A social media tag multi-objective combination recommendation method and system

Abstract

The application discloses a social media tag multi-target combination recommendation method and system, and relates to the technical field of data processing.The video record including the parent tag, the sub-tag and the video attribute data is read, the video record is divided according to the parent tag to obtain an analysis unit, the affinity degree of the parent tag and the sub-tag in the analysis unit is statistically analyzed, and the affinity level is generated, the sub-tags in the same analysis unit are combined two by two, the co-occurrence number of the tags and the appearance number of the single tag are counted, the tag correlation weight is calculated, and the tag correlation matrix is constructed; the trend intensity, the commercial concentration and the stability of the play quantity of the sub-tag are calculated according to the video attribute data, the comprehensive recommendation score is obtained in combination with the affinity score; the number of the recommended tags, the grade ratio constraint, the internal correlation constraint and the approximate tag constraint are set, the candidate sub-tags are selected according to the comprehensive recommendation score, and the candidate sub-tags meeting the constraint are taken as the recommended tags, so that the accuracy and the applicability of the tag combination recommendation are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for multi-objective combination recommendation of social media tags. Background Technology

[0002] With the development of short video platforms and social media content distribution mechanisms, hashtags have gradually become important data identifiers for content to enter specific traffic pools. When creators publish video content, they usually need to configure multiple tags for the video so that the platform can identify the content theme based on the tags and distribute the content to the corresponding audience.

[0003] Existing tag selection methods primarily include platform search recommendations, third-party tag query tools, and manual selection based on experience. Platform search recommendations typically return relevant tags based on overall platform usage or popularity, resulting in a relatively singular display dimension. Third-party tools often focus on statistics such as usage and popularity changes for individual tags, making it difficult to reflect the overall status when multiple tags are used together. Manual selection relies on creator experience, is easily influenced by subjective judgment, and struggles to adapt to changes in the platform's content ecosystem. Some existing recommendation methods generate tag lists based on keyword similarity, co-occurrence relationships, or historical usage, but the output usually requires users to further filter and combine tags, failing to directly address the actual usage needs of a single content post. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for multi-objective combination recommendation of social media tags to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A multi-objective recommendation method for social media tags, comprising the following steps: Read video recordings, which include parent tags, child tags, and video attribute data; Furthermore, the parent tag is a tag contained in the video in the video record, and the child tags are other tags contained in the video in the video record besides the parent tag; the video attribute data includes the number of times the video has been played and whether it has commercial external links attached.

[0006] Among them, tags refer to structured text identifiers attached to social media video records and used to characterize the theme, content attributes, posting context, or target audience of the video content.

[0007] Commercial backlinks refer to links that are attached to the homepage of a social media account, a page associated with the video content, or a redirect location that is allowed by the platform, and that can point to a product purchase page, service reservation page, brand promotion page, transaction conversion page, or other pages for commercial purposes. The video recordings are divided into analysis units according to the parent tag. The affinity between the parent tag and the child tag in the analysis unit is statistically analyzed to generate the affinity level and construct the tag association matrix. Furthermore, the process of generating affinity levels includes: Video records with the same parent tag are grouped into the same analysis unit; The number of video records appearing for each sub-tag in the statistical analysis unit is used as the affinity in that analysis unit. The sub-labels in the analysis unit are arranged in descending order of affinity to obtain the affinity sequence; the sub-labels in the affinity sequence are divided into several affinity levels according to the preset partitioning rules.

[0008] Furthermore, the process of constructing the tag association matrix includes: Within the same analysis unit, all sub-tags in each video record are combined in pairs to obtain the tag pairs for that video record; The number of video records appearing in the analysis unit for each tag pair is counted as the tag co-occurrence number; and the number of video records appearing in the analysis unit for each of the two sub-tags in the tag pair is counted as the single tag occurrence number. The tag co-occurrence number corresponds to the single tag occurrence number. The difference between the sum of the occurrence counts of the two corresponding single tags and the number of tag co-occurrences is taken as the joint occurrence count of the tags, and the ratio of the number of tag co-occurrences to the total number of tag co-occurrences is taken as the tag association weight. Specifically, the process of calculating tag association weights includes: ; Where a and b represent two child tags in the same tag pair, N a N represents the number of video records in the analysis unit where sub-label 'a' appears. b N represents the number of video records in the analysis unit where sub-tag b appears. ab W represents the number of video records in which sub-tags a and b appear together in the same video record. ab Represents the weight of tag association; A tag association matrix is ​​generated according to the tag association weights corresponding to each tag pair. The tag association matrix is ​​a data structure that uses the pairwise combination of sub-tags in the same analysis unit as the index and the tag association weights of the corresponding tag pairs as the matrix values.

[0009] The business indicator data of the sub-tags in the video attribute data calculation and analysis unit are used to score the sub-tags based on the affinity level and business indicator data to obtain the comprehensive recommendation score of the sub-tags. Constraints are constructed in combination with the tag association matrix, and tag recommendations are made based on the comprehensive recommendation score and the constraints.

[0010] Furthermore, the process of calculating the business indicator data of the sub-tags in the analysis unit includes: The business metrics data for sub-tags include the trend strength, commercial concentration, and playback stability of the sub-tags; The trend strength is the proportion of the number of video records carrying the corresponding sub-tags to the total number of video records in the analysis unit; The commercial concentration is the ratio of the number of video records carrying the corresponding sub-tag and having commercial external links to the total number of video records carrying that sub-tag. The stability of the playback volume is the ratio of the average playback volume of video records carrying the corresponding sub-tag to the maximum playback volume of video records carrying that sub-tag.

[0011] Furthermore, the process of obtaining the comprehensive recommendation score for sub-tags includes: Different affinity levels are preset with corresponding affinity scores. The comprehensive recommendation score is calculated by weighting the affinity score of the sub-tag, trend strength, commercial concentration, and playback stability, and using the sub-tag as the basis.

[0012] Specifically, the process of calculating the comprehensive recommendation score includes: ; Among them, S i A represents the overall recommendation score for the i-th sub-tag. i T represents the affinity score of the i-th sub-label. i C represents the trend strength of the i-th sub-label. i P represents the commercial concentration of the i-th sub-label. i The play count stability represents the play count stability of the i-th sub-tag, and w1, w2, w3, and w4 represent the preset weight coefficients corresponding to affinity rating, trend strength, commercial concentration, and play count stability, respectively. Specifically, the preset weighting coefficients are adjusted according to different content objectives; The content target refers to the data items entered by the user before tag recommendation, which represent the business orientation expected to be achieved in this video release. It is used to determine the degree of participation of different evaluation dimensions in the calculation of the comprehensive recommendation score. The content objectives include brand exposure objectives, user growth objectives, and direct conversion objectives. Among them, brand exposure objectives correspond to the release needs of increasing the reach of content and the intensity of theme display; user growth objectives correspond to the release needs of increasing the stability of the content audience and the foundation for account follower conversion; and direct conversion objectives correspond to the release needs of increasing commercial guidance capabilities and transaction conversion opportunities. When calculating the comprehensive recommendation score for a sub-tag, the corresponding weight coefficients are matched according to the content goals specified by the user, so that affinity score, trend strength, commercial concentration and playback stability participate in the weighted calculation according to their respective weights. When the content goals change, the weight coefficients corresponding to affinity score, trend strength, commercial concentration and playback stability are adjusted accordingly, so that the same sub-tag can get different comprehensive recommendation scores under different content goals, and the tag recommendation results correspond to the content publishing needs specified by the user.

[0013] Furthermore, the process of recommending tags based on the comprehensive recommendation score and the constraints includes: Set the number of recommended tags; The constraints include grade matching constraints, internal association constraints, and approximate label constraints; The proportion of different affinity levels is preset as the level matching constraint. Sub-labels in the same analysis unit are selected as candidate sub-labels in order of comprehensive recommendation score from high to low. The affinity level of the candidate sub-labels is obtained in real time. When the proportion of the number of candidate sub-labels of a certain affinity level to the number of recommended labels reaches the corresponding level matching constraint, the candidate sub-labels corresponding to that affinity level are excluded in the subsequent selection process. When selecting sub-labels in the same analysis unit as candidate sub-labels, the average of the label association weights of the new candidate sub-label and other candidate sub-labels is calculated in real time as the candidate association value; when the candidate association value is less than the preset internal association constraint, the candidate sub-label is excluded, and the next sub-label is selected as the candidate sub-label in the order of comprehensive recommendation score from high to low. When selecting sub-labels in the same analysis unit as candidate sub-labels, the new candidate sub-labels are combined with other candidate sub-labels in pairs to obtain in-pair label pairs. The character edit distance between the two candidate sub-labels in the in-pair label pairs is calculated. When there is a character edit distance less than the preset approximate label constraint, the trend strength of the two candidate sub-labels corresponding to the character edit distance is compared. The candidate sub-label with the lower trend strength is eliminated, and the next sub-label is selected as a candidate sub-label in order of comprehensive recommendation score from high to low. If the trend strength of the two candidate sub-tags corresponding to the edit distance of the character is the same, other indicators such as the comprehensive recommendation score can be compared until one of the candidate sub-tags is selected. When the number of candidate sub-tags reaches the number of recommended tags, and the new candidate sub-tags satisfy the level matching constraint, internal association constraint, and approximate tag constraint, the candidate sub-tags are used as recommended tags for tag recommendation.

[0014] A multi-objective combination recommendation system for social media tags, the system includes a record reading module, an affinity construction module, and a recommendation execution module; The output of the record reading module is connected to the input of the affinity building module; the output of the affinity building module is connected to the input of the recommendation execution module.

[0015] The recording reading module is used to receive video recordings to be processed and to determine the parent tag, child tag, and video attribute data in the video recordings; The affinity construction module is used to classify video records according to parent tags to form analysis units, and to obtain the affinity, affinity level and tag association matrix between parent tags and child tags in the analysis unit; The recommendation execution module is used to generate tag recommendation results based on video attribute data, affinity level, business indicator data, comprehensive recommendation score, tag association matrix, and constraints.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention divides video records into analysis units according to parent tags, and statistically analyzes the affinity between sub-tags and parent tags in the analysis unit. Then, the sub-tags are divided into affinity levels, so that tag recommendation no longer depends solely on the popularity of a single tag, but can determine the recommendation basis based on the relationship between tags within the same content range, thereby improving the data targeting of tag screening.

[0017] 2. This invention constructs a tag association matrix and introduces internal association constraints during the tag recommendation process to determine the tag association weights between candidate sub-tags. This ensures that multiple tags entering the recommendation results have a common occurrence relationship, reducing the combination discrepancy problem caused by simply stacking tags based on the comprehensive recommendation score, and improving the content consistency among recommended tags.

[0018] 3. This invention calculates trend strength, commercial concentration, and playback stability, and combines them with affinity rating to form a comprehensive recommendation score. Then, it filters recommended tags through level matching constraints, internal association constraints, and similar tag constraints, so that the recommendation results simultaneously consider tag activity, commercial attributes, playback stability, and text similarity relationships, thereby improving the applicability of tag combinations. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a multi-objective combination recommendation method for social media tags according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example: Figure 1 As shown, the present invention provides a technical solution, a method for multi-objective combination recommendation of social media tags, the method comprising the following steps: Read video recordings, which include parent tags, child tags, and video attribute data; The parent tag is a tag contained in the video in the video record, and the child tags are other tags contained in the video in the video record besides the parent tag; the video attribute data includes the number of times the video has been played and whether it has commercial external links attached.

[0022] In this embodiment, for example, the video recording is set to 60, the parent tag is #coffee, and the child tags include #latte, #espresso, #coffeeshop, and #coffeebeans; In the video attribute data, the number of views ranges from 1,200 to 68,000, and the number of video records with commercial external links is 14. The video recordings are divided into analysis units according to the parent tag. The affinity between the parent tag and the child tag in the analysis unit is statistically analyzed to generate the affinity level and construct the tag association matrix. The process of generating affinity levels includes: Video records with the same parent tag are grouped into the same analysis unit; The number of video records appearing for each sub-tag in the statistical analysis unit is used as the affinity in that analysis unit. The sub-labels in the analysis unit are arranged in descending order of affinity to obtain the affinity sequence; the sub-labels in the affinity sequence are divided into several affinity levels according to the preset partitioning rules.

[0023] In this embodiment, 60 video recordings are grouped into the same analysis unit according to the parent tag #coffee, and the affinity of the sub-tags in the analysis unit is statistically analyzed. #latte corresponds to 32, #espresso corresponds to 26, #coffee shop corresponds to 21, and #coffee bean corresponds to 15. For example, the default affinity level is 2, #latte and #espresso are classified into the first affinity level, and #coffeeshop and #coffeebeans are classified into the second affinity level; the number of affinity levels can be adjusted according to the affinity sequence length and the number of recommended tags; The process of constructing a tag association matrix includes: Within the same analysis unit, all sub-tags in each video record are combined in pairs to obtain the tag pairs for that video record; The number of video records appearing in the analysis unit for each tag pair is counted as the tag co-occurrence number; and the number of video records appearing in the analysis unit for each of the two sub-tags in the tag pair is counted as the single tag occurrence number. The tag co-occurrence number corresponds to the single tag occurrence number. The difference between the sum of the occurrence counts of the two corresponding single tags and the number of tag co-occurrences is taken as the joint occurrence count of the tags, and the ratio of the number of tag co-occurrences to the total number of tag co-occurrences is taken as the tag association weight. A tag association matrix is ​​generated according to the tag association weights corresponding to each tag pair. The tag association matrix is ​​a data structure that uses the pairwise combination of sub-tags in the same analysis unit as the index and the tag association weights of the corresponding tag pairs as the matrix values.

[0024] In this embodiment, for example, the number of occurrences of the single tags #latte and #espresso are 32 and 26 respectively, the number of co-occurrences is 15, and the tag association weight is 0.35; the number of occurrences of the single tags #latte and #coffee are 32 and 21 respectively, the number of co-occurrences is 13, and the tag association weight is 0.33; the number of occurrences of the single tags #espresso and #coffee beans are 26 and 15 respectively, the number of co-occurrences is 9, and the tag association weight is 0.28; the number of occurrences of the single tags #espresso and #coffee are 26 and 21 respectively, the number of co-occurrences is 10, and the tag association weight is 0.27. The business indicator data of the sub-tags in the video attribute data calculation and analysis unit are used to score the sub-tags based on the affinity level and business indicator data to obtain the comprehensive recommendation score of the sub-tags. Constraints are constructed in combination with the tag association matrix, and tag recommendations are made based on the comprehensive recommendation score and the constraints.

[0025] The process of calculating the business indicator data of sub-tags in the analysis unit includes: The business metrics data for sub-tags include the trend strength, commercial concentration, and playback stability of the sub-tags; The trend strength is the proportion of the number of video records carrying the corresponding sub-tags to the total number of video records in the analysis unit; The commercial concentration is the ratio of the number of video records carrying the corresponding sub-tag and having commercial external links to the total number of video records carrying that sub-tag. The stability of the playback volume is the ratio of the average playback volume of video records carrying the corresponding sub-tag to the maximum playback volume of video records carrying that sub-tag.

[0026] In this embodiment, for example, the trend strength of #Latte is 0.53, the commercial concentration is 0.16, and the playback stability is 0.44; The trend strength for espresso is 0.43, the commercial concentration is 0.23, and the play count stability is 0.41. The trend strength of #Coffee Shop is 0.35, the commercial concentration is 0.10, and the playback stability is 0.52; The trend strength of the coffee bean is 0.25, the commercial concentration is 0.33, and the playback stability is 0.46.

[0027] The process of obtaining the comprehensive recommendation score for sub-tags includes: Different affinity levels are preset with corresponding affinity scores. The comprehensive recommendation score is calculated by weighting the affinity score of the sub-tag, trend strength, commercial concentration, and playback stability, and using the sub-tag as the basis.

[0028] In this embodiment, for example, the content target specified by the user is the brand exposure target, and the number of recommended tags is 3. The specific number of recommended tags can be determined according to the platform's publishing restrictions and the user's publishing habits. Under the brand exposure goal, the weight coefficients are set as w1=0.30, w2=0.35, w3=0.15, and w4=0.20. The specific weight coefficients can be adjusted according to the importance of the evaluation dimensions corresponding to the content goals. The sum of the weight coefficients is 1. For example, the affinity score corresponding to the first affinity level is preset to 1, and the affinity score corresponding to the second affinity level is 0.5; the comprehensive recommendation score of the sub-tags is calculated according to the brand exposure target, #Latte is 0.60, #Espresso is 0.57, #Coffee Shop is 0.39, and #Coffee Beans is 0.35; The process of recommending tags based on the comprehensive recommendation score and the constraints includes: Set the number of recommended tags; The constraints include grade matching constraints, internal association constraints, and approximate label constraints; The proportion of different affinity levels is preset as the level matching constraint. Sub-labels in the same analysis unit are selected as candidate sub-labels in order of comprehensive recommendation score from high to low. The affinity level of the candidate sub-labels is obtained in real time. When the proportion of the number of candidate sub-labels of a certain affinity level to the number of recommended labels reaches the corresponding level matching constraint, the candidate sub-labels corresponding to that affinity level are excluded in the subsequent selection process. When selecting sub-labels in the same analysis unit as candidate sub-labels, the average of the label association weights of the new candidate sub-label and other candidate sub-labels is calculated in real time as the candidate association value; when the candidate association value is less than the preset internal association constraint, the candidate sub-label is excluded, and the next sub-label is selected as the candidate sub-label in the order of comprehensive recommendation score from high to low. When selecting sub-labels in the same analysis unit as candidate sub-labels, the new candidate sub-labels are combined with other candidate sub-labels in pairs to obtain in-pair label pairs. The character edit distance between the two candidate sub-labels in the in-pair label pairs is calculated. When there is a character edit distance less than the preset approximate label constraint, the trend strength of the two candidate sub-labels corresponding to the character edit distance is compared. The candidate sub-label with the lower trend strength is eliminated, and the next sub-label is selected as a candidate sub-label in order of comprehensive recommendation score from high to low. When the number of candidate sub-tags reaches the number of recommended tags, and the new candidate sub-tags satisfy the level matching constraint, internal association constraint, and approximate tag constraint, the candidate sub-tags are used as recommended tags for tag recommendation.

[0029] In this embodiment, for example, the level matching constraint is set to the first affinity level not exceeding 0.67 and the second affinity level not exceeding 0.34; after conversion according to the number of recommended tags 3, the target number of the first affinity level is 2 and the target number of the second affinity level is 1. The specific ratio is configured by rounding according to the number of recommended tags and the affinity sequence distribution. For example, the internal association constraint is set to 0.25, which is lower than the tag association weights of #latte and #espresso, #latte and #coffee shop, and #espresso and #coffee beans. This can retain sub-tags with stable co-occurrence relationships. The specific association threshold is determined according to the median or quantile of the tag association weights. For example, the approximate label constraint is set to a character edit distance of less than 2. This threshold is suitable for the text approximation recognition of short Chinese labels. The specific approximation distance threshold can be adjusted according to the sub-label text length and language type.

[0030] #Latte and #Espresso are selected as candidate sub-tags. Both belong to the first affinity level. Once the number of sub-tags with the first affinity level reaches 2, the sub-tags corresponding to the first affinity level will be excluded when selecting further. #Coffee shop is a candidate sub-tag, belonging to the second affinity level, with 3 candidate sub-tags. #Coffee beans are not a recommended tag. The tag association weight between candidate sub-tags is 0.25, and there are no tag pairs with a character edit distance of less than 2. If there are tag pairs with a character edit distance of less than 2, compare the trend strength of the corresponding two candidate sub-tags and exclude the candidate sub-tag with lower trend strength; if the trend strength is the same, compare the comprehensive recommendation score; if the comprehensive recommendation scores are the same, retain the candidate sub-tag with the higher comprehensive recommendation score. In this embodiment, the final recommended tags are #latte, #espresso, and #coffee shop.

[0031] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A multi-objective combination recommendation method for social media tags, characterized in that: The method includes the following steps: Read video recordings, which include parent tags, child tags, and video attribute data; The video records are divided into analysis units according to the parent tag. The affinity between the parent tag and the child tag in the analysis unit is statistically analyzed to generate the affinity level and construct the tag association matrix. The business indicator data of the sub-tags in the video attribute data calculation and analysis unit are used to score the sub-tags based on the affinity level and business indicator data to obtain the comprehensive recommendation score of the sub-tags. Constraints are constructed in combination with the tag association matrix, and tag recommendations are made based on the comprehensive recommendation score and the constraints.

2. The social media tag multi-objective combination recommendation method according to claim 1, characterized in that: The parent tag is a tag contained in the video in the video record, and the child tags are other tags contained in the video in the video record besides the parent tag; the video attribute data includes the number of times the video has been played and whether it has commercial external links attached.

3. The social media tag multi-objective combination recommendation method according to claim 2, characterized in that, The process of generating affinity levels includes: Video records with the same parent tag are grouped into the same analysis unit; The number of video records appearing for each sub-tag in the statistical analysis unit is used as the affinity in that analysis unit. The sub-labels in the analysis unit are arranged in descending order of affinity to obtain the affinity sequence; the sub-labels in the affinity sequence are divided into several affinity levels according to the preset partitioning rules.

4. The social media tag multi-objective combination recommendation method according to claim 3, characterized in that, The process of constructing a tag association matrix includes: Within the same analysis unit, all sub-tags in each video record are combined in pairs to obtain the tag pairs for that video record; The number of video records appearing in the analysis unit for each tag pair is counted as the tag co-occurrence number; and the number of video records appearing in the analysis unit for each of the two sub-tags in the tag pair is counted as the single tag occurrence number. The tag co-occurrence number corresponds to the single tag occurrence number. The difference between the sum of the occurrence counts of the two corresponding single tags and the number of tag co-occurrences is taken as the joint occurrence count of the tags, and the proportion of the number of tag co-occurrences to the total number of tag co-occurrences is taken as the tag association weight. A tag association matrix is ​​generated according to the tag association weights corresponding to each tag pair. The tag association matrix is ​​a data structure that uses the pairwise combination of sub-tags in the same analysis unit as the index and the tag association weights of the corresponding tag pairs as the matrix values.

5. The social media tag multi-objective combination recommendation method according to claim 4, characterized in that, The process of calculating the business indicator data of sub-tags in the analysis unit includes: The business metrics data for sub-tags include the trend strength, commercial concentration, and playback stability of the sub-tags; The trend strength is the proportion of the number of video records carrying the corresponding sub-tags to the total number of video records in the analysis unit; The commercial concentration is the ratio of the number of video records carrying the corresponding sub-tag and having commercial external links to the total number of video records carrying that sub-tag. The stability of the playback volume is the ratio of the average playback volume of video records carrying the corresponding sub-tag to the maximum playback volume of video records carrying that sub-tag.

6. The social media tag multi-objective combination recommendation method according to claim 5, characterized in that, The process of obtaining the comprehensive recommendation score for sub-tags includes: Different affinity levels are preset with corresponding affinity scores. The comprehensive recommendation score is calculated by weighting the affinity score of the sub-tag, trend strength, commercial concentration, and playback stability, and using the sub-tag as the basis.

7. The social media tag multi-objective combination recommendation method according to claim 6, characterized in that, The process of recommending tags based on the comprehensive recommendation score and the constraints includes: Set the number of recommended tags; The constraints include grade matching constraints, internal association constraints, and approximate label constraints; The proportion of different affinity levels is preset as the level matching constraint. Sub-labels in the same analysis unit are selected as candidate sub-labels in order of comprehensive recommendation score from high to low. The affinity level of the candidate sub-labels is obtained in real time. When the proportion of the number of candidate sub-labels of a certain affinity level to the number of recommended labels reaches the corresponding level matching constraint, the candidate sub-labels corresponding to that affinity level are excluded in the subsequent selection process. When selecting sub-labels in the same analysis unit as candidate sub-labels, the average of the label association weights of the new candidate sub-label and other candidate sub-labels is calculated in real time as the candidate association value; when the candidate association value is less than the preset internal association constraint, the candidate sub-label is excluded, and the next sub-label is selected as the candidate sub-label in the order of comprehensive recommendation score from high to low. When selecting sub-labels in the same analysis unit as candidate sub-labels, the new candidate sub-labels are combined with other candidate sub-labels in pairs to obtain in-pair label pairs. The character edit distance between the two candidate sub-labels in the in-pair label pairs is calculated. When there is a character edit distance less than the preset approximate label constraint, the trend strength of the two candidate sub-labels corresponding to the character edit distance is compared. The candidate sub-label with the lower trend strength is eliminated, and the next sub-label is selected as a candidate sub-label in order of comprehensive recommendation score from high to low. When the number of candidate sub-tags reaches the number of recommended tags, and the new candidate sub-tags satisfy the level matching constraint, internal association constraint, and approximate tag constraint, the candidate sub-tags are used as recommended tags for tag recommendation.

8. A social media tag multi-objective combination recommendation system, applied to the social media tag multi-objective combination recommendation method according to any one of claims 1-8, characterized in that: The system includes a record reading module, an affinity building module, and a recommended execution module; The output of the record reading module is connected to the input of the affinity building module; the output of the affinity building module is connected to the input of the recommendation execution module.

9. A social media tag multi-objective combination recommendation system according to claim 8, characterized in that: The recording reading module is used to receive video recordings to be processed and to determine the parent tag, child tag, and video attribute data in the video recordings; The affinity construction module is used to classify video records according to parent tags to form analysis units, and to obtain the affinity, affinity level and tag association matrix between parent tags and child tags in the analysis unit; The recommendation execution module is used to generate tag recommendation results based on video attribute data, affinity level, business indicator data, comprehensive recommendation score, tag association matrix, and constraints.

Images