Video search term recommendation method, device, electronic device and storage medium

Abstract

The present disclosure relates to a method, device, electronic device and storage medium for recommending video search terms. The method includes: for a target video, extracting conceptual entities associated with the video content of the target video and the entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the target video; inputting the video content text and the associated knowledge graph of the target video into a pre-trained search term generation model to obtain at least one candidate search term; inputting each candidate search term and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each candidate search term; determining the search term to be recommended for the target video based on the candidate search term whose relevance label meets preset conditions; the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video. The present disclosure can be used to generate search terms in combination with the associated knowledge graph, thereby improving the relevance of the search term and the video content.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for recommending video search terms. Background Art

[0002] At present, when users watch videos, by generating search terms for the videos they watch and recommending them to users, users can be guided to initiate searches, increase exposure opportunities for corresponding search products, and meet users' in-depth consumption demands. However, the traditional method can only extract the brand names involved in the video to generate the corresponding search terms, resulting in a limited application range of the generated search terms, weak relevance of video content, and lack of diversity in search terms.

[0003] Therefore, there is a problem in the related art that the search terms generated for a video have a low degree of relevance to the video content of the video. Summary of the invention

[0004] The present disclosure provides a video search term recommendation method, device, electronic device and storage medium to at least solve the problem in the related art that the search terms generated for a video are less relevant to the video content of the video. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, a method for recommending video search terms is provided, comprising:

[0006] For a target video, extract conceptual entities associated with the video content of the target video and entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the target video;

[0007] Inputting the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term;

[0008] Inputting each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the relevance between the candidate search term and the video content of the target video;

[0009] According to the candidate search terms whose relevance tags meet preset conditions, a search term to be recommended for the target video is determined; the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video.

[0010] In a possible implementation, determining the search term to be recommended for the target video according to the candidate search terms whose relevance tags satisfy a preset condition includes:

[0011] The candidate search words whose relevance tags meet the preset conditions are used as target search words;

[0012] Abnormal search words in at least one of the target search words are filtered to obtain the search words to be recommended; the abnormal search words are determined according to preset business requirements and / or playback platform rules.

[0013] In a possible implementation, filtering abnormal search terms in at least one of the target search terms to obtain the search terms to be recommended includes:

[0014] If the target search term includes a designated word, the target search term is determined to be the abnormal search term; the designated word is determined based on preset business requirements and playback platform rules;

[0015] Deleting the specified word in the abnormal search term to obtain a modified search term;

[0016] The modified search term and the non-abnormal search term are used as the search term to be recommended; the non-abnormal search term is a target search term that does not contain the designated term.

[0017] In a possible implementation, filtering abnormal search terms in at least one of the target search terms to obtain the search terms to be recommended includes:

[0018] If the target search term contains a preset word, and / or the word representation concept of the target search term matches the preset abnormal representation concept, the target search term is determined to be the abnormal search term; the preset word and the preset abnormal representation concept are both determined based on preset business requirements and playback platform rules;

[0019] The abnormal search term is deleted from at least one of the target search terms to obtain the search term to be recommended.

[0020] In a possible implementation, the pre-trained search term generation model has a pre-trained encoder and a pre-trained decoder, and the video content text of the target video and the associated knowledge graph are input into the pre-trained search term generation model to obtain at least one candidate search term, including:

[0021] Inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph;

[0022] The encoding result is input into the pre-trained decoder to obtain at least one candidate search word; the candidate search word is obtained by decoding the fused encoding result between the first encoding result and the second encoding result through the pre-trained decoder.

[0023] In a possible implementation, the step of inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result includes:

[0024] Splicing a preset search word control code with the video content text to obtain a spliced ​​text; the search word control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search word; the keyword control code is used to control whether the candidate search word contains a keyword corresponding to the keyword control code;

[0025] The concatenated text and the associated knowledge graph are input into the pre-trained encoder to obtain an encoding result.

[0026] In a possible implementation, before the step of inputting the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term, the step further includes:

[0027] Acquire first training sample data; each first training sample data includes a video content text of a first sample video and a first sample search term of the first sample video, wherein the number of clicks of a user account corresponding to the first sample search term is greater than a preset click number threshold;

[0028] For the first sample video, extract conceptual entities associated with the video content of the first sample video and entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the first sample video;

[0029] Based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term, the search term generation model to be trained is trained to obtain the pre-trained search term generation model.

[0030] In a possible implementation, before the step of inputting each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms, the method further includes:

[0031] Acquire second training sample data; each second training sample data includes a video content text of a second sample video, a second sample search term of the second sample video and its corresponding relevance label, a random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels;

[0032] Based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, the search term evaluation model to be trained is trained to obtain the pre-trained search term evaluation model.

[0033] According to a second aspect of an embodiment of the present disclosure, a video search term recommendation device is provided, comprising:

[0034] The associated knowledge graph generating unit is configured to extract concept entities associated with the video content of the target video and entity relationships between the concept entities from a preset knowledge graph for the target video, and generate an associated knowledge graph corresponding to the target video;

[0035] A candidate search term obtaining unit is configured to input the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term;

[0036] A relevance label obtaining unit is configured to input each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the relevance between the candidate search term and the video content of the target video;

[0037] The unit for determining the search term to be recommended is configured to execute the candidate search terms that meet preset conditions according to the relevance tags to determine the search term to be recommended for the target video; the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video.

[0038] In one possible implementation, the unit for determining the search term to be recommended is specifically configured to execute the candidate search term whose relevance tag meets preset conditions as the target search term; filter the abnormal search term in at least one of the target search terms to obtain the search term to be recommended; the abnormal search term is determined according to preset business requirements and / or playback platform rules.

[0039] In one possible implementation, the unit for determining the search term to be recommended is further specifically configured to execute: if the target search term contains a specified term, then determine that the target search term is the abnormal search term; the specified term is determined based on preset business requirements and playback platform rules; delete the specified term in the abnormal search term to obtain a modified search term; use the modified search term and the non-abnormal search term as the search term to be recommended; the non-abnormal search term is the target search term that does not contain the specified term.

[0040] In one possible implementation, the unit for determining the search term to be recommended is further configured to execute: if the target search term contains a preset word, and / or the word representation concept of the target search term matches the preset abnormal representation concept, then the target search term is determined to be the abnormal search term; the preset word and the preset abnormal representation concept are both determined based on preset business requirements and playback platform rules; and the abnormal search term is deleted from at least one of the target search terms to obtain the search term to be recommended.

[0041] In one possible implementation, the pre-trained search term generation model has a pre-trained encoder and a pre-trained decoder, and the candidate search term obtaining unit is specifically configured to execute inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph; the encoding result is input into the pre-trained decoder to obtain at least one candidate search term; the candidate search term is obtained by decoding a fusion encoding result between the first encoding result and the second encoding result through the pre-trained decoder.

[0042] In one possible implementation, the candidate search term obtaining unit is further configured to execute splicing of a preset search term control code with the video content text to obtain a spliced ​​text; the search term control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search term; the keyword control code is used to control whether the candidate search term contains a keyword corresponding to the keyword control code; the spliced ​​text and the associated knowledge graph are input into the pre-trained encoder to obtain an encoding result.

[0043] In a possible implementation, the video search term recommendation device further includes:

[0044] A first training sample data acquisition unit is specifically configured to execute acquisition of first training sample data; each first training sample data includes a video content text of a first sample video and a first sample search term of the first sample video, wherein the number of clicks of a user account corresponding to the first sample search term is greater than a preset click number threshold;

[0045] The associated knowledge graph generating unit corresponding to the first sample video is specifically configured to extract concept entities associated with the video content of the first sample video and entity relationships between the concept entities from a preset knowledge graph for the first sample video, and generate an associated knowledge graph corresponding to the first sample video;

[0046] The search term generation model training unit is specifically configured to execute training on the search term generation model to be trained based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term to obtain the pre-trained search term generation model.

[0047] In a possible implementation, the video search term recommendation device further includes:

[0048] The second training sample data acquisition unit is specifically configured to execute acquisition of second training sample data; each second training sample data includes a video content text of a second sample video, a second sample search term of the second sample video and its corresponding relevance label, a random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels;

[0049] The search term evaluation model training unit is specifically configured to execute training of the search term evaluation model to be trained based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, to obtain the pre-trained search term evaluation model.

[0050] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the video search term recommendation method as described in the first aspect or any possible implementation manner of the first aspect is implemented.

[0051] According to a fourth aspect of an embodiment of the present disclosure, a storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the video search term recommendation method as described in the first aspect or any possible implementation manner of the first aspect is implemented.

[0052] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, the program product comprising a computer program, the computer program being stored in a readable storage medium, at least one processor of a device reading and executing the computer program from the readable storage medium, so that the device executes the video search term recommendation method described in any one of the embodiments of the first aspect.

[0053] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0054] The scheme disclosed in the present invention, for a target video, extracts conceptual entities associated with the video content of the target video and the entity relationships between the conceptual entities from a preset knowledge graph, generates an associated knowledge graph corresponding to the target video, inputs the video content text and the associated knowledge graph of the target video into a pre-trained search term generation model to obtain at least one candidate search term, and then inputs each candidate search term and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each candidate search term, the relevance label is used to characterize the degree of relevance between the candidate search term and the video content of the target video, and then determines the search term to be recommended for the target video based on the candidate search term whose relevance label meets preset conditions, and the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video, so that search terms can be generated based on the video content in combination with the associated knowledge graph extracted from the knowledge graph, thereby improving the relevance of the generated search terms and the video content, and determining the search term to be recommended for the video based on the relevance label can ensure the relevance and effectiveness of the search term without human intervention.

[0055] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0057] Figure 1 The figure is a diagram showing an application environment of a method for recommending video search terms according to an exemplary embodiment.

[0058] Figure 2 The figure is a flowchart of a method for recommending video search terms according to an exemplary embodiment.

[0059] Figure 3a The figure is a schematic diagram of a video search term display interface according to an exemplary embodiment.

[0060] Figure 3bThe figure is a flowchart of a video search term recommendation example according to an exemplary embodiment.

[0061] Figure 4 The figure is a flowchart of another method for recommending video search terms according to an exemplary embodiment.

[0062] Figure 5 The figure is a block diagram of a video search term recommendation device according to an exemplary embodiment.

[0063] Figure 6 The figure is a diagram showing the internal structure of an electronic device according to an exemplary embodiment. DETAILED DESCRIPTION

[0064] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0065] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure.

[0066] It should also be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0067] The video search term recommendation method provided by the present disclosure can be applied to Figure 1 In the application environment shown. The user terminal 110 interacts with the server 120 through the network. The server 120 can determine the recommended search words for the target video played in the user terminal 110, and then the user terminal 110 can display the recommended search words in the played target video interface to guide the user account corresponding to the user terminal 110 to perform a search operation after accessing the target video. In actual applications, the user terminal 110 can include but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices, and the server 120 can be implemented by an independent server or a server cluster composed of multiple servers.

[0068] Figure 2 is a flowchart of a method for recommending video search terms according to an exemplary embodiment. Figure 2 As shown, this method can be used to Figure 1 The server 120 includes the following steps.

[0069] In step S210, for the target video, conceptual entities associated with the video content of the target video and entity relationships between the conceptual entities are extracted from a preset knowledge graph to generate an associated knowledge graph corresponding to the target video;

[0070] The target video may be a video for which search terms are to be generated. For example, during the playback of a short video, a search term may be generated for the short video.

[0071] As an example, the preset knowledge graph may be a semantic network pre-established based on the relevant content involved in the short video, which may have multiple nodes and edges connecting the nodes, each node representing an entity or a concept, and the edges representing various semantic relationships between entities / concepts.

[0072] In practical applications, for the target video for which the search term is to be generated, conceptual entities associated with the video content of the target video and the entity relationships between the conceptual entities can be extracted from the preset knowledge graph. The extracted conceptual entities and the entity relationships between the conceptual entities can then be used as the associated knowledge graph corresponding to the target video to further generate search terms in combination with the video content text of the target video.

[0073] In one example, for a test drive video of a certain car model series name x (i.e., the target video), video-related knowledge such as (car model series name x, brand, brand name y), (car model series name x, manufacturer, manufacturer z car), (car model series name x, engine type, turbocharged), (brand name y, is a, car brand) can be extracted according to the preset knowledge graph as the associated knowledge graph of the test drive video of the certain car model series name x, where the conceptual entities can be car model series name x, brand name y, etc., and the entity relationship between the conceptual entities, such as the brand corresponding to car model series name x is brand name y.

[0074] In step S220, the video content text and the associated knowledge graph of the target video are input into a pre-trained search term generation model to obtain at least one candidate search term;

[0075] As an example, the video content text can be extracted from the target video through a text extraction module. For example, a video content text may include video voice text, picture text, short video text meta-information, i.e. title, description, cover image, etc., which is not specifically limited in this embodiment.

[0076] In actual applications, for the target video for which the search term is to be generated, the video content text of the target video and the associated knowledge graph extracted from the preset knowledge graph can be input into the pre-trained search term generation model, and then at least one candidate search term for the target video can be obtained.

[0077] Specifically, the text content, video title, video voice text, video picture text, etc. contained in the video cover image of the target video can be used as the video content text, and then a pre-trained search term generation model can be used to generate at least one candidate search term based on the video content text and the extracted associated knowledge graph corresponding to the target video, such as a batch of search terms that match the short video content.

[0078] In one example, the pre-trained search term generation model architecture adopts the idea of ​​text generation enhanced by knowledge graph. By incorporating knowledge into the traditional generation model structure, it can make use of the rich semantic information between entities and their entity relationships in the knowledge graph, which can be more conducive to the search term generation task.

[0079] In step S230, each candidate search word and video content text is input into a pre-trained search word evaluation model to obtain a relevance label corresponding to each candidate search word; the relevance label is used to characterize the relevance between the candidate search word and the video content of the target video;

[0080] Among them, the relevance tag may include multiple level tags corresponding to multiple relevance levels. Different relevance tags can represent different relevance levels between the candidate search terms and the video content of the target video. For example, the relevance tag setting method can be a strong relevance tag, a weak relevance tag, an irrelevant tag, or level 2, level 1, or level 0. No specific restrictions are made in this embodiment.

[0081] After obtaining at least one candidate search term, each candidate search term and video content text can be input into a pre-trained search term evaluation model. By scoring the relevance between each candidate search term and the video content, a relevance label corresponding to each candidate search term can be obtained. Based on the relevance label, the degree of relevance of each candidate search term to the video content of the target video can be determined.

[0082] In step S240, based on the candidate search terms whose relevance tags meet preset conditions, the recommended search terms for the target video are determined; the recommended search terms are used to guide the user account to perform a search operation after accessing the target video.

[0083] After obtaining the relevance tags corresponding to each candidate search term, the candidate search terms corresponding to the relevance tags that meet the preset conditions can be screened out, and then through search term filtering processing, the recommended search terms for the target video can be determined based on the candidate search terms whose relevance tags meet the preset conditions, so as to guide the user account to perform a search operation after accessing the target video based on the recommended search terms.

[0084] For example, in order to ensure the relevance of the generated search terms, the candidate search terms corresponding to the strongly correlated tags and the candidate search terms corresponding to the weakly correlated tags can be used as the candidate search terms whose relevance tags meet the preset conditions, and the candidate search terms corresponding to the irrelevant tags will not enter the subsequent search term filtering process.

[0085] In an optional embodiment, after obtaining the recommended search terms for the target video, the recommended search terms can be displayed to users watching the target video through the search term recommendation area in the interface playing the target video, and then based on the recommended search terms, users can be guided to initiate searches after accessing the target video, thereby fully tapping the users' demand for video consumption, lowering the users' search threshold, expanding the traffic of video searches, and increasing exposure opportunities for related products searched for.

[0086] In one example, if Figure 3a As shown, in the scenario where users consume videos using a short video platform, when the user is watching the currently playing video, by mounting a search PLC (Programmable Logic Controller), that is, a search term recommendation area, in the video playing interface, based on the displayed search term xxx, the user can be guided to initiate a search behavior after accessing the video.

[0087] Compared with the traditional method of only extracting the brand names involved in the video to generate corresponding search terms, since the videos containing brand names only account for a small part of the total videos, most videos cannot generate their corresponding search terms; and the brand name only conveys part of the information in the short video. The search terms generated by only extracting the brand name lack deep connection with the short video content. Using only the brand name as a search term fails to reflect the user's real search needs. Moreover, short videos may involve many aspects of content, and different users are more interested in different aspects of the content. Using only the brand name as a search term cannot reflect the diversity of video content.

[0088] The technical solution of this embodiment is adopted to generate candidate search terms by combining the video content text and the related knowledge graph extracted from the knowledge graph, and determine the recommended search terms for the video from the candidate search terms according to the relevance tags, thereby avoiding limiting the video scope and being able to generate content-related search terms for any type of short video, thereby improving the adaptability of the search terms and the information conveyed by the short video, making the generated search terms closely related to the short video content, and for each short video, multiple related search terms reflecting different content angles of the short video can be generated, thereby enhancing the diversity of search terms.

[0089] In the above-mentioned video search term recommendation method, by extracting conceptual entities associated with the video content of the target video and the entity relationships between the conceptual entities from a preset knowledge graph for the target video, an associated knowledge graph corresponding to the target video is generated, the video content text and the associated knowledge graph of the target video are input into a pre-trained search term generation model to obtain at least one candidate search term, and then each candidate search term and the video content text are input into a pre-trained search term evaluation model to obtain a relevance label corresponding to each candidate search term, and then the search term to be recommended for the target video is determined based on the candidate search term whose relevance label meets preset conditions. In this way, search terms can be generated based on the video content combined with the associated knowledge graph extracted from the knowledge graph, which improves the relevance of the generated search terms and the video content, and the search terms to be recommended for the video are determined based on the relevance labels, which can ensure the relevance and effectiveness of the search terms without human intervention.

[0090] In an exemplary embodiment, based on candidate search words whose relevance tags meet preset conditions, search words to be recommended for a target video are determined, including: taking candidate search words whose relevance tags meet preset conditions as target search words; filtering abnormal search words in at least one target search word to obtain search words to be recommended; abnormal search words are determined based on preset business requirements and / or playback platform rules.

[0091] In a specific implementation, based on the relevance tags corresponding to each candidate search word, the candidate search words corresponding to the relevance tags that meet preset conditions can be screened out as target search words, and then abnormal search words can be determined based on preset business requirements and / or playback platform rules. Furthermore, by filtering the abnormal search words in at least one target search word, the search words to be recommended can be obtained.

[0092] For example, filtering and rewriting rules can be customized according to business needs, or they can be preset to ensure that the generated search terms comply with the unified rules of the platform (i.e., preset business needs and / or playback platform rules). Then, abnormal search terms can be identified for candidate search terms corresponding to strongly related tags and candidate search terms corresponding to weakly related tags (i.e., target search terms), and the search terms to be recommended can be obtained by filtering the abnormal search terms.

[0093] The technical solution of this embodiment, by taking candidate search terms whose relevance tags meet preset conditions as target search terms, and then filtering abnormal search terms in at least one target search term to obtain search terms to be recommended, can ensure the relevance and security of search terms without human intervention and reduce the risk of search terms being explicit.

[0094] In an exemplary embodiment, abnormal search terms in at least one target search term are filtered to obtain search terms to be recommended, including: if the target search term contains a specified term, the target search term is determined to be an abnormal search term; the specified term is determined based on preset business requirements and playback platform rules; the specified term in the abnormal search term is deleted to obtain a modified search term; the modified search term and a non-abnormal search term are used as search terms to be recommended; the non-abnormal search term is a target search term that does not contain the specified term.

[0095] In actual applications, if it is detected that a target search term contains a specified word, then the target search term can be determined to be an abnormal search term, and then the abnormal search term can be rewritten. By deleting the specified word in the abnormal search term to the modified search term, the modified search term and the non-abnormal search term can be used as search terms to be recommended.

[0096] Specifically, rewriting rules can be customized according to business needs. Based on the rewriting rules, when it is detected that a target search term contains a geographic word (i.e., a specified word), it can be determined as an abnormal search term, and the geographic word in the abnormal search term can be deleted to rewrite the abnormal search term. For example, for abnormal search terms containing geographic words such as "Heilongjiang", the geographic word "Heilongjiang" in the abnormal search term can be deleted.

[0097] The technical solution of this embodiment is to determine that the target search term is an abnormal search term if the target search term contains the specified term, and then delete the specified term in the abnormal search term to obtain a modified search term, and then use the modified search term and the non-abnormal search term as the search term to be recommended. The post-processing logic based on the search term rewriting can ensure the relevance and effectiveness of the search term without human intervention.

[0098] In an exemplary embodiment, abnormal search terms in at least one target search term are filtered to obtain search terms to be recommended, including: if the target search term contains preset words, and / or the word representation concept of the target search term matches the preset abnormal representation concept, then the target search term is determined to be an abnormal search term; the preset words and the preset abnormal representation concept are both determined based on preset business requirements and playback platform rules; and the abnormal search terms are deleted from at least one target search term to obtain the search term to be recommended.

[0099] In actual applications, if it is detected that a target search word contains preset words, and / or the word representation concept of the target search word matches the preset abnormal representation concept, then the target search word can be determined to be an abnormal search word, and then the abnormal search word can be filtered and deleted. By deleting the abnormal search word, the target search word after deleting the abnormal search word can be used as a search word to be recommended.

[0100] Specifically, you can customize the filtering rules according to business needs. Based on the filtering rules, when it is detected that a target search term contains a banned word (i.e., a preset word), it can be determined as an abnormal search term, and then the abnormal search term is filtered and deleted. Or, in order to ensure that the generated search terms comply with the unified rules of the platform, when it is detected that the word representation concept of a target search term matches the preset abnormal representation concept, it can be determined as an abnormal search term, and then the abnormal search term is filtered and deleted.

[0101] For example, abnormal search terms containing the banned word "advertisement" can be directly filtered and deleted; abnormal search terms that are judged to be "pornography" or "politics" related concepts (i.e., preset abnormal representation concepts) can also be directly filtered and deleted.

[0102] The technical solution of this embodiment is to determine that the target search term is an abnormal search term if the target search term contains preset words and / or the word representation concept of the target search term matches the preset abnormal representation concept, and then delete the abnormal search term from at least one target search term to obtain the search term to be recommended. It can ensure the relevance and effectiveness of the search term without human intervention based on the post-processing logic of search term filtering.

[0103] In an exemplary embodiment, a pretrained search term generation model has a pretrained encoder and a pretrained decoder, and the video content text and the associated knowledge graph of the target video are input into the pretrained search term generation model to obtain at least one candidate search term, including: inputting the video content text and the associated knowledge graph into the pretrained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph; the encoding result is input into the pretrained decoder to obtain at least one candidate search term; the candidate search term is obtained by decoding a fusion encoding result between the first encoding result and the second encoding result through the pretrained decoder.

[0104] In a specific implementation, the pre-trained search term generation model may adopt Bert2Bert based on Transformer (a model based on encoder-decoder, i.e., encoding-decoding structure), and the pre-trained search term generation model may have a pre-trained encoder and a pre-trained decoder, and the relevant parameters of the encoder and the decoder may be initialized by pre-trained Bert (a pre-trained model with strong generalization ability).

[0105] Based on the pre-trained search term generation model, by inputting the video content text and the associated knowledge graph into the pre-trained encoder, the video content text can be encoded to obtain a first encoding result, and the associated knowledge graph can be encoded to obtain a second encoding result. Then, the pre-trained decoder can be used to decode the fused encoding result between the first encoding result and the second encoding result to obtain at least one candidate search term.

[0106] In one example, the video content text can be encoded during the encoding process. For example, by incorporating video text information of multiple modalities, including video titles, voice text (ASR), and image text (OCR), etc., the video content understanding can be enhanced. Knowledge graph enhancement can also be performed by incorporating prior knowledge in the knowledge graph (i.e., associated knowledge graph), the video semantic information can be enriched. For example, conceptual entities associated with the video content and the entity relationships between the conceptual entities are extracted from the knowledge graph as prior knowledge, and then the prior knowledge is encoded separately as a semantic context to constrain the generation of search terms in the decoding process and improve the quality of search term generation.

[0107] In yet another example, during the decoding process, diversity control can be performed, that is, sampling can be carried out using the sampling method of the Top-p decoding strategy to increase the diversity of the generated results; duplicate character control can also be performed, that is, by penalizing duplicate text segments to mitigate the problem of the model generating duplicate segments and avoid generating search terms similar to "used second-hand car market". It can directly set the probability of characters that may form duplicate segments to 0 in each step of the decoding process, and thus avoid sampling such characters during sampling. For example, if the decoder has generated the prefix "used second", the probability of the character "second" can be set to 0; word length control can also be performed, that is, by changing the numbers in the input word length control code, the length of the generated search term and whether it contains specific keywords can be controlled; relevance control can also be performed, that is, during the decoding process, by integrating the prior knowledge in the knowledge graph and the video content text, the generated words can be constrained, improving the semantic relevance between the generated search terms and the video content.

[0108] The technical solution of this embodiment obtains an encoding result by inputting the video content text and the associated knowledge graph into a pre-trained encoder, and then inputs the encoding result into a pre-trained decoder to obtain at least one candidate search term. Based on diversity control, duplicate character control, word length control, and relevance control, the diversity and readability of the generated search terms are improved, the word length can be customized, and specific keywords can be included, enhancing the relevance of the generated search terms.

[0109] In an exemplary embodiment, inputting the video content text and the associated knowledge graph into a pre-trained encoder to obtain an encoding result includes: splicing a preset search term control code with the video content text to obtain the spliced text; the search term control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search term; the keyword control code is used to control whether the candidate search term contains a keyword corresponding to the keyword control code; inputting the spliced text and the associated knowledge graph into a pre-trained encoder to obtain an encoding result.

[0110] In one example, for the requirement in the business scenario that the length of the search term cannot be too long and the product marketing need, such as the search term needs to contain a specific keyword (brand name), it can be based on a preset search term control code, which can include a word length control code and a keyword control code. Among them, the word length control code can be used to control the word length of the candidate search term, and the keyword control code can be used to control whether the candidate search term contains a keyword corresponding to the keyword control code. By splicing the search term control code with the video content text, the spliced text can be obtained, and then inputting the spliced text and the associated knowledge graph into a pre-trained encoder, an encoding result can be obtained.

[0111] For example, the concatenated text can be expressed as follows:

[0112] [Start]6[Separator]Brand Namexx[Separator]1[Separator]XXXXXXYYYYY

[0113] Among them, the number 6 between the start symbol and the first separator can represent the length of the generated search term (i.e., the length of the candidate search term), the brand name xx between the first separator and the second separator can be the keyword content contained in the search term (i.e., the keyword corresponding to the keyword control code), the number between the second separator and the third separator can indicate whether the keyword is contained (such as 1 for yes and 0 for no), and the XXXXXXYYYYY after the third separator can be the text content of the short video (i.e., the video content text).

[0114] The technical solution of this embodiment is to obtain a spliced ​​text by splicing a preset search term control code with the video content text, and then input the spliced ​​text and the associated knowledge graph into a pre-trained encoder to obtain an encoding result, which can finely control the word length of the generated search term and whether it contains specified keywords.

[0115] In an exemplary embodiment, before the step of inputting the video content text and the associated knowledge graph of the target video into the pre-trained search term generation model to obtain at least one candidate search term, it also includes: obtaining first training sample data; each first training sample data includes the video content text of the first sample video, the first sample search term of the first sample video, and the number of user account clicks corresponding to the first sample search term is greater than a preset click count threshold; for the first sample video, extracting conceptual entities associated with the video content of the first sample video and the entity relationships between the conceptual entities from the preset knowledge graph to generate an associated knowledge graph corresponding to the first sample video; based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term, training the search term generation model to be trained to obtain a pre-trained search term generation model.

[0116] In one example, the first training sample data can be obtained through the click log of online user short video searches. The click log is a record of the user's behavior of clicking on short videos after searching in the background of the short video platform. Based on the click log, the search terms associated with each short video and the number of clicks on the search terms can be obtained. Then, for each short video, based on its search term click data, a search term with a higher number of clicks (i.e., the first sample search term) can be selected, such as a click count greater than 100 (i.e., a preset click count threshold). Then, the video content text of the first sample video and the first sample search term of the first sample video can be used as training sample data for the pre-trained search term generation model.

[0117] For example, the input data and output data of the search term generation model to be trained can be as follows:

[0118]

[0119] The technical solution of this embodiment obtains the first training sample data, and then for the first sample video, extracts the conceptual entities associated with the video content of the first sample video and the entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the first sample video, and then trains the search term generation model to be trained based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term to obtain a pre-trained search term generation model. Based on the log data, high-quality search terms can be screened out according to the number of clicks to construct the first training sample data, without the need for manual labeling costs, thereby improving the pre-training efficiency of the search term generation model.

[0120] In an exemplary embodiment, before the step of inputting each candidate search word and video content text into a pre-trained search word evaluation model to obtain a relevance label corresponding to each candidate search word, it also includes: obtaining second training sample data; each second training sample data includes the video content text of the second sample video, the second sample search word of the second sample video and its corresponding relevance label, and the random search word and its corresponding relevance label, and different relevance labels correspond to different degrees of relevance; based on the video content text of the second sample video, the second sample search word and its corresponding relevance label, and the random search word and its corresponding relevance label, the search word evaluation model to be trained is trained to obtain a pre-trained search word evaluation model.

[0121] In one example, by screening out multiple videos that have both high-click search terms and low-click search terms (i.e., second sample videos), the labels of the high-click search terms can be set to "strong correlation" or "weak correlation" (i.e., the second sample search terms and their corresponding correlation labels), and for each video, a search term can be randomly sampled from the search term set, and its label can be set to "irrelevant" (i.e., a random search term and its corresponding correlation label), and then the video content text of the second sample video, the second sample search terms and their corresponding correlation labels, and the random search terms and their corresponding correlation labels can be used as training sample data for a pre-trained search term evaluation model.

[0122] For example, the relevance label configuration of the search term evaluation model to be trained can be as follows:

[0123]

[0124]

[0125] In another example, the search term evaluation model may adopt a pre-trained BERT, and the search term evaluation model to be trained (three-classification model) may be obtained by fine-tuning on the constructed training data.

[0126] The technical solution of this embodiment obtains the second training sample data, and then trains the search term evaluation model to be trained based on the video content text of the second sample video, the second sample search terms and their corresponding relevance labels, the random search terms and their corresponding relevance labels, to obtain a pre-trained search term evaluation model. The second training sample data can be constructed based on the difference in click distribution and random sampling, and relevance labeling can be automatically performed, thereby improving the pre-training efficiency of the search term evaluation model.

[0127] In order to enable those skilled in the art to better understand the above steps, the embodiments of the present disclosure are illustrated below by using an example, but it should be understood that the embodiments of the present disclosure are not limited thereto.

[0128] like Figure 3b As shown, in the process of video search term recommendation, three modules may be included: a search term generation module, a search term evaluation module, and a search term filtering module. In process S001, based on the search term generation module (i.e., a pre-trained search term generation model), a search term (i.e., at least one candidate search term) may be generated according to the short video text content (i.e., the video content text of the target video) in combination with the associated knowledge graph; in process S002, based on the search term evaluation module (i.e., a pre-trained search term evaluation model), the relevance between each search term and the short video obtained in process S001 may be scored to obtain a relevance label corresponding to each search term (i.e., a relevance label corresponding to each candidate search term); in process S003 and process S004, based on the search term filtering module, the search terms with strong relevance labels and weak relevance labels (i.e., target search terms) obtained in process S002 may be rewritten / filtered and risk control filtered according to preset business requirements and / or playback platform rules, thereby obtaining short video search terms (i.e., search terms to be recommended) to guide the user account to perform a search operation after accessing the short video.

[0129] Figure 4 is a flowchart of another method for recommending video search terms according to an exemplary embodiment. Figure 4 As shown, this method can be used to Figure 1 The server 120 includes the following steps.

[0130] In step S410, first training sample data is obtained; each first training sample data includes the video content text of the first sample video and the first sample search term of the first sample video, and the number of user account clicks corresponding to the first sample search term is greater than the preset click count threshold. In step S420, for the first sample video, conceptual entities associated with the video content of the first sample video and the entity relationships between the conceptual entities are extracted from the preset knowledge graph to generate the associated knowledge graph corresponding to the first sample video. In step S430, based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term, the search term generation model to be trained is trained to obtain the pre-trained search term generation model. In step S440, second training sample data is obtained; each second training sample data includes the video content text of the second sample video, the second sample search term of the second sample video and its corresponding relevance label, and the random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels. In step S450, the search term evaluation model to be trained is trained based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, to obtain the pre-trained search term evaluation model. In step S460, for the target video, conceptual entities associated with the video content of the target video and the entity relationships between the conceptual entities are extracted from the preset knowledge graph to generate an associated knowledge graph corresponding to the target video. In step S470, the video content text of the target video and the associated knowledge graph are input into the pre-trained search term generation model to obtain at least one candidate search term. In step S480, each of the candidate search terms and the video content text are input into the pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the degree of relevance between the candidate search term and the video content of the target video. In step S490, based on the candidate search terms whose relevance tags meet the preset conditions, a recommended search term for the target video is determined; the recommended search term is used to guide the user account to perform a search operation after accessing the target video. It should be noted that the specific definition of the above steps can refer to the specific definition of a video search term recommendation method above, which will not be repeated here.

[0131] It should be understood that although Figure 1 , Figure 4 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 , Figure 4 At least part of the steps may include multiple steps or multiple stages. These steps or stages are not necessarily performed at the same time, but can be performed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed in turn or alternately with other steps or at least part of the steps or stages in other steps.

[0132] It can be understood that the same / similar parts between the various embodiments of the above method in this specification can refer to each other, and each embodiment focuses on the differences from other embodiments. For related points, please refer to the description of other method embodiments.

[0133] Figure 5 FIG. 1 is a block diagram of a video search term recommendation device according to an exemplary embodiment. Figure 5 , the device comprises:

[0134] The associated knowledge graph generating unit 501 is configured to extract concept entities associated with the video content of the target video and entity relationships between the concept entities from a preset knowledge graph for the target video, and generate an associated knowledge graph corresponding to the target video;

[0135] A candidate search term obtaining unit 502 is configured to input the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term;

[0136] The relevance label obtaining unit 503 is configured to input each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the relevance between the candidate search term and the video content of the target video;

[0137] The recommended search term determination unit 504 is configured to execute the candidate search terms that meet preset conditions according to the relevance tags to determine the recommended search terms for the target video; the recommended search terms are used to guide the user account to perform a search operation after accessing the target video.

[0138] In one possible implementation, the search term determination unit 504 for recommendation is specifically configured to execute the candidate search term whose relevance tag meets preset conditions as the target search term; filter the abnormal search term in at least one of the target search terms to obtain the search term for recommendation; the abnormal search term is determined according to preset business requirements and / or playback platform rules.

[0139] In one possible implementation, the search term determination unit 504 to be recommended is further specifically configured to execute: if the target search term contains a specified term, then determine that the target search term is the abnormal search term; the specified term is determined based on preset business requirements and playback platform rules; delete the specified term in the abnormal search term to obtain a modified search term; use the modified search term and the non-abnormal search term as the search term to be recommended; the non-abnormal search term is the target search term that does not contain the specified term.

[0140] In one possible implementation, the to-be-recommended search term determination unit 504 is further specifically configured to execute: if the target search term contains a preset term, and / or the term representation concept of the target search term matches a preset abnormal representation concept, then the target search term is determined to be the abnormal search term; the preset term and the preset abnormal representation concept are both determined based on preset business requirements and playback platform rules; and the abnormal search term is deleted from at least one of the target search terms to obtain the to-be-recommended search term.

[0141] In one possible implementation, the pre-trained search term generation model has a pre-trained encoder and a pre-trained decoder, and the candidate search term obtaining unit 502 is specifically configured to execute inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph; the encoding result is input into the pre-trained decoder to obtain at least one candidate search term; the candidate search term is obtained by decoding a fusion encoding result between the first encoding result and the second encoding result through the pre-trained decoder.

[0142] In one possible implementation, the candidate search term obtaining unit 502 is further specifically configured to execute splicing of a preset search term control code with the video content text to obtain a spliced ​​text; the search term control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search term; the keyword control code is used to control whether the candidate search term contains a keyword corresponding to the keyword control code; the spliced ​​text and the associated knowledge graph are input into the pre-trained encoder to obtain an encoding result.

[0143] In a possible implementation, the video search term recommendation device further includes:

[0144] A first training sample data acquisition unit is specifically configured to execute acquisition of first training sample data; each first training sample data includes a video content text of a first sample video and a first sample search term of the first sample video, wherein the number of clicks of a user account corresponding to the first sample search term is greater than a preset click number threshold;

[0145] The associated knowledge graph generating unit corresponding to the first sample video is specifically configured to extract concept entities associated with the video content of the first sample video and entity relationships between the concept entities from a preset knowledge graph for the first sample video, and generate an associated knowledge graph corresponding to the first sample video;

[0146] The search term generation model training unit is specifically configured to execute training on the search term generation model to be trained based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term to obtain the pre-trained search term generation model.

[0147] In a possible implementation, the video search term recommendation device further includes:

[0148] The second training sample data acquisition unit is specifically configured to execute acquisition of second training sample data; each second training sample data includes a video content text of a second sample video, a second sample search term of the second sample video and its corresponding relevance label, a random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels;

[0149] The search term evaluation model training unit is specifically configured to execute training of the search term evaluation model to be trained based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, to obtain the pre-trained search term evaluation model.

[0150] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0151] Figure 6 is a block diagram of an electronic device 600 for a video search term recommendation method according to an exemplary embodiment. For example, the electronic device 600 may be a server. Figure 6, the electronic device 600 includes a processing component 620, which further includes one or more processors, and a memory resource represented by a memory 622 for storing instructions executable by the processing component 620, such as an application. The application stored in the memory 622 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 620 is configured to execute the instructions to perform the above method.

[0152] The electronic device 600 may further include a power supply component 624 configured to perform power management of the electronic device 600, a wired or wireless network interface 626 configured to connect the electronic device 600 to a network, and an input / output (I / O) interface 628. The electronic device 600 may operate based on an operating system stored in the memory 622, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or the like.

[0153] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 622 including instructions, and the above instructions can be executed by a processor of the electronic device 600 to perform the above method. The storage medium can be a computer-readable storage medium, for example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0154] In an exemplary embodiment, a computer program product is further provided. The computer program product includes instructions. The instructions can be executed by a processor of the electronic device 600 to complete the above method.

[0155] It should be noted that the above-mentioned devices, electronic devices, computer-readable storage media, computer program products, etc. may also include other implementation methods according to the description of the method embodiments. The specific implementation methods can refer to the description of the relevant method embodiments, which will not be described one by one here.

[0156] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0157] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A video search term recommendation method, characterized in that: The method comprises: For a target video, extract conceptual entities associated with the video content of the target video and entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the target video; Inputting the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term; Inputting each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the relevance between the candidate search term and the video content of the target video; According to the candidate search terms whose relevance tags meet preset conditions, a search term to be recommended for the target video is determined; the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video.

2. The method according to claim 1, characterized in that The step of determining the search term to be recommended for the target video according to the candidate search terms whose relevance tags satisfy the preset conditions comprises: The candidate search words whose relevance tags meet the preset conditions are used as target search words; Abnormal search words in at least one of the target search words are filtered to obtain the search words to be recommended; the abnormal search words are determined according to preset business requirements and / or playback platform rules.

3. The method according to claim 2, characterized in that The filtering of abnormal search terms in at least one of the target search terms to obtain the search terms to be recommended includes: If the target search term includes a designated word, the target search term is determined to be the abnormal search term; the designated word is determined based on preset business requirements and playback platform rules; Deleting the specified word in the abnormal search term to obtain a modified search term; The modified search term and the non-abnormal search term are used as the search term to be recommended; the non-abnormal search term is a target search term that does not contain the designated term.

4. The method according to claim 2, characterized in that: The filtering of abnormal search terms in at least one of the target search terms to obtain the search terms to be recommended includes: If the target search term contains a preset word, and / or the word representation concept of the target search term matches the preset abnormal representation concept, the target search term is determined to be the abnormal search term; the preset word and the preset abnormal representation concept are both determined based on preset business requirements and playback platform rules; The abnormal search term is deleted from at least one of the target search terms to obtain the search term to be recommended.

5. The method according to claim 1, characterized in that: The pre-trained search term generation model has a pre-trained encoder and a pre-trained decoder, and the video content text of the target video and the associated knowledge graph are input into the pre-trained search term generation model to obtain at least one candidate search term, including: Inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph; The encoding result is input into the pre-trained decoder to obtain at least one candidate search word; the candidate search word is obtained by decoding the fused encoding result between the first encoding result and the second encoding result through the pre-trained decoder.

6. The method according to claim 5, characterized in that The step of inputting the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result includes: Splicing a preset search word control code with the video content text to obtain a spliced ​​text; the search word control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search word; the keyword control code is used to control whether the candidate search word contains a keyword corresponding to the keyword control code; The concatenated text and the associated knowledge graph are input into the pre-trained encoder to obtain an encoding result.

7. The method according to claim 1, characterized in that Before the step of inputting the video content text of the target video and the associated knowledge graph into the pre-trained search term generation model to obtain at least one candidate search term, the method further includes: Acquire first training sample data; each first training sample data includes a video content text of a first sample video and a first sample search term of the first sample video, wherein the number of clicks of a user account corresponding to the first sample search term is greater than a preset click number threshold; For the first sample video, extract conceptual entities associated with the video content of the first sample video and entity relationships between the conceptual entities from a preset knowledge graph to generate an associated knowledge graph corresponding to the first sample video; Based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term, the search term generation model to be trained is trained to obtain the pre-trained search term generation model.

8. The method according to any one of claims 1 to 7, characterized in that: Before the step of inputting each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms, the method further includes: Acquire second training sample data; each second training sample data includes a video content text of a second sample video, a second sample search term of the second sample video and its corresponding relevance label, a random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels; Based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, the search term evaluation model to be trained is trained to obtain the pre-trained search term evaluation model.

9. A video search term recommendation device, characterized in that: include: The associated knowledge graph generating unit is configured to extract concept entities associated with the video content of the target video and entity relationships between the concept entities from a preset knowledge graph for the target video, and generate an associated knowledge graph corresponding to the target video; A candidate search term obtaining unit is configured to input the video content text of the target video and the associated knowledge graph into a pre-trained search term generation model to obtain at least one candidate search term; A relevance label obtaining unit is configured to input each of the candidate search terms and the video content text into a pre-trained search term evaluation model to obtain a relevance label corresponding to each of the candidate search terms; the relevance label is used to characterize the relevance between the candidate search term and the video content of the target video; The unit for determining the search term to be recommended is configured to execute the candidate search terms that meet preset conditions according to the relevance tags to determine the search term to be recommended for the target video; the search term to be recommended is used to guide the user account to perform a search operation after accessing the target video.

10. The device according to claim 9, characterized in that The to-be-recommended search word determination unit is specifically configured to execute the candidate search word whose relevance tag meets the preset condition as the target search word; Abnormal search words in at least one of the target search words are filtered to obtain the search words to be recommended; the abnormal search words are determined according to preset business requirements and / or playback platform rules.

11. The device according to claim 10, characterized in that The to-be-recommended search term determination unit is further specifically configured to determine that the target search term is the abnormal search term if the target search term contains a designated term; the designated term is determined based on preset business requirements and playback platform rules; Deleting the specified word in the abnormal search term to obtain a modified search term; The modified search term and the non-abnormal search term are used as the search term to be recommended; the non-abnormal search term is a target search term that does not contain the designated term.

12. The device according to claim 10, characterized in that The to-be-recommended search term determination unit is further specifically configured to determine that the target search term is the abnormal search term if the target search term contains a preset term and / or the term representation concept of the target search term matches a preset abnormal representation concept; The preset words and the preset abnormal representation concepts are determined based on preset business requirements and playback platform rules; The abnormal search term is deleted from at least one of the target search terms to obtain the search term to be recommended.

13. The device according to claim 9, characterized in that The pre-trained search term generation model has a pre-trained encoder and a pre-trained decoder, and the candidate search term obtaining unit is specifically configured to input the video content text and the associated knowledge graph into the pre-trained encoder to obtain an encoding result; the encoding result includes a first encoding result obtained by encoding the video content text and a second encoding result obtained by encoding the associated knowledge graph; The encoding result is input into the pre-trained decoder to obtain at least one candidate search word; the candidate search word is obtained by decoding the fused encoding result between the first encoding result and the second encoding result through the pre-trained decoder.

14. The device according to claim 13, characterized in that The candidate search term obtaining unit is further specifically configured to execute splicing of a preset search term control code with the video content text to obtain a spliced ​​text; the search term control code includes a word length control code and a keyword control code; the word length control code is used to control the word length of the candidate search term; the keyword control code is used to control whether the candidate search term contains a keyword corresponding to the keyword control code; the spliced ​​text and the associated knowledge graph are input into the pre-trained encoder to obtain an encoding result.

15. The device according to claim 9, characterized in that The video search term recommendation device further includes: A first training sample data acquisition unit is specifically configured to execute acquisition of first training sample data; each first training sample data includes a video content text of a first sample video and a first sample search term of the first sample video, wherein the number of clicks of a user account corresponding to the first sample search term is greater than a preset click number threshold; The associated knowledge graph generating unit corresponding to the first sample video is specifically configured to extract concept entities associated with the video content of the first sample video and entity relationships between the concept entities from a preset knowledge graph for the first sample video, and generate an associated knowledge graph corresponding to the first sample video; The search term generation model training unit is specifically configured to execute training on the search term generation model to be trained based on the video content text of the first sample video, the associated knowledge graph corresponding to the first sample video, and the first sample search term to obtain the pre-trained search term generation model.

16. The device according to any one of claims 9 to 15, characterized in that The video search term recommendation device further includes: The second training sample data acquisition unit is specifically configured to execute acquisition of second training sample data; each second training sample data includes a video content text of a second sample video, a second sample search term of the second sample video and its corresponding relevance label, a random search term and its corresponding relevance label, and different relevance labels correspond to different relevance levels; The search term evaluation model training unit is specifically configured to execute training of the search term evaluation model to be trained based on the video content text of the second sample video, the second sample search term and its corresponding relevance label, the random search term and its corresponding relevance label, to obtain the pre-trained search term evaluation model.

17. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the video search term recommendation method as described in any one of claims 1 to 8.

18. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the video search term recommendation method as described in any one of claims 1 to 8.

19. A computer program product, comprising instructions, characterized in that: When the instruction is executed by a processor of an electronic device, the electronic device is enabled to execute the video search term recommendation method as described in any one of claims 1 to 8.

Images