AI intelligent marketing public praise monitoring method based on social platform
By acquiring multi-dimensional evaluation data from social platforms, standardizing the data, calculating relevance scores, and configuring importance weights using a self-attention mechanism, the inaccuracy and inflexibility issues of existing marketing reputation monitoring technologies are resolved, enabling precise strategy execution and resource optimization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING SENBO MINGDE MARKETING TECH CO LTD
- Filing Date
- 2026-02-26
- Publication Date
- 2026-05-15
AI Technical Summary
Existing marketing reputation monitoring and processing strategies rely on human experience and lack automation mechanisms, resulting in inaccurate and inflexible processing, and weak correlation between evaluation and marketing characteristics.
By acquiring multi-dimensional evaluation data from social media platforms, performing standardized preprocessing, calculating relevance scores, configuring importance weights using a self-attention mechanism, generating importance ranking results, and comparing and analyzing them with a preset marketing strategy table to determine improvement strategies.
It improves the accuracy and flexibility of marketing feature strategies, ensuring that resources are focused on highly relevant and high-impact features, and achieves accuracy in importance ranking and precision in strategy application.
Smart Images
Figure CN122048418A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marketing reputation monitoring technology, and in particular to an AI-powered intelligent marketing reputation monitoring method based on social platforms. Background Technology
[0002] With the deep integration of social media and e-commerce, brands generally regard social media platforms as the main battleground for word-of-mouth marketing. The current mainstream approach involves capturing multimodal content such as text, images, and short videos containing brand keywords; then determining the sentiment polarity of the content and outputting positive / negative / neutral labels; finally, constructing a popularity index based on likes, shares, and comments, and ultimately presenting the "sentiment-popularity" two-dimensional monitoring results in a dashboard format. Some companies further manually label topics with sentiment scores below a threshold as "negative word-of-mouth" and trigger public relations action tickets, requiring the operations team to conduct crisis public relations or launch positive content to offset the negative sentiment. For the identification of marketing features such as price, packaging, logistics, customer service, and efficacy, existing solutions mostly use keyword matching or LDA topic models, clustering posts into preset topics and then calculating the sentiment mean under each topic to roughly determine which features need optimization.
[0003] However, existing technologies rely on human experience to generate improvement strategies and lack automated mechanisms that link them to importance ranking, resulting in inaccurate processing strategies. In addition, the weak correlation between evaluation and marketing characteristics means that operators can only rely on experience to judge the direction of improvement, resulting in poor flexibility. Summary of the Invention
[0004] The purpose of this invention is to provide an AI-powered intelligent marketing reputation monitoring method based on social platforms, which solves the problems of inaccurate processing strategies and poor processing flexibility in existing technologies.
[0005] To achieve the above objectives, this invention provides an AI-powered intelligent marketing reputation monitoring method based on social platforms, comprising the following steps: S1. Obtain multi-dimensional evaluation data of target marketing objects on social platforms, and perform standardized preprocessing on the evaluation data to obtain a standardized evaluation dataset; S2. Evaluate the relevance between each evaluation item in the standardized evaluation dataset and the corresponding marketing feature information, and calculate the relevance score between each evaluation item and the marketing feature information; S3. Based on the relevance score, select marketing feature information that is higher than the preset relevance threshold as marketing features to be improved; S4. Based on the self-attention mechanism, assign importance weights to the marketing features to be improved, obtain the importance weight values of each marketing feature to be improved, evaluate the importance based on the importance weight values, and generate the importance ranking results based on the importance evaluation results; S5. Compare and analyze the importance ranking results with the preset marketing strategy table to determine the improvement strategies corresponding to each marketing feature to be improved, and implement the improvement strategies in order of importance from high to low.
[0006] In some embodiments of this application, in step S1, multi-dimensional evaluation data of the target marketing object on the social platform is obtained, and the evaluation data is preprocessed to standardize it to obtain a standardized evaluation dataset, including: S11. Extract multi-dimensional evaluation data of target marketing objects from social platforms based on API interfaces. The multi-dimensional evaluation data includes the number of likes, the number of reposts, the number of followers, the sentiment tags of comments, and the user's authentication status. S12. Perform standardized preprocessing on the multi-dimensional evaluation data, and standardize the number of likes, reposts and followers using z-score, mapping them uniformly to the [0, 1] interval; One-hot encoding was used for comment sentiment tags and user authentication status, and missing values were filled with the median. The standardized evaluation dataset is obtained by sorting the results by timestamp based on the standardized preprocessing results and generating unique evaluation IDs.
[0007] In some embodiments of this application, in S2, the relevance assessment of each evaluation item in the standardized evaluation dataset with the corresponding marketing feature information is performed, and the relevance score between each evaluation item and the marketing feature information is calculated, including: S21. Extract marketing feature information corresponding to the target marketing object from historical marketing logs, and normalize the marketing feature information to obtain a standard feature vector. S22. For each evaluation item based on the standardized evaluation dataset, a pre-trained language model is used to extract the text semantic vector, and the feature is fused with the standard feature vector. The cosine similarity algorithm is then used to calculate the relevance score.
[0008] In some embodiments of this application, in S2, the marketing feature information includes price, product function index, promotion investment index, and scenario-based feature index.
[0009] In some embodiments of this application, in S2, the expression for calculating the relevance score using the cosine similarity algorithm is as follows: ; in, and All are L2 norm. Let j be the baseline vector of the marketing feature. Let i be the fusion vector of the i-th evaluation item. The relevance score; The expression is: ; in, For text semantic vectors, For price perception value, For product functionality index, To promote the input index, For contextualized feature index; The expression is: ; in, This corresponds to the target value of the text semantic vector. For the price target value, The target value for the product functionality index. To promote the target value of the input index, This is the target value for the scenario-based feature index.
[0010] In some embodiments of this application, in step S3, marketing feature information that is higher than a preset relevance threshold is selected based on the relevance score and included as marketing features to be improved: S31. For any marketing feature j, if the maximum value of all evaluation items and relevance scores is lower than the preset relevance threshold, then the marketing feature is determined to be a candidate improvement feature. S32. Calculate the proportion of evaluation items with relevance scores higher than the preset relevance threshold out of the total number of evaluations. ,like If so, the marketing feature is determined to be a candidate improvement feature; S33. Take the union of the candidate improvement features obtained in steps S31 and S32, and remove the marketing features that do not conform to the company's future marketing goals to obtain the marketing features to be improved.
[0011] In some embodiments of this application, in step S4, the importance weights of the marketing features to be improved are configured based on a self-attention mechanism, and the importance weight values of each marketing feature to be improved are obtained as follows: The standard feature vector of each marketing feature to be improved is input into the feedforward network and uniformly mapped to a feature embedding vector of dimension d. ; Multiply each feature embedding vector by the trainable parameter matrix to obtain the query matrix Q, the key matrix K, and the value matrix V. By using scaled dot product attention, we obtain an m×m self-attention weight matrix. Where m is the total number of marketing features to be improved; For the self-attention weight matrix Calculate the average attention score for each feature relative to the other features by averaging across all columns. ,Will The value vectors of the corresponding features are weighted and fused together, and then Sigmoid normalization is performed. The importance weight values are then configured based on the normalization results.
[0012] In some embodiments of this application, in step S4, importance is evaluated based on importance weight values, and an importance ranking result is generated based on the importance evaluation result, including: The importance weight values of each marketing feature to be improved are determined as Level 1, Level 2, and Level 3 according to the preset segmentation standard. Within the same level, they are arranged in descending order of importance weight value. If the weights are the same, they are arranged in ascending order of feature coverage evaluation quantity and negative sentiment ratio.
[0013] The advantages and beneficial effects of this invention compared to the prior art are: 1. This invention introduces a quantifiable relevance score between standardized preprocessed evaluation items and marketing features, and sets a threshold filter to ensure that only features that truly trigger negative word-of-mouth enter the subsequent process; then, it uses a self-attention mechanism to allocate importance weights to the selected features, so that resources are concentrated on high relevance and high impact, and finally calls the preset marketing strategy table in order of importance, thereby improving the accuracy of importance ranking and the accuracy of applying improved marketing feature strategies.
[0014] 2. This invention parameterizes the "relevance threshold", "self-attention weight matrix" and "marketing strategy table", which can update the weights in real time after receiving new data. Users can quickly switch strategy packages according to scenarios such as new product launch, major promotion, and routine maintenance, which effectively improves the flexibility of strategy processing.
[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0016] Figure 1 This is a flowchart of an AI-powered intelligent marketing reputation monitoring method based on a social platform, as described in an embodiment of the present invention. Detailed Implementation
[0017] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.
[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0019] like Figure 1 As shown, this invention provides an AI-powered intelligent marketing reputation monitoring method based on social platforms, comprising the following steps: S1. Obtain multi-dimensional evaluation data of target marketing objects on social platforms, and perform standardized preprocessing on the evaluation data to obtain a standardized evaluation dataset.
[0020] S2. Evaluate the relevance between each evaluation item in the standardized evaluation dataset and the corresponding marketing feature information, and calculate the relevance score between each evaluation item and the marketing feature information.
[0021] S3. Based on the relevance score, select marketing feature information that is higher than the preset relevance threshold as marketing features to be improved.
[0022] S4. Based on the self-attention mechanism, assign importance weights to the marketing features to be improved, obtain the importance weight values of each marketing feature to be improved, evaluate the importance based on the importance weight values, and generate the importance ranking results based on the importance evaluation results.
[0023] S5. Compare and analyze the importance ranking results with the preset marketing strategy table to determine the improvement strategies corresponding to each marketing feature to be improved, and implement the improvement strategies in order of importance from high to low.
[0024] This invention introduces a quantifiable relevance score between standardized preprocessed evaluation items and marketing features, and sets a threshold filter to ensure that only features that truly trigger negative word-of-mouth enter the subsequent process. Then, it uses a self-attention mechanism to allocate importance weights to the selected features, so that resources are concentrated on high relevance and high impact. Finally, it calls the preset marketing strategy table in order of importance, which improves the accuracy of importance ranking and the accuracy of applying improved marketing feature strategies.
[0025] In some embodiments of this application, in step S1, multi-dimensional evaluation data of the target marketing object on the social platform is obtained, and the evaluation data is preprocessed to standardize it to obtain a standardized evaluation dataset, including: S11. Extract multi-dimensional evaluation data of target marketing objects from social platforms based on API interfaces. The multi-dimensional evaluation data includes the number of likes, the number of reposts, the number of followers, the sentiment tags of comments, and the user's authentication status. S12. Perform standardized preprocessing on the multi-dimensional evaluation data, and standardize the number of likes, reposts and followers using z-score, mapping them uniformly to the [0, 1] interval; One-hot encoding was used for comment sentiment tags and user authentication status, and missing values were filled with the median. The standardized evaluation dataset is obtained by sorting the results by timestamp based on the standardized preprocessing results and generating unique evaluation IDs.
[0026] In some embodiments, the original data scraped consisted of 28,600 short video comments from Douyin and 12,400 image and text note comments from Xiaohongshu, totaling 41,000 comments. An example field was: {text: "I got closed comedones after applying this for three days, it's too oily!", like: 420, repost: 38, fan: 12,000, emotion: -1, verified: 1, timestamp: 2024053008:12}. This field was preprocessed to generate a unique review ID: EV_20240530_0812_0001. The standardized vector was [0.68, 0.55, 0.71, 1, 0, 0, 1, 0]. A total of 41,000 standardized vectors were obtained within 48 hours, forming the dataset.
[0027] In some embodiments of this application, in S2, the relevance assessment of each evaluation item in the standardized evaluation dataset with the corresponding marketing feature information is performed, and the relevance score between each evaluation item and the marketing feature information is calculated, including: S21. Extract marketing feature information corresponding to the target marketing object from historical marketing logs, and normalize the marketing feature information to obtain a standard feature vector. S22. For each evaluation item based on the standardized evaluation dataset, a pre-trained language model is used to extract the text semantic vector, and the feature is fused with the standard feature vector. The cosine similarity algorithm is then used to calculate the relevance score.
[0028] In some embodiments of this application, in S2, the marketing feature information includes price, product function index, promotion investment index, and scenario-based feature index.
[0029] In some embodiments of this application, in S2, the expression for calculating the relevance score using the cosine similarity algorithm is as follows: ; in, and All are L2 norm. Let j be the baseline vector of the marketing feature. Let i be the fusion vector of the i-th evaluation item. The relevance score; The expression is: ; in, For text semantic vectors, For price perception value, For product functionality index, To promote the input index, For contextualized feature index; The expression is: ; in, This corresponds to the target value of the text semantic vector. For the price target value, The target value for the product functionality index. To promote the target value of the input index, This is the target value for the scenario-based feature index.
[0030] In some embodiments, marketing features (after historical log normalization): Price F1 = [0.42, 0.00, 0.00, 0.00]; Product function F2=[0.00, 0.88, 0.00, 0.00]; Promotional investment F3 = [0.00, 0.00, 0.75, 0.00]; Contextualized F4 = [0.00, 0.00, 0.00, 0.66]; Extract the text vector from EV_20240530_0812_0001, perform pre-training, and obtain the fused vector V. i =[0.63, 0.10, 0.82, 0.05, 0.08], the cosine similarity is calculated to be 0.954, and the similarity with price is 0.12, promotion is 0.06, and scene is 0.07. A full calculation was performed, with 41k features × 4 features = 164k scores, an average time of 6ms per feature, and a total time of 984s to complete.
[0031] In some embodiments of this application, in step S3, marketing feature information that is higher than a preset relevance threshold is selected based on the relevance score and included as marketing features to be improved: S31. For any marketing feature j, if the maximum value of all evaluation items and relevance scores is lower than the preset relevance threshold, then the marketing feature is determined to be a candidate improvement feature. S32. Calculate the proportion of evaluation items with relevance scores higher than the preset relevance threshold out of the total number of evaluations. ,like If so, the marketing feature is determined to be a candidate improvement feature; S33. Take the union of the candidate improvement features obtained in steps S31 and S32, and remove the marketing features that do not conform to the company's future marketing goals to obtain the marketing features to be improved.
[0032] In some embodiments, the preset relevance threshold is 0.45, and the statistics are as follows: 2100 posts, or 5.1%, had prices higher than 0.45 and were not included in the shortlist. Functionality: 19,600 entries, accounting for 47.8%, and max=0.954>0.45, thus qualifying; Promotions: 1800 items, accounting for 4.4%, shortlisted; Scenario: 2300, accounting for 5.6%, not included.
[0033] The core selling point of the promotion is "repair function," so "function" is retained. Although "promotion" is included, its budget is limited and it is not core, so it is temporarily placed in the observation pool. The final list of features to be improved is: product functions.
[0034] In some embodiments of this application, in step S4, the importance weights of the marketing features to be improved are configured based on a self-attention mechanism, and the importance weight values of each marketing feature to be improved are obtained as follows: The standard feature vector of each marketing feature to be improved is input into the feedforward network and uniformly mapped to a feature embedding vector of dimension d. ; Multiply each feature embedding vector by the trainable parameter matrix to obtain the query matrix Q, the key matrix K, and the value matrix V. By using scaled dot product attention, we obtain an m×m self-attention weight matrix. Where m is the total number of marketing features to be improved; For the self-attention weight matrix Calculate the average attention score for each feature relative to the other features by averaging across all columns. ,Will The value vectors of the corresponding features are weighted and fused together, and then Sigmoid normalization is performed. The importance weight values are then configured based on the normalization results.
[0035] In some embodiments of this application, in step S4, importance is evaluated based on importance weight values, and an importance ranking result is generated based on the importance evaluation result, including: The importance weight values of each marketing feature to be improved are determined as Level 1, Level 2, and Level 3 according to the preset segmentation standard. Within the same level, they are arranged in descending order of importance weight value. If the weights are the same, they are arranged in ascending order of feature coverage evaluation quantity and negative sentiment ratio.
[0036] In some embodiments, since there is only one feature, the self-attention degenerates to its own weight of 1.0, which is ≈0.952 after Sigmoid normalization. The importance is segmented as follows: >0.8 is level 1. The ranking result is: [{feature: product function, weight: 0.952, level: level 1}].
[0037] In some embodiments, correction S5 is performed: Strategy Mapping: Strategy package budget capped at 250,000 yuan, execution window 48 hours, channels: Douyin feed 60% + Xiaohongshu 40%; Matching priority: Priority = 0.952 × 46% (negative percentage) × 1.37 (popularity coefficient) = 0.60 (highest priority); Scheduling: Starts at 00:00 on June 1st, with a daily budget cap of 125,000; CMS automatically pulls templates to generate 120 short videos and 300 images and texts; the ad delivery system bids at 0.35 yuan / time CPC, with an expected exposure of 21 million; Tracking: Data retrieved at 0:00 on June 3rd showed 18,200 new reviews. The percentage of negative reviews related to the function decreased from 46% to 27%, a drop of 19% (>5%), which met the criteria. The system determined that no further revisions were needed, and this round of closed-loop processing was completed. The remaining budget of 32,000 was released to the general pool.
[0038] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. In case of any inconsistency, the meaning set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the scope of this application.
[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for monitoring marketing reputation based on social platforms using AI, characterized in that, Includes the following steps: S1. Obtain multi-dimensional evaluation data of target marketing objects on social platforms, and perform standardized preprocessing on the evaluation data to obtain a standardized evaluation dataset; S2. Evaluate the relevance between each evaluation item in the standardized evaluation dataset and the corresponding marketing feature information, and calculate the relevance score between each evaluation item and the marketing feature information. S3. Based on the relevance score, select marketing feature information that is higher than the preset relevance threshold as marketing features to be improved; S4. Based on the self-attention mechanism, assign importance weights to the marketing features to be improved, obtain the importance weight values of each marketing feature to be improved, evaluate the importance based on the importance weight values, and generate the importance ranking results based on the importance evaluation results. S5. Compare and analyze the importance ranking results with the preset marketing strategy table to determine the improvement strategies corresponding to each marketing feature to be improved, and implement the improvement strategies in order of importance from high to low.
2. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 1, characterized in that, In step S1, multi-dimensional evaluation data of the target marketing object on the social platform is obtained, and the evaluation data is preprocessed to standardize it to obtain a standardized evaluation dataset, including: S11. Extract multi-dimensional evaluation data of target marketing objects from social platforms based on API interfaces. The multi-dimensional evaluation data includes the number of likes, the number of reposts, the number of followers, the sentiment tags of comments, and the user's authentication status. S12. Perform standardized preprocessing on the multi-dimensional evaluation data, and standardize the number of likes, reposts and followers using z-score, mapping them uniformly to the [0, 1] interval; One-hot encoding was used for comment sentiment tags and user authentication status, and missing values were filled with the median. The standardized evaluation dataset is obtained by sorting the results by timestamp based on the standardized preprocessing results and generating unique evaluation IDs.
3. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 2, characterized in that, In step S2, the correlation between each evaluation item in the standardized evaluation dataset and its corresponding marketing feature information is assessed, and the correlation score between each evaluation item and the marketing feature information is calculated, including: S21. Extract marketing feature information corresponding to the target marketing object from historical marketing logs, and normalize the marketing feature information to obtain a standard feature vector; S22. For each evaluation item based on the standardized evaluation dataset, a pre-trained language model is used to extract the text semantic vector, and the feature is fused with the standard feature vector. The cosine similarity algorithm is then used to calculate the relevance score.
4. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 3, characterized in that, In S2, the marketing feature information includes price, product function index, promotion investment index, and scenario-based feature index.
5. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 4, characterized in that, In S2, the expression for calculating the relevance score using the cosine similarity algorithm is as follows: ； in, and All are L2 norm. Let j be the baseline vector of the marketing feature. Let i be the fusion vector of the i-th evaluation item. The relevance score; The expression is: ； in, For text semantic vectors, For price perception value, For product functionality index, To promote the input index, For contextualized feature indices; The expression is: ； in, This corresponds to the target value of the text semantic vector. For the price target value, The target value for the product functionality index. To promote the target value of the input index, This is the target value for the scenario-based feature index.
6. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 5, characterized in that, In step S3, marketing feature information that is higher than a preset relevance threshold is selected based on the relevance score and included as marketing features to be improved: S31. For any marketing feature j, if the maximum value of all evaluation items and relevance scores is lower than the preset relevance threshold, then the marketing feature is determined to be a candidate improvement feature. S32. Calculate the proportion of evaluation items with relevance scores higher than the preset relevance threshold out of the total number of evaluations. ,like If so, the marketing feature is determined to be a candidate improvement feature; S33. Take the union of the candidate improvement features obtained in steps S31 and S32, and remove the marketing features that do not conform to the company's future marketing goals to obtain the marketing features to be improved.
7. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 6, characterized in that, In step S4, importance weights are assigned to the marketing features to be improved based on a self-attention mechanism, resulting in importance weight values for each marketing feature to be improved, including: The standard feature vector of each marketing feature to be improved is input into the feedforward network and uniformly mapped to a feature embedding vector of dimension d. ; Multiply each feature embedding vector by the trainable parameter matrix to obtain the query matrix Q, the key matrix K, and the value matrix V. By using scaled dot product attention, we obtain an m×m self-attention weight matrix. Where m is the total number of marketing features to be improved; For the self-attention weight matrix Calculate the average attention score for each feature relative to the other features by averaging across all columns. ,Will The value vectors of the corresponding features are weighted and fused together, and then Sigmoid normalization is performed. The importance weight values are then configured based on the normalization results.
8. The AI-powered intelligent marketing reputation monitoring method based on a social platform according to claim 7, characterized in that, In step S4, importance is assessed based on importance weight values, and an importance ranking result is generated based on the importance assessment result, including: The importance weight values of each marketing feature to be improved are determined as Level 1, Level 2, and Level 3 according to the preset segmentation standard. Within the same level, they are arranged in descending order of importance weight value. If the weights are the same, they are arranged in ascending order of feature coverage evaluation quantity and negative sentiment ratio.