CN116955775A

CN116955775A - Label recommending method, device, equipment, storage medium and program product

Info

Publication number: CN116955775A
Application number: CN202211504401.1A
Authority: CN
Inventors: 康战辉
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2022-11-28
Filing date: 2022-11-28
Publication date: 2023-10-27

Abstract

The embodiment of the application provides a label recommending method, device, equipment, storage medium and program product, and relates to the technical field of artificial intelligence and recommendation. The method comprises the following steps: acquiring a plurality of reference tag sequences, wherein the reference tag sequences are tag sequences of reference media content, and the tag sequences are used for representing a plurality of tags of the media content and arrangement sequences among the plurality of tags; generating an adjacency graph based on a plurality of reference tag sequences, wherein nodes in the adjacency graph are used for representing tags in the reference tag sequences, and edges in the adjacency graph are connected with two nodes corresponding to adjacent tags in the reference tag sequences; acquiring at least one keyword of target media content; a recommendation tag for the target media content is determined based on the adjacency graph and at least one keyword for the target media content. The technical scheme provided by the embodiment of the application can improve the determination efficiency of the label of the media content.

Description

Label recommending method, device, equipment, storage medium and program product

Technical Field

The embodiment of the application relates to the technical field of artificial intelligence and recommendation, in particular to a label recommendation method, device, equipment, storage medium and program product.

Background

The current social platform has more and more supported functions, and can conveniently and rapidly release media contents such as videos, pictures and the like for other users to browse.

In the related art, when a user issues media content such as video and pictures, the user can manually tag the media content to help the media content obtain more click through.

In the related art described above, a user needs to think about what tag is added to media content based on his own experience, and the determination of the tag is inefficient.

Disclosure of Invention

The embodiment of the application provides a tag recommendation method, a device, equipment, a storage medium and a program product, which can improve the determination efficiency of tags of media contents. The technical scheme is as follows:

according to an aspect of an embodiment of the present application, there is provided a tag recommendation method, including:

acquiring a plurality of reference tag sequences, wherein the reference tag sequences are tag sequences of reference media content, and the tag sequences are used for representing a plurality of tags of the media content and arrangement sequences among the plurality of tags;

generating an adjacency graph based on the plurality of reference tag sequences, wherein nodes in the adjacency graph are used for representing tags in the reference tag sequences, and edges in the adjacency graph are connected with two nodes corresponding to adjacent tags in the reference tag sequences;

Acquiring at least one keyword of target media content;

a recommendation tag for the target media content is determined based on the adjacency graph and at least one keyword for the target media content.

According to an aspect of an embodiment of the present application, there is provided a tag recommendation apparatus, including:

the tag acquisition module is used for acquiring a plurality of reference tag sequences, wherein the reference tag sequences are tag sequences of reference media contents, and the tag sequences are used for representing a plurality of tags of the media contents and arrangement sequences among the plurality of tags;

an adjacency graph generating module, configured to generate an adjacency graph based on the multiple reference tag sequences, where nodes in the adjacency graph are used to represent tags in the reference tag sequences, and edges in the adjacency graph connect two nodes corresponding to adjacent tags in the reference tag sequences;

the keyword acquisition module is used for acquiring at least one keyword of the target media content;

and the label determining module is used for determining a recommendation label of the target media content based on the adjacency graph and at least one keyword of the target media content.

According to an aspect of an embodiment of the present application, there is provided a computer device including a processor and a memory, in which a computer program is stored, the computer program being loaded and executed by the processor to implement the tag recommendation method described above.

According to an aspect of an embodiment of the present application, there is provided a computer-readable storage medium having stored therein a computer program loaded and executed by a processor to implement the tag recommendation method described above.

According to an aspect of an embodiment of the present application, there is provided a computer program product comprising a computer program stored in a computer readable storage medium. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program so that the computer device performs the tag recommendation method described above.

The technical scheme provided by the embodiment of the application can have the following beneficial effects:

the recommendation label matched with the media content is automatically determined based on the adjacency graph and the keywords of the media content by acquiring the reference label sequence and generating the adjacency graph based on the reference label sequence, so that the label determination efficiency of the media content is improved.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.

Drawings

FIG. 1 is a schematic diagram of a label recommendation method according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a tag recommendation system according to an embodiment of the present application;

FIG. 3 is a flowchart of a tag recommendation method according to one embodiment of the present application;

FIG. 4 is a schematic diagram of a tag sequence provided by one embodiment of the present application;

FIG. 5 is a flowchart of a label recommendation method according to another embodiment of the present application;

FIG. 6 is a schematic illustration of a random walk provided by one embodiment of the present application;

FIG. 7 is a block diagram of a tag recommendation device provided by an embodiment of the present application;

FIG. 8 is a block diagram of a tag recommendation device according to another embodiment of the present application;

FIG. 9 is a block diagram of a computer device provided in one embodiment of the application.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the application. Rather, they are merely examples of methods consistent with aspects of the application as detailed in the accompanying claims.

As shown in fig. 1, in the embodiment of the present application, a plurality of reference tag sequences 110 corresponding to a plurality of reference media contents are acquired. Wherein the plurality of reference media content is superior media content published by one or more users; among media contents distributed by one user, a plurality of media contents may be referred to as reference media contents. Based on the plurality of reference tag sequences 110, an adjacency graph 120 is constructed, the nodes of the adjacency graph 120 are used for representing the tags in the reference tag sequences 110, and the directed edges of the adjacency graph 120 are used for representing the arrangement order of adjacent tags corresponding to the nodes in the reference tag sequences 110. Determining a plurality of random walk sequences 130 from the adjacency graph 120 based on a random walk algorithm; based on the Skip Gram algorithm, at least one candidate tag sequence is determined from the plurality of random walk sequences 130. Based on the candidate tag sequence, a recommendation tag that is adapted to the target media content may be obtained.

Artificial intelligence (Artificial Intelligence, AI) is the theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and extend human intelligence, sense the environment, acquire knowledge and use the knowledge to obtain optimal results. In other words, artificial intelligence is an integrated technology of computer science that attempts to understand the essence of intelligence and to produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence, i.e. research on design principles and implementation methods of various intelligent machines, enables the machines to have functions of sensing, reasoning and decision.

The artificial intelligence technology is a comprehensive subject, and relates to the technology with wide fields, namely the technology with a hardware level and the technology with a software level. Artificial intelligence infrastructure technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and other directions.

Computer Vision (CV) is a science of studying how to "look" a machine, and more specifically, to replace a human eye with a camera and a Computer to perform machine Vision such as recognition and measurement on a target, and further perform graphic processing to make the Computer process an image more suitable for human eye observation or transmission to an instrument for detection. As a scientific discipline, computer vision research-related theory and technology has attempted to build artificial intelligence systems that can acquire information from images or multidimensional data. Computer vision techniques typically include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content/behavior recognition, three-dimensional object reconstruction, 3D techniques, virtual reality, augmented reality, synchronous positioning, and map construction, among others, as well as common biometric recognition techniques such as face recognition, fingerprint recognition, and others.

Key technologies to the speech technology (Speech Technology) are automatic speech recognition technology (ASR) and speech synthesis technology (TTS) and voiceprint recognition technology. The method can enable the computer to listen, watch, say and feel, is the development direction of human-computer interaction in the future, and voice becomes one of the best human-computer interaction modes in the future.

With research and advancement of artificial intelligence technology, research and application of artificial intelligence technology is being developed in various fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned, automatic driving, unmanned aerial vehicles, robots, smart medical treatment, smart customer service, etc., and it is believed that with the development of technology, artificial intelligence technology will be applied in more fields and with increasing importance value.

The scheme provided by the embodiment of the application relates to the technologies of computer vision, voice, machine learning and the like of artificial intelligence, such as recognizing or extracting characters from images through the computer vision technology, recognizing text contents displayed through audio through the voice technology and the like, and is specifically described by the following embodiments.

Referring to fig. 2, a schematic diagram of an implementation environment provided by an embodiment of the present application is shown, where the implementation environment may be implemented as a tag recommendation system. As shown in fig. 2, the system 20 may include: a terminal device 11.

The terminal device 11 has installed and running therein a target application program, such as a client of the target application program. Optionally, the client has a user account logged in. The terminal is an electronic device with data computing, processing and storage capabilities. The terminal may be a smart phone, a tablet computer, a PC (Personal Computer ), a wearable device, etc., which is not limited by the embodiment of the present application. The target application may be a social application, such as a text social application, a photo social application, a video social application, an audio social application, and the like, to which embodiments of the application are not limited in detail. The target application may also be any application with tag recommendation functionality, such as a payment application, a video application, a music application, a shopping application, a news application, a game application, an instant messaging application, and the like. The method provided by the embodiment of the present application may be that the execution subject of each step is the terminal device 11, such as a client running in the terminal device 11.

In some embodiments, the system 20 further includes a server 12, where the server 12 establishes a communication connection (e.g., a network connection) with the terminal device 11, and the server 12 is configured to provide background services for the target application. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing cloud computing service.

The steps of the embodiment of the present application may be performed by the terminal device 11 alone, may be performed by the server 12 alone, or may be performed alternately by the terminal device 11 and the server 12, which is not particularly limited in the embodiment of the present application.

Referring to fig. 3, a flowchart of a tag recommendation method according to an embodiment of the application is shown. In this embodiment, this method is applied to the system 20 described above for illustration. The method may include the following steps (310-340):

at step 310, a plurality of reference tag sequences are obtained.

In some embodiments, the reference tag sequence is a tag sequence that references the media content, the tag sequence being used to represent a plurality of tags of the media content and an order of arrangement between the plurality of tags.

In some embodiments, a user may publish media content through a client of a target application for other users to browse. After media content released by a user is pushed to clients of other users, the other users can browse the media content through the clients and perform operations such as praying, commenting, forwarding, collecting, bullet screen, appreciation (such as coin-in, cash-in, virtual gift-out) and the like on the media content. Wherein praise refers to emotion of support, approval, encouragement, appreciation and the like for media content presentation; of course, the user may also perform operations such as clicking, reporting, shielding, etc. on the media content to indicate negative emotions such as dissatisfaction, objection, aversion, etc. of the user to the media content. One or more data of click quantity, browsing quantity, praying quantity, comment quantity, forwarding quantity, collection quantity, bullet screen quantity, appreciation quantity, click quantity, reporting quantity and shielding quantity corresponding to the media content form feedback data corresponding to the media content.

The feedback data may be used to indicate how popular the corresponding media content is with the users, so that the feedback data may represent better performing media content (e.g., media content with a click through to a first threshold, praise to a second threshold)And the like) can be considered to be a relatively reasonable tag sequence capable of better improving the pushing accuracy of the media content, so that the media content can be determined as the reference media content, and the tag sequence of the reference media content is used as the reference tag sequence for providing reference when determining tags or tag sequences of other media content. That is, the reference tag sequence of the acquired reference media content may be considered a preferred tag sequence that may provide a positive reference value.

In some embodiments, the media content may be divided into multiple types, for example, the media content may be literal media content such as segments, news, novels, popular science articles, recipes, and the like; the media content can also be picture media content such as cartoon, photographic works, art works, calligraphic works and the like; the media content can also be audio-type media content such as dubbing works, songs, music pieces (such as piano music, violin music, etc.), looks, talk shows, audio books, etc.; the media content may also be video-like media content such as news videos, science popularization videos, food videos, and the like. Of course, the media content may also correspond to two or more types of media content described above.

In some embodiments, the tags in the tag sequence are arranged from a smaller range of words to a larger range of words. As shown in fig. 4, the media content 13 is video content related to program code, and the tags in the tag sequence 14 of the media content 13 are in turn: the labels are focused and then diverged from a smaller range to a larger range, so that a user can conveniently read the labels layer by layer in a progressive way according to the label sequence 14, and the labels are conveniently drained through the automatic label recommending system, so that forward feedback data such as the click quantity, the browse quantity and the like of media contents are improved.

At step 320, an adjacency graph is generated based on the plurality of reference tag sequences.

In some embodiments, as shown in fig. 1, the nodes in the adjacency graph 120 are used to represent labels in multiple reference label sequences 110, and the edges in the adjacency graph 120 connect two nodes corresponding to adjacent labels in the reference label sequences 120.

In some embodiments, the direction of the edge between the nodes corresponding to the two labels in the adjacency graph is determined by the arrangement order of the two labels in the reference label sequence, so that the relative relationship between focusing and diverging between the labels is embodied, thereby helping to better learn the arrangement order between the labels of the reference media content, and further optimizing the recommended labels of the target media content. For example, as shown in fig. 1, in the tag sequence corresponding to the user 1, the tags corresponding to the node D and the node a are adjacent tags, and the tag corresponding to the node a is after the tag corresponding to the node D; then for adjacency graph 120 there is an edge 15 established between node D and node a with the direction of edge 15 being from node D to node a. For another example, in the tag sequences corresponding to the user 1 and the user 3, the tags corresponding to the node a and the node B are adjacent tags, and in the tag sequence corresponding to the user 1, the tag corresponding to the node B follows the tag corresponding to the node a, and in the tag sequence corresponding to the user 3, the tag corresponding to the node a follows the tag corresponding to the node B; then for adjacency graph 120, there is established between node a and node B both an edge 16 directed from a to B and an edge 17 directed from B to a; of course, the edges 16 and 17 may be combined into a single bi-directional edge.

In some embodiments, this step further comprises the sub-steps of:

1. acquiring a first click quantity, a second click quantity and a third click quantity for a second label and a third label contained in the adjacency graph; wherein the first click rate refers to the click rate of the reference media content containing the second label, the second click rate refers to the click rate of the reference media content containing the third label, and the third click rate refers to the click rate of the reference media content containing the second label and the third label;

2. determining a fourth click volume according to the first click volume, the second click volume and the third click volume; the fourth click quantity is obtained by adding the second click quantity to the first click quantity and subtracting the third click quantity from the first click quantity;

3. comparing the third click quantity with the fourth click quantity to obtain a correlation coefficient between the second label and the third label;

4. and establishing edges between nodes corresponding to the second label and the third label in the adjacency graph under the condition that the correlation coefficient between the second label and the third label is larger than or equal to a fourth threshold value and the second label and the third label are adjacent labels in the reference label sequence.

In some embodiments, some tags have a higher heat (which may be considered hot tags) for some periods of time, such as tags associated with social hot spots for that period of time, may occur in a sequence of tags for many media content that are less relevant to the tag. That is, there are some reference media contents to which the hot spot tag is added only to promote the exposure and click quantity, and the relevance to the hot spot tag is not great in practice. Then edges between the hot tag and other low-relevance tags need to be removed, i.e., some edges in the adjacency graph are filtered out.

It will be appreciated that two tags of low relevance, which co-occur in the same tag sequence, should be much less frequently than the respective occurrence of the two tags in the tag sequence. Thus, for two adjacent tags, such as the second tag and the third tag, the first click rate, the second click rate, and the third click rate are obtained respectively, and a fourth click rate is calculated, where the fourth click rate is used to represent a sum of the click rate of the reference media content corresponding to the reference tag sequence that includes only the second tag but does not include the third tag and the click rate of the reference media content corresponding to the reference tag sequence that includes only the third tag but does not include the second tag, among the plurality of reference sequences. And comparing the third click quantity with the fourth click quantity to obtain a ratio, namely a correlation coefficient between the second label and the third label. Reference is made to the following calculation formula:

wherein a represents the second label, B represents the third label, and J (a, B) represents the correlation coefficient between the second label and the third label.

When the correlation coefficient is larger than or equal to the fourth threshold value, the correlation between the second label and the third label is larger, and an edge is established between nodes corresponding to the second label and the third label; and under the condition that the correlation coefficient is smaller than the fourth threshold value, the correlation between the second label and the third label is smaller, and then the edges between the nodes corresponding to the second label and the third label are removed, or the edges between the nodes corresponding to the second label and the third label are not established. The fourth threshold may refer to 0.1, 0.28, 0.3, 0.4, 0.5, 0.65, 0.7, and so on, which may be specifically set by a person skilled in the relevant arts according to actual situations, and the embodiment of the present application is not limited thereto specifically.

For example, if the fourth threshold is 0.1, the first click amount corresponding to the second label is 1241 times, the second click amount corresponding to the third label is 20 times, and the third click amount corresponding to the second label and the third label is 15 times, the correlation coefficient between the second label and the third label is: 15/(1241+20-15) ≡ 0.012,0.012 is smaller than the fourth threshold value 0.1, which means that the correlation between the second tag and the third tag is smaller, and should not appear in the tag sequence of the same media content, the edge between the nodes corresponding to the second tag and the third tag is removed, or the edge is not established between the nodes corresponding to the second tag and the third tag.

In the above embodiment, the correlation coefficient between the second tag and the third tag is calculated based on the click quantity of the reference media content corresponding to the two tags; in addition, the correlation coefficient between the second tag and the third tag can be calculated based on the number of the reference media contents corresponding to the two tags; the correlation coefficient between the second tag and the third tag can also be calculated based on other feedback data corresponding to the second tag and the third tag, such as the correlation coefficient between the second tag and the third tag based on browsing amount or praise amount corresponding to the second tag and the third tag; the correlation coefficient between the second tag and the third tag may also be calculated based on various feedback data corresponding to the two. For specific calculation methods of other correlation coefficients, reference may be made to the above embodiments, and details are not repeated here.

In some embodiments, edges in the adjacency graph may also be filtered based on the collaborative filtering method of UserCF.

In the embodiment, based on calculating the correlation coefficient between the second label and the third label, edges between nodes corresponding to labels with low correlation are removed, so that an adjacent structure diagram is optimized, accuracy of the adjacent structure diagram is improved, and accuracy of the finally obtained recommended label of the target media content is improved.

Step 330, at least one keyword of the target media content is obtained.

In some embodiments, the target media content is media content for which the tag sequence has not been finalized, needs modification or addition, such as media content to be published. In some embodiments, the target media content may also be published media content, and the user may modify the tags in the tag sequence of the published media content and the order between the tags.

In some embodiments, keywords related to the target media content may be directly obtained based on the target media content, and/or extracted by image recognition, audio recognition, or the like. The keywords of the target media content may be one or more, and the keywords may be used to represent the subject matter of the target media content to be expressed, the keywords of the introduced content, etc.

In some embodiments, at least one keyword of the target media content is obtained, including at least one of:

1. keywords are determined based on the title of the target media content.

In some embodiments, the title of the target media content is segmented to obtain keywords. Further, in the process of word segmentation of the title, nonsensical words are filtered out. Wherein, nonsensical words can be' and other auxiliary words; nonsensical words may also be "although", "but", "and the like; the nonsensical word may also be a pronoun of "we", "he", "here", etc. For example, for the title "place where we grow-S city", the corresponding keywords may be "grow", "S city", "place"; and the words of "we" and "we" can be filtered out, i.e. not used as keywords.

2. Keywords are determined based on the document content contained in the target media content.

In some embodiments, where the target media content is video content or audio content, some text content may be included in the target media content, such as notes, conversations, and the like. And segmenting the text content to obtain the keywords of the target media content. The word segmentation process may refer to the above and will not be described here again.

In some embodiments, in the case of longer text, if the number of characters in the text reaches a certain threshold (e.g., 50 characters), the word with higher occurrence frequency in the text may be preferentially determined as the keyword. In some embodiments, the document content is divided into a plurality of chapters or paragraphs, and the title of each chapter or paragraph may be determined as a keyword.

3. Keywords are determined based on text content extracted from an image of the target media content.

In some embodiments, the target media content is video content or image content, and the video frames may be considered as images that make up the video content. Some text content may also be present in the image, and keywords for the target media content may be determined by word segmentation or other means based on the text content. For example, some video frames of the video content are pasted with pictures of the chicken breast, text content of chicken breast is displayed beside the pictures of the chicken breast, and then the words of chicken breast can be extracted by means of image recognition and the like, and the chicken breast is determined to be a keyword.

4. Keywords are determined by analyzing text content generated by images in the target media content.

In some embodiments, the target media content is video content or an image, and the video frame may be considered to be an image that makes up the video content. The content in the image can be identified by performing image recognition on the image, and the identified content of the image is determined as a keyword. For example, if it is recognized by image recognition or the like that a picture of a chicken breast in the image of the target media content exists, the "chicken breast" may be determined as a keyword.

In some embodiments, for target media content in the form of images, image content occupying a larger area in the image is preferentially determined as keywords; for target media content in the video form, the corresponding image content with longer video duration is preferentially determined as a keyword.

5. And determining keywords based on the content partitions corresponding to the target media content.

In some private settings, a user may select a corresponding content partition of the target media content in the platform, such as a partition of food, numbers, sports, knowledge, etc., during the distribution of the target media content. After the user selects to determine the content partition corresponding to the target media content, the name of the content partition may be determined as a keyword of the target media content.

6. Keywords are determined based on the comments of the target media content.

In some embodiments, after the target media content is posted, users of the corresponding platform may post comments based on the target media content, and may consider that the comments have a greater relevance to the target media content, and may determine keywords based on the comments of the target media content. Reference is made to the above for specific embodiments, and details are not repeated here.

In this embodiment, the keywords of the target media content may be determined in a plurality of ways, so that the richness of at least one keyword is improved, and the recommendation label of the target media content obtained based on the richness is also more abundant and diverse.

Step 340, determining a recommendation label for the target media content based on the adjacency graph and at least one keyword of the target media content.

In some embodiments, since at least one keyword is strongly related to the target media content, tags with a higher similarity to the at least one keyword may be selected from the adjacency graph and determined as recommended tags for the target media content.

In some embodiments, the obtained recommended label of the target media content may be displayed in a candidate area of the user interface, so that the user can determine whether to use the recommended label by himself, thereby improving the label determination efficiency and simultaneously enabling the degree of coincidence between the label of the media content and the mind of the user to be as high as possible; the recommendation label can be automatically determined as the label of the target media content, namely the target media content is automatically labeled, so that the determination efficiency of the label is improved, and the release efficiency of the media content is improved.

In summary, according to the technical scheme provided by the embodiment of the application, the adjacency graph is generated based on the reference tag sequence by acquiring the reference tag sequence, and the recommendation tag adapted to the media content is automatically determined based on the adjacency graph and the keywords of the media content, so that the tag determination efficiency of the media content is improved.

In some possible implementations, as shown in fig. 5, step 340 in the embodiment of fig. 3 described above may include the following sub-steps (341-344):

in step 341, a graph vector corresponding to each tag in the adjacency graph is obtained, where the graph vector is used to represent a tag sequence with the tag as a center tag.

In some embodiments, a tag sequence corresponding to each tag in the adjacency graph is obtained, and a graph vector corresponding to each tag is determined based on the tags in the tag sequence and the arrangement sequence between the tags. For example, different labels may be represented as different label elements, between which there may be edge elements for representing the direction of edges between nodes to which the labels correspond, the label elements and the edge elements together constituting a graph vector of a label sequence centered around the respective label.

In some embodiments, this step 341 further includes the steps of:

1. in the adjacency graph, obtaining a random tag sequence taking the first tag as a center tag through random walk;

2. selecting a random tag sequence meeting a first condition from a plurality of random tag sequences taking the first tag as a center tag as a tag sequence corresponding to the first tag; wherein the first condition comprises: the product of the occurrence probabilities of other tags except the first tag in the random tag sequence is maximum;

3. and carrying out vectorization representation on the label sequence corresponding to the first label to obtain a graph vector corresponding to the first label.

Wherein, the occurrence probability refers to: the probability that a tag will appear at a particular location in the random tag sequence of the center tag that is the first tag. That is, in the case of a tag centered on a certain determined tag, the probability that other tags appear at different positions in the random tag sequence may be different. For example, since the label "skin disease" is more focused than the label "medical", the probability that the label "skin disease" is located before the label "medical" is higher than the probability that the label "medical" is located before the label "skin disease".

In some embodiments, for the first tag, a plurality of random tag sequences corresponding to the first tag are determined based on a random walk algorithm (such as a deep algorithm), and a tag located in a middle position of the random tag sequences is a center tag. In some embodiments, v _i Optimizing front and rear labels v for center label _i-w ,…,v _i-1 ,v _i+1 ,…,v _i+w Based on the assumption that these tags appear independent of each other, then the tag at the center is determined as v _i In the case of the tag, the probability of occurrence of the tag before and after it is equal to the product of the probability of occurrence of each tag in the tag before and after it. The product can be expressed as the following formula:

wherein, the liquid crystal display device comprises a liquid crystal display device,v _j other tags than the center tag are shown.

In some embodiments, as shown in FIG. 6, for adjacency graph 18, the slave tag v ₄ Starting, according to the directed edge of the adjacent graph, a random walk sequence is obtained by random walk, wherein the random walk sequence is as follows: v ₄ ，v ₃ ，v ₁ ，v ₅ …, where v ₁ As a center tag, tag v ₃ And tag v ₅ Is the center label v ₁ Is a front-to-back tag of (c). Based on skip-gram language model, through label v ₁ Word vector prediction of (a) its front and rear labels, maximizing label v ₃ And tag v ₅ In label v ₁ Is the probability product under the center label. Based on the independence assumption, the conditional probability of each front and rear label at the node corresponding to the center label is irrelevant to other nodes, and then the form of the respective conditional probability of the front and rear labels can be written. In some embodiments, due to Difficult to calculate directly, hierarchical Softmax (hierarchical Softmax) 19 can be used to make a two-class, i.e. predict a given tag v _i Down label v _j Probability of occurrence. Hierarchical Softmax corresponds to logV binary classifiers.

In the above embodiment, by maximizing the product of the occurrence probabilities of the front and rear labels, the most likely label sequence that occurs under the condition that the center label is determined can be obtained, that is, the optimal label sequence under the condition is obtained, so that the universality of the obtained label sequence is improved as much as possible, and only the improvement of the label sequence candidate and the recommended label is facilitated.

In step 342, similarity calculation is performed on the image quantities to obtain the image similarity between the label sequences.

In some embodiments, graph similarity is used to represent the degree of similarity between the tags contained in the two tag sequences and the order in which the tags are arranged. In some embodiments, the image similarity between the image amounts may be cosine similarity (i.e. consin similarity) between vectors, or euclidean distance between vectors, which is not limited in the embodiments of the present application.

At step 343, at least one candidate tag sequence is determined from the tag sequences corresponding to each tag in the adjacent graph based on the graph similarity between the plurality of tag sequences.

In some embodiments, the graph similarity between each tag sequence and the other tag sequences is known by the graph similarity between each tag sequence and the other tag sequences, and at least one candidate tag sequence is determined based on the graph similarity.

In some embodiments, selecting a tag sequence satisfying a second condition from tag sequences respectively corresponding to the tags in the adjacency graph to obtain at least one candidate tag sequence; wherein the second condition comprises: the graph similarity to the first number or first ratio of other tag sequences is greater than or equal to the threshold value. By meeting the second condition, the graph similarity between the selected candidate tag sequence and at least part of the tag sequences can be ensured to be higher, so that the tag sequences which are not similar to other tag sequences or are similar to only a small part of the tag sequences are prevented from being selected as the candidate tag sequences, the overall quality of the obtained at least one candidate tag sequence is ensured, and the universality of the at least one candidate tag sequence for different media contents is improved.

In some embodiments, the first number may be 1, 5, 20, 30, etc., the first ratio may be 1/10, 1/8, 1/3, 1/2, etc., and the threshold value may be 0.05, 0.15, 0.2, 0.3, etc. The specific values of the first number, the first ratio and the threshold value may be set by a related person according to actual situations, which is not specifically limited in the embodiment of the present application.

In some embodiments, at least one candidate tag sequence may constitute a set of candidate tag sequences.

In step 344, a recommendation tag for the target media content is determined based on the at least one candidate tag sequence and the at least one keyword for the target media content.

The at least one candidate tag sequence selected in the above step is not adapted to the target media content, and thus, further selection of tags in the at least one candidate tag sequence based on the target media content (e.g., based on at least one keyword of the target media content) is required to select and determine recommended tags suitable for the target media content.

In some embodiments, obtaining semantic similarity between each tag in the at least one candidate tag sequence and each keyword; and determining the recommended label of the target media content from at least one candidate label sequence based on the semantic similarity between each label and each keyword. And screening the tags in the candidate tag sequence by adopting the keywords of the target media content, so that a recommended tag matched with the target media content is obtained, and the matching degree of the recommended tag and the target media content is ensured.

In some embodiments, a word vector for each tag and a word vector for each keyword are obtained; and carrying out similarity calculation on the word vector of each label and the word vector of each keyword to obtain semantic similarity between each label and each keyword. The word vector of each tag and the word vector of each keyword can be obtained based on a word2vec algorithm, or can be obtained based on Bert or even an algorithm added with a supervised semantic similarity labeling finetune; the semantic similarity between each label and each keyword may be cosine similarity between each label and each keyword, or may be represented by euclidean distance or other forms, which is not specifically limited in the embodiment of the present application.

In some embodiments, determining a recommended tag for the target media content from the at least one candidate tag sequence based on semantic similarity between each tag and each keyword, respectively, includes at least one of:

mode one:

and determining the label with the semantic similarity with any one keyword being greater than or equal to a first threshold value as a recommended label in labels contained in at least one candidate label sequence.

In some embodiments, a tag may be determined to be a recommended tag as long as the semantic similarity between the tag and any one of the keywords of the target media content is greater than a first threshold. Alternatively, the first threshold may be 0.7, or may be other values, which may be specifically set by a relevant technician according to the actual situation, which is not specifically limited in the embodiment of the present application. By the method, each keyword of the target media content can be matched with the corresponding recommendation label as much as possible, and therefore richness of the recommendation label is improved.

Mode two:

and determining the label with the average value of the semantic similarity with at least one keyword being greater than or equal to a second threshold value as a recommended label from labels contained in the at least one candidate label sequence.

In some embodiments, for the tags included in the at least one candidate tag sequence, the average value of semantic similarity between the tag sequence and all keywords of the target media content is required to be greater than or equal to a second threshold value, so that the tag sequence can be determined to be the recommended tag of the target media content, and therefore certain correlation between each recommended tag and all keywords is ensured, and matching and accuracy of the obtained recommended tag and the target media content are improved. Alternatively, the specific value of the second threshold may be set by a skilled person according to the actual situation, which is not particularly limited in the embodiment of the present application.

In some embodiments, in the event that the number of at least one keyword is less than or equal to a fifth threshold, a recommendation label for the target media content is determined in a second manner.

Mode three:

and determining the label with the average value of the semantic similarity with the second number or the second proportion of keywords being greater than or equal to a third threshold value as a recommended label in the labels contained in the at least one candidate label sequence.

In some embodiments, a tag may be determined to be a recommended tag whenever the similarity between the tag and a portion of the keywords reaches a third threshold. Therefore, for the case that at least one keyword comprises a plurality of different aspects, the recommendation labels respectively matched with the keywords in different aspects can be obtained with high probability, and the richness and the comprehensiveness of the recommendation labels are improved.

The specific values of the second number, the second ratio, and the third threshold may be set by a person skilled in the relevant art according to the actual situation, which is not specifically limited in the embodiment of the present application.

In some embodiments, if the manners of determining the recommendation label for the target media content include the second and third manners described above, the third threshold is greater than the second threshold; if the ways of determining the recommendation label of the target media content include the first way and the second way, the first threshold is greater than the second threshold.

In some embodiments, the recommendation tags for the target media content are obtained by one or more of the above-described determinations.

In summary, according to the technical solution provided in the embodiments of the present application, through similarity calculation, a better tag sequence is extracted from an adjacency graph obtained based on a reference tag sequence, so as to obtain at least one candidate tag sequence, where the at least one candidate tag sequence may be obtained in advance through the pre-execution step; and then, the tags in the at least one candidate tag sequence can be used for determining the recommended tags of the plurality of media contents, so that the determination efficiency of the recommended tags of the media contents is further improved.

In some possible implementations, the number of recommendation tags for the target media content is multiple; and sequencing the plurality of recommended labels in the plurality of reference label sequences from low to high according to the occurrence times of the plurality of recommended labels, and obtaining the recommended label sequence of the target media content.

In the above implementation, to avoid recommendation tags being individual stand alone tags, after the recommendation tags for the target media content are determined, the tags may be ordered and presented in a certain order (e.g., from focused to divergent). Because the divergent tags appear in the reference tag sequence more times than the focused tags, the number of times that a plurality of recommended tags appear in the plurality of reference tag sequences can be ordered from low to high, so that a recommended tag sequence of the target media content is obtained, and the plurality of tags in the recommended tag sequence are focused to be divergent, so that the sequence of the recommended tag sequence of the target media content is improved, and forward feedback data such as click quantity or browsing quantity of the target media content is facilitated to be improved.

In some possible implementations, the number of recommendation tags for the target media content is multiple; after determining the recommendation label for the target media content based on the adjacency graph and the at least one keyword of the target media content, further comprising: mapping a plurality of recommended labels to nodes in the adjacency graph respectively, wherein the recommended labels are matched with labels corresponding to the mapped nodes; and determining the arrangement sequence among a plurality of recommendation tags based on the directed edges between the nodes to which the recommendation tags are mapped, and obtaining a recommendation tag sequence of the target media content.

In the above implementation, after obtaining the plurality of recommended labels, in order to arrange the plurality of recommended labels in the preferred label sequences of the reference label sequences, the plurality of recommended labels may be mapped into the adjacency graph. Because the adjacency graph can represent the relevance and the sequence between the labels through the directed edges, the arrangement sequence among a plurality of recommended labels can be determined according to the directed edges between the nodes to which the recommended labels are mapped, so that a recommended label sequence which is adaptive to the target media content and contains labels with characteristics from focusing to diverging is obtained, and forward feedback data such as click quantity or browsing quantity of the target media content can be promoted.

In some embodiments, the obtained recommended tag sequence can be directly determined as the tag sequence of the target media content, so that the time for determining the tag sequence is reduced, and the tag determination efficiency of the target media content is improved.

In some embodiments, the cosine value between the two vectors a, b can be found by the Euclidean dot product formula as follows:

a·＝‖a‖‖b‖cosθ

thus, given two vectors A, B, the cosine similarity between the two can be derived from the dot product of the vectors and the vector length as follows:

wherein A is _i Representing a word vector corresponding to an ith tag in at least one recommended tag sequence, A _i And representing a word vector corresponding to an ith keyword in the at least one keyword, wherein similarity represents cosine similarity.

The cosine similarity ranges from [ -1,1], where-1 means that the directions of the two vectors are exactly opposite, 1 means that the directions of the two vectors are exactly the same, 0 means that the two vectors are generally independent, and 0 means that the similarity between the two vectors is at an intermediate level in the present application.

As shown in the following table 1, after the technical scheme provided by the embodiment of the application is adopted, recall rate, exposure duty ratio and click rate corresponding to the search recommendation label are all optimized. That is, by adopting the technical scheme provided by the embodiment of the application, the media content can obtain better recommended labels and better release effects.

TABLE 1

The following are examples of the apparatus of the present application that may be used to perform the method embodiments of the present application. For details not disclosed in the embodiments of the apparatus of the present application, please refer to the embodiments of the method of the present application.

Referring to fig. 7, a block diagram of a tag recommendation device according to an embodiment of the application is shown. The device has the function of realizing the label recommending method example, and the function can be realized by hardware or can be realized by executing corresponding software by hardware. The device may be the terminal device described above, or may be provided on the terminal device. The apparatus 700 may include: a tag acquisition module 710, an adjacency graph generation module 720, a keyword acquisition module 730, and a tag determination module 740.

The tag obtaining module 710 is configured to obtain a plurality of reference tag sequences, where the reference tag sequences are tag sequences of reference media content, and the tag sequences are used to represent a plurality of tags of the media content and an arrangement order among the plurality of tags.

The adjacency graph generating module 720 is configured to generate an adjacency graph based on the multiple reference tag sequences, where nodes in the adjacency graph are used to represent tags in the reference tag sequences, and edges in the adjacency graph connect two nodes corresponding to adjacent tags in the reference tag sequences.

The keyword obtaining module 730 is configured to obtain at least one keyword of the target media content.

The tag determination module 740 is configured to determine a recommendation tag of the target media content based on the adjacency graph and at least one keyword of the target media content.

In some embodiments, as shown in fig. 8, the tag determination module 740 includes: a vector acquisition sub-module 741, a similarity calculation sub-module 742, and a label determination sub-module 743.

The vector obtaining sub-module 741 is configured to obtain a graph vector corresponding to each tag in the adjacency graph, where the graph vector is used to represent a tag sequence with the tag as a center tag.

The similarity calculation submodule 742 is configured to calculate a similarity between two pairs of the plurality of image amounts, so as to obtain a graph similarity between two pairs of the plurality of tag sequences, where the graph similarity is used to represent a degree of similarity between tags included in two tag sequences and an arrangement order of the tags.

The tag determination submodule 743 is configured to determine at least one candidate tag sequence from tag sequences corresponding to each tag in the adjacency graph, based on the graph similarity between the plurality of tag sequences.

The tag determination submodule 743 is further configured to determine a recommended tag for the target media content based on the at least one candidate tag sequence and at least one keyword for the target media content.

In some embodiments, as shown in fig. 8, the vector acquisition sub-module 741 is configured to:

in the adjacency graph, a random tag sequence taking the first tag as a center tag is obtained through random walk;

selecting a random tag sequence meeting a first condition from a plurality of random tag sequences taking the first tag as a center tag as a tag sequence corresponding to the first tag; wherein the first condition includes: the product of the occurrence probabilities of other tags except the first tag in the random tag sequence is maximum;

and carrying out vectorization representation on the label sequence corresponding to the first label to obtain a graph vector corresponding to the first label.

In some embodiments, as shown in fig. 8, the tag determination submodule 743 is configured to:

selecting a tag sequence meeting a second condition from tag sequences corresponding to the tags in the adjacency graph respectively to obtain at least one candidate tag sequence;

Wherein the second condition includes: the graph similarity to the first number or first ratio of other tag sequences is greater than or equal to the threshold value.

acquiring semantic similarity between each label in the at least one candidate label sequence and each keyword;

and determining the recommended label of the target media content from the label sequence of at least one candidate based on the semantic similarity between each label and each keyword.

determining a label with the semantic similarity with any keyword being greater than or equal to a first threshold value in labels contained in the at least one candidate label sequence as the recommended label;

determining a label with the average value of the semantic similarity with the at least one keyword being greater than or equal to a second threshold value as the recommended label;

and determining the label with the average value of the semantic similarity with the keywords of the second number or the second proportion being greater than or equal to a third threshold value as the recommended label.

In some embodiments, the adjacency graph generating module 720 is configured to:

acquiring a first click quantity, a second click quantity and a third click quantity for a second label and a third label contained in the adjacency graph; wherein the first click rate refers to the click rate of the reference media content containing the second tag, the second click rate refers to the click rate of the reference media content containing the third tag, and the third click rate refers to the click rate of the reference media content containing the second tag and the third tag;

determining a fourth click quantity according to the first click quantity, the second click quantity and the third click quantity; wherein the fourth click rate is obtained by adding the first click rate to the second click rate and subtracting the third click rate from the second click rate;

comparing the third click rate with the fourth click rate to obtain a correlation coefficient between the second label and the third label;

and establishing edges between nodes corresponding to the second tag and the third tag in the adjacency graph under the condition that the correlation coefficient between the second tag and the third tag is larger than or equal to a fourth threshold value and the second tag and the third tag are adjacent tags in the reference tag sequence.

In some embodiments, the direction of the edge between the nodes corresponding to the two labels in the adjacency graph is determined by the arrangement order of the two labels in the reference label sequence.

In some embodiments, the number of recommendation tags for the target media content is a plurality; as shown in fig. 8, the apparatus 700 further includes: the tag sequence determination module 750.

The tag sequence determining module 750 is configured to sort the plurality of recommended tags according to the number of times that the plurality of recommended tags appear in the plurality of reference tag sequences from low to high, so as to obtain a recommended tag sequence of the target media content.

In some embodiments, the number of recommendation tags for the target media content is a plurality; as shown in fig. 8, the apparatus 700 further includes: the tag mapping module 760.

The label mapping module 760 is configured to map the plurality of recommended labels to nodes in the adjacency graph, where the recommended labels are matched with labels corresponding to the mapped nodes.

The tag sequence determining module 750 is further configured to determine an arrangement order among a plurality of recommended tags based on the directed edges between the nodes to which the recommended tags are mapped, so as to obtain a recommended tag sequence of the target media content.

In some embodiments, the keyword obtaining module 730 is configured to:

determining the keywords based on the title of the target media content;

determining the keywords based on the document content contained in the target media content;

determining the keywords based on text content extracted from the image of the target media content;

determining the keywords by analyzing text content generated by images in the target media content;

determining the keywords based on the content partitions corresponding to the target media content;

and determining the keywords based on the comments of the target endosome content.

It should be noted that, in the apparatus provided in the foregoing embodiment, when implementing the functions thereof, only the division of the foregoing functional modules is used as an example, in practical application, the foregoing functional allocation may be implemented by different functional modules, that is, the internal structure of the device is divided into different functional modules, so as to implement all or part of the functions described above. In addition, the apparatus and the method embodiments provided in the foregoing embodiments belong to the same concept, and specific implementation processes of the apparatus and the method embodiments are detailed in the method embodiments and are not repeated herein.

Referring to fig. 9, a block diagram of a computer device according to an embodiment of the present application is shown. The computer device is used for implementing the label recommending method provided in the embodiment. Specifically, the present application relates to a method for manufacturing a semiconductor device.

The computer apparatus 900 includes a CPU (Central Processing Unit ) 901, a system Memory 904 including a RAM (Random Access Memory ) 902 and a ROM (Read-Only Memory) 903, and a system bus 905 connecting the system Memory 904 and the central processing unit 901. The computer device 900 also includes a basic I/O (Input/Output) system 906, which helps to transfer information between various devices within the computer, and a mass storage device 907, for storing an operating system 913, application programs 914, and other program modules 915.

The basic input/output system 906 includes a display 908 for displaying information and an input device 909, such as a mouse, keyboard, etc., for user input of information. Wherein the display 908 and the input device 909 are connected to the central processing unit 901 via an input output controller 910 connected to the system bus 905. The basic input/output system 906 can also include an input/output controller 910 for receiving and processing input from a number of other devices, such as a keyboard, mouse, or electronic stylus. Similarly, the input-output controller 910 also provides output to a display screen, a printer, or other type of output device.

The mass storage device 907 is connected to the central processing unit 901 through a mass storage controller (not shown) connected to the system bus 905. The mass storage device 907 and its associated computer-readable media provide non-volatile storage for the computer device 900. That is, the mass storage device 907 may include a computer readable medium (not shown) such as a hard disk or CD-ROM (Compact Disc Read-Only Memory) drive.

The computer readable medium may include computer storage media and communication media without loss of generality. Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory, erasable programmable read-only memory), flash memory or other solid state memory, CD-ROM, DVD (Digital Video Disc, high density digital video disc) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. Of course, those skilled in the art will recognize that the computer storage medium is not limited to the one described above. The system memory 904 and mass storage device 907 described above may be collectively referred to as memory.

According to various embodiments of the application, the computer device 900 may also operate by being connected to a remote computer on a network, such as the Internet. I.e., the computer device 900 may be connected to the network 912 through a network interface unit 911 coupled to the system bus 905, or other types of networks or remote computer systems (not shown) may be coupled using the network interface unit 911.

In an exemplary embodiment, a computer readable storage medium is also provided, in which a computer program is stored which, when being executed by a processor, implements the above-mentioned tag recommendation method.

Alternatively, the computer-readable storage medium may include: ROM (Read-Only Memory), RAM (Random-Access Memory), SSD (Solid State Drives, solid State disk), optical disk, or the like. The random access memory may include ReRAM (Resistance Random Access Memory, resistive random access memory) and DRAM (Dynamic Random Access Memory ), among others.

In an exemplary embodiment, a computer program product is also provided, the computer program product comprising a computer program stored in a computer readable storage medium. The processor of the computer device reads the computer program from the computer-readable storage medium, and the processor executes the computer program so that the computer device performs the tag recommendation method described above.

It should be noted that, before collecting relevant data of a user and during collecting relevant data of a user (for example, reference media content, target media content, document content, comments and the like mentioned in the present application), the present application may display a prompt interface, a popup window or output voice prompt information, where the prompt interface, the popup window or the voice prompt information is used to prompt the user to collect relevant data currently, so that the present application only starts to execute the relevant step of obtaining relevant data of the user after obtaining the confirmation operation sent by the user to the prompt interface or the popup window, otherwise (i.e., when the confirmation operation sent by the user to the prompt interface or the popup window is not obtained), ends the relevant step of obtaining relevant data of the user, i.e., does not obtain relevant data of the user. In other words, all user data collected by the present application is collected with the consent and authorization of the user, and the collection, use and processing of relevant user data requires compliance with relevant laws and regulations and standards of the relevant country and region.

It should be understood that references herein to "a plurality" are to two or more. "and/or", describes an association relationship of an association object, and indicates that there may be three relationships, for example, a and/or B, and may indicate: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship.

The foregoing description of the exemplary embodiments of the application is not intended to limit the application to the particular embodiments disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the application.

Claims

1. A label recommendation method, the method comprising:

acquiring at least one keyword of target media content;

2. The method of claim 1, wherein the determining a recommendation tag for the target media content based on the adjacency graph and at least one keyword for the target media content comprises:

Obtaining graph vectors corresponding to all the labels in the adjacent graph respectively, wherein the graph vectors are used for representing a label sequence taking the label as a central label;

performing similarity calculation on the plurality of image quantities to obtain graph similarity between the plurality of tag sequences, wherein the graph similarity is used for representing the similarity between tags contained in the two tag sequences and the arrangement sequence of the tags;

determining at least one candidate tag sequence from tag sequences corresponding to each tag in the adjacent graph based on graph similarity between the tag sequences;

a recommendation tag for the target media content is determined based on the at least one candidate tag sequence and at least one keyword for the target media content.

3. The method according to claim 2, wherein the obtaining the graph vectors respectively corresponding to the labels in the adjacency graph includes:

4. The method according to claim 2, wherein determining at least one candidate tag sequence from tag sequences respectively corresponding to each tag in the adjacency graph based on graph similarity between the plurality of tag sequences comprises:

5. The method of claim 2, wherein the determining the recommendation tag for the target media content based on the at least one candidate tag sequence and at least one keyword for the target media content comprises:

6. The method of claim 5, wherein determining the recommended tags for the target media content from the at least one candidate tag sequence based on semantic similarity between the respective tags and the respective keywords, comprises at least one of:

7. The method of any of claims 1 to 6, wherein the generating an adjacency graph based on the plurality of reference tag sequences comprises:

8. The method of any one of claims 1 to 6, wherein the direction of the edge between the nodes corresponding to two labels in the adjacency graph is determined by the order in which the two labels are arranged in the reference label sequence.

9. The method of any one of claims 1 to 6, wherein the number of recommended tags for the target media content is a plurality;

after determining the recommendation label of the target media content based on the adjacency graph and at least one keyword of the target media content, the method further comprises:

And sequencing the occurrence times of the plurality of recommended labels in the plurality of reference label sequences from low to high to obtain the recommended label sequence of the target media content.

10. The method of any one of claims 1 to 6, wherein the number of recommended tags for the target media content is a plurality;

mapping a plurality of recommended labels to nodes in the adjacency graph respectively, wherein the recommended labels are matched with labels corresponding to the mapped nodes;

and determining the arrangement sequence among a plurality of recommendation tags based on the directed edges between the nodes to which the recommendation tags are mapped, and obtaining a recommendation tag sequence of the target media content.

11. The method according to any one of claims 1 to 6, wherein the obtaining at least one keyword of the target media content comprises at least one of:

determining the keywords based on the title of the target media content;

12. A tag recommendation device, the device comprising:

13. A computer device, characterized in that it comprises a processor and a memory in which a computer program is stored, which computer program is loaded and executed by the processor to implement the tag recommendation method according to any of the preceding claims 1 to 11.

14. A computer readable storage medium, characterized in that the computer readable storage medium has stored therein a computer program, which is loaded and executed by a processor to implement the tag recommendation method according to any of the preceding claims 1 to 11.

15. A computer program product, characterized in that it comprises a computer program stored in a computer readable storage medium, from which a processor reads and executes the computer program to implement the tag recommendation method according to any one of claims 1 to 11.