Comment processing methods, devices, electronic devices, storage media, and computer program products

Abstract

This disclosure relates to a comment processing method, apparatus, electronic device, storage medium, and computer program product. The method includes: extracting core entity words and core trigger words from comments on a video to be analyzed; determining a first comment score for the comment based on the core entity words, core trigger words, and a first set of target words corresponding to a target video category to which the video to be analyzed belongs; determining a second comment score for the comment based on the core entity words, core trigger words, and a second set of target words corresponding to the target video category; determining a third comment score for the comment based on a set of target comments corresponding to the target video category; determining a target comment score based on the first, second, and third comment scores; and identifying the target comment from the comments to be analyzed based on the target comment score. Using this method, the accuracy of identifying the target comment can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and more particularly to a comment processing method, apparatus, electronic device, storage medium, and computer program product. Background Technology

[0002] With the development of internet technology, videos are emerging in endless streams online, and the comment sections of these videos often contain targeted comments, such as popular comments.

[0003] In related technologies, current methods for identifying target comments generally rely on the number of likes or replies to a video's comments to determine the target comment. However, this method only considers the comment's own information, resulting in a relatively singular calculation dimension and thus a low accuracy rate in identifying the target comment. Summary of the Invention

[0004] This disclosure provides a comment processing method, apparatus, electronic device, storage medium, and computer program product to at least solve the problem of low accuracy in identifying target comments in related technologies. The technical solution of this disclosure is as follows:

[0005] According to a first aspect of the present disclosure, a comment processing method is provided, comprising:

[0006] Extract the core entity words and core trigger words corresponding to the comments in the video to be analyzed;

[0007] Based on the core entity words, the core trigger words, and the first set of target words corresponding to the target video category to which the video to be analyzed belongs, a first comment score is determined for the comment to be analyzed; the first set of target words includes candidate core entity words and candidate core trigger words of the target candidate comment; the video category of the sample video corresponding to the target candidate comment is the same as the target video category;

[0008] Based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category, the second comment score of the comment to be analyzed is determined; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; the topic category of the target candidate topic matches the target video category;

[0009] A third comment score is determined for the comment to be analyzed based on the target comment set corresponding to the target video category; the target comment set includes the target candidate comments.

[0010] The target comment score for the comment to be analyzed is determined based on the first comment score, the second comment score, and the third comment score.

[0011] The target comment is determined from the comments to be analyzed based on the target comment score.

[0012] In an exemplary embodiment, extracting the core entity words and core trigger words corresponding to the comments to be analyzed from the comments to be analyzed in the video to be analyzed includes:

[0013] Obtain video information from the video to be analyzed;

[0014] Based on the video information, the comments to be analyzed in the video to be analyzed are adjusted to obtain the processed comments; the processed comments are used to represent the comments obtained after adjusting the pronouns in the comments to be analyzed to the actual content corresponding to the pronouns;

[0015] The processed comments are subjected to word extraction processing to obtain the core entity words and core trigger words corresponding to the processed comments. The core entity words and core trigger words corresponding to the processed comments are used as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0016] In an exemplary embodiment, determining the first comment score of the comment to be analyzed based on the core entity word, the core trigger word, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs includes:

[0017] From the candidate core entity words, determine the target core entity words that match the core entity words, and from the candidate core trigger words, determine the target core trigger words that match the core trigger words;

[0018] Query the matching relationship between the candidate core entity words and their importance scores to obtain the importance score corresponding to the target core entity word; and query the matching relationship between the candidate core trigger words and their importance scores to obtain the importance score corresponding to the target core trigger word.

[0019] The importance scores corresponding to the target core entity words and the target core trigger words are fused to obtain a first fused importance score, which is then used as the first comment score of the comment to be analyzed.

[0020] In an exemplary embodiment, before determining the first comment score of the comment to be analyzed based on the core entity words, the core trigger words, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs, the method further includes:

[0021] Obtain the target video category to which the video to be analyzed belongs;

[0022] From the first preset word set, a first word set corresponding to the target video category is determined, and the first word set corresponding to the target video category is taken as the first target word set; the first preset word set includes multiple first word sets corresponding to video categories; each candidate core entity word and candidate core trigger word in the first word set is matched with a corresponding importance score.

[0023] In an exemplary embodiment, the first preset word set is constructed in the following manner:

[0024] Construct a candidate comment set based on the comments of sample videos under each video category;

[0025] Extract candidate core entity words and candidate core trigger words from the candidate comments in the candidate comment set;

[0026] Based on the video category of the sample video corresponding to the candidate comment, the candidate core entity words and candidate core trigger words of the candidate comment are clustered to obtain a first word set corresponding to multiple video categories, and the importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category are determined.

[0027] The first preset word set is constructed based on the first word set corresponding to the multiple video categories.

[0028] In an exemplary embodiment, constructing a candidate comment set based on the comments of sample videos under each video category includes:

[0029] Obtain interactive data of comments for sample videos under each video category;

[0030] From the comments of sample videos under each video category, select the initial comments whose interaction data meets the first preset interaction conditions;

[0031] Obtain the comment time corresponding to the initial comment, and the reply time of the reply comment corresponding to the initial comment;

[0032] From the initial comments, filter out the intermediate comments whose comment time and reply time both meet the preset time conditions;

[0033] The intermediate comments are clustered to obtain multiple intermediate comment sets, and the interaction score of the comments in each intermediate comment set is determined based on the interaction data of the comments in each intermediate comment set.

[0034] From each of the intermediate comment sets, candidate comments whose interaction scores meet the second preset interaction conditions are selected, and the candidate comment set is constructed based on the candidate comments.

[0035] In an exemplary embodiment, before extracting the candidate core entity words and candidate core trigger words of the candidate comments in the candidate comment set, the method further includes:

[0036] If the candidate comment set includes a first candidate comment whose comment length is greater than a first preset length, extract the summary information corresponding to the first candidate comment, and use the summary information corresponding to the first candidate comment as the simplified candidate comment corresponding to the first candidate comment;

[0037] Based on the simplified candidate comments corresponding to the first candidate comments, the first candidate comments in the candidate comment set are updated to obtain the updated candidate comment set;

[0038] The extraction of candidate core entity words and candidate core trigger words from the candidate comment set includes:

[0039] Extract the candidate core entity words and candidate core trigger words from the candidate comments in the updated candidate comment set.

[0040] In one exemplary embodiment, after constructing a candidate comment set based on the comments of sample videos under each video category, the method further includes:

[0041] Based on the video category of the sample video corresponding to the candidate comments in the candidate comment set, the candidate comments are clustered to obtain comment sets corresponding to multiple video categories, and the importance score of the candidate comments in the comment set corresponding to each video category is determined; the importance score of each candidate comment is determined by the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment;

[0042] Construct a preset comment set based on the comment sets corresponding to the multiple video categories;

[0043] Before determining the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category, the method further includes:

[0044] From the preset comment set, determine the comment set corresponding to the target video category, and use the comment set corresponding to the target video category as the target comment set.

[0045] In an exemplary embodiment, determining the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category includes:

[0046] From the target candidate comments, identify candidate comments whose similarity to the comment to be analyzed is greater than a preset similarity, and take the candidate comments whose similarity to the comment to be analyzed is greater than the preset similarity as similar comments to the comment to be analyzed;

[0047] Query the matching relationship between the target candidate comments and the importance scores to obtain the importance scores of the similar comments;

[0048] The importance scores of the similar comments are fused to obtain a third fused importance score, which is then used as the third comment score of the comment to be analyzed.

[0049] In an exemplary embodiment, determining the second comment score of the comment to be analyzed based on the core entity word, the core trigger word, and a second target word set corresponding to the target video category includes:

[0050] From the candidate generalized entity words, a target generalized entity word that matches the core entity word is determined, and from the candidate generalized trigger words, a target generalized trigger word that matches the core trigger word is determined;

[0051] Query the matching relationship between the candidate generalized entity words and the importance scores to obtain the importance score corresponding to the target generalized entity word; and query the matching relationship between the candidate generalization trigger words and the importance scores to obtain the importance score corresponding to the target generalization trigger words.

[0052] The importance scores corresponding to the target generalized entity words and the target generalized trigger words are fused to obtain a second fused importance score, which is then used as the second comment score of the comment to be analyzed.

[0053] In an exemplary embodiment, before determining the second comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category, the method further includes:

[0054] Determine the topic category that matches the target video category;

[0055] From the second preset word set, a second word set corresponding to the topic category is determined, and the second word set corresponding to the topic category is used as the second target word set corresponding to the target video category; the second preset word set includes multiple second word sets corresponding to topic categories; each candidate generalized entity word and candidate generalized trigger word in the second word set is matched with a corresponding importance score.

[0056] In an exemplary embodiment, the second preset word set is constructed in the following manner:

[0057] Based on the topics in each sample list, construct a candidate topic set;

[0058] Extract candidate core entity words and candidate core trigger words from the candidate topics in the candidate topic set;

[0059] The candidate core entity words and candidate core trigger words of the candidate topics are respectively subjected to word generalization processing to obtain the candidate generalized entity words and candidate generalized trigger words of the candidate topics.

[0060] According to the topic category of the candidate topics, the candidate generalized entity words and candidate generalization trigger words of the candidate topics are clustered to obtain a second word set corresponding to multiple topic categories, and the importance scores of the candidate generalized entity words and candidate generalization trigger words in the second word set corresponding to each topic category are determined.

[0061] The second preset word set is constructed based on the second word set corresponding to the multiple topic categories.

[0062] In an exemplary embodiment, constructing a candidate topic set based on topics in each sample list includes:

[0063] Obtain the ranking weights for each sample list;

[0064] From the sample lists, select the target list whose list weight meets the preset list conditions;

[0065] The topics in the target list are clustered to obtain multiple topic sets, and the topic weights of the topics in each topic set are determined according to the list weights of the target list to which the topics in each topic set belong.

[0066] From each topic set, candidate topics whose topic weights satisfy preset topic conditions are selected, and a candidate topic set is constructed based on the candidate topics.

[0067] In one exemplary embodiment, before extracting the core entity words and core trigger words of the candidate topics in the candidate topic set, the method further includes:

[0068] If the candidate topic set includes a first candidate topic with a topic length greater than a second preset length, extract the summary information corresponding to the first candidate topic, and use the summary information corresponding to the first candidate topic as the simplified candidate topic corresponding to the first candidate topic.

[0069] Based on the simplified candidate topic corresponding to the first candidate topic, the first candidate topic in the candidate topic set is updated to obtain the updated candidate topic set.

[0070] The extraction of core entity words and core trigger words from the candidate topics set includes:

[0071] Extract the core entity words and core trigger words of the candidate topics in the updated candidate topic set.

[0072] In an exemplary embodiment, determining the target comment score of the comment to be analyzed based on the first comment score, the second comment score, and the third comment score includes:

[0073] Obtain the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, and the third preset weight corresponding to the third comment score;

[0074] According to the first preset weight, the second preset weight, and the third preset weight, the first comment score, the second comment score, and the third comment score are fused to obtain a fused comment score, and the fused comment score is used as the target comment score of the comment to be analyzed.

[0075] In an exemplary embodiment, before fusing the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain the fused comment score, the method further includes:

[0076] Obtain the interaction data of the comments to be analyzed;

[0077] Based on the interaction data of the comments to be analyzed, determine the fourth comment score of the comments to be analyzed, and obtain the fourth preset weight corresponding to the fourth comment score;

[0078] The step of fusing the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain a fused comment score includes:

[0079] The first comment score, the second comment score, the third comment score, and the fourth comment score are merged according to the first preset weight, the second preset weight, the third preset weight, and the fourth preset weight to obtain a merged comment score.

[0080] In one exemplary embodiment, determining the target comment from the comments to be analyzed based on the target comment score includes:

[0081] The comments to be analyzed are sorted in descending order of the target comment scores to obtain the sorted comments to be analyzed.

[0082] From the sorted comments to be analyzed, the top N comments are selected as target comments; where N is a positive integer.

[0083] According to a second aspect of the present disclosure, a comment processing apparatus is provided, comprising:

[0084] The word extraction unit is configured to extract the core entity words and core trigger words corresponding to the comments to be analyzed from the comments to be analyzed in the video to be analyzed;

[0085] The first determining unit is configured to determine a first comment score for the comment to be analyzed based on the core entity words, the core trigger words, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs; the first set of target words includes candidate core entity words and candidate core trigger words of the target candidate comment; the video category of the sample video corresponding to the target candidate comment is the same as the target video category;

[0086] The second determining unit is configured to determine a second comment score for the comment to be analyzed based on the core entity words, the core trigger words, and a second target word set corresponding to the target video category; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; the topic category of the target candidate topic matches the target video category;

[0087] The third determining unit is configured to determine a third comment score for the comment to be analyzed based on a target comment set corresponding to the target video category; the target comment set includes the target candidate comments;

[0088] The fourth determining unit is configured to determine the target comment score of the comment to be analyzed based on the first comment score, the second comment score, and the third comment score.

[0089] The comment determination unit is configured to determine the target comment from the comments to be analyzed based on the target comment score.

[0090] In an exemplary embodiment, the word extraction unit is further configured to: acquire video information of the video to be analyzed; adjust the comments to be analyzed in the video to be analyzed based on the video information to obtain processed comments; the processed comments represent comments obtained by adjusting pronouns in the comments to be analyzed to the actual content corresponding to the pronouns; perform word extraction processing on the processed comments to obtain core entity words and core trigger words corresponding to the processed comments, and use the core entity words and core trigger words corresponding to the processed comments as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0091] In an exemplary embodiment, the first determining unit is further configured to perform the following operations: determining a target core entity word matching the core entity word from the candidate core entity words, and determining a target core trigger word matching the core trigger word from the candidate core trigger words; querying the matching relationship between the candidate core entity words and importance scores to obtain the importance score corresponding to the target core entity word, and querying the matching relationship between the candidate core trigger words and importance scores to obtain the importance score corresponding to the target core trigger word; performing a fusion process on the importance score corresponding to the target core entity word and the importance score corresponding to the target core trigger word to obtain a first fused importance score, and using the first fused importance score as the first comment score of the comment to be analyzed.

[0092] In an exemplary embodiment, the apparatus further includes a set determination unit configured to perform the following: obtaining the target video category to which the video to be analyzed belongs; determining a first word set corresponding to the target video category from a first preset word set, and using the first word set corresponding to the target video category as the first target word set; the first preset word set includes multiple first word sets corresponding to video categories; and each candidate core entity word and candidate core trigger word in the first word set are matched with corresponding importance scores.

[0093] In an exemplary embodiment, the apparatus further includes a first construction unit configured to perform the following operations: constructing a candidate comment set based on comments of sample videos under each video category; extracting candidate core entity words and candidate core trigger words from the candidate comments in the candidate comment set; performing clustering processing on the candidate core entity words and candidate core trigger words of the candidate comments according to the video category of the sample videos corresponding to the candidate comments, obtaining a first word set corresponding to multiple video categories, and determining the importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category; and constructing a first preset word set based on the first word sets corresponding to the multiple video categories.

[0094] In an exemplary embodiment, the first construction unit is further configured to: acquire interactive data of comments from sample videos under each video category; filter initial comments from the comments of sample videos under each video category whose interactive data satisfies a first preset interaction condition; acquire the comment time corresponding to the initial comment and the reply time of the reply comment corresponding to the initial comment; filter intermediate comments from the initial comments whose comment time and reply time both satisfy a preset time condition; perform clustering processing on the intermediate comments to obtain multiple intermediate comment sets, and determine the interaction score of the comments in each intermediate comment set based on the interactive data of the comments in each intermediate comment set; filter candidate comments from each intermediate comment set whose interaction scores satisfy a second preset interaction condition, and construct the candidate comment set based on the candidate comments.

[0095] In an exemplary embodiment, the apparatus further includes a first simplification unit configured to, when the candidate comment set includes a first candidate comment whose comment length is greater than a first preset length, extract summary information corresponding to the first candidate comment and use the summary information corresponding to the first candidate comment as a simplified candidate comment corresponding to the first candidate comment; and update the first candidate comment in the candidate comment set according to the simplified candidate comment corresponding to the first candidate comment to obtain an updated candidate comment set.

[0096] The first building unit is further configured to extract candidate core entity words and candidate core trigger words from the candidate comments in the updated candidate comment set.

[0097] In one exemplary embodiment, the apparatus further includes a third construction unit configured to perform clustering processing on the candidate comments according to the video categories of the sample videos corresponding to the candidate comments in the candidate comment set, to obtain comment sets corresponding to multiple video categories, and to determine the importance score of the candidate comments in the comment sets corresponding to each video category; the importance score of each candidate comment is determined by the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment; and a preset comment set is constructed based on the comment sets corresponding to the multiple video categories.

[0098] The device further includes a set query unit, configured to determine, from the preset set of comments, a set of comments corresponding to the target video category, as the target set of comments.

[0099] In an exemplary embodiment, the third determining unit is further configured to perform the following operations: determining candidate comments from the target candidate comments whose similarity to the comment to be analyzed is greater than a preset similarity, and taking the candidate comments whose similarity to the comment to be analyzed is greater than the preset similarity as similar comments of the comment to be analyzed; querying the matching relationship between the target candidate comments and importance scores to obtain the importance scores of the similar comments; performing a fusion process on the importance scores of each similar comment to obtain a third fused importance score, and taking the third fused importance score as the third comment score of the comment to be analyzed.

[0100] In an exemplary embodiment, the second determining unit is further configured to perform the following operations: determining a target generalized entity word matching the core entity word from the candidate generalized entity words; determining a target generalized trigger word matching the core trigger word from the candidate generalized trigger words; querying the matching relationship between the candidate generalized entity words and importance scores to obtain the importance score corresponding to the target generalized entity word; querying the matching relationship between the candidate generalized trigger words and importance scores to obtain the importance score corresponding to the target generalized trigger word; performing a fusion process on the importance score corresponding to the target generalized entity word and the importance score corresponding to the target generalized trigger word to obtain a second fused importance score; and using the second fused importance score as the second comment score of the comment to be analyzed.

[0101] In one exemplary embodiment, the apparatus further includes a set filtering unit configured to perform the following: determining a topic category that matches the target video category; determining a second word set corresponding to the topic category from a second preset word set, and using the second word set corresponding to the topic category as a second target word set corresponding to the target video category; the second preset word set includes multiple second word sets corresponding to topic categories; and each candidate generalized entity word and candidate generalized trigger word in each second word set is matched with a corresponding importance score.

[0102] In an exemplary embodiment, the apparatus further includes a second construction unit configured to perform the following operations: constructing a candidate topic set based on topics in each sample list; extracting candidate core entity words and candidate core trigger words from the candidate topics in the candidate topic set; performing word generalization processing on the candidate core entity words and candidate core trigger words of the candidate topics respectively to obtain candidate generalized entity words and candidate generalized trigger words of the candidate topics; performing clustering processing on the candidate generalized entity words and candidate generalized trigger words of the candidate topics according to the topic categories of the candidate topics to obtain a second word set corresponding to multiple topic categories, and determining the importance scores of the candidate generalized entity words and candidate generalized trigger words in the second word set corresponding to each topic category; and constructing a second preset word set based on the second word sets corresponding to the multiple topic categories.

[0103] In an exemplary embodiment, the second construction unit is further configured to perform the following operations: obtaining the ranking weights of each sample ranking list; selecting target ranking lists from the sample ranking lists whose ranking weights satisfy preset ranking conditions; performing clustering processing on the topics in the target ranking lists to obtain multiple topic sets, and determining the topic weights corresponding to the topics in each topic set according to the ranking weights of the target ranking lists to which the topics in each topic set belong; selecting candidate topics from each topic set whose topic weights satisfy preset topic conditions, and constructing the candidate topic set based on the candidate topics.

[0104] In an exemplary embodiment, the apparatus further includes a second simplification unit, configured to perform the following steps when the candidate topic set includes a first candidate topic with a topic length greater than a second preset length: extract summary information corresponding to the first candidate topic, and use the summary information corresponding to the first candidate topic as a simplified candidate topic corresponding to the first candidate topic; update the first candidate topic in the candidate topic set according to the simplified candidate topic corresponding to the first candidate topic to obtain an updated candidate topic set.

[0105] The second building unit is also configured to extract the core entity words and core trigger words of the candidate topics in the updated candidate topic set.

[0106] In an exemplary embodiment, the fourth determining unit is further configured to perform the following steps: obtaining a first preset weight corresponding to the first comment score, a second preset weight corresponding to the second comment score, and a third preset weight corresponding to the third comment score; and performing a fusion process on the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain a fused comment score, which is used as the target comment score for the comment to be analyzed.

[0107] In an exemplary embodiment, the fourth determining unit is further configured to perform the following actions: acquiring interaction data of the comment to be analyzed; determining a fourth comment score of the comment to be analyzed based on the interaction data of the comment to be analyzed, and acquiring a fourth preset weight corresponding to the fourth comment score; and performing a fusion process on the first comment score, the second comment score, the third comment score, and the fourth comment score according to the first preset weight, the second preset weight, the third preset weight, and the fourth preset weight to obtain a fused comment score.

[0108] In one exemplary embodiment, the comment determination unit is further configured to sort the comments to be analyzed in descending order of the target comment scores to obtain sorted comments to be analyzed; and to select the top N comments from the sorted comments to be analyzed as target comments; where N is a positive integer.

[0109] According to a third aspect of the present disclosure, an electronic device is provided, comprising:

[0110] processor;

[0111] Memory used to store the processor's executable instructions;

[0112] The processor is configured to execute the instructions to implement the comment processing method as described in any embodiment of the first aspect.

[0113] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided that, when instructions in the computer-readable storage medium are executed by a processor of an electronic device, enables the electronic device to perform a comment processing method as described in any embodiment of the first aspect.

[0114] According to a fifth aspect of the present disclosure, a computer program product is provided, the computer program product including instructions that, when executed by a processor of an electronic device, enable the electronic device to perform a comment processing method as described in any embodiment of the first aspect.

[0115] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects:

[0116] By extracting core entity words and core trigger words from the comments of the video to be analyzed, and determining the first comment score based on the core entity words, core trigger words, and a first target word set corresponding to the target video category to which the video to be analyzed belongs; the first target word set includes candidate core entity words and candidate core trigger words of the target candidate comments; the video category of the sample video corresponding to the target candidate comment is the same as the target video category; then, determining the second comment score based on the core entity words, core trigger words, and a second target word set corresponding to the target video category; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; the topic category of the target candidate topic matches the target video category; next, determining the third comment score based on the target comment set corresponding to the target video category; the target comment set includes target candidate comments; finally, determining the target comment score based on the first, second, and third comment scores, and identifying the target comment from the comments to be analyzed based on the target comment score. In this way, when identifying target comments from the comments of the video to be analyzed, comprehensively considering the core entity words and core trigger words corresponding to the comment, the first set of target words corresponding to the target video category to which the video to be analyzed belongs, the second set of target words corresponding to the target video category, and the set of target comments corresponding to the target video category, it is beneficial to determine the comment score of the comment to be analyzed from multiple dimensions. Combining the comment scores from multiple dimensions to determine the target comment score of the comment to be analyzed helps to improve the accuracy of the calculation of the target comment score of the comment to be analyzed. This makes the target comments selected based on the target comment score more accurate, thereby improving the accuracy of target comment identification and avoiding the shortcomings of considering only the single dimension of the comment's own information, which leads to a low accuracy of target comment identification.

[0117] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0118] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0119] Figure 1 This is a flowchart illustrating a comment processing method according to an exemplary embodiment.

[0120] Figure 2 This is a flowchart illustrating the steps of constructing a first preset word set according to an exemplary embodiment.

[0121] Figure 3 This is a flowchart illustrating the steps of determining a third comment score for a comment to be analyzed, according to an exemplary embodiment.

[0122] Figure 4 This is a flowchart illustrating the steps of constructing a second preset word set according to an exemplary embodiment.

[0123] Figure 5 This is a flowchart illustrating another comment processing method according to an exemplary embodiment.

[0124] Figure 6 This is a flowchart illustrating a short video hot comment mining method based on a large model, according to an exemplary embodiment.

[0125] Figure 7 This is a block diagram illustrating a comment processing apparatus according to an exemplary embodiment.

[0126] Figure 8 This is a block diagram illustrating an electronic device according to an exemplary embodiment. Detailed Implementation

[0127] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0128] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0129] 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 used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0130] Figure 1 This is a flowchart illustrating a comment processing method according to an exemplary embodiment, such as... Figure 1As shown, this comment processing method is used in a server; it is understood that this method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. The terminal can be, but is not limited to, various personal computers, laptops, smartphones, and tablets, and the server can be a standalone server or a server cluster consisting of multiple servers. In this exemplary embodiment, the method includes the following steps:

[0131] In step S110, the core entity words and core trigger words corresponding to the comments to be analyzed are extracted from the comments to be analyzed in the video to be analyzed.

[0132] Among them, the video to be analyzed refers to the video from which the target comments (such as popular comments) need to be identified, such as videos and short videos on the Internet, videos and short videos on video applications, etc.

[0133] Among them, the comments to be analyzed refer to the comments on the video to be analyzed, specifically the comments in the comment section of the video to be analyzed, such as "So funny", "The male lead's acting is very good", "The scenery in this place is so beautiful", etc.

[0134] Among them, core entity words refer to entities or keywords that are of significant importance or represent the theme in the text, such as names of people, places, and organizations. The core entity words corresponding to the comment to be analyzed refer to the key entities in the comment, specifically the subject and object of the event in the core event triple of the comment, such as names of people, places, and organizations.

[0135] Among them, core trigger words refer to words in the text that trigger a certain event or relationship, such as "buy," "like," and "dislike." The core trigger words corresponding to the comments to be analyzed refer to the key trigger words in the comments to be analyzed, specifically the event predicates in the core event triples of the comments to be analyzed, such as "buy" and "like."

[0136] In natural language processing, a core event triple is a data structure used to represent events, typically consisting of three elements: the event subject, the event predicate, and the event object. For example, if the comment to be analyzed is "A dislikes durian," then the core entities are "A" and "durian," and the core trigger word is "dislikes." If the comment to be analyzed is "C also bought a cake," then the core entities are "C" and "cake," and the core trigger word is "buy."

[0137] For example, the server first obtains the comments to be analyzed from the video to be analyzed, and then uses a large model to perform natural language processing on the comments to extract core entity words and core trigger words from the comments. The large model can be a commonly used large language model, such as the licensed ChatGPT or GPT4.

[0138] In step S120, the first comment score of the comment to be analyzed is determined based on the core entity words, core trigger words, and the first target word set corresponding to the target video category to which the video to be analyzed belongs; the first target word set includes candidate core entity words and candidate core trigger words of the target candidate comment; the video category of the sample video corresponding to the target candidate comment is the same as the target video category.

[0139] The target video category refers to the video category to which the video to be analyzed belongs, such as sports, games, food, finance, etc.

[0140] Among them, the candidate core entity words of the target candidate comments refer to the key entities in the target candidate comments, specifically the event subject and event object in the core event triple of the target candidate comments, such as names of people, places, and organizations. The candidate core trigger words of the target candidate comments refer to the key trigger words in the target candidate comments, specifically the event predicates in the core event triples of the target candidate comments, such as "buy" and "like".

[0141] Each video category corresponds to a first word set, which includes candidate core entity words and candidate core trigger words for candidate comments that belong to the same video category as the corresponding sample video. It should be noted that candidate comments refer to pre-selected comments, such as... Figure 6 The comments shown are from the popular comment candidate pool; and the videos corresponding to the candidate comments are called sample videos.

[0142] The first target word set refers to the set of words whose corresponding video category is the same as the target video category. Specifically, it includes candidate core entity words and candidate core trigger words of candidate comments whose corresponding sample video category is the same as the target video category. Here, candidate comments whose corresponding sample video category is the same as the target video category are referred to as target candidate comments.

[0143] Among them, the first comment score of the comment to be analyzed refers to the first popularity score, which is specifically the matching score of the core entity word / core trigger word in the same category. It can be obtained by adding the importance score of the target core entity word (i.e. the target core entity word that matches the core entity word among the candidate core entity words in the first target word set) to the importance score of the target core trigger word (i.e. the target core trigger word that matches the core trigger word among the candidate core trigger words in the first target word set).

[0144] For example, the server obtains the target video category to which the video to be analyzed belongs, and determines the first set of words corresponding to the target video category to which the video to be analyzed belongs from the first set of words corresponding to multiple video categories, as the first target word set; and determines the first comment score of the comment to be analyzed based on the core entity words, core trigger words, and candidate core entity words and candidate core trigger words in the first target word set.

[0145] In step S130, the second comment score of the comment to be analyzed is determined based on the core entity words, core trigger words, and the second target word set corresponding to the target video category; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; the topic category of the target candidate topic matches the target video category.

[0146] Among them, the candidate generalized entity words of the target candidate topic refer to the generalized entity words of the candidate core entity words of the target candidate topic, specifically the associated entity words of the candidate core entity words of the target candidate topic, such as names of people, places, and organizations. The candidate generalized trigger words of the target candidate topic refer to the generalized trigger words of the candidate core trigger words of the target candidate topic, specifically the associated trigger words of the candidate core trigger words in the target candidate topic, such as "buy" and "like".

[0147] Among them, the candidate core entity words of the target candidate topic refer to the key entities in the target candidate topic, specifically the event subject and event object in the core event triple of the target candidate topic, such as names of people, places, and organizations. The candidate core trigger words of the target candidate topic refer to the key trigger words in the target candidate topic, specifically the event predicates in the core event triple of the target candidate topic, such as "buy" and "like".

[0148] The topic category refers to the category to which the topic belongs, such as sports, games, food, finance, etc.

[0149] Each topic category corresponds to a second word set, which includes candidate generalized entity words and candidate generalization trigger words for candidate topics that are in the same topic category. It should be noted that candidate topics refer to pre-selected topics, such as... Figure 6 The topics shown are from the popular topic candidate pool.

[0150] The second target word set refers to the set of words that match the corresponding topic category and the target video category. Specifically, it refers to the set of words that are the same as or similar to the corresponding topic category and the target video category. This set includes candidate generalized entity words and candidate generalization trigger words for candidate topics that match (e.g., are the same as or similar to) the corresponding topic category and the target video category. Here, candidate topics that match (e.g., are the same as or similar to) the corresponding topic category and the target video category are referred to as target candidate topics.

[0151] Among them, the second comment score to be analyzed refers to the second popularity score, which is specifically the matching score of generalized entity words / generalized trigger words in the same category. It can be obtained by adding the importance score of the target generalized entity word (i.e. the target generalized entity word that matches the core entity word among the candidate generalized entity words in the second target word set) to the importance score of the target generalized trigger word (i.e. the target generalized trigger word that matches the core trigger word among the candidate generalized trigger words in the second target word set).

[0152] For example, the server determines the second set of words corresponding to the topic category that matches the target video category to which the video to be analyzed belongs from the second set of words corresponding to multiple topic categories, and uses it as the second target word set; based on the core entity words, core trigger words, and candidate generalized entity words and candidate generalized trigger words in the second target word set, the server determines the second comment score of the comment to be analyzed.

[0153] In step S140, the third comment score of the comment to be analyzed is determined based on the target comment set corresponding to the target video category; the target comment set includes target candidate comments.

[0154] Each video category corresponds to a set of comments, and the set of comments for each video category includes candidate comments for the corresponding sample video that are in the same video category.

[0155] The target comment set refers to the set of comments whose corresponding video type is the same as the target video category. Specifically, it includes candidate comments whose corresponding sample video category is the same as the target video category, i.e., it specifically includes target candidate comments.

[0156] Among them, the third comment score of the comment to be analyzed refers to the third popularity score, which is specifically the matching score of popular simplified comments in the same category. It can be obtained by adding the importance scores of candidate comments in the target comment set that are similar to the comment to be analyzed (such as candidate comments in the target comment set whose similarity to the comment to be analyzed is greater than the preset similarity).

[0157] For example, the server determines the set of comments corresponding to the target video category to which the video to be analyzed belongs from the set of comments corresponding to multiple video categories, and uses this as the target comment set; based on the candidate comments in the target comment set that are similar to the comment to be analyzed, the server determines the third comment score of the comment to be analyzed.

[0158] In step S150, the target comment score for the comment to be analyzed is determined based on the first comment score, the second comment score, and the third comment score.

[0159] The target comment score refers to the final comment score of the comment to be analyzed, such as the final popularity score.

[0160] For example, the server performs a weighted summation of the first comment score, the second comment score, and the third comment score to obtain the target comment score for the comment to be analyzed.

[0161] In step S160, the target comments are determined from the comments to be analyzed based on the target comment score.

[0162] Here, "target comments" refers to the comments that are ultimately selected, specifically those whose target comment scores meet preset criteria, such as the top 10 comments in terms of target comment score or comments whose target comment score is greater than the preset score. In practical scenarios, target comments refer to popular comments.

[0163] For example, the server selects comments from the comments to be analyzed that meet the preset comment criteria and uses them as target comments.

[0164] In the above comment processing method, the core entity words and core trigger words corresponding to the comments to be analyzed are extracted from the comments to be analyzed of the video to be analyzed. A first comment score is determined based on the core entity words, core trigger words, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs. The first set of target words includes candidate core entity words and candidate core trigger words of the target candidate comments. The video category of the sample video corresponding to the target candidate comment is the same as the target video category. Then, a second comment score is determined based on the core entity words, core trigger words, and a second set of target words corresponding to the target video category. The second set of target words includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic, and the topic category of the target candidate topic matches the target video category. Next, a third comment score is determined based on the target comment set corresponding to the target video category. The target comment set includes target candidate comments. Finally, a target comment score is determined based on the first, second, and third comment scores, and the target comment is identified from the comments to be analyzed based on the target comment score. In this way, when identifying target comments from the comments of the video to be analyzed, comprehensively considering the core entity words and core trigger words corresponding to the comment, the first set of target words corresponding to the target video category to which the video to be analyzed belongs, the second set of target words corresponding to the target video category, and the set of target comments corresponding to the target video category, it is beneficial to determine the comment score of the comment to be analyzed from multiple dimensions. Combining the comment scores from multiple dimensions to determine the target comment score of the comment to be analyzed helps to improve the accuracy of the calculation of the target comment score of the comment to be analyzed. This makes the target comments selected based on the target comment score more accurate, thereby improving the accuracy of target comment identification and avoiding the shortcomings of considering only the single dimension of the comment's own information, which leads to a low accuracy of target comment identification.

[0165] In an exemplary embodiment, step S110, which extracts the core entity words and core trigger words corresponding to the comments to be analyzed from the comments to be analyzed in the video to be analyzed, specifically includes the following: obtaining video information of the video to be analyzed; adjusting the comments to be analyzed in the video to be analyzed based on the video information to obtain the processed comments; the processed comments are used to represent the comments obtained after adjusting the pronouns in the comments to be analyzed to the actual content corresponding to the pronouns; performing word extraction processing on the processed comments to obtain the core entity words and core trigger words corresponding to the processed comments, and using the core entity words and core trigger words corresponding to the processed comments as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0166] The video information to be analyzed refers to the contextual information of the video, specifically including the video title, video OCR (Optical Character Recognition), video ASR (Automatic Speech Recognition), and target video category.

[0167] Among these, the adjustment processing refers to the pronoun resolution processing. Pronoun resolution processing involves using the video information of the video to be analyzed to restore the pronouns (such as "this" and "that") in the comments to their corresponding actual content, facilitating the extraction of core entity words and core trigger words in the next step. Specifically, adjusting the pronouns in the comments to their corresponding actual content means restoring the pronouns in the comments to their actual content.

[0168] For example, the server identifies the video to be analyzed, obtains the video information of the video to be analyzed, and uses the video information to perform referential resolution processing on the comments to be analyzed in the video to be analyzed, so as to restore the pronouns in the comments to the actual content corresponding to the pronouns, thereby obtaining the processed comments; then, the server uses a large model to perform relevant word extraction processing on the processed comments to obtain the core entity words and core trigger words corresponding to the processed comments, which are used as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0169] The technical solution provided in this disclosure first adjusts the comments to be analyzed based on the video information of the video to be analyzed, obtaining processed comments. Then, word extraction is performed on the processed comments to obtain the core entity words and core trigger words corresponding to the processed comments, which are used as the core entity words and core trigger words corresponding to the comments to be analyzed. In this way, after adjusting the comments to be analyzed, the core entity words and core trigger words are extracted, which helps to improve the accuracy of the extraction of the core entity words and core trigger words corresponding to the comments to be analyzed.

[0170] In an exemplary embodiment, step S120, which determines the first comment score of the comment to be analyzed based on the core entity words, core trigger words, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs, specifically includes the following: determining the target core entity words that match the core entity words from the candidate core entity words, and determining the target core trigger words that match the core trigger words from the candidate core trigger words; querying the matching relationship between the candidate core entity words and importance scores to obtain the importance score corresponding to the target core entity words, and querying the matching relationship between the candidate core trigger words and importance scores to obtain the importance score corresponding to the target core trigger words; performing a fusion process on the importance score corresponding to the target core entity words and the importance score corresponding to the target core trigger words to obtain a first fused importance score, and using the first fused importance score as the first comment score of the comment to be analyzed.

[0171] Here, "target core entity word" refers to a candidate core entity word in the first target word set whose match degree with the core entity word is greater than a first preset match degree. "Target core trigger word" refers to a candidate core trigger word in the first target word set whose match degree with the core trigger word is greater than a second preset match degree.

[0172] In this context, each candidate core entity word in the first target word set is matched with a corresponding importance score, meaning there is a matching relationship between candidate core entity words and their importance scores. Therefore, the importance score corresponding to the target core entity word can be directly obtained. Similarly, each candidate core trigger word in the first target word set is matched with a corresponding importance score, meaning there is a matching relationship between candidate core trigger words and their importance scores. Therefore, the importance score corresponding to the target core trigger word can be directly obtained.

[0173] Specifically, the first fusion importance score is obtained by adding the importance score corresponding to the target core entity word and the importance score corresponding to the target core trigger word.

[0174] For example, the server extracts word vectors of core entity words and word vectors of each candidate core entity word in the first target word set using a deep learning model. Based on the word vectors of the core entity words and each candidate core entity word in the first target word set, the server calculates the matching degree between the core entity word and each candidate core entity word in the first target word set. Then, from the candidate core entity words in the first target word set, the server selects candidate core entity words whose matching degree with the core entity word is greater than a first preset matching degree, and these are designated as target core entity words. Similarly, the server extracts word vectors of core trigger words and word vectors of each candidate core trigger word in the first target word set using a deep learning model. Based on the word vectors of the core trigger words and each candidate core trigger word in the first target word set, the server calculates the matching degree between the core trigger word and each candidate core trigger word in the first target word set. Then, from the candidate core trigger words in the first target word set, the server selects candidate core trigger words whose matching degree with the core trigger word is greater than a second preset matching degree, and these are designated as target core trigger words. Next, the server queries the matching relationship between candidate core entity words and importance scores to obtain the importance score corresponding to the target core entity word, and queries the matching relationship between candidate core trigger words and importance scores to obtain the importance score corresponding to the target core trigger word. The importance score corresponding to the target core entity word is added to the importance score corresponding to the target core trigger word to obtain the first fused importance score, which is used as the first comment score of the comment to be analyzed.

[0175] The technical solution provided in this disclosure first identifies target core entity words matching the core entity words from candidate core entity words, and identifies target core trigger words matching the core trigger words from candidate core trigger words. Then, it obtains the importance scores corresponding to the target core entity words and the target core trigger words, and fuses these scores to obtain the first comment score for the comment to be analyzed. Thus, when calculating the first comment score, comprehensively considering both the target core entity words matching the core entity words and their corresponding importance scores in the first target word set, and the target core trigger words matching the core trigger words and their corresponding importance scores, helps improve the accuracy of determining the first comment score.

[0176] In an exemplary embodiment, before determining the first comment score of the comment to be analyzed based on the core entity words, core trigger words, and the first target word set corresponding to the target video category to which the video to be analyzed belongs, step S120 further includes a step of determining the first target word set corresponding to the target video category. Specifically, this includes: obtaining the target video category to which the video to be analyzed belongs; determining the first word set corresponding to the target video category from the first preset word set, and using the first word set corresponding to the target video category as the first target word set; the first preset word set includes multiple first word sets corresponding to video categories; and each candidate core entity word and candidate core trigger word in the first word set is matched with a corresponding importance score.

[0177] Here, the first preset word set refers to a preset word set that includes multiple video categories corresponding to the first word set, such as... Figure 6 The set of weighted popular core entity words / core trigger words shown.

[0178] For example, the server first identifies the video to be analyzed to obtain the target video category to which the video to be analyzed belongs. Then, from the first preset word set which includes a first word set corresponding to multiple video categories, the server selects the first word set corresponding to the target video category and uses it as the first target word set.

[0179] The technical solution provided in this embodiment first obtains the target video category to which the video to be analyzed belongs, and then determines the first target word set corresponding to the target video category from the first preset word set. In this way, by determining the first target word set corresponding to the target video category in advance, it is beneficial to comprehensively consider the core entity words and core trigger words of the comment to be analyzed, as well as the first target word set corresponding to the target video category, when determining the first comment score, so that the final determination of the first comment score is more accurate.

[0180] In one exemplary embodiment, such as Figure 2 As shown, the comment processing method provided in this disclosure also includes the step of constructing a first preset word set, which can be implemented through the following steps:

[0181] In step S210, a candidate comment set is constructed based on the comments of sample videos under each video category.

[0182] In step S220, candidate core entity words and candidate core trigger words are extracted from the candidate comments in the candidate comment set.

[0183] In step S230, according to the video category of the sample video corresponding to the candidate comment, the candidate core entity words and candidate core trigger words of the candidate comment are clustered to obtain the first word set corresponding to multiple video categories, and the importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category are determined.

[0184] In step S240, a first preset word set is constructed based on the first word set corresponding to multiple video categories.

[0185] The sample videos refer to videos within the site, such as videos within a preset time period (e.g., the past month), excluding the videos to be analyzed. The video category refers to the category to which the video belongs, such as sports, games, food, finance, etc.

[0186] Among them, the candidate comment set refers to the comment set composed of candidate comments initially selected from the comments of sample videos under each video category, such as... Figure 6 The pool of popular comment candidates is shown.

[0187] Each candidate core entity word and each candidate core trigger word in each first word set is matched with a corresponding importance score; moreover, the importance score is used to characterize the importance of each candidate core entity word and each candidate core trigger word in the first word set to which they belong, and can be calculated by the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm.

[0188] The clustering process, which involves grouping candidate core entity words and candidate core trigger words of candidate comments according to the video category of the sample videos corresponding to the candidate comments, aims to combine the candidate core entity words and candidate core trigger words of candidate comments whose corresponding sample videos belong to the same video category, and to remove duplicate candidate core entity words and duplicate candidate core trigger words, ultimately forming a first word set.

[0189] For example, the server first obtains comments from sample videos under each video category, then filters the comments from the sample videos under each video category multiple times to obtain candidate comments, and constructs a candidate comment set based on the candidate comments; next, the server uses a large model to perform relevant word extraction processing on the candidate comments in the candidate comment set to obtain the core entity words and core trigger words corresponding to the candidate comments, which serve as the candidate core entity words and candidate core trigger words of the candidate comments; then, the server performs clustering processing on the candidate core entity words and candidate core trigger words of the candidate comments according to the video category of the sample videos corresponding to the candidate comments, so as to combine the candidate core entity words and candidate core trigger words of the candidate comments whose corresponding sample videos belong to the same video category, thereby obtaining a first word set corresponding to multiple video categories, and calculates the importance score of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category using the TF-IDF algorithm; finally, based on the first word sets corresponding to multiple video categories, a first preset word set including the first word sets corresponding to multiple video categories is constructed.

[0190] The technical solution provided in this disclosure first constructs a candidate comment set based on the comments of sample videos under each video category, and extracts candidate core entity words and candidate core trigger words from the candidate comment sets. Then, according to the video category of the sample videos corresponding to the candidate comments, the candidate core entity words and candidate core trigger words of the candidate comments are clustered to obtain a first word set corresponding to multiple video categories. The importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category are determined. Finally, a first preset word set is constructed based on the first word sets corresponding to multiple video categories. In this way, by using the comments of sample videos under each video category to construct a first preset word set including the first word sets corresponding to multiple video categories, it is beneficial to comprehensively consider the first target word set in the first preset word set that corresponds to the target video category to which the video to be analyzed belongs when calculating the first comment score of the comment to be analyzed, thereby further improving the accuracy of determining the first comment score.

[0191] In an exemplary embodiment, step S210, which constructs a candidate comment set based on the comments of sample videos under each video category, specifically includes the following: obtaining interaction data of the comments of sample videos under each video category; selecting initial comments whose interaction data meets a first preset interaction condition from the comments of sample videos under each video category; obtaining the comment time corresponding to the initial comment and the reply time of the reply comment corresponding to the initial comment; selecting intermediate comments from the initial comments whose comment time and reply time both meet preset time conditions; performing clustering processing on the intermediate comments to obtain multiple intermediate comment sets, and determining the interaction score of the comments in each intermediate comment set based on the interaction data of the comments in each intermediate comment set; selecting candidate comments whose interaction scores meet a second preset interaction condition from each intermediate comment set, and constructing a candidate comment set based on the candidate comments.

[0192] The interaction data for comments refers to the number of likes, dislikes, and replies. The interaction data meets the first preset interaction condition when the number of likes, dislikes, and replies all meet the corresponding thresholds, such as the number of likes being greater than threshold A, the number of dislikes being less than threshold B, and the number of replies being greater than threshold C.

[0193] The initial comments refer to the comments in the sample videos under each video category whose corresponding interaction data meets the first preset interaction condition.

[0194] The comment time corresponding to the initial comment refers to the time when the initial comment was published; the reply time corresponding to the reply comment refers to the time when the reply comment was published.

[0195] The condition that both the comment time and the reply time meet the preset time conditions means that both the comment time and the reply time meet the preset time range, such as both the comment time and the reply time being within the preset time range.

[0196] Intermediate comments refer to comments in the initial comments whose comment time and the reply time of the corresponding reply comments both meet the preset time conditions.

[0197] Each set of intermediate comments includes similar intermediate comments, specifically those with semantically similar meanings. The comment's interaction score measures its importance and is calculated based on the number of likes, dislikes, and replies. For example, the comment's interaction score = number of likes + number of replies - number of dislikes.

[0198] In this context, "interaction score meeting the second preset interaction condition" means having the highest interaction score. Candidate comments refer to the intermediate comments in each set of intermediate comments whose corresponding interaction scores meet the second preset interaction condition.

[0199] For example, the server first obtains the interaction data of comments on sample videos under each video category, and then filters out comments from the comments of sample videos under each video category whose number of likes, dislikes, and replies all meet the corresponding thresholds, as initial comments whose interaction data meets the first preset interaction condition; then it obtains the comment time corresponding to the initial comment and the reply time of the reply comment corresponding to the initial comment, and filters out comments from the initial comments whose comment time and reply time both meet the preset time range, as intermediate comments whose comment time and reply time both meet the preset time condition; then it uses a deep learning model (such as SimCSE, Sentence-Ber) to... The sentence embedding model (such as t) extracts sentence vectors from intermediate comments and uses these vectors to cluster the intermediate comments, grouping semantically similar comments together to obtain multiple intermediate comment sets. Then, based on the number of likes, dislikes, and replies to comments in each intermediate comment set, an interaction score is calculated: interaction score = number of likes + number of replies - number of dislikes. Next, the comments with the highest interaction scores are selected from each intermediate comment set as candidate comments that meet the second preset interaction condition. Finally, a candidate comment set including the candidate comments is constructed based on these candidate comments.

[0200] The technical solution provided in this disclosure first selects initial comments from the comments of sample videos under each video category whose interaction data meets a first preset interaction condition. Then, from the initial comments, it selects intermediate comments whose comment time and reply time both meet preset time conditions. Next, it performs clustering processing on the intermediate comments to obtain multiple intermediate comment sets. From each intermediate comment set, it selects candidate comments whose interaction scores meet a second preset interaction condition. Finally, it constructs a candidate comment set based on the candidate comments. In this way, by using interaction data, comment time and reply time, and interaction scores to perform layer-by-layer filtering of the comments of sample videos under each video category, it is beneficial to improve the accuracy of candidate comment selection, thereby improving the accuracy of candidate comment set construction.

[0201] In an exemplary embodiment, step S220, before extracting the candidate core entity words and candidate core trigger words of the candidate comments in the candidate comment set, further includes a step of obtaining an updated candidate comment set. Specifically, this includes: if the candidate comment set includes a first candidate comment with a comment length greater than a first preset length, extracting the summary information corresponding to the first candidate comment, and using the summary information as a simplified candidate comment; updating the first candidate comments in the candidate comment set based on the simplified candidate comments, to obtain the updated candidate comment set. Therefore, step S220, extracting the candidate core entity words and candidate core trigger words of the candidate comments in the candidate comment set, specifically includes: extracting the candidate core entity words and candidate core trigger words of the candidate comments in the updated candidate comment set.

[0202] The first candidate comment refers to a candidate comment whose length is greater than the first preset length, specifically a candidate comment that is too long.

[0203] The summary information corresponding to the first candidate comment refers to the general information about the first candidate comment. Furthermore, extracting the summary information of the first candidate comment is done to simplify target candidate comments that are too long.

[0204] Here, the simplified candidate comment corresponding to the first candidate comment refers to the summary information of the first candidate comment. Updating the first candidate comment in the candidate comment set is to replace the first candidate comment with the corresponding simplified candidate comment.

[0205] The updated candidate comment set includes candidate comments that have not been updated, as well as simplified candidate comments corresponding to the first candidate comment.

[0206] For example, the server identifies the comment length of candidate comments in the candidate comment set. If it identifies a first candidate comment whose comment length is greater than a first preset length, it simplifies the first candidate comment using a large model to obtain the summary information corresponding to the first candidate comment and uses it as the simplified candidate comment corresponding to the first candidate comment. Then, it updates the first candidate comment in the candidate comment set with the corresponding simplified candidate comment to obtain the updated candidate comment set. Finally, it performs relevant word extraction processing on the candidate comments in the candidate comment set using a large model to obtain the candidate core entity words and candidate core trigger words of the candidate comments.

[0207] The technical solution provided in this disclosure first obtains simplified candidate comments corresponding to the first candidate comments in the candidate comment set whose comment length is greater than a first preset length. Then, it updates the first candidate comments in the candidate comment set with the corresponding simplified candidate comments, obtaining an updated candidate comment set. Finally, it extracts the candidate core entity words and candidate core trigger words from the candidate comments in the updated candidate comment set. In this way, by first simplifying the first candidate comments in the candidate comment set whose comment length is greater than the first preset length, and then extracting the candidate core entity words and candidate core trigger words from them, the extracted candidate core entity words and candidate core trigger words can be more accurate.

[0208] In an exemplary embodiment, step S210, after constructing a candidate comment set based on the comments of sample videos under each video category, further includes a step of constructing a preset comment set. Specifically, this includes: clustering the candidate comments according to the video category of the sample videos corresponding to the candidate comments in the candidate comment set, obtaining comment sets corresponding to multiple video categories, and determining the importance score of the candidate comments in each video category's comment set; the importance score of each candidate comment is determined by the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment; and constructing a preset comment set based on the comment sets corresponding to multiple video categories. Similarly, step S140, before determining the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category, further includes a step of determining the target comment set corresponding to the target video category. Specifically, this includes: determining the comment set corresponding to the target video category from the preset comment set, and using the comment set corresponding to the target video category as the target comment set.

[0209] The clustering process, which involves grouping candidate comments according to the video category of the sample videos corresponding to the candidate comments in the candidate comment set, aims to combine candidate comments whose corresponding sample videos belong to the same video category, remove duplicate candidate comments, and ultimately form a comment set.

[0210] The importance score of a candidate comment is used to measure the importance of a candidate comment. Specifically, it is determined by the importance scores of the candidate core entity words and candidate core trigger words included in the candidate comment. For example, it can be obtained by adding the importance scores of the candidate core entity words and candidate core trigger words included in the candidate comment.

[0211] Among them, the preset comment set refers to a preset set that includes comment sets corresponding to multiple video categories, such as... Figure 6 The set of weighted, popular, simplified comments shown.

[0212] For example, after constructing the candidate comment set, the server clusters the candidate comments according to the video category of the sample videos corresponding to the candidate comments in the candidate comment set, so as to group the candidate comments whose corresponding sample videos belong to the same video category together, thereby obtaining comment sets corresponding to multiple video categories. Then, the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment in each comment set corresponding to the video category are added together to obtain the importance score of each candidate comment. Finally, based on the comment sets corresponding to multiple video categories, a preset comment set including the comment sets corresponding to multiple video categories is constructed. Further, the server selects the comment set corresponding to the target video category from the preset comment set including the comment sets corresponding to multiple video categories, and uses it as the target comment set.

[0213] The technical solution provided in this disclosure first clusters the candidate comments according to the video categories of the sample videos corresponding to the candidate comments in the candidate comment set, obtaining comment sets corresponding to multiple video categories, and determining the importance score of the candidate comments in the comment sets corresponding to each video category. Then, based on the comment sets corresponding to multiple video categories, a preset comment set is constructed. Finally, the comment set corresponding to the target video category is determined from the preset comment set as the target comment set. In this way, by determining the target comment set corresponding to the target video category in advance, it is beneficial to comprehensively consider the target comment set corresponding to the target video category when determining the third comment score, making the final determination of the third comment score more accurate.

[0214] In one exemplary embodiment, such as Figure 3 As shown, step S140 above, which determines the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category, can be implemented through the following steps:

[0215] In step S310, candidate comments with a similarity greater than a preset similarity to the comment to be analyzed are identified from the target candidate comments, and these candidate comments with a similarity greater than the preset similarity to the comment to be analyzed are taken as similar comments to the comment to be analyzed.

[0216] In step S320, the matching relationship between the target candidate comments and the importance scores is queried to obtain the importance scores of similar comments.

[0217] In step S330, the importance scores of each similar comment are fused to obtain a third fused importance score, and the third fused importance score is used as the third comment score of the comment to be analyzed.

[0218] The preset similarity score can be 0.8, 0.9, etc., and can be adjusted according to the actual situation. The similarity between the comment to be analyzed and the target candidate comments in the target comment set is used to measure the degree of similarity between the comment to be analyzed and the target candidate comments in the target comment set.

[0219] Among them, similar comments refer to candidate comments in the target comment set whose similarity to the comment to be analyzed is greater than the preset similarity.

[0220] Each target candidate comment in the target comment set is matched with a corresponding importance score, meaning there is a matching relationship between the target candidate comment and the importance score. Therefore, the importance score corresponding to similar comments can be obtained directly.

[0221] The third fusion importance score is obtained by adding the importance scores of similar comments together.

[0222] For example, the server extracts the sentence vector of the comment to be analyzed and the sentence vector of each target candidate comment in the target comment set using a deep learning model. Based on the sentence vectors of the comment to be analyzed and each target candidate comment in the target comment set, the server calculates the similarity between the comment to be analyzed and each target candidate comment in the target comment set. Then, from the target candidate comments in the target comment set, candidate comments with a similarity greater than a preset similarity are selected as similar comments to the comment to be analyzed. Next, the server queries the matching relationship between the target candidate comments and their importance scores to obtain the importance scores of the similar comments to the comment to be analyzed. The server then sums the importance scores of the similar comments to obtain a third fused importance score, which is used as the third comment score of the comment to be analyzed.

[0223] The technical solution provided in this disclosure first identifies similar comments to the comment to be analyzed from the target candidate comments in the target comment set corresponding to the target video category. Then, it obtains the importance scores of these similar comments. Finally, it performs a fusion process on the importance scores of the similar comments to obtain a third fused importance score, which serves as the third comment score for the comment to be analyzed. In this way, when calculating the third comment score, the comprehensive consideration of similar comments in the target comment set and their corresponding importance scores helps improve the accuracy of determining the third comment score.

[0224] In an exemplary embodiment, step S130, which determines the second comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category, specifically includes the following: determining the target generalized entity words that match the core entity words from the candidate generalized entity words, and determining the target generalized trigger words that match the core trigger words from the candidate generalized trigger words; querying the matching relationship between the candidate generalized entity words and the importance scores to obtain the importance score corresponding to the target generalized entity words, and querying the matching relationship between the candidate generalized trigger words and the importance scores to obtain the importance score corresponding to the target generalized trigger words; performing a fusion process on the importance scores corresponding to the target generalized entity words and the importance scores corresponding to the target generalized trigger words to obtain a second fused importance score, and using the second fused importance score as the second comment score of the comment to be analyzed.

[0225] Among them, the target generalized entity word refers to the candidate generalized entity word in the second target word set whose matching degree with the core entity word is greater than the third preset matching degree. The target generalized trigger word refers to the candidate generalized trigger word in the second target word set whose matching degree with the core trigger word is greater than the fourth preset matching degree.

[0226] In the second target word set, each candidate generalized entity word is matched with a corresponding importance score, meaning there is a matching relationship between candidate generalized entity words and importance scores. Therefore, the importance score corresponding to the target generalized entity word can be directly obtained. Similarly, each candidate generalization trigger word in the second target word set is matched with a corresponding importance score, meaning there is a matching relationship between candidate generalization trigger words and importance scores. Therefore, the importance score corresponding to the target generalization trigger word can be directly obtained.

[0227] The second fusion importance score is obtained by adding the importance score corresponding to the target generalized entity word and the importance score corresponding to the target generalization trigger word.

[0228] For example, the server extracts the word vectors of the core entity word and the word vectors of each candidate generalized entity word in the second target word set using a deep learning model. Based on the word vectors of the core entity word and each candidate generalized entity word in the second target word set, the server calculates the matching degree between the core entity word and each candidate generalized entity word in the second target word set. Then, from the candidate generalized entity words in the second target word set, the server selects candidate generalized entity words whose matching degree with the core entity word is greater than a third preset matching degree, and uses these as target generalized entity words. Similarly, the server extracts the word vectors of the core trigger word and the word vectors of each candidate generalized trigger word in the second target word set using a deep learning model. Based on the word vectors of the core trigger word and each candidate generalized trigger word in the second target word set, the server calculates the matching degree between the core trigger word and each candidate generalized trigger word in the second target word set. Then, from the candidate generalized trigger words in the second target word set, the server selects candidate generalized trigger words whose matching degree with the core trigger word is greater than a fourth preset matching degree, and uses these as target generalized trigger words. Next, the server queries the matching relationship between candidate generalized entity words and importance scores to obtain the importance score corresponding to the target generalized entity word, and queries the matching relationship between candidate generalized trigger words and importance scores to obtain the importance score corresponding to the target generalized trigger word. The importance score corresponding to the target generalized entity word is added to the importance score corresponding to the target generalized trigger word to obtain the second fusion importance score, which is used as the second comment score of the comment to be analyzed.

[0229] The technical solution provided in this disclosure first identifies target generalized entity words matching the core entity word from candidate generalized entity words, and identifies target generalized trigger words matching the core trigger word from candidate generalized trigger words. Then, it obtains the importance scores corresponding to the target generalized entity words and the target generalized trigger words, and fuses these scores to obtain a second comment score for the comment to be analyzed. Thus, when calculating the second comment score, comprehensively considering both the target generalized entity words matching the core entity word and their corresponding importance scores in the second target word set, as well as the target generalized trigger words matching the core trigger word and their corresponding importance scores, helps improve the accuracy of determining the second comment score.

[0230] In an exemplary embodiment, before determining the second comment score of the comment to be analyzed based on the core entity words, core trigger words, and the second target word set corresponding to the target video category, step S130 further includes a step of determining the second target word set corresponding to the topic category. Specifically, this includes: determining the topic category that matches the target video category; determining the second word set corresponding to the topic category from the second preset word set, and using the second word set corresponding to the topic category as the second target word set corresponding to the target video category; the second preset word set includes multiple second word sets corresponding to topic categories; and each candidate generalized entity word and candidate generalized trigger word in each second word set is matched with a corresponding importance score.

[0231] Among them, the topic category that matches the target video category refers to the topic category that is the same as or similar to the target video category.

[0232] The second preset word set refers to a preset word set that includes multiple topic categories corresponding to second word sets, such as... Figure 6 The set of weighted popular generalized entity words / generalization trigger words shown.

[0233] For example, the server first determines the topic category that matches the target video category from multiple topic categories, and then filters out the second word set corresponding to the determined topic category from the second preset word set that includes the second word set corresponding to multiple topic categories, and uses it as the second target word set corresponding to the target video category.

[0234] The technical solution provided in this disclosure first determines the topic category that matches the target video category, and then determines the second word set corresponding to the topic category from the second preset word set, which serves as the second target word set corresponding to the target video category. In this way, by determining the second target word set corresponding to the target video category in advance, it is beneficial to comprehensively consider the core entity words and core trigger words of the comment to be analyzed, as well as the second target word set corresponding to the target video category, when determining the second comment score, so that the final determination of the second comment score is more accurate.

[0235] In one exemplary embodiment, such as Figure 4 As shown, the comment processing method provided in this application also includes the step of constructing a second preset word set, which can be implemented through the following steps:

[0236] In step S410, a candidate topic set is constructed based on the topics in each sample list.

[0237] In step S420, candidate core entity words and candidate core trigger words of candidate topics in the candidate topic set are extracted.

[0238] In step S430, word generalization processing is performed on the candidate core entity words and candidate core trigger words of the candidate topics to obtain the candidate generalized entity words and candidate generalized trigger words of the candidate topics.

[0239] In step S440, the candidate generalized entity words and candidate generalization trigger words of the candidate topics are clustered according to the topic category of the candidate topics to obtain the second word set corresponding to multiple topic categories, and the importance scores of the candidate generalized entity words and candidate generalization trigger words in the second word set corresponding to each topic category are determined.

[0240] In step S450, a second preset word set is constructed based on the second word set corresponding to multiple topic categories.

[0241] Among them, each sample list refers to the internal list and the external list, such as the list of each platform within a preset time period (such as the past week).

[0242] Each sample list includes multiple topics. Topics refer to trending information, such as "Xiao Bing breaks a record" or "A certain loan will have a new interest rate starting next year."

[0243] Among them, the candidate topic set refers to the topic set composed of candidate topics initially selected from the topics in each sample list, such as... Figure 6 The pool of popular topics shown.

[0244] Among them, word generalization processing refers to word association processing, which is used to determine the corresponding associated words. For example, it is used to determine the associated entity words of the candidate core entity words of the candidate topic and the associated entity words of the candidate core trigger words of the candidate topic. Specifically, it can be implemented through a large model.

[0245] In this context, each candidate generalized entity word and each candidate generalization trigger word in each second word set is matched with a corresponding importance score; moreover, the importance score is used to characterize the importance of each candidate generalized entity word and each candidate generalization trigger word in its respective second word set, which can be specifically calculated using the TF-IDF algorithm.

[0246] The clustering process, which involves grouping candidate generalized entity words and candidate generalized trigger words according to the topic category of the candidate topics, aims to combine candidate generalized entity words and candidate generalized trigger words of candidate topics belonging to the same topic category, and to remove duplicate candidate generalized entity words and duplicate candidate generalized trigger words, ultimately forming a second word set.

[0247] For example, the server first obtains topics from each sample list, then filters the topics from each sample list multiple times to obtain candidate topics, and constructs a candidate topic set based on the candidate topics; then, the server uses a large model to perform relevant word extraction processing on the candidate topics in the candidate topic set to obtain the core entity words and core trigger words corresponding to the candidate topics, which are used as candidate core entity words and candidate core trigger words of the candidate topics; then, the server uses a large model to perform word generalization processing on the candidate core entity words and candidate core trigger words of the candidate topics to obtain generalized entity words of the candidate core entity words and generalized trigger words of the candidate core trigger words, which are used as candidate generalized entity words and candidate generalized trigger words of the candidate topics. Next, the server performs clustering processing on the candidate generalized entity words and candidate generalized trigger words of the candidate topics according to their topic categories, so as to combine the candidate generalized entity words and candidate generalized trigger words of candidate topics that belong to the same topic category together, thereby obtaining a second word set corresponding to multiple topic categories. The importance score of the candidate generalized entity words and candidate generalized trigger words in the second word set corresponding to each topic category is calculated by using the TF-IDF algorithm. Finally, based on the second word sets corresponding to multiple topic categories, a second preset word set including the second word sets corresponding to multiple topic categories is constructed.

[0248] The technical solution provided in this disclosure first constructs a candidate topic set based on the topics in each sample list, and extracts candidate core entity words and candidate core trigger words from the candidate topic set. These are then subjected to word generalization processing to obtain candidate generalized entity words and candidate generalized trigger words. Next, according to the topic category of the candidate topics, the candidate generalized entity words and candidate generalized trigger words are clustered to obtain a second word set corresponding to multiple topic categories. The importance scores of the candidate generalized entity words and candidate generalized trigger words in the second word set corresponding to each topic category are determined. Finally, a second preset word set is constructed based on the second word sets corresponding to multiple topic categories. This method of constructing a second preset word set, including second word sets corresponding to multiple topic categories, using the topics in each sample list, facilitates the subsequent calculation of the second comment score by comprehensively considering the second target word set in the second preset word set that corresponds to the target video category to which the video to be analyzed belongs, further improving the accuracy of determining the second comment score.

[0249] In an exemplary embodiment, step S410, which constructs a candidate topic set based on the topics in each sample list, specifically includes the following: obtaining the list weights of each sample list; selecting target lists from each sample list whose list weights meet preset list conditions; performing clustering processing on the topics in the target list to obtain multiple topic sets, and determining the topic weights corresponding to the topics in each topic set based on the list weights of the target list to which the topics in each topic set belong; selecting candidate topics from each topic set whose topic weights meet preset topic conditions, and constructing a candidate topic set based on the candidate topics.

[0250] The ranking weights of each sample list are used to measure the importance of each sample list and are preset, but can be adjusted according to the actual situation.

[0251] In this context, "list weight meets preset list conditions" means that the list weight meets the corresponding weight threshold, such as the list weight being greater than the weight threshold. Alternatively, "list weight meets preset list conditions" could also mean that the list weight ranks in the top 90%.

[0252] Among them, the target list refers to the sample list whose corresponding list weight meets the preset list conditions.

[0253] Each topic set includes similar topics, specifically those with semantic similarities. The topic weight is used to measure the topic's importance, specifically calculated based on the ranking weight of the target ranking list to which the topic belongs. For example, the topic weight equals the ranking weight of the target ranking list to which the topic belongs. If a topic appears on multiple target ranking lists simultaneously, its topic weight is the sum of the ranking weights of each target ranking list to which the topic belongs.

[0254] In this context, a topic with the highest weight meets the preset topic criteria. Candidate topics refer to topics in each topic set whose corresponding topic weights meet the preset topic criteria.

[0255] For example, the server first obtains the ranking weights of each sample ranking list, and then selects the sample ranking lists whose ranking weights meet the preset ranking conditions as the target ranking lists. Next, it extracts the sentence vectors of the topics in the target ranking lists using a deep learning model, and uses the sentence vectors of the topics in the target ranking lists to perform clustering processing on the topics in the target ranking lists, so as to aggregate topics with similar semantics together, thereby obtaining multiple topic sets. Then, it adds up the ranking weights of the target ranking lists to which each topic in each topic set belongs, and obtains the topic weights corresponding to each topic in each topic set. Then, it selects the topic with the highest topic weight from each topic set as the candidate topics whose topic weights meet the preset topic conditions. Finally, it constructs a candidate topic set including the candidate topics based on the candidate topics.

[0256] The technical solution provided in this disclosure first selects target lists from various sample lists whose list weights meet preset list conditions. Then, it clusters the topics in the target lists to obtain multiple topic sets. From each topic set, it selects candidate topics whose topic weights meet preset topic conditions. Finally, it constructs a candidate topic set based on the candidate topics. In this way, by utilizing list weights and topic weights to perform layer-by-layer filtering of topics in each sample list, it is beneficial to improve the accuracy of candidate topic selection, thereby improving the accuracy of candidate topic set construction.

[0257] In an exemplary embodiment, step S420, before extracting the core entity words and core trigger words of the candidate topics in the candidate topic set, further includes a step of obtaining an updated candidate topic set. Specifically, this includes: if the candidate topic set includes a first candidate topic with a topic length greater than a second preset length, extracting the summary information corresponding to the first candidate topic, and using the summary information as a simplified candidate topic; updating the first candidate topics in the candidate topic set based on the simplified candidate topics to obtain the updated candidate topic set. Therefore, step S420, extracting the core entity words and core trigger words of the candidate topics in the candidate topic set, specifically includes: extracting the core entity words and core trigger words of the candidate topics in the updated candidate topic set.

[0258] The first candidate topic refers to a candidate topic whose length is greater than the second preset length, specifically a candidate topic with an excessively long length.

[0259] The summary information corresponding to the first candidate topic refers to the general information about the first candidate topic. Furthermore, extracting the summary information for the first candidate topic is to simplify first candidate topics that are too long.

[0260] Here, the simplified candidate topic corresponding to the first candidate topic refers to the summary information corresponding to the first candidate topic. Updating the first candidate topic in the candidate topic set is to replace the first candidate topic with the corresponding simplified candidate topic.

[0261] The updated candidate topic set includes candidate topics that have not been updated, as well as simplified candidate topics corresponding to the first candidate topic.

[0262] For example, the server identifies the topic length of the candidate topics in the candidate topic set. If the candidate topic set includes a first candidate topic with a topic length greater than a second preset length, the server simplifies the first candidate topic using a large model to obtain the summary information corresponding to the first candidate topic and uses it as the simplified candidate topic corresponding to the first candidate topic. Then, the server updates the first candidate topic in the candidate topic set to the corresponding simplified candidate topic to obtain the updated candidate topic set. Finally, the server performs relevant word extraction processing on the candidate topics in the candidate topic set using a large model to obtain the candidate core entity words and candidate core trigger words of the candidate topics.

[0263] The technical solution provided in this disclosure first obtains simplified candidate topics corresponding to first candidate topics in the candidate topic set whose topic length is greater than a second preset length. Then, it updates the first candidate topics in the candidate topic set with the corresponding simplified candidate topics, obtaining an updated candidate topic set. Finally, it extracts candidate core entity words and candidate core trigger words from the candidate topics in the updated candidate topic set. This simplification process, where the first candidate topics in the candidate topic set whose topic length is greater than the second preset length are simplified before extracting the candidate core entity words and candidate core trigger words, makes the extracted candidate core entity words and candidate core trigger words more accurate.

[0264] In an exemplary embodiment, step S150, which determines the target comment score of the comment to be analyzed based on the first comment score, the second comment score, and the third comment score, specifically includes the following: obtaining the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, and the third preset weight corresponding to the third comment score; performing a fusion process on the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain a fused comment score, and using the fused comment score as the target comment score of the comment to be analyzed.

[0265] Among them, the first preset weight, the second preset weight, and the third preset weight are all preset weights that can be adjusted according to the actual situation.

[0266] Here, "fusion processing" refers to weighted summation. The fusion comment score is the score obtained after weighted summation.

[0267] For example, the server retrieves the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, and the third preset weight corresponding to the third comment score from the local database; then, it performs a fusion process on the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain the fused comment score, that is, fused comment score = first preset weight × first comment score + second preset weight × second comment score + third preset weight × third comment score; finally, the fused comment score is used as the target comment score for the comment to be analyzed.

[0268] The technical solution provided in this disclosure integrates the first comment score, the second comment score, and the third comment score according to the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, and the third preset weight corresponding to the third comment score, to obtain the target comment score of the comment to be analyzed. In this way, when calculating the target comment score, the first comment score and its corresponding first preset weight, the second comment score and its corresponding second preset weight, and the third comment score and its corresponding third preset weight are comprehensively considered, which helps to improve the accuracy of the target comment score calculation and makes the target comments subsequently filtered by the target comment score more accurate.

[0269] In an exemplary embodiment, before fusing the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain the fused comment score, the method further includes a step of determining a fourth comment score for the comment to be analyzed. This step specifically includes: acquiring interaction data of the comment to be analyzed; determining the fourth comment score of the comment to be analyzed based on the interaction data, and acquiring the fourth preset weight corresponding to the fourth comment score. Therefore, fusing the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain the fused comment score specifically includes: fusing the first comment score, the second comment score, the third comment score, and the fourth comment score according to the first preset weight, the second preset weight, the third preset weight, and the fourth preset weight to obtain the fused comment score.

[0270] The fourth comment score refers to the comment interaction score, which is calculated based on the interaction data of the comments to be analyzed, such as the number of likes, dislikes, and replies to the comments.

[0271] The fourth preset weight is a preset weight that can be adjusted according to the actual situation.

[0272] For example, the server first obtains the interaction data of the comments to be analyzed, and then calculates the fourth comment score of the comments to be analyzed according to the following formula:

[0273] H I =e λN

[0274] Among them, H I λ represents the fourth comment score, λ represents the preset adjustment parameter, and N = number of likes + number of replies - number of dislikes.

[0275] Next, the server performs a fusion process on the first comment score, second comment score, third comment score, and fourth comment score according to the first preset weight, second preset weight, third preset weight, and fourth preset weight to obtain the fused comment score, i.e., fused comment score = first preset weight × first comment score + second preset weight × second comment score + third preset weight × third comment score + fourth preset weight × fourth comment score; finally, the fused comment score is used as the target comment score for the comment to be analyzed.

[0276] For example, the server can directly calculate the target comment score for the comment to be analyzed using the following formula:

[0277] H = W c ×H c +W t ×H t +W s ×H s +W I ×H I

[0278] Where H represents the target review score, W c H represents the first preset weight. c W represents the score of the first comment. t H represents the second preset weight. t W represents the second comment score. s H represents the third preset weight. s W represents the third comment score. I H represents the fourth preset weight. I This indicates the score for the fourth comment.

[0279] The technical solution provided in this disclosure integrates the first comment score, the second comment score, the third comment score, and the fourth comment score according to the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, the third preset weight corresponding to the third comment score, and the fourth preset weight corresponding to the fourth comment score, to obtain the target comment score of the comment to be analyzed. Thus, by comprehensively considering the first comment score and its corresponding first preset weight, the second comment score and its corresponding second preset weight, the third comment score and its corresponding third preset weight, and the fourth comment score and its corresponding fourth preset weight when calculating the target comment score, the accuracy of the target comment score calculation is improved. This facilitates the accurate selection of the target comment from the comments to be analyzed based on the target comment score, thereby improving the accuracy of the target comment determination.

[0280] In an exemplary embodiment, step S160, which determines the target comments from the comments to be analyzed based on the target comment score, specifically includes the following: sorting the comments to be analyzed in descending order of the target comment score to obtain sorted comments to be analyzed; selecting the top N comments from the sorted comments to be analyzed as target comments; where N is a positive integer.

[0281] Where N can be any positive integer, such as 5, 10, 15, etc., and can be adjusted according to the actual situation.

[0282] For example, the server sorts the comments to be analyzed in descending order of the target comment scores, and then selects the top N comments from the sorted comments to be analyzed, and uses the top N comments as the target comments; for example, if there are 100 sorted comments to be analyzed, the top 10 comments are selected from them.

[0283] The technical solution provided in this disclosure first uses the target comment score to sort the comments to be analyzed, and then selects the top N comments from the sorted comments to be analyzed as target comments, which helps to improve the accuracy of target comment determination.

[0284] Figure 5 This is a flowchart illustrating another comment processing method according to an exemplary embodiment, such as... Figure 5 As shown, this comment processing method is used in the terminal and includes the following steps:

[0285] In step S501, video information of the video to be analyzed is obtained.

[0286] In step S502, the comments to be analyzed in the video to be analyzed are adjusted according to the video information to obtain the processed comments; the processed comments are used to represent the comments obtained after adjusting the pronouns in the comments to be analyzed to the actual content corresponding to the pronouns.

[0287] In step S503, word extraction processing is performed on the processed comments to obtain the core entity words and core trigger words corresponding to the processed comments, and the core entity words and core trigger words corresponding to the processed comments are used as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0288] In step S504, the target video category to which the video to be analyzed belongs is obtained; from the first preset word set, the first word set corresponding to the target video category is determined, and the first word set corresponding to the target video category is used as the first target word set; the first preset word set includes the first word sets corresponding to multiple video categories; the candidate core entity words and candidate core trigger words in each first word set are matched with corresponding importance scores.

[0289] In step S505, target core entity words matching the core entity words are determined from the candidate core entity words in the first target word set, and target core trigger words matching the core trigger words are determined from the candidate core trigger words in the first target word set.

[0290] In step S506, the matching relationship between candidate core entity words and importance scores is queried to obtain the importance score corresponding to the target core entity word, and the matching relationship between candidate core trigger words and importance scores is queried to obtain the importance score corresponding to the target core trigger word; the importance scores corresponding to the target core entity words and the importance scores corresponding to the target core trigger words are fused to obtain the first fused importance score, and the first fused importance score is used as the first comment score of the comment to be analyzed.

[0291] In step S507, a topic category matching the target video category is determined; a second word set corresponding to the topic category is determined from the second preset word set, and the second word set corresponding to the topic category is used as the second target word set corresponding to the target video category; the second preset word set includes multiple second word sets corresponding to topic categories; each candidate generalized entity word and candidate generalized trigger word in the second word set is matched with a corresponding importance score.

[0292] In step S508, target generalized entity words that match the core entity words are determined from the candidate generalized entity words in the second target word set, and target generalized trigger words that match the core trigger words are determined from the candidate generalized trigger words in the second target word set.

[0293] In step S509, the matching relationship between candidate generalized entity words and importance scores is queried to obtain the importance score corresponding to the target generalized entity word, and the matching relationship between candidate generalized trigger words and importance scores is queried to obtain the importance score corresponding to the target generalized trigger word; the importance scores corresponding to the target generalized entity word and the importance scores corresponding to the target generalized trigger word are fused to obtain a second fused importance score, and the second fused importance score is used as the second comment score of the comment to be analyzed.

[0294] In step S510, a comment set corresponding to the target video category is determined from the preset comment set, and the comment set corresponding to the target video category is taken as the target comment set; the preset comment set includes comment sets corresponding to multiple video categories; the candidate comments in each comment set are matched with corresponding importance scores.

[0295] In step S511, candidate comments with a similarity greater than a preset similarity to the comment to be analyzed are determined from the target comment set corresponding to the target video category, and these candidate comments with a similarity greater than the preset similarity to the comment to be analyzed are taken as similar comments to the comment to be analyzed.

[0296] In step S512, the matching relationship between the target candidate comments and the importance scores is queried to obtain the importance scores of similar comments; the importance scores of each similar comment are fused to obtain the third fused importance score, and the third fused importance score is used as the third comment score of the comment to be analyzed.

[0297] In step S513, the first preset weight corresponding to the first comment score, the second preset weight corresponding to the second comment score, and the third preset weight corresponding to the third comment score are obtained; the interaction data of the comment to be analyzed is obtained; based on the interaction data of the comment to be analyzed, the fourth comment score of the comment to be analyzed is determined, and the fourth preset weight corresponding to the fourth comment score is obtained.

[0298] In step S514, the first comment score, the second comment score, the third comment score, and the fourth comment score are fused according to the first preset weight, the second preset weight, the third preset weight, and the fourth preset weight to obtain a fused comment score, which is then used as the target comment score for the comment to be analyzed.

[0299] In step S515, the comments to be analyzed are sorted in descending order of the target comment scores to obtain the sorted comments to be analyzed; from the sorted comments to be analyzed, the top N comments are selected as the target comments; N is a positive integer.

[0300] In the above-mentioned comment processing method, when identifying target comments from the comments to be analyzed in the video to be analyzed, the method comprehensively considers the core entity words and core trigger words corresponding to the comment to be analyzed, the first set of target words corresponding to the target video category to which the video to be analyzed belongs, the second set of target words corresponding to the target video category, and the set of target comments corresponding to the target video category. This is beneficial for determining the comment score of the comment to be analyzed from multiple dimensions, and combining the comment scores from multiple dimensions to determine the target comment score of the comment to be analyzed. This helps to improve the accuracy of the calculation of the target comment score of the comment to be analyzed, thereby making the target comments selected based on the target comment score more accurate, thus improving the accuracy of target comment identification. This avoids the defect of low accuracy in identifying target comments due to only considering the single dimension of the comment's own information.

[0301] To more clearly illustrate the comment processing method provided in this disclosure, a specific embodiment is described below. In one embodiment, as... Figure 6 As shown, this disclosure also provides a method for mining popular short video comments based on a large model. This method comprehensively considers the characteristics of open-domain text in short videos, taking into account the comment content itself, the overall comment ecosystem within the platform, and the hot topic ecosystem both inside and outside the platform. It scores the popularity of comments based on comment semantics, comment interaction, and the relevance of the comment and the short video content to hot topics. Specifically, it includes the following two main modules:

[0302] 1. Hotspot mining based on large models

[0303] refer to Figure 6 The popular candidate pool mainly includes two types: one is the popular comment candidate pool based on in-site comments, and the other is the popular topic candidate pool based on in-site and out-of-site hot topics.

[0304] The process of building a popular comment candidate pool is as follows: (1) For comments on videos on the site, a comment pool is built based on comment interaction data (including the number of likes, dislikes, and comment replies) and thresholds (that is, for comments on all videos on the site, those that meet all three conditions are retained, such as first judging the number of dislikes, then judging the number of likes, and finally judging the number of comment replies); (2) Based on this, a sliding time window (t1-t2) is set to filter comments whose time of occurrence and related interactions are not in this window (both conditions must be met to retain them); (3) Based on a deep learning model, the sentence vectors (embedding) of the comments are extracted and clustered and filtered. For comments within the same cluster (representing semantic similarity), the comments with the highest interaction data (number of likes + number of comment replies - number of dislikes) are retained and enter the popular comment candidate pool.

[0305] The process of building a hot topic candidate pool is as follows: (1) Monitor relevant lists inside and outside the site (such as lists of major portals outside the site); (2) Sort the obtained lists by pre-configured list weights (the weights of those more relevant to the lists inside the site are higher) and filter out the bottom 10%; (3) Extract the sentence vectors (embeddings) of the topics in the filtered lists based on a deep learning model, and perform clustering and filtering. For the same cluster (representing semantic similarity), retain the topic with the highest list weight (the list weight corresponding to each topic is the weight of the list to which it belongs, and the list weights corresponding to topics that appear in multiple lists are summed and used for sorting) and enter the hot topic candidate pool.

[0306] The weight calculation process is as follows: (1) For popular comments that are too long in the popular comment candidate pool, the large model will be used to simplify them first; for topic content that is too long in the hot topic candidate pool, the large model will be used to simplify it first; (2) For the hot comment candidate pool, the large model is used to extract the core entity words and core trigger words (trigger words refer to the predicates in the core event triples extracted by the large model); for the hot topic candidate pool, the large model is used to extract the core entity words and core trigger words in the topic title, and the large model is used to generalize (associate) based on the extracted core entity words and core trigger words to obtain generalized related words; (3) For the core entity words and trigger words mined from the hot comment candidate pool, the large model will be used to extract the core entity words and core trigger words. c First, comments are aggregated based on the short video category (e.g., sports, games). After aggregation, the importance of each word is calculated using the TF-IDF algorithm. ci For the comment set S containing the aforementioned core entity words and trigger words... C The weight of each comment is the sum of the importance of the aforementioned words it contains; the core entity words and core trigger words set w are extracted and generalized from the popular list. t Similarly, the importance of each word is first calculated using the TF-IDF algorithm based on the category of the list (such as sports, entertainment, etc.). ti .

[0307] Module 1 above yields three sets: S of weighted popular (simplified) comments. c Weighted core entity words / trigger word set w c Weighted popular generalized entity words / trigger words set w t .

[0308] 2. Calculation of comment popularity score

[0309] refer to Figure 6Module 2, Popularity Score, is primarily used to calculate the popularity score of comments in the short video field. The input consists of actual online comments and the context of the video (video title, ASR, OCR). The video context mainly includes the video title, OCR, ASR, and the video's category (category tags, such as finance, food, sports). Comments mainly include the comment text and the text of any replies to the comments. The main calculation steps are as follows:

[0310] (1) Dereference resolution: Dereference the current comment by using the relevant context of the video, that is, restore the pronouns in the comment (such as this, that, etc.) to the actual content, so as to facilitate the next step of relevant extraction.

[0311] (2) Extraction of core entity words / trigger words: Use a large model to extract the core entity words / trigger words of the text after the referential resolution.

[0312] (3) Matching core entity words / trigger words within the same category: Find the core entity words and trigger words mined in Module 1 for the short video category where the comment is located. c If the total number of matched words is M, then the final matching score for the core entity words / trigger words of the same category is:

[0313] (4) Matching generalized entity words / trigger words within the same category: For the short video category where the comment is located, find the set of core weighted popular generalized entity words / trigger words mined in Module 1. t The matching is performed, and the total number of matched words is N. The final score for the generalized entity word / trigger word matching of the same category is:

[0314] (5) Simplified comment matching of popular comments in the same category: Find the corresponding category in Module 1 for the short video category and mine the weighted popular (simplified) comments S. c For sentence-level matching, similarity is calculated using embedding. First, S... c The comments in the dataset are used to extract an embedding set E using a deep learning model. Then, the same deep learning model is used to extract the current comment's embedding, denoted as e. The similarity between the current e and each embedding in the aforementioned set E is calculated. Let K be the number of similar comments with a similarity greater than 0.9. The final simplified comment matching score for popular comments in the same category is then calculated as follows:

[0315] (6) Calculation of comment popularity score: In addition to the scores in (3)-(5) above, each comment also has a comment interaction score H. I =e λNWhere N is the number of likes + number of replies - number of dislikes, and λ is a preset adjustment parameter. The final popularity score for a comment is H = W. c ×H c +W t ×H t +W s ×H s +W I ×H I W c W t W s W I The preset empirical parameters.

[0316] The popularity score of comments is calculated using Module 1 and Module 2. After calculating the popularity score of all comments under the same video, the comments are sorted and the top 10 popular comments are selected to achieve the goal of finding the most popular comments under the video.

[0317] The above-mentioned short video hot comment mining method based on large models can achieve the following technical effects: (1) Comment ecosystem construction: After scoring the popularity of comments, it can be applied in scenarios such as comment transfer, sorting, and filtering to help build the comment area ecosystem, cultivate users' mindset of posting high-quality comments, and thus further improve the comment area dwell time and comment penetration rate. (2) Short video recommendation: By mining hot comments, the quality signal of the comment area can be better used in the distribution and recommendation of short videos (the popular comment score is used as a recommendation dimension), thereby further improving the App (Application) usage time, DAU (Daily Active User), etc.

[0318] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0319] It is understood that the same / similar parts between the various embodiments of the methods described above in this specification can be referred to each other. Each embodiment focuses on the differences from other embodiments, and relevant parts can be referred to the description of other method embodiments.

[0320] Based on the same inventive concept, this disclosure also provides a comment processing apparatus for implementing the comment processing method described above.

[0321] Figure 7 This is a block diagram illustrating a comment processing apparatus according to an exemplary embodiment. (Refer to...) Figure 7 The device includes a word extraction unit 710, a first determination unit 720, a second determination unit 730, a third determination unit 740, a fourth determination unit 750, and a comment determination unit 760.

[0322] The word extraction unit 710 is configured to extract the core entity words and core trigger words corresponding to the comments to be analyzed from the comments to be analyzed in the video to be analyzed.

[0323] The first determining unit 720 is configured to determine the first comment score of the comment to be analyzed based on the core entity words, core trigger words, and a first target word set corresponding to the target video category to which the video to be analyzed belongs; the first target word set includes candidate core entity words and candidate core trigger words of the target candidate comment; the video category of the sample video corresponding to the target candidate comment is the same as the target video category.

[0324] The second determining unit 730 is configured to determine the second comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; the topic category of the target candidate topic matches the target video category.

[0325] The third determining unit 740 is configured to determine the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category; the target comment set includes target candidate comments.

[0326] The fourth determining unit 750 is configured to determine the target comment score of the comment to be analyzed based on the first comment score, the second comment score, and the third comment score.

[0327] The comment determination unit 760 is configured to determine the target comment from the comments to be analyzed based on the target comment score.

[0328] In an exemplary embodiment, the word extraction unit 710 is further configured to perform the following actions: acquiring video information of the video to be analyzed; adjusting the comments to be analyzed in the video to be analyzed based on the video information to obtain processed comments; the processed comments are used to represent the comments obtained after adjusting the pronouns in the comments to be analyzed to the actual content corresponding to the pronouns; performing word extraction processing on the processed comments to obtain the core entity words and core trigger words corresponding to the processed comments, and using the core entity words and core trigger words corresponding to the processed comments as the core entity words and core trigger words corresponding to the comments to be analyzed.

[0329] In an exemplary embodiment, the first determining unit 720 is further configured to perform the following operations: determining a target core entity word that matches a core entity word from candidate core entity words, and determining a target core trigger word that matches a core trigger word from candidate core trigger words; querying the matching relationship between candidate core entity words and importance scores to obtain the importance score corresponding to the target core entity word, and querying the matching relationship between candidate core trigger words and importance scores to obtain the importance score corresponding to the target core trigger word; performing a fusion process on the importance score corresponding to the target core entity word and the importance score corresponding to the target core trigger word to obtain a first fused importance score, and using the first fused importance score as the first comment score of the comment to be analyzed.

[0330] In an exemplary embodiment, the comment processing apparatus further includes a set determination unit configured to perform the following: obtaining the target video category to which the video to be analyzed belongs; determining a first set of words corresponding to the target video category from a first preset set of words, and using the first set of words corresponding to the target video category as a first target set of words; the first preset set of words includes multiple first set of words corresponding to video categories; and each candidate core entity word and candidate core trigger word in the first set of words is matched with a corresponding importance score.

[0331] In an exemplary embodiment, the comment processing apparatus further includes a first construction unit configured to perform the following operations: constructing a candidate comment set based on comments from sample videos under each video category; extracting candidate core entity words and candidate core trigger words from the candidate comments in the candidate comment set; performing clustering processing on the candidate core entity words and candidate core trigger words of the candidate comments according to the video category of the sample videos corresponding to the candidate comments, obtaining a first word set corresponding to multiple video categories, and determining the importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category; and constructing a first preset word set based on the first word sets corresponding to multiple video categories.

[0332] In an exemplary embodiment, the first construction unit is further configured to perform the following actions: acquiring interactive data of comments on sample videos under each video category; filtering initial comments whose interactive data satisfies a first preset interaction condition from the comments on the sample videos under each video category; acquiring the comment time corresponding to the initial comment and the reply time of the reply comment corresponding to the initial comment; filtering intermediate comments from the initial comments whose comment time and reply time both satisfy a preset time condition; performing clustering processing on the intermediate comments to obtain multiple intermediate comment sets, and determining the interaction score of the comments in each intermediate comment set based on the interactive data of the comments in each intermediate comment set; filtering candidate comments whose interaction scores satisfy a second preset interaction condition from each intermediate comment set, and constructing a candidate comment set based on the candidate comments.

[0333] In an exemplary embodiment, the comment processing apparatus further includes a first simplification unit, configured to perform the following steps when the candidate comment set includes a first candidate comment whose comment length is greater than a first preset length: extracting summary information corresponding to the first candidate comment and using the summary information corresponding to the first candidate comment as a simplified candidate comment corresponding to the first candidate comment; updating the first candidate comment in the candidate comment set according to the simplified candidate comment corresponding to the first candidate comment to obtain an updated candidate comment set;

[0334] The first building unit is also configured to extract candidate core entity words and candidate core trigger words from the candidate comments in the updated candidate comment set.

[0335] In an exemplary embodiment, the comment processing apparatus further includes a third construction unit configured to perform clustering processing on the candidate comments according to the video categories of the sample videos corresponding to the candidate comments in the candidate comment set, to obtain comment sets corresponding to multiple video categories, and to determine the importance score of the candidate comments in the comment set corresponding to each video category; the importance score of each candidate comment is determined by the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment; and a preset comment set is constructed based on the comment sets corresponding to multiple video categories.

[0336] The comment processing device also includes a set query unit, configured to determine the comment set corresponding to the target video category from a preset comment set, and use the comment set corresponding to the target video category as the target comment set.

[0337] In an exemplary embodiment, the third determining unit 740 is further configured to perform the following operations: determining candidate comments from the target candidate comments whose similarity to the comment to be analyzed is greater than a preset similarity, and taking the candidate comments whose similarity to the comment to be analyzed is greater than the preset similarity as similar comments to the comment to be analyzed; querying the matching relationship between the target candidate comments and the importance score to obtain the importance score of the similar comments; performing a fusion process on the importance scores of each similar comment to obtain a third fused importance score, and taking the third fused importance score as the third comment score of the comment to be analyzed.

[0338] In an exemplary embodiment, the second determining unit 730 is further configured to perform the following operations: determining a target generalized entity word that matches the core entity word from the candidate generalized entity words; determining a target generalized trigger word that matches the core trigger word from the candidate generalized trigger words; querying the matching relationship between the candidate generalized entity words and importance scores to obtain the importance score corresponding to the target generalized entity word; querying the matching relationship between the candidate generalized trigger words and importance scores to obtain the importance score corresponding to the target generalized trigger word; performing a fusion process on the importance score corresponding to the target generalized entity word and the importance score corresponding to the target generalized trigger word to obtain a second fused importance score; and using the second fused importance score as the second comment score of the comment to be analyzed.

[0339] In an exemplary embodiment, the comment processing apparatus further includes a set filtering unit configured to perform the following actions: determining a topic category that matches a target video category; determining a second set of words corresponding to the topic category from a second preset set of words, and using the second set of words corresponding to the topic category as the second target set of words corresponding to the target video category; the second preset set of words includes multiple second set of words corresponding to topic categories; and each candidate generalized entity word and candidate generalized trigger word in each second set of words is matched with a corresponding importance score.

[0340] In an exemplary embodiment, the comment processing apparatus further includes a second construction unit configured to perform the following operations: constructing a candidate topic set based on topics in each sample list; extracting candidate core entity words and candidate core trigger words from the candidate topic set; performing word generalization processing on the candidate core entity words and candidate core trigger words of the candidate topics respectively to obtain candidate generalized entity words and candidate generalized trigger words of the candidate topics; performing clustering processing on the candidate generalized entity words and candidate generalized trigger words of the candidate topics according to the topic categories of the candidate topics to obtain a second word set corresponding to multiple topic categories, and determining the importance scores of the candidate generalized entity words and candidate generalized trigger words in the second word set corresponding to each topic category; and constructing a second preset word set based on the second word sets corresponding to multiple topic categories.

[0341] In an exemplary embodiment, the second construction unit is further configured to perform the following operations: obtaining the ranking weights of each sample ranking list; selecting target ranking lists from each sample ranking list whose ranking weights meet preset ranking conditions; performing clustering processing on the topics in the target ranking list to obtain multiple topic sets, and determining the topic weights corresponding to the topics in each topic set according to the ranking weights of the target ranking lists to which the topics in each topic set belong; selecting candidate topics from each topic set whose topic weights meet preset topic conditions, and constructing a candidate topic set based on the candidate topics.

[0342] In an exemplary embodiment, the comment processing apparatus further includes a second simplification unit, configured to perform the following steps when the candidate topic set includes a first candidate topic with a topic length greater than a second preset length: extracting summary information corresponding to the first candidate topic and using the summary information corresponding to the first candidate topic as a simplified candidate topic corresponding to the first candidate topic; updating the first candidate topic in the candidate topic set according to the simplified candidate topic corresponding to the first candidate topic to obtain an updated candidate topic set.

[0343] The second building unit is also configured to extract the core entity words and core trigger words of the candidate topics in the updated candidate topic set.

[0344] In an exemplary embodiment, the fourth determining unit 750 is further configured to perform the following operations: obtain a first preset weight corresponding to the first comment score, a second preset weight corresponding to the second comment score, and a third preset weight corresponding to the third comment score; perform a fusion process on the first comment score, the second comment score, and the third comment score according to the first preset weight, the second preset weight, and the third preset weight to obtain a fused comment score, and use the fused comment score as the target comment score of the comment to be analyzed.

[0345] In an exemplary embodiment, the fourth determining unit 750 is further configured to perform the following: acquiring interactive data of the comment to be analyzed; determining a fourth comment score of the comment to be analyzed based on the interactive data of the comment to be analyzed, and acquiring a fourth preset weight corresponding to the fourth comment score; and performing a fusion process on the first comment score, the second comment score, the third comment score, and the fourth comment score according to the first preset weight, the second preset weight, the third preset weight, and the fourth preset weight to obtain a fused comment score.

[0346] In one exemplary embodiment, the comment determination unit 760 is further configured to sort the comments to be analyzed in descending order of the target comment scores to obtain sorted comments to be analyzed; and to select the top N comments from the sorted comments to be analyzed as target comments; where N is a positive integer.

[0347] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0348] Each module in the aforementioned comment processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0349] Figure 8 This is a block diagram illustrating an electronic device 800 for implementing a comment processing method according to an exemplary embodiment. For example, the electronic device 800 may be a server. (Refer to...) Figure 8 The electronic device 800 includes a processing component 820, which further includes one or more processors, and memory resources represented by memory 822 for storing instructions, such as application programs, that can be executed by the processing component 820. The application programs stored in memory 822 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 820 is configured to execute instructions to perform the methods described above.

[0350] Electronic device 800 may further include: a power supply component 824 configured to perform power management of electronic device 800, a wired or wireless network interface 826 configured to connect electronic device 800 to a network, and an input / output (I / O) interface 828. Electronic device 800 may operate on an operating system stored in memory 822, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, or similar.

[0351] In one exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory 822 including instructions, which can be executed by a processor of an electronic device 800 to perform the above-described method. The storage medium may be a computer-readable storage medium, such as a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device.

[0352] In one exemplary embodiment, a computer program product is also provided, the computer program product including instructions that can be executed by a processor of an electronic device 800 to perform the above-described method.

[0353] It should be noted that the above-mentioned apparatus, electronic equipment, computer-readable storage medium, computer program product, etc., may also include other implementation methods according to the description of the method embodiments. For specific implementation methods, please refer to the description of the relevant method embodiments, which will not be elaborated here.

[0354] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the claims.

[0355] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A comment processing method, characterized in that, include: Extract the core entity words and core trigger words corresponding to the comments in the video to be analyzed; Based on the core entity words, the core trigger words, and the first set of target words corresponding to the target video category to which the video to be analyzed belongs, the first comment score of the comment to be analyzed is determined; the first set of target words includes candidate core entity words and candidate core trigger words of the target candidate comment. The sample video corresponding to the target candidate comment has the same video category as the target video category; Based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category, the second comment score of the comment to be analyzed is determined; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; The topic category of the target candidate topic matches the target video category; A third comment score is determined for the comment to be analyzed based on the target comment set corresponding to the target video category; the target comment set includes the target candidate comments. The target comment score for the comment to be analyzed is determined based on the first comment score, the second comment score, and the third comment score. The target comment is determined from the comments to be analyzed based on the target comment score.

2. The method according to claim 1, characterized in that, The process of extracting core entity words and core trigger words corresponding to the comments in the video to be analyzed includes: Obtain video information from the video to be analyzed; Based on the video information, the comments to be analyzed in the video to be analyzed are adjusted to obtain the processed comments; the processed comments are used to represent the comments obtained after adjusting the pronouns in the comments to be analyzed to the actual content corresponding to the pronouns; The processed comments are subjected to word extraction processing to obtain the core entity words and core trigger words corresponding to the processed comments. The core entity words and core trigger words corresponding to the processed comments are used as the core entity words and core trigger words corresponding to the comments to be analyzed.

3. The method according to claim 1, characterized in that, The step of determining the first comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the first set of target words corresponding to the target video category to which the video to be analyzed belongs includes: From the candidate core entity words, determine the target core entity words that match the core entity words, and from the candidate core trigger words, determine the target core trigger words that match the core trigger words; Query the matching relationship between the candidate core entity words and their importance scores to obtain the importance score corresponding to the target core entity word; and query the matching relationship between the candidate core trigger words and their importance scores to obtain the importance score corresponding to the target core trigger word. The importance scores corresponding to the target core entity words and the target core trigger words are fused to obtain a first fused importance score, which is then used as the first comment score of the comment to be analyzed.

4. The method according to claim 3, characterized in that, Before determining the first comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the first set of target words corresponding to the target video category to which the video to be analyzed belongs, the process further includes: Obtain the target video category to which the video to be analyzed belongs; From the first preset word set, a first word set corresponding to the target video category is determined, and the first word set corresponding to the target video category is taken as the first target word set; the first preset word set includes multiple first word sets corresponding to video categories; each candidate core entity word and candidate core trigger word in the first word set is matched with a corresponding importance score.

5. The method according to claim 4, characterized in that, The first preset word set is constructed in the following manner: Construct a candidate comment set based on the comments of sample videos under each video category; Extract candidate core entity words and candidate core trigger words from the candidate comments in the candidate comment set; Based on the video category of the sample video corresponding to the candidate comment, the candidate core entity words and candidate core trigger words of the candidate comment are clustered to obtain a first word set corresponding to multiple video categories, and the importance scores of the candidate core entity words and candidate core trigger words in the first word set corresponding to each video category are determined. The first preset word set is constructed based on the first word set corresponding to the multiple video categories.

6. The method according to claim 5, characterized in that, The process of constructing a candidate comment set based on the comments of sample videos under each video category includes: Obtain interactive data of comments for sample videos under each video category; From the comments of sample videos under each video category, select the initial comments whose interaction data meets the first preset interaction conditions; Obtain the comment time corresponding to the initial comment, and the reply time of the reply comment corresponding to the initial comment; From the initial comments, filter out the intermediate comments whose comment time and reply time both meet the preset time conditions; The intermediate comments are clustered to obtain multiple intermediate comment sets, and the interaction score of the comments in each intermediate comment set is determined based on the interaction data of the comments in each intermediate comment set. From each of the intermediate comment sets, candidate comments whose interaction scores meet the second preset interaction conditions are selected, and the candidate comment set is constructed based on the candidate comments.

7. The method according to claim 5, characterized in that, Before extracting the candidate core entity words and candidate core trigger words of the candidate comments in the candidate comment set, the process also includes: If the candidate comment set includes a first candidate comment whose comment length is greater than a first preset length, extract the summary information corresponding to the first candidate comment, and use the summary information corresponding to the first candidate comment as the simplified candidate comment corresponding to the first candidate comment; Based on the simplified candidate comments corresponding to the first candidate comments, the first candidate comments in the candidate comment set are updated to obtain the updated candidate comment set; The extraction of candidate core entity words and candidate core trigger words from the candidate comment set includes: Extract the candidate core entity words and candidate core trigger words from the candidate comments in the updated candidate comment set.

8. The method according to claim 5, characterized in that, After constructing a candidate comment set based on the comments of sample videos under each video category, the following steps are also included: Based on the video category of the sample video corresponding to the candidate comments in the candidate comment set, the candidate comments are clustered to obtain comment sets corresponding to multiple video categories, and the importance score of the candidate comments in the comment set corresponding to each video category is determined; the importance score of each candidate comment is determined by the importance scores of the candidate core entity words and candidate core trigger words included in each candidate comment; Construct a preset comment set based on the comment sets corresponding to the multiple video categories; Before determining the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category, the method further includes: From the preset comment set, determine the comment set corresponding to the target video category, and use the comment set corresponding to the target video category as the target comment set.

9. The method according to claim 8, characterized in that, The step of determining the third comment score of the comment to be analyzed based on the target comment set corresponding to the target video category includes: From the target candidate comments, identify candidate comments whose similarity to the comment to be analyzed is greater than a preset similarity, and take the candidate comments whose similarity to the comment to be analyzed is greater than the preset similarity as similar comments to the comment to be analyzed; Query the matching relationship between the target candidate comments and the importance scores to obtain the importance scores of the similar comments; The importance scores of the similar comments are fused to obtain a third fused importance score, which is then used as the third comment score of the comment to be analyzed.

10. The method according to claim 1, characterized in that, The step of determining the second comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category includes: From the candidate generalized entity words, a target generalized entity word that matches the core entity word is determined, and from the candidate generalized trigger words, a target generalized trigger word that matches the core trigger word is determined; Query the matching relationship between the candidate generalized entity words and the importance scores to obtain the importance score corresponding to the target generalized entity word; and query the matching relationship between the candidate generalization trigger words and the importance scores to obtain the importance score corresponding to the target generalization trigger words. The importance scores corresponding to the target generalized entity words and the target generalized trigger words are fused to obtain a second fused importance score, which is then used as the second comment score of the comment to be analyzed.

11. The method according to claim 10, characterized in that, Before determining the second comment score of the comment to be analyzed based on the core entity words, the core trigger words, and the second target word set corresponding to the target video category, the process further includes: Determine the topic category that matches the target video category; From the second preset word set, a second word set corresponding to the topic category is determined, and the second word set corresponding to the topic category is used as the second target word set corresponding to the target video category; the second preset word set includes multiple second word sets corresponding to topic categories; each candidate generalized entity word and candidate generalized trigger word in the second word set is matched with a corresponding importance score.

12. The method according to claim 11, characterized in that, The second preset word set is constructed in the following manner: Based on the topics in each sample list, construct a candidate topic set; Extract candidate core entity words and candidate core trigger words from the candidate topics in the candidate topic set; The candidate core entity words and candidate core trigger words of the candidate topics are respectively subjected to word generalization processing to obtain the candidate generalized entity words and candidate generalized trigger words of the candidate topics. According to the topic category of the candidate topics, the candidate generalized entity words and candidate generalization trigger words of the candidate topics are clustered to obtain a second word set corresponding to multiple topic categories, and the importance scores of the candidate generalized entity words and candidate generalization trigger words in the second word set corresponding to each topic category are determined. The second preset word set is constructed based on the second word set corresponding to the multiple topic categories.

13. The method according to claim 12, characterized in that, The process of constructing a candidate topic set based on the topics in each sample list includes: Obtain the ranking weights for each sample list; From the sample lists, select the target list whose list weight meets the preset list conditions; The topics in the target list are clustered to obtain multiple topic sets, and the topic weights of the topics in each topic set are determined according to the list weights of the target list to which the topics in each topic set belong. From each topic set, candidate topics whose topic weights satisfy preset topic conditions are selected, and a candidate topic set is constructed based on the candidate topics.

14. The method according to claim 12, characterized in that, Before extracting the core entity words and core trigger words of the candidate topics in the candidate topic set, the following steps are also included: If the candidate topic set includes a first candidate topic with a topic length greater than a second preset length, extract the summary information corresponding to the first candidate topic, and use the summary information corresponding to the first candidate topic as the simplified candidate topic corresponding to the first candidate topic. Based on the simplified candidate topic corresponding to the first candidate topic, the first candidate topic in the candidate topic set is updated to obtain the updated candidate topic set. The extraction of core entity words and core trigger words from the candidate topics set includes: Extract the core entity words and core trigger words of the candidate topics in the updated candidate topic set.

15. A comment processing device, characterized in that, include: The word extraction unit is configured to extract the core entity words and core trigger words corresponding to the comments to be analyzed from the comments to be analyzed in the video to be analyzed; The first determining unit is configured to determine a first comment score for the comment to be analyzed based on the core entity words, the core trigger words, and a first set of target words corresponding to the target video category to which the video to be analyzed belongs; the first set of target words includes candidate core entity words and candidate core trigger words of the target candidate comments; The sample video corresponding to the target candidate comment has the same video category as the target video category; The second determining unit is configured to determine a second comment score for the comment to be analyzed based on the core entity words, the core trigger words, and a second target word set corresponding to the target video category; the second target word set includes candidate generalized entity words and candidate generalized trigger words of the target candidate topic; The topic category of the target candidate topic matches the target video category; The third determining unit is configured to determine a third comment score for the comment to be analyzed based on a set of target comments corresponding to the target video category. The target comment set includes the target candidate comments; The fourth determining unit is configured to determine the target comment score of the comment to be analyzed based on the first comment score, the second comment score, and the third comment score. The comment determination unit is configured to determine the target comment from the comments to be analyzed based on the target comment score.

16. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the comment processing method as described in any one of claims 1 to 14.

17. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to perform the comment processing method as described in any one of claims 1 to 14.

18. A computer program product, the computer program product comprising instructions, characterized in that, When the instruction is executed by the processor of the electronic device, the electronic device is able to perform the comment processing method as described in any one of claims 1 to 14.

Images