Social media sentiment analysis system and method based on semantic understanding
By introducing deep learning and real-time feedback mechanisms, the social media sentiment analysis system solves the problem of existing systems ignoring user behavior and interests, realizes personalized sentiment analysis and real-time monitoring, and improves the accuracy of sentiment analysis and platform efficiency.
Patent Information
- Application Number
- CN202510826194.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-19
- Publication Date
- 2025-11-18
AI Technical Summary
Existing social media sentiment analysis systems ignore users' historical behavior and interests, making it difficult to capture complex emotional expressions. Furthermore, they lack real-time feedback mechanisms, resulting in inaccurate and unpersonalized analysis results.
By introducing deep learning technology, user behavior analysis, and real-time feedback mechanisms, and combining them with natural language processing, a social media sentiment analysis system based on semantic understanding is constructed. This system includes modules for data collection, sentiment analysis, personalization, and real-time feedback. Personalized sentiment analysis and real-time monitoring are performed through deep learning models and user feature modeling.
It improves the accuracy and personalization of sentiment analysis, supports content recommendation, advertising optimization and public opinion management, enhances user experience and platform business benefits, and can respond to changes in user emotions in a timely manner to prevent public opinion crises.
Smart Images
Figure CN120974017A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of sentiment analysis, in particular to a social media sentiment analysis system and method based on semantic understanding. BACKGROUND
[0002] With the rapid development of social media, more and more users share their personal opinions, emotions and experiences through the platform. A large amount of user-generated content such as comments, posts, microblogs and short videos on social media platforms provides valuable data sources for sentiment analysis. The purpose of social media sentiment analysis is to understand and classify the sentiment tendency in social media text through natural language processing technology, helping enterprises or platforms make more accurate user behavior prediction, content recommendation and advertisement optimization. However, existing sentiment analysis systems usually only consider the sentiment tendency of the text itself, ignoring individual factors such as user historical behavior and interest preferences. At the same time, social media content has rich contextual information and emotional details, and traditional sentiment analysis methods often have difficulty capturing these complex emotional expressions.
[0003] To solve these problems, the social media sentiment analysis system based on semantic understanding can more accurately analyze text sentiment and provide personalized customized analysis results by introducing deep learning technology, user behavior analysis and real-time feedback mechanism. This system not only can make content recommendations based on sentiment analysis results, but also can monitor public opinion trends on social media platforms in real time, and timely discover and intervene negative emotions. SUMMARY
[0004] To solve the above technical problems, the social media sentiment analysis system and method based on semantic understanding are provided, which solve the above problems.
[0005] To achieve the above purposes, the technical scheme adopted by the present application is: The social media sentiment analysis system and method based on semantic understanding include: Data acquisition module: used for collecting text data in social media platform and preprocessing Sentiment analysis module: the sentiment analysis module is electrically connected with the data acquisition module, and the sentiment analysis module is used for sentiment classification of text data based on natural language processing technology combined with deep learning model, and outputs sentiment analysis results; Personalized customization module: the personalized customization module is electrically connected with the sentiment analysis module, and the personalized customization module is used for personalized customization of sentiment analysis results based on user historical social behavior and interest field data; Real-time feedback module: The real-time feedback module is electrically connected with the personalized customization module, the sentiment analysis module and the data collection module, and is used for feeding back the sentiment analysis result to the social media platform in real time to support subsequent user behavior prediction, content recommendation and advertisement optimization.
[0006] Preferably, the sentiment analysis module comprises: A preprocessing unit for cleaning, removing noise and word segmentation preprocessing operation on social media text data; A model training unit for training the sentiment analysis model based on a deep learning algorithm to identify sentiment information in the text; A sentiment classification unit for classifying the preprocessed text data and outputting sentiment labels and sentiment scores; Wherein, the sentiment score calculation formula is:
[0007] In the formula, is the score of the sentiment category , and is the predicted probability of the sentiment category of the given text .
[0008] Preferably, the personalized customization module comprises: A user feature modeling unit for establishing a user's emotional feature model and performing sentiment preference analysis based on user historical social behavior data; Wherein, the user preference modeling formula is:
[0009] In the formula, is the user's sentiment preference score, is the user's preference score in the th emotional field, is the weight of the emotional field, indicating the user's attention to the field; A sentiment customization unit adjusts the output of the sentiment analysis model according to the user's personalized needs and emotional preferences to improve the accuracy of sentiment classification.
[0010] Preferably, the real-time feedback module comprises: A content recommendation unit for pushing content and advertisements that meet the user's emotional preferences based on real-time sentiment analysis results; An public opinion monitoring unit for real-time monitoring of public opinion trends on social media platforms, and timely discovering and intervening negative emotions.
[0011] The method for sentiment analysis of social media based on semantic understanding comprises: Collecting text data in a social media platform, preprocessing the data based on natural language processing techniques, extracting keywords and sentiment lexicon; Based on the sentiment lexicon, combined with a deep learning model for sentiment classification, analyzing the sentiment tendency in the text; Constructing a sentiment analysis model, training based on user semantic sentiment labels, capturing sentiment features in the text, and outputting corresponding sentiment classification results; Combine the sentiment analysis results with user features to customize the sentiment analysis; Based on the analysis results, real-time feedback to the social media platform.
[0012] Preferably, the construction of the sentiment analysis model, based on the user's semantic sentiment label, captures the sentiment features in the text, and outputs the corresponding sentiment classification results, specifically includes: Based on the pre-training language model in deep learning, pre-train the social media text data, and further fine-tune the model for sentiment analysis; Based on the sentiment lexicon, construct word vectors, map sentiment lexicon to sentiment space, and construct feature vectors with sentiment context; Use bidirectional long short-term memory network to model text sequence, cross-learn sentiment lexicon and context information, and accurately capture sentiment tendency; Combine the advantages of convolutional neural network and LSTM to build a hybrid model; During the training process, use sentiment labels as target output, combine cross-entropy loss function for training, and ensure the accuracy of sentiment classification.
[0013] Preferably, the individualization of the sentiment analysis results specifically includes: Based on user historical social behavior data, establish user sentiment feature model, capture user's emotional performance in different situations; Combine the user's interest field and social circle information to speculate the user's emotional preference in a specific topic, and customize the sentiment analysis results; Use collaborative filtering algorithm to compare and analyze the emotional performance of similar user groups, and improve the accuracy of individualized analysis results; Adjust the output of the sentiment analysis model by analyzing the user's feedback on different emotional content.
[0014] Preferably, the text data in the social media platform includes: User comments, posts, microblogs and text descriptions of short videos in the social media platform; Based on the sentiment keywords and context information in the text, combined with the event topics in the social media, mine the hidden sentiment tendency in the text; The text sentiment is labeled by using label data provided by a social media platform, including topic labels and sentiment labels, and a sentiment analysis model is trained.
[0015] Preferably, the training method of the sentiment analysis model specifically comprises: A training data set is constructed, including social media text data labeled with sentiment labels, and is divided into a training set, a validation set and a test set; The training data set is expanded by a data enhancement technique to increase the generalization ability of the sentiment analysis model; The accuracy of sentiment classification is improved by combining supervised learning and semi-supervised learning, and the adaptability of the model in different industry scenarios is improved by transfer learning technology; The parameters of the sentiment analysis model are optimized using a hybrid optimization algorithm to minimize the error of sentiment classification.
[0016] Preferably, the specific way of real-time feedback of the sentiment analysis result comprises: The sentiment analysis result and the sentiment score are fed back to the social media platform in real time; Based on the real-time sentiment analysis result, personalized content recommendation is carried out, including pushing articles, videos and advertisements that meet the user's emotional preference; The sentiment analysis result is used to monitor the social public opinion of the platform in real time, discover potential negative emotions and intervene in time.
[0017] Compared with the prior art, the beneficial effects of the present application are: The social media sentiment analysis system based on semantic understanding proposed in the present application introduces deep learning, personalized customization, real-time feedback and public opinion monitoring modules on the basis of existing sentiment analysis technology, which can more accurately identify emotional information in text and provide personalized customized results. This not only improves the accuracy of sentiment analysis, but also effectively supports content recommendation, advertisement optimization and public opinion management, providing more valuable analysis services for social media platforms. At the same time, the real-time feedback and dynamic adjustment capability of the system ensures that the platform can respond to changes in user emotions in time, further improving user experience and business benefits of the platform. BRIEF DESCRIPTION OF DRAWINGS
[0018] Figure 1 The system framework diagram of the present application; Figure 2 The step flow framework diagram of the present application. DETAILED DESCRIPTION
[0019] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only as examples, and other obvious modifications can be thought of by those skilled in the art.
[0020] Referring to Figure 1 The social media sentiment analysis system and method based on semantic understanding includes: Data collection module: for collecting text data in social media platforms and preprocessing it Sentiment analysis module: the sentiment analysis module is electrically connected with the data collection module, and the sentiment analysis module is used for sentiment classification of text data based on natural language processing technology combined with deep learning model, and outputs sentiment analysis results; Personalized customization module: the personalized customization module is electrically connected with the sentiment analysis module, and the personalized customization module is used for personalized customization of sentiment analysis results based on user's historical social behavior and interest field data; Real-time feedback module: the real-time feedback module is electrically connected with the personalized customization module, the sentiment analysis module and the data collection module, and the real-time feedback module is used for real-time feedback of sentiment analysis results to social media platforms to support subsequent user behavior prediction, content recommendation and advertisement optimization.
[0021] The sentiment analysis module includes: Preprocessing unit, for cleaning, removing noise and word segmentation preprocessing operation of social media text data; Model training unit: based on deep learning algorithm, the sentiment analysis model is trained to identify the sentiment information in the text; Sentiment classification unit: for sentiment classification of preprocessed text data, and outputting sentiment label and sentiment score; Wherein, the sentiment score calculation formula is:
[0022] In the formula, is the score of the sentiment category , is the prediction probability of the sentiment category of the given text By analyzing the user's historical social behavior and interest field data, the personalized customization module can adjust the analysis results based on the sentiment preference of each user. This personalized sentiment analysis not only improves the accuracy of sentiment classification, but also effectively supports personalized content recommendation and advertisement placement, thereby improving user satisfaction and platform business benefits.
[0023] The personalized customization module includes: User feature modeling unit: for establishing user's sentiment feature model, and performing sentiment preference analysis based on user historical social behavior data; Wherein, the user preference modeling formula is:
[0024] wherein, is the user's emotional preference score, is the user's preference score in the emotional domain, is the weight of the emotional domain, indicating the user's attention to the domain; The emotional customization unit adjusts the output of the emotional analysis model according to the user's personalized needs and emotional preferences to improve the accuracy of emotional classification. The deep learning-based emotional analysis model can accurately identify emotional features in text. By using advanced pre-trained language models, bidirectional long short-term memory networks, and convolutional neural networks, the system can deeply understand the emotional expression in the text, improving the accuracy and generalization ability of emotional classification.
[0025] The real-time feedback module includes: Content recommendation unit: based on real-time emotional analysis results, push content and advertisements that meet the user's emotional preferences; Public opinion monitoring unit: real-time monitoring of public opinion trends on social media platforms, timely detection of negative emotions and intervention.
[0026] Referring to Figure 2 The social media emotional analysis method based on semantic understanding includes: Collecting text data from social media platforms, pre-processing the data based on natural language processing technology, extracting keywords and emotional vocabulary; Based on emotional vocabulary, combined with deep learning model for emotional classification, analyze the emotional tendency of the text; Constructing an emotional analysis model, training based on user semantic emotional labels, capturing emotional features in text, and outputting corresponding emotional classification results; Combine emotional analysis results with user features to customize emotional analysis; Based on the analysis results, real-time feedback to the social media platform.
[0027] Constructing an emotional analysis model, training based on user semantic emotional labels, capturing emotional features in text, and outputting corresponding emotional classification results specifically includes: Based on the pre-trained language model in deep learning, pre-train the social media text data, and further fine-tune the model for emotional analysis; Based on emotional vocabulary, construct word vectors, map emotional vocabulary to emotional space, and construct feature vectors with emotional context; Use bidirectional long short-term memory network to model text sequence, cross-learn emotional vocabulary and context information, and accurately capture emotional tendency; Combine the advantages of convolutional neural networks and LSTM to build a hybrid model. During training, use sentiment labels as target output and combine cross-entropy loss function to ensure the accuracy of sentiment classification.
[0028] The personalized customization of sentiment analysis results specifically includes: Establish a user sentiment feature model based on historical social behavior data to capture users' emotional expressions in different situations. Combine users' interest areas and social circle information to infer users' emotional preferences on specific topics and customize sentiment analysis results. Use collaborative filtering algorithms to compare and analyze the emotional expressions of similar user groups to improve the accuracy of personalized analysis results. Adjust the output of the sentiment analysis model by analyzing users' feedback on different emotional content.
[0029] Text data in social media platforms includes: User comments, posts, microblogs, and text descriptions of short videos on social media platforms. Based on emotional keywords and contextual information in the text, combined with event topics in social media, extract hidden emotional tendencies in the text. Use the label data provided by social media platforms, including topic labels and emotional labels, to label text sentiment and assist in training the sentiment analysis model.
[0030] The training method of the sentiment analysis model specifically includes: Build a training data set, including social media text data labeled with emotional labels, and divide it into training set, validation set and test set. Expand the training data set through data augmentation techniques to increase the generalization ability of the sentiment analysis model. Combine supervised learning and semi-supervised learning to improve the accuracy of sentiment classification, and use transfer learning techniques to improve the adaptability of the model in different industry scenarios. Use hybrid optimization algorithms to optimize the parameters of the sentiment analysis model to minimize sentiment classification errors.
[0031] The specific ways of real-time feedback of sentiment analysis results include: Real-time feedback of sentiment analysis results and sentiment scores to social media platforms. Based on real-time sentiment analysis results, conduct personalized content recommendation, including pushing articles, videos and advertisements that meet users' emotional preferences. Use sentiment analysis results to monitor social public opinion in real time, discover potential negative emotions and intervene in time. By combining the text data, user comments, posts and label data and other information of the social media platform, the system can effectively mine the sentiment tendency hidden in the text, and through the label data to assist the training of the sentiment analysis model, the precision of the sentiment analysis model is improved.
[0032] In summary, the advantages of the present application are: The sentiment analysis model based on deep learning can accurately identify the emotional features in the text, and by using the advanced pre-training language model, the bidirectional long short-term memory network and the convolutional neural network hybrid model, the system can deeply understand the emotional expression in the text, improve the accuracy and generalization ability of the sentiment classification; Through the analysis of the user's historical social behavior and interest field data, the personalized customization module can adjust the analysis results based on the emotional preference of each user, this personalized sentiment analysis not only improves the accuracy of sentiment classification, but also effectively supports personalized content recommendation and advertisement placement, thereby improving the user's satisfaction and the platform's business benefits; The real-time feedback module ensures that the sentiment analysis results can be quickly fed back to the social media platform, supporting real-time public opinion monitoring and intervention, the system can dynamically adjust the recommended content according to the sentiment analysis results, push articles, videos and advertisements that meet the user's emotional preferences, effectively improving the platform user's interaction rate; The public opinion monitoring unit can track the sentiment trend on the social media in real time, automatically detect potential negative emotions and intervene in time, which helps the social platform to take prompt measures when the user's emotions appear adverse fluctuations, avoiding the spread of public opinion crisis; By combining the text data, user comments, posts and label data and other information of the social media platform, the system can effectively mine the sentiment tendency hidden in the text, and through the label data to assist the training of the sentiment analysis model, the precision of the sentiment analysis model is improved. Combined with the personalized analysis results for content recommendation and advertisement placement, it helps to improve the user experience, reduce the interference of advertising information, so that users can see more content that meets their emotional needs, thereby enhancing user stickiness and platform activity.
[0033] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above examples, and the above examples and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.
Claims
1. A social media sentiment analysis system based on semantic understanding, characterized in that, include: Data Acquisition Module: Used to collect text data from social media platforms and preprocess it. Sentiment Analysis Module: The sentiment analysis module is electrically connected to the data acquisition module. The sentiment analysis module is used to classify the sentiment of text data based on natural language processing technology and combined with a deep learning model, and output the sentiment analysis results. Personalized Customization Module: The personalized customization module is electrically connected to the sentiment analysis module. The personalized customization module is used to personalize the sentiment analysis results based on the user's historical social behavior and interest area data. Real-time feedback module: The real-time feedback module is electrically connected to the personalization module, the sentiment analysis module, and the data acquisition module. The real-time feedback module is used to feed the sentiment analysis results back to the social media platform in real time to support subsequent user behavior prediction, content recommendation, and advertising optimization.
2. The social media sentiment analysis system based on semantic understanding according to claim 1, characterized in that, The sentiment analysis module includes: The preprocessing unit is used for cleaning, noise removal, and word segmentation preprocessing of social media text data; Model training unit: Trains the sentiment analysis model based on deep learning algorithms to identify sentiment information in text; Sentiment classification unit: used to classify the preprocessed text data according to sentiment and output sentiment labels and sentiment scores; The formula for calculating the emotional score is as follows: In the formula, It is an emotion category The score, It is a given text Emotional categories The predicted probability.
3. The social media sentiment analysis system based on semantic understanding according to claim 2, characterized in that, The personalized customization module includes: User Feature Modeling Unit: Used to build a user's sentiment feature model and perform sentiment preference analysis based on the user's historical social behavior data; The user preference modeling formula is as follows: In the formula, It is the user's sentiment preference score. Is the user in the Preference ratings in the emotional domain, It represents the weight of this emotional domain, indicating the user's level of attention to this domain; The emotion customization unit adjusts the output of the emotion analysis model based on the user's personalized needs and emotional preferences to improve the accuracy of emotion classification.
4. The social media sentiment analysis system based on semantic understanding according to claim 3, characterized in that, The real-time feedback module includes: Content recommendation unit: Based on real-time sentiment analysis results, pushes content and advertisements that match users' sentiment preferences; Public opinion monitoring unit: Monitors public opinion trends on social media platforms in real time, promptly identifies negative sentiments, and intervenes accordingly.
5. A social media sentiment analysis method based on semantic understanding, characterized in that, include: Text data is collected from social media platforms, preprocessed using natural language processing technology, and keywords and sentiment words are extracted. Based on sentiment vocabulary, combined with deep learning models, sentiment classification is performed to analyze the sentiment tendency in the text. Construct a sentiment analysis model, train it based on the user's semantic sentiment tags, capture the sentiment features in the text, and output the corresponding sentiment classification results; By combining sentiment analysis results with user characteristics, personalized sentiment analysis can be achieved. Based on the analysis results, feedback is sent to social media platforms in real time.
6. The social media sentiment analysis method based on semantic understanding according to claim 5, characterized in that, The construction of the sentiment analysis model, trained based on the user's semantic sentiment tags, captures sentiment features in the text and outputs corresponding sentiment classification results, specifically includes: Pre-training of social media text data is performed based on pre-trained language models in deep learning, and the model is further fine-tuned by sentiment analysis. Based on emotional vocabulary, word vectors are constructed, and emotional vocabulary is mapped to the emotional space to construct feature vectors with emotional context. A bidirectional long short-term memory network is used to model text sequences, and sentiment tendencies are accurately captured through cross-learning of sentiment words and contextual information. By combining the advantages of convolutional neural networks and LSTM, a hybrid model is constructed. During training, sentiment labels are used as the target output, and training is performed using the cross-entropy loss function to ensure the accuracy of sentiment classification.
7. The social media sentiment analysis method based on semantic understanding according to claim 6, characterized in that, The personalized customization of the sentiment analysis results specifically includes: A user emotional characteristic model is built based on users' historical social behavior data to capture users' emotional expressions in different situations. By combining users' areas of interest and social circle information, we can infer users' emotional preferences on specific topics and customize the sentiment analysis results. Collaborative filtering algorithms are used to compare and analyze the emotional expressions of similar user groups, thereby improving the accuracy of personalized analysis results; By analyzing user feedback on different emotional content, the output of the sentiment analysis model can be adjusted.
8. The social media sentiment analysis method based on semantic understanding according to claim 7, characterized in that, The text data in the social media platform includes: Text descriptions of user comments, posts, tweets, and short videos on social media platforms; Based on the emotional keywords and contextual information in the text, combined with the topics of events on social media, we can uncover the hidden emotional tendencies in the text. By utilizing tag data provided by social media platforms, including topic tags and sentiment tags, the sentiment of text is labeled and used to assist in training the sentiment analysis model.
9. The social media sentiment analysis method based on semantic understanding according to claim 8, characterized in that, The training method for the sentiment analysis model specifically includes: Construct a training dataset, which includes social media text data labeled with sentiment tags, and divide it into training, validation and test sets; Data augmentation techniques can be used to expand the training dataset and increase the generalization ability of the sentiment analysis model. By combining supervised and semi-supervised learning methods, the accuracy of sentiment classification can be improved, and the adaptability of the model in different industry scenarios can be enhanced through transfer learning techniques. Hybrid optimization algorithms are used to optimize the parameters of the sentiment analysis model, thereby minimizing the error in sentiment classification.
10. The social media sentiment analysis method based on semantic understanding according to claim 9, characterized in that, The specific methods for real-time feedback of the sentiment analysis results include: The sentiment analysis results and sentiment scores are fed back to social media platforms in real time; Based on real-time sentiment analysis results, personalized content recommendations are made, including pushing articles, videos and advertisements that match users' sentiment preferences; By using sentiment analysis results, we can monitor social media sentiment on the platform in real time, identify potential negative emotions, and intervene in a timely manner.