[0109] In other embodiments, the above terminal device or server may be a node in a distributed system. Among them, the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting the multiple nodes through network communication. Among them, the nodes may form a peer-to-peer (P2P, Peer To Peer) network, and any form of computing device, such as a server, terminal, and other electronic devices, can become a node in the blockchain system by joining the peer-to-peer network.

[0110] According to one aspect of the present application, there is provided a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided in the above various optional implementation manners. Among them, the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0111] Optionally, in this embodiment, the above computer-readable storage medium may be configured to store a computer program for executing the following steps:

[0112] S1, obtain a set of target description information, where the set of target description information has a one-to-one correspondence with a set of media resources, and each piece of the target description information includes the description information in the original description information of a media resource except for the target label, and the original description information of each media resource includes the target label;

[0113] S2, determine a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information;

[0114] S3, perform clustering processing on the set of content vectors to obtain a clustering result;

[0115] S4, identify the target label according to the clustering result to obtain a label identification result.

[0116] Optionally, in this embodiment, those of ordinary skill in the art can understand that all or part of the steps in the above various methods can be completed by a program instructing the relevant hardware of the terminal device, and the program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disc, etc.

[0117] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0118] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above computer-readable storage media. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0119] In the above embodiments of the present invention, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0120] In the several embodiments provided by the present application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the units or modules can be in electrical or other forms.

[0121] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0122] In addition, the functional units in the various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0123] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A label recognition method, characterized in that, Including: Determining a set of target media resources identified by a target label to be recognized; Obtaining a set of target description information that has a one-to-one correspondence with the target media resources, where each piece of the target description information includes the description information in the original description information of a media resource except the target label, and the original description information of each media resource includes the target label; Determining a set of content vectors according to the words in the set of target description information, where the set of content vectors has a one-to-one correspondence with the set of target description information; Performing clustering processing on the set of content vectors to obtain a clustering result; Recognizing the target label according to the clustering result to obtain a label recognition result for indicating the validity of the target label.

2. The method according to claim 1, wherein The performing clustering processing on the set of content vectors to obtain a clustering result includes: Performing clustering processing on the set of content vectors according to the distance between each two content vectors in the set of content vectors to obtain the clustering result, where the set of content vectors includes at least two content vectors.

3. The method according to claim 2, characterized in that The performing clustering processing on the set of content vectors according to the distance between each two content vectors in the set of content vectors to obtain the clustering result includes: Repeatedly performing the following steps until all the content vectors in the set of content vectors are processed: According to the distance between each two content vectors in the set of content vectors, determining a current core content vector among the unprocessed content vectors in the set of content vectors; According to the distance between each two content vectors in the set of content vectors, determining a first set of content vectors among the unprocessed content vectors in the set of content vectors, where the relationship between the content vectors in the first set of content vectors and the current core content vector is density-reachable, and the current core content vector and the first set of content vectors form a clustering cluster obtained by clustering the set of content vectors.

4. The method according to claim 3, characterized in that, The determining a current core content vector among the unprocessed content vectors in the set of content vectors according to the distance between each two content vectors in the set of content vectors includes: Selecting a current content vector to be processed among the unprocessed content vectors in the set of content vectors; When there is a second set of content vectors among the unprocessed content vectors in the set of content vectors and the number of content vectors in the second set of content vectors is greater than or equal to a first preset threshold, determining the current content vector as the current core content vector, where the distance between the content vectors in the second set of content vectors and the current content vector is less than or equal to a second preset threshold.

5. The method according to claim 3, wherein The determining a first set of content vectors among the unprocessed content vectors in the set of content vectors according to the distance between each two content vectors in the set of content vectors includes: Among the unprocessed content vectors in the set of content vectors, determine a set of core content vectors, where the set of core content vectors includes the current core content vector, each content vector in the set of core content vectors is a core content vector, and the distance between each content vector in the set of core content vectors and at least one content vector in the set of core content vectors is less than or equal to a second preset threshold; Among the unprocessed content vectors in the set of content vectors, determine the first set of content vectors, where the distance between the content vectors in the first set of content vectors and at least one content vector in the set of core content vectors is less than or equal to a second preset threshold.

6. The method according to claim 1, characterized in that, Determining a set of content vectors according to the words in the set of target description information includes: When the set of target description information includes N pieces of target description information, perform word segmentation on each piece of target description information respectively to obtain N word sets, where N is a natural number greater than 1; Convert the N word sets into content vectors respectively to obtain N content vectors, where the set of content vectors includes the N content vectors.

7. The method according to any one of claims 1 to 6, characterized in that, Identifying the target label according to the clustering result to obtain a label recognition result for indicating the validity of the target label includes: Determine the clustering dispersion degree of the target label according to the clustering result; Identify the target label according to the clustering dispersion degree of the target label to obtain the label recognition result.

8. The method according to claim 7, wherein Determining the clustering dispersion degree of the target label according to the clustering result includes: Determine the clustering dispersion degree of the target label according to the number of clustering clusters obtained by clustering the set of content vectors, where the clustering result includes the number of clustering clusters obtained by clustering the set of content vectors.

9. The method according to claim 8, wherein Determining the clustering dispersion degree of the target label according to the number of clustering clusters obtained by clustering the set of content vectors includes: Determine the clustering dispersion degree S of the target label through the following formula: S = D × tag_idf × Cst / FC Where Cst represents the number of clustering clusters obtained by clustering the set of content vectors, FC represents the number of media resources in the set of media resources, D represents the variance of the number of clustering clusters relative to the target mean, each number of clustering clusters is the number of clusters obtained by clustering the content vectors of a media resource set, the original description information of the media resources in each media resource set includes the same one label, and tag_idf represents the discrimination degree of the target label in the set of media resources.

10. The method according to claim 9, characterized in that, The method further includes: Determine the discrimination degree tag_idf through the following formula: Where SFC represents the total number of media resources on the preset platform, and the set of media resources is the media resources on the preset platform.

11. The method according to claim 9, wherein The method further includes: Determine the variance through the following formula: where m is used to represent the number of the clustering clusters, m is a natural number greater than 1, and x j represents the number of content vectors in the j-th clustering cluster among the m clustering clusters obtained by clustering the group of content vectors.

12. The method according to claim 7, wherein Identifying the target label according to the clustering result to obtain a label recognition result for indicating the validity of the target label includes: When the clustering dispersion degree is greater than a third preset threshold, the target label is identified to obtain a first identification result, where the label identification result includes the first identification result, and the first identification result is used to indicate that the target label is an invalid label; When the clustering dispersion degree is less than the third preset threshold, the target label is identified to obtain a second identification result, where the label identification result includes the second identification result, and the second identification result is used to indicate that the target label is a valid label.

13. A label recognition device, characterized in that, Comprising: An acquisition module, configured to determine a group of target media resources identified by a target label to be identified; Acquire a group of target description information having a one-to-one correspondence with the target media resources, where each piece of the target description information includes the description information other than the target label in the original description information of a media resource, and the original description information of each media resource includes the target label; A determination module, configured to determine a group of content vectors according to the words in the group of target description information, where the group of content vectors has a one-to-one correspondence with the group of target description information; A processing module, configured to perform clustering processing on the group of content vectors to obtain a clustering result; An identification module, configured to identify the target label according to the clustering result to obtain a label identification result for indicating the validity of the target label.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, where the program, when running, executes the method described in any one of claims 1 to 12.

15. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to execute the method described in any one of claims 1 to 12 through the computer program.

Images