Video comment auditing method and related device

Abstract

The embodiment of the invention provides a video comment auditing method and a related device, and relates to the technical field of artificial intelligence. According to the method, the target keyword corresponding to the to-be-audited video is extracted by using the large language model, and the network comments are collected based on the target keyword, so that the correlation between the network comments and the to-be-audited video is ensured. The association relationship among network new words, keywords and emotional tendencies is stored by utilizing the knowledge graph to enrich semantic information among vocabularies, and multi-dimensional emotional tendency scoring is performed on network comments by utilizing a large language model and the knowledge graph, so that the emotional tendencies of network evaluation are evaluated more comprehensively, misjudgment and missed examination are reduced, and the evaluation efficiency is improved. Whether the network comments are shielded or not is judged according to the multi-dimensional emotional tendency score, so that the accuracy, flexibility and efficiency of video comment auditing are improved, and the content management level and user experience of a video platform are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a video comment review method and related devices. Background Art

[0002] With the rapid development of Internet technology, more and more users are using short videos for their daily life, study and work. If the comments on the short video platform contain illegal information, it will affect the user experience and the reputation of the short video platform. At present, manual or keyword filtering is usually used to review video comments, which has the problem of low review efficiency or poor accuracy. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a video comment review method and related devices, which can improve the efficiency and accuracy of video comment review.

[0004] In order to achieve the above purpose, the technical solution adopted by the embodiment of the present invention is as follows:

[0005] In a first aspect, the present invention provides a video comment review method, the method comprising:

[0006] Using a large language model, a target keyword corresponding to the video to be reviewed is obtained according to the video to be reviewed, and online comments are collected based on the target keyword;

[0007] Using the large language model, based on the video to be reviewed, the online comments and the knowledge graph, the sentiment tendency scores of each online comment in multiple dimensions are obtained; the knowledge graph is used to store the association relationship between new online words, keywords and sentiment tendencies; the new online words are words that are weakly associated with the keywords and have a high frequency of occurrence;

[0008] Calculate the review score of the online review according to each of the sentiment tendency scores and the corresponding preset weights;

[0009] When the review score exceeds a threshold, the online comment will be blocked.

[0010] In an optional implementation manner, the use of a large language model to obtain target keywords corresponding to the video to be reviewed based on the video to be reviewed includes:

[0011] Extracting frames from the video to be reviewed to obtain image frames;

[0012] Obtaining text information based on the audio in the video to be reviewed;

[0013] Inputting the image frame and the text information into the large language model to generate summary information corresponding to the video to be reviewed;

[0014] The summary information is input into the large language model, key information in the summary information is extracted, and target keywords corresponding to the video to be reviewed are obtained.

[0015] In an optional implementation, after the step of collecting online comments based on the target keyword, the method further includes:

[0016] Using the large language model, extracting target network new words from the network comments;

[0017] When the sentiment tendency annotated based on the target network new word is received, the knowledge graph is updated according to the target keyword, the target network new word and the received sentiment tendency.

[0018] In an optional implementation manner, the calculation of the review score of the online review according to each of the sentiment tendency scores and the corresponding preset weights includes:

[0019] For each of the online comments, calculating the product of each of the sentiment tendency scores and the corresponding preset weight;

[0020] The review score of the online review is obtained by summing up all the products.

[0021] In an optional embodiment, the method further comprises:

[0022] The emotional tendency with the largest product of the emotional tendency score and the preset weight is determined as the shielding reason.

[0023] In a second aspect, the present invention provides a video comment review device, the device comprising:

[0024] A processing module, used to obtain target keywords corresponding to the video to be reviewed based on the video to be reviewed by using a large language model, and collect online comments based on the target keywords;

[0025] The review module is used to use the large language model to obtain the sentiment tendency scores of each online comment in multiple dimensions according to the video to be reviewed, the online comments and the knowledge graph; the knowledge graph is used to store the association relationship between new online words, keywords and sentiment tendencies; the new online words are words that are weakly associated with the keywords and have a high frequency of occurrence;

[0026] The shielding module is used to calculate the review score of the online comment according to each of the sentiment tendency scores and the corresponding preset weights; when the review score exceeds a threshold, the online comment is shielded.

[0027] In an optional implementation, the processing module is used to extract frames from the video to be reviewed to obtain image frames;

[0028] Obtaining text information based on the audio in the video to be reviewed;

[0029] Inputting the image frame and the text information into the large language model to generate summary information corresponding to the video to be reviewed;

[0030] The summary information is input into the large language model, key information in the summary information is extracted, and target keywords corresponding to the video to be reviewed are obtained.

[0031] In a third aspect, the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the video comment review method described in any of the aforementioned implementations.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a video comment review method as described in any one of the aforementioned embodiments.

[0033] In a fifth aspect, the present invention provides a program product, which, when executed by a processor, implements a video comment review method as described in any one of the aforementioned embodiments.

[0034] Compared with the prior art, the video comment review method and related device provided by the embodiment of the present invention use a large language model to extract the target keywords corresponding to the video to be reviewed, and collect online comments based on the target keywords to ensure the relevance of the online comments to the video to be reviewed. The knowledge graph is used to store the association between new network words, keywords and sentiment tendencies to enrich the semantic information between words, and the large language model and knowledge graph are used to score the sentiment tendencies of network comments in multiple dimensions, so as to more comprehensively evaluate the sentiment tendencies of network evaluations, reduce misjudgments and missed reviews, so as to determine whether to block network comments based on the multi-dimensional sentiment tendency scores, thereby improving the accuracy, flexibility and efficiency of video comment review, and further improving the content management level and user experience of the video platform.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0037] Figure 1 A schematic diagram of a process flow of a video comment review method provided by an embodiment of the present invention is shown.

[0038] Figure 2 Another flow chart of the video comment review method provided by an embodiment of the present invention is shown.

[0039] Figure 3 Another flow chart of the video comment review method provided by an embodiment of the present invention is shown.

[0040] Figure 4 A block diagram of a video comment review device provided by an embodiment of the present invention is shown.

[0041] Figure 5 A block diagram of an electronic device provided by an embodiment of the present invention is shown.

[0042] Icons: 100 - electronic device; 110 - memory; 120 - processor; 130 - communication module; 400 - video comment review device; 401 - processing module; 402 - review module; 403 - shielding module. DETAILED DESCRIPTION

[0043] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0044] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0045] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0046] Comments on short video platforms may contain illegal information, which will affect user experience and platform reputation. The inventors have found that traditional comment review methods mainly include manual review and keyword filtering review, which have the problems of low review efficiency or low accuracy.

[0047] In order to improve the review efficiency, an intelligent review system can be used to review comments. However, since existing intelligent systems find it difficult to accurately understand popular online information, new online words, or the different meanings of words in a specific context, there is a problem of low review accuracy.

[0048] Based on this, the video comment review method and related device provided by the embodiment of the present invention use a large language model to extract the target keywords corresponding to the video to be reviewed, and collect online comments based on the target keywords to ensure the relevance of the online comments to the video to be reviewed. The knowledge graph is used to store the association between new network words, keywords and sentiment tendencies to enrich the semantic information between words, and the large language model and knowledge graph are used to score the sentiment tendencies of network comments in multiple dimensions, so as to more comprehensively evaluate the sentiment tendencies of network evaluations, reduce misjudgments and missed reviews, so as to determine whether to block network comments based on the multi-dimensional sentiment tendency scores, thereby improving the accuracy, flexibility and efficiency of video comment review, and further improving the content management level and user experience of the video platform.

[0049] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0050] Please refer to Figure 1 , Figure 1 A schematic diagram of a process flow of a video comment review method provided by an embodiment of the present invention is shown. The method comprises the following steps:

[0051] Step S10, using a large language model, obtains target keywords corresponding to the video to be reviewed based on the video to be reviewed, and collects online comments based on the target keywords.

[0052] In the embodiment of the present invention, a large language model is used to understand the content of the video to be reviewed, and keywords in the video to be reviewed, namely target keywords, are extracted. The target keywords represent the main content and spirit of the video to be reviewed.

[0053] The target keyword is used as the search term to collect online comments related to the target keyword from the preset network platform (such as social media, forums, news websites, etc.). Specifically, crawler technology or application programming interface (Application Programming Interface, referred to as: API) can be used to obtain online comments corresponding to the target keyword to ensure the diversity and comprehensiveness of online comments.

[0054] It should be noted that in order to further improve the accuracy and efficiency of the review, the collected online comments can be preprocessed (including data cleaning and standardization) to obtain preprocessed online comments, and then the preprocessed online comments can be reviewed to ensure that the preprocessed online comments are more relevant to the content of the video to be reviewed.

[0055] Among them, data cleaning includes deleting irrelevant information, noise and duplicate content. Irrelevant information is content that is not strongly related to the target keyword. Standardization processing includes format standardization such as unified punctuation and uppercase conversion.

[0056] Step S20, using a large language model, based on the video to be reviewed, online comments and knowledge graph, obtain the sentiment tendency score of each online comment in multiple dimensions; the knowledge graph is used to store the association between new online words, keywords and sentiment tendencies; new online words are words that are weakly associated with keywords and appear frequently.

[0057] In an embodiment of the present invention, the preprocessed online comments, the video to be reviewed and the knowledge graph are transmitted to the large language model, and the large language model is instructed to review the preprocessed online comments based on the video to be reviewed and the knowledge graph. The large language model scores the preprocessed online comments in multiple dimensions according to the sentiment tendency type to obtain multiple sentiment tendency scores.

[0058] Among them, the sentiment tendency type is pre-set, and the sentiment tendency type is the type of illegal information, including negative comments, political discussions, threats to social order, bad comments and indecent comments. The sentiment tendency score of each sentiment tendency type can be any score from 1 to 100, and the higher the score, the stronger the tendency.

[0059] Step S30, calculating the review score of the online review according to each sentiment tendency score and the corresponding preset weight.

[0060] Step S40, when the review score exceeds a threshold, the online comments are blocked.

[0061] In the embodiment of the present invention, the sentiment tendency type corresponds to the preset weight one by one, that is, each sentiment tendency type is set with a corresponding preset weight. According to the sentiment tendency score corresponding to each sentiment tendency type and the preset weight, the review score of the network comment is calculated. When the review score does not exceed the threshold, the network comment is reviewed and the network comment is displayed normally. When the review score exceeds the threshold, the network comment is not reviewed and the network comment is blocked and cannot be displayed normally.

[0062] In summary, the video comment review method provided by the embodiment of the present invention uses a large language model to extract target keywords corresponding to the video to be reviewed, and collects online comments based on the target keywords to ensure the relevance of the online comments to the video to be reviewed. The knowledge graph is used to store the association between new network words, keywords and sentiment tendencies to enrich the semantic information between words, and the large language model and knowledge graph are used to score the sentiment tendencies of network comments in multiple dimensions, so as to more comprehensively evaluate the sentiment tendencies of network evaluations, reduce misjudgments and missed reviews, and determine whether to block network comments based on the multi-dimensional sentiment tendency scores, thereby improving the accuracy, flexibility and efficiency of video comment review, and further improving the content management level and user experience of the video platform.

[0063] Optionally, the following is a possible implementation method for extracting keywords from the video to be reviewed. Figure 2 , Figure 1 The sub-steps of step S10 may include:

[0064] Step S101, extracting frames from the video to be reviewed to obtain image frames.

[0065] In an embodiment of the present invention, image processing technology, such as convolutional neural network, is used to extract image frames from the video to be reviewed. Assuming that the frame rate of the video to be reviewed is 29 frames per second, the human eye's visual change perception is basically around 200 milliseconds, that is, an image is extracted every 6 frames. The currently extracted image is compared with each pixel of the previous image. If the proportion of pixels with color value changes exceeds a preset ratio (for example, 30%), the current image is retained, and the retained image is the image frame.

[0066] Step S102, obtaining text information based on the audio in the video to be reviewed.

[0067] In the embodiment of the present invention, the automatic speech recognition (ASR) technology is used to convert the audio in the video to be reviewed into text to obtain corresponding text information.

[0068] Step S103: input the image frame and text information into the large language model to generate summary information corresponding to the video to be reviewed.

[0069] In an embodiment of the present invention, the image frame and text information are input into the large language model, and the large language model is instructed to extract corresponding summary information according to the image frame and text information, wherein the summary information is a general description of the content of the video to be reviewed.

[0070] Step S104: input the summary information into the large language model, extract key information from the summary information, and obtain target keywords corresponding to the video to be reviewed.

[0071] In the embodiment of the present invention, keywords are key points in the summary information, and keywords should cover the theme, main content, and key points that may cause discussion of the video. The summary information is input into the large language model, and the large language model is instructed to extract keywords representing the core content from the summary information to obtain the target keywords corresponding to the video to be reviewed.

[0072] As a possible implementation, for example, the video to be reviewed is a video discussing the cancellation of the concert of star A in city B. Then, the summary information output after the analysis of the large language model can be that the concert of star A in city B was cancelled, and the reason for the cancellation was that there was an unreported process during the performance, and the rainbow flag was displayed. The target keywords can be star A, city B, or rainbow flag.

[0073] Step S105, collecting online comments based on the target keywords.

[0074] It can be seen that the embodiment of the present invention uses a large language model to extract summary information of the video to be reviewed, and then extracts keywords corresponding to the video to be reviewed based on the summary information, which can narrow the content scope where the keywords are located, thereby improving the accuracy of keyword extraction.

[0075] Optionally, the following is a possible implementation method for building a knowledge graph. Figure 3 , Figure 1 After step S10, the following steps are also included:

[0076] Step S60, using the large language model to extract target network new words from network comments.

[0077] In the embodiment of the present invention, the network comments are input into the large language model, and the large language model is instructed to identify new network words from the network conclusions based on the existing vocabulary, and obtain the target new network words corresponding to the network comments. Among them, the new network words include network buzzwords, slang, and terms in specific fields.

[0078] Step S70, when the sentiment tendency based on the target network new word annotation is received, the knowledge graph is updated according to the target keyword, the target network new word and the received sentiment tendency.

[0079] In an embodiment of the present invention, a vocabulary library can be used to store the correspondence between new network words and sentiment tendencies. After extracting the target new network words corresponding to the network comments, the target new network words are saved in the vocabulary library. When the business personnel check the newly added target new network words in the vocabulary library, the newly added target new network words are marked, and the sentiment tendencies corresponding to the target new network words are sent to the electronic device.

[0080] When the electronic device receives the sentiment tendency corresponding to the target new network word, it saves the received sentiment tendency in the position corresponding to the target new network word in the vocabulary library, and builds an association relationship based on the target keyword, the target new network word and the received sentiment tendency, and updates it to the knowledge graph.

[0081] As a possible implementation method, for example, LangChain is used to build an updated knowledge graph, and the target keywords, target network new words, received sentiment tendencies and knowledge graph are input into LangChain, and LangChain is instructed to bind the target keywords, target network new words and received sentiment tendencies with an association relationship so as to update the knowledge graph.

[0082] The nodes in the knowledge graph represent words (keywords and new words on the Internet), and the edges represent the relationships between words, such as hyponymy, synonymy, etc. The knowledge graph can capture the meanings of words between words and in specific contexts, providing richer semantic information for the large language model. In the knowledge graph, triples are set for keywords, new words on the Internet, and sentiment tendencies, and each pair is represented as a triple of entity-relationship-entity. The knowledge graph can be imported into the JanusGraph graph database for indexing so that it can be used for indexing during the review of the large language model.

[0083] It should be noted that the knowledge graph is pre-created and is empty. When the big language model identifies new words from online comments, it will add new words, keywords and sentiment to the knowledge graph, gradually enriching the knowledge graph so that the big language model can review online comments in combination with the knowledge graph containing new words, which can effectively improve the accuracy and efficiency of video comment review. At the same time, the knowledge graph can also be used to train the big language model so that it can more accurately identify keywords and new words.

[0084] It is worth mentioning that two threads (i.e., workflows) can be created and executed in parallel. One workflow is used to update the vocabulary and knowledge graph to improve the timeliness of knowledge graph updates. The other workflow is used for the large language model to review online comments based on the timely updated knowledge graph to obtain more accurate sentiment tendency scores, thereby improving the accuracy and efficiency of video comment review. Figure 3 shown.

[0085] It can be seen that the embodiment of the present invention only requires manual participation in the annotation of new network words, and does not require manual participation in the review of network comments, which significantly reduces the demand for manual review and reduces operating costs.

[0086] Optionally, the following is a possible implementation method for building a knowledge graph. Figure 2 , Figure 1 The sub-steps of step S30 may include:

[0087] Step S301, for each online comment, calculate the product of each sentiment tendency score and the corresponding preset weight.

[0088] Step S302, summing up all products to obtain the review score of the online review.

[0089] In an embodiment of the present invention, the sentiment tendency score and preset weight corresponding to each network evaluation are obtained, each sentiment tendency score and the corresponding preset authority are multiplied, and all products corresponding to each network evaluation are added together to obtain the review score of each network rating.

[0090] Optionally, the following provides a possible implementation method for how to set the blocking reason. Figure 3 , the video comment review method also includes the following steps:

[0091] Step S50, determining the emotional tendency with the largest product of the emotional tendency score and the preset weight as the shielding reason.

[0092] In the embodiment of the present invention, after calculating the product of each sentiment tendency score and the corresponding preset weight, the sentiment tendency (ie, sentiment tendency type) with the largest product is used as the reason for blocking the network comment.

[0093] Based on the same inventive concept, the basic principles and technical effects of the video comment review device provided by the embodiment of the present invention are the same as those of the above-mentioned embodiment. For the sake of brief description, for parts not mentioned in this embodiment, reference may be made to the corresponding contents in the above-mentioned embodiment.

[0094] Please refer to Figure 4 , Figure 4 A schematic block diagram of a video comment review device 400 provided in an embodiment of the present invention. The video comment review device 400 includes a processing module 401 , a review module 402 and a shielding module 403 .

[0095] Processing module 401, for obtaining target keywords corresponding to the video to be reviewed based on the video to be reviewed by using a large language model, and collecting online comments based on the target keywords;

[0096] The review module 402 is used to use a large language model to obtain the sentiment tendency scores of each online comment in multiple dimensions based on the video to be reviewed, online comments and knowledge graphs; the knowledge graph is used to store the association between new online words, keywords and sentiment tendencies; new online words are words that are weakly associated with keywords and appear frequently;

[0097] The shielding module 403 is used to calculate the review score of the online comment according to each sentiment tendency score and the corresponding preset weight; when the review score exceeds a threshold, the online comment is shielded.

[0098] In summary, the video comment review device provided by the embodiment of the present invention uses a large language model to extract target keywords corresponding to the video to be reviewed, and collects online comments based on the target keywords to ensure the relevance of the online comments to the video to be reviewed. The knowledge graph is used to store the association between new network words, keywords and sentiment tendencies to enrich the semantic information between words, and the large language model and knowledge graph are used to score the sentiment tendencies of online comments in multiple dimensions, so as to more comprehensively evaluate the sentiment tendencies of network evaluations, reduce misjudgments and missed reviews, and determine whether to block online comments based on the multi-dimensional sentiment tendency scores, thereby improving the accuracy, flexibility and efficiency of video comment review, and further improving the content management level and user experience of the video platform.

[0099] Optionally, the processing module 401 is used to extract frames from the video to be reviewed to obtain image frames; obtain text information based on the audio in the video to be reviewed; input the image frames and text information into the large language model to generate summary information corresponding to the video to be reviewed; input the summary information into the large language model, extract key information from the summary information, and obtain target keywords corresponding to the video to be reviewed.

[0100] Optionally, the processing module 401 is also used to extract target network new words from network comments using a large language model; when a sentiment tendency based on the target network new words is received, the knowledge graph is updated according to the target keywords, the target network new words and the received sentiment tendency.

[0101] Optionally, the shielding module 403 is specifically used to calculate the product of each sentiment tendency score and the corresponding preset weight for each online comment; and sum up all the products to obtain the review score of the online comment.

[0102] Optionally, the shielding module 403 is further configured to determine the emotional tendency with the largest product of the emotional tendency score and the preset weight as the shielding reason.

[0103] Please refer to Figure 5, is a block diagram of an electronic device 100 provided in an embodiment of the present invention. The electronic device 100 includes a memory 110, a processor 120, and a communication module 130. The memory 110, the processor 120, and the communication module 130 are electrically connected to each other directly or indirectly to achieve data transmission or interaction. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0104] The memory 110 is used to store programs or data. The memory 110 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0105] The processor 120 is used to read / write data or programs stored in the memory 110 and execute corresponding functions. For example, when the computer program stored in the memory 110 is executed by the processor 120, the video comment review method disclosed in the above embodiments can be implemented.

[0106] The communication module 130 is used to establish a communication connection between the electronic device 100 and other communication terminals through a network, and to send and receive data through the network.

[0107] It should be understood that Figure 5 The structure shown is only a schematic diagram of the structure of the electronic device 100. The electronic device 100 may also include Figure 5 More or fewer components as shown, or with Figure 5 Different configurations are shown. Figure 5 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0108] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by the processor 120, the video comment review method disclosed in the above embodiments is implemented.

[0109] The embodiment of the present invention further provides a program product, which, when executed by the processor 120, implements the video comment review method disclosed in the above embodiments.

[0110] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0111] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0112] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0113] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A video comment review method, characterized in that: The method comprises: Using a large language model, a target keyword corresponding to the video to be reviewed is obtained according to the video to be reviewed, and online comments are collected based on the target keyword; Using the large language model, based on the video to be reviewed, the online comments and the knowledge graph, the sentiment tendency scores of each online comment in multiple dimensions are obtained; the knowledge graph is used to store the association relationship between new online words, keywords and sentiment tendencies; the new online words are words that are weakly associated with the keywords and have a high frequency of occurrence; Calculate the review score of the online review according to each of the sentiment tendency scores and the corresponding preset weights; When the review score exceeds a threshold, the online comment will be blocked.

2. The video review review method according to claim 1, characterized in that: The method of using the large language model to obtain target keywords corresponding to the video to be reviewed according to the video to be reviewed includes: Extracting frames from the video to be reviewed to obtain image frames; Obtaining text information based on the audio in the video to be reviewed; Inputting the image frame and the text information into the large language model to generate summary information corresponding to the video to be reviewed; The summary information is input into the large language model, key information in the summary information is extracted, and target keywords corresponding to the video to be reviewed are obtained.

3. The video review audit method according to claim 1, characterized in that: After the step of collecting online comments based on the target keyword, the method further includes: Using the large language model, extracting target network new words from the network comments; When the sentiment tendency annotated based on the target network new word is received, the knowledge graph is updated according to the target keyword, the target network new word and the received sentiment tendency.

4. The video comment review method according to claim 1, characterized in that: The step of calculating the review score of the online review according to each of the sentiment tendency scores and the corresponding preset weights includes: For each of the online comments, calculating the product of each of the sentiment tendency scores and the corresponding preset weight; The review score of the online review is obtained by summing up all the products.

5. The video comment review method according to claim 1, characterized in that: The method further comprises: The emotional tendency with the largest product of the emotional tendency score and the preset weight is determined as the shielding reason.

6. A video comment review device, characterized in that: The device comprises: A processing module, used to obtain target keywords corresponding to the video to be reviewed based on the video to be reviewed by using a large language model, and collect online comments based on the target keywords; The review module is used to use the large language model to obtain the sentiment tendency scores of each online comment in multiple dimensions according to the video to be reviewed, the online comments and the knowledge graph; the knowledge graph is used to store the association relationship between new online words, keywords and sentiment tendencies; the new online words are words that are weakly associated with the keywords and have a high frequency of occurrence; The shielding module is used to calculate the review score of the online comment according to each of the sentiment tendency scores and the corresponding preset weights; when the review score exceeds a threshold, the online comment is shielded.

7. The video comment review device according to claim 6, characterized in that: The processing module is used to extract frames from the video to be reviewed to obtain image frames; Obtaining text information based on the audio in the video to be reviewed; Inputting the image frame and the text information into the large language model to generate summary information corresponding to the video to be reviewed; The summary information is input into the large language model, key information in the summary information is extracted, and target keywords corresponding to the video to be reviewed are obtained.

8. An electronic device, characterized in that: It includes a processor and a memory, the memory stores a computer program that can be executed by the processor, and the processor can execute the computer program to implement the video comment review method described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the video comment review method as described in any one of claims 1-5 is implemented.

10. A program product, characterized in that When the program product is executed by a processor, the video comment review method as described in any one of claims 1-5 is implemented.

Images