A social media negative sentiment recognition method based on generative artificial intelligence

Abstract

This invention discloses a method for identifying negative emotions on social media based on generative artificial intelligence, belonging to the field of artificial intelligence. The method includes: collecting text data from social media, including posts, comments, and messages; preprocessing the collected text data using a generative artificial intelligence model, the preprocessing steps including text cleaning, word segmentation, stop word removal, and semantic annotation; training a negative emotion recognition model using the preprocessed text data; and using the trained negative emotion recognition model to identify negative emotions in the text data from social media. Simultaneously, it also achieves emotion recognition for image, audio, and video data based on multimodal data fusion technology. This invention improves the accuracy, real-time performance, efficiency, and comprehensiveness of negative emotion recognition in social media, providing users with better mental health protection.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence. Specifically, this invention relates to a method for recognizing negative emotions on social media based on generative artificial intelligence. Background Technology

[0002] Natural Language Processing (NLP) is an important branch of artificial intelligence that aims to enable computers to understand, interpret, and generate human language for effective human-computer communication. Sentiment analysis, as a key application area of ​​NLP, involves extracting emotional information from text and identifying and processing its emotional tendencies.

[0003] In today's internet age, social media has become an important platform for people to express emotions, share experiences, and express opinions. With the rapid growth of social media content, effectively identifying and analyzing users' emotional states, especially negative emotions, from massive amounts of data has become an important research and application area. This invention provides an innovative method that, by combining generative artificial intelligence technology, can more accurately and efficiently identify and analyze negative emotions in social media, which is of great significance for maintaining a healthy online environment and promoting users' mental health.

[0004] With the rapid development of internet technology, social media has become a major platform for the public to exchange information and express emotions. However, the widespread dissemination of negative emotions on social media can negatively impact users' mental health. Therefore, accurately and efficiently identifying and analyzing negative emotions on social media is of great significance for maintaining a healthy online environment and promoting user well-being.

[0005] Traditional sentiment analysis methods are primarily based on rules or statistical models, such as sentiment dictionaries and basic text analysis techniques. These methods have limitations when dealing with complex, ambiguous, or implicit sentiment expressions, especially in dynamic and informal linguistic environments like social media, where they struggle to achieve efficient and accurate recognition. To improve accuracy and efficiency, researchers have begun exploring deep learning-based methods, which can learn and simulate the human sentiment recognition process, handling more complex emotional and semantic information.

[0006] Generative artificial intelligence, especially deep learning models such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs), has been widely applied to sentiment analysis of text, image, and speech data. These models, by learning from large amounts of data, are able to better understand and interpret human language and emotions, thereby improving the accuracy and efficiency of sentiment recognition.

[0007] However, even deep learning-based methods face challenges such as the quality and quantity of training data, the model's generalization ability, and real-time processing capabilities. Furthermore, the multimodal nature of social media data (a combination of text, images, audio, and video) places even greater demands. Therefore, a comprehensive approach that considers these factors is needed to effectively handle negative sentiment information in social media. Summary of the Invention

[0008] This invention aims to overcome the shortcomings of existing technologies by using innovative technical solutions and algorithm optimizations to improve the real-time performance, comprehensiveness, accuracy, and efficiency of negative emotion recognition, thereby enhancing the user experience of social media platforms and the health of the online environment.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: a method for identifying negative sentiment on social media based on generative artificial intelligence, the method comprising the following steps:

[0010] S1. Collect text data from social media, including posts, comments, and messages from social media;

[0011] S2. Use a generative artificial intelligence model to preprocess the collected text data. The preprocessing steps include text cleaning, word segmentation, stop word removal, and semantic annotation.

[0012] S3. Train a negative sentiment recognition model using preprocessed text data;

[0013] S4. Use a trained negative sentiment recognition model to identify negative sentiment in text data from social media.

[0014] Furthermore, in step S2, the generative artificial intelligence model is a generative adversarial network (GAN), which includes a generator and a discriminator. The generator generates synthetic text data, and the discriminator distinguishes between the generated synthetic text data and real text data. The generator G and the discriminator D are trained using the following formula:

[0015] G(z) = Generate synthetic text data.

[0016] D(x) = Determines whether the input data x is real text data.

[0017] Where z is the input noise vector of the generator, and x is the text data input to the discriminator. Furthermore, the generative adversarial network is trained using the following loss function:

[0018]

[0019] Where, p data(x) is the distribution of the real data, p z (z) is the distribution of the generator input noise, and E represents the desired operation.

[0020] Furthermore, in step S3, the negative emotion recognition model is a deep neural network, which includes convolutional neural networks, recurrent neural networks, and long short-term memory networks. The convolutional neural network is used to extract local features from the text data, as expressed by the following formula:

[0021]

[0022] in, These are elements in the feature map after the l-th convolutional layer, where σ is the activation function and b (l) It is a bias term. The weights of the convolution kernel, x i+m,j+n These are elements of the input feature map; the recurrent neural network and long short-term memory network are used to process time series information in text data, and the recurrent neural network is expressed by the following formula:

[0023] s t =f(U·x t +W·s t-1 )

[0024] o t =g(V·s) t )

[0025] Among them, s t In the hidden state at time t, o t It is the output layer, x t The input at time t is defined by U, V, and W, which are weight matrices, and f and g are activation functions. The Long Short-Term Memory (LSTM) network is expressed by the following formula:

[0026] f t =σ(W f ·[h t-1 ,x t ]+b f ),

[0027] i t =σ(W i ·[h t-1 ,x t ]+b i ),

[0028]

[0029]

[0030] o t =σ(Wo ·[h t-1 ,x t ]+b o ), h t =o t *tanh(C t ),

[0031] Among them, f t i t o t These are the activation values ​​for the forget gate, input gate, and output gate, respectively. It is a candidate cell state, C t It represents the cell state at the current time step, h. t It represents the hidden state, where W and b are the weight matrix and bias term, respectively, and * denotes the Hadamard product.

[0032] Furthermore, in step S3, a weighted loss function is applied when training the negative emotion recognition model to balance the uneven distribution of various emotion samples in the dataset. The weighted loss function is expressed by the following formula:

[0033]

[0034] in, Here, y is the loss function, and y is the true label. These are predicted labels, where C is the total number of categories, and w is the total number of categories. c It is the weight corresponding to category c, y c and These are the values ​​of category c in the true label and the predicted label, respectively.

[0035] Furthermore, in step S3, a negative sentiment recognition model is trained using support vector machines, decision trees, and random forest algorithms, wherein:

[0036] (1) The support vector machine is used to find the optimal decision boundary in high-dimensional space to distinguish different sentiment categories, as expressed by the following formula:

[0037]

[0038] sty i (w·φ(x i )+b)≥1-ξ i ,ξ i ≥0;

[0039] Where w is the weight vector, b is the bias term, and φ(x) is the weight vector. i ξ is a function that maps input data to a high-dimensional space, C is the regularization parameter, and ξ is the regularization parameter. i It is a slack variable;

[0040] (2) The decision tree classifies the data by splitting the dataset layer by layer to form a tree-like decision path;

[0041] (3) The random forest consists of multiple decision trees, and the final classification result is determined by a voting mechanism.

[0042] Furthermore, step S4 includes:

[0043] (1) Real-time data stream processing: Build a real-time data stream processing framework to capture text data streams from social media platforms in real time;

[0044] (2) Negative sentiment recognition: The trained negative sentiment recognition model is used to analyze the real-time captured text data to determine the sentiment attributes of the text, expressed by the following formula:

[0045]

[0046] Where E(x) represents the sentiment score of text x, w i It is word x i The weights, S(x) i ) is the word x i Emotional polarity score;

[0047] (3) Provide real-time feedback and intervention mechanisms based on the results of negative emotion identification;

[0048] (4) Data update and model optimization: The negative emotion recognition model is optimized by new data collected during real-time data stream processing. The optimization includes adjusting model parameters and optimizing the training set.

[0049] Furthermore, to further optimize the method, the method also includes comprehensively analyzing emotional information in text, image, audio, and video data based on multimodal data fusion technology, with the following steps:

[0050] (1) Multimodal data processing: Processing data of different modalities in social media and extracting emotional features from the data, wherein the data of different modalities includes text, image, audio and video data;

[0051] (2) Data fusion: Data fusion technology is used to combine data features from different modalities, which can be expressed by the following formula:

[0052] F = γ1·F text +γ2·F image +γ3·F audio +γ4·F video ,

[0053] Where F is the fused feature vector, F text F imageF audio and F video These are the feature vectors for text, image, audio, and video, respectively, and γ1, γ2, γ3, and γ4 are the fusion weights for each modality of data;

[0054] (3) Comprehensive sentiment analysis: Based on the fused feature vector, a negative sentiment recognition model is used to perform comprehensive sentiment analysis.

[0055] Based on the above optimizations, step S4 further includes negative emotion recognition of image, audio, and video data from social media, specifically as follows: For image data, a convolutional neural network is used to extract features from the image data. The convolutional neural network automatically learns image features through multi-layer filters and can identify negative emotional expressions in images, such as sad expressions or somber tones, which can be expressed by the following formula:

[0056] f(x) = σ(W*x + b),

[0057] Where f(x) is the feature extraction function, W is the weight matrix of the convolution kernel, * denotes the convolution operation, b is the bias term, and σ is the activation function. For audio data processing, a recurrent neural network is used to analyze the tone and rhythm of speech in the audio data to identify the emotional coloring in the speech. For example, a fast and sharp speech rhythm may express anger or anxiety. For video data, the aforementioned image and audio data processing methods are combined to analyze the visual and auditory information in the video and comprehensively identify negative emotions in the video. Specifically, video data processing involves multimodal data analysis, requiring the simultaneous processing of visual and auditory information and their combination for comprehensive judgment.

[0058] Furthermore, the method includes analyzing user behavior data in social media using a negative emotion recognition model, and determining whether a user has negative emotions based on the analysis results, as follows:

[0059] (1) Collection of user behavior data: Collect user behavior data on social media, including posting frequency, interaction patterns and online time;

[0060] (2) Behavioral pattern analysis: Apply machine learning algorithms to perform pattern analysis on user behavior data to identify behavioral patterns that indicate negative emotions. The machine learning algorithms include cluster analysis and sequence pattern mining techniques.

[0061] (3) Negative emotion prediction: Combine user behavior patterns and data analysis results related to users to comprehensively predict users' negative emotions. Data related to users includes text, pictures, audio and video.

[0062] (4) Application of prediction results: Based on the prediction results, provide corresponding support and intervention measures for users with negative emotions. These measures include pushing positive content to users, recommending professional psychological counseling services to users, managing and supervising social media platforms, and promptly deleting harmful information based on the negative emotion identification results.

[0063] The technical effects of this invention are as follows: This invention adopts advanced generative artificial intelligence technology and combines deep learning algorithms to achieve efficient and accurate identification of negative emotions in social media, enhances the adaptability and robustness of the model, and is suitable for the rapidly changing social media environment. Specifically: (1) This invention adopts deep learning models such as convolutional neural networks (CNN), recurrent neural networks (RNN) and long short-term memory networks (LSTM). These models can effectively process complex text data and extract emotional features from it; (2) In the data preprocessing stage, this invention adopts text cleaning, word segmentation, stop word removal and part-of-speech tagging and other technologies, which effectively improves the quality and efficiency of subsequent model training; (3) In order to solve the problem of imbalanced training data, this invention adopts a weighted loss function and gives higher weights to the fewer emotion categories to improve the overall recognition performance of the model; (4) This invention not only processes text data, but also processes multimodal data such as images, audio and video. By comprehensively analyzing the emotional information in these data, it provides more comprehensive negative emotion recognition; (5) This invention can monitor the data flow of social media platforms in real time, identify and respond to negative emotions in a timely manner, and effectively reduce the spread and impact of bad information. Attached Figure Description

[0064] Figure 1 This is a flowchart of a convolutional neural network for a social media negative sentiment recognition method based on generative artificial intelligence, according to the present invention.

[0065] Figure 2 This is a schematic diagram of the basic recurrent neural network for a social media negative sentiment recognition method based on generative artificial intelligence according to the present invention.

[0066] Figure 3 This is a schematic diagram of a recurrent neural network unfolded chronologically, representing a social media negative sentiment recognition method based on generative artificial intelligence according to the present invention. Detailed Implementation

[0067] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings, in order to help those skilled in the art to have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention, and to facilitate its implementation.

[0068] This invention uses a computer as a platform to identify negative emotions present in social media. This invention is efficient and flexible, adaptable to the ever-changing social media environment, and provides users and platforms with accurate negative emotion identification and timely response mechanisms. Specifically, a method for identifying negative emotions on social media based on generative artificial intelligence includes the following steps:

[0069] S1. Collect text data from social media, including posts, comments, and messages. The text data must also include timestamps and user identification information for subsequent sentiment trend analysis and user behavior research.

[0070] S2. The collected text data is preprocessed using a generative artificial intelligence model. The preprocessing steps include text cleaning, word segmentation, stop word removal, and semantic annotation. During this process, the text data is transformed into a format that the model can process, such as vectors or other mathematical representations. Further, in step S2, the generative artificial intelligence model is a generative adversarial network (GAN), which includes a generator and a discriminator. The generator generates synthetic text data that is similar to real social media text data; the discriminator distinguishes between the generated synthetic text data and real text data. The generator G and discriminator D are trained using the following formula:

[0071] G(z) = Generate synthetic text data.

[0072] D(x) = Determines whether the input data x is real text data.

[0073] Where z is the input noise vector of the generator, and x is the text data input to the discriminator. During the training of the generative adversarial network (GAN), the generator and discriminator are optimized adversarially. The generator aims to produce text data that is as realistic as possible to deceive the discriminator, while the discriminator aims to accurately distinguish between real and generated data. The GAN is trained using the following loss function:

[0074]

[0075] Where, p data (x) is the distribution of the real data, p z (z) represents the distribution of the generator input noise, and E represents the desired operation. This generative adversarial network not only enhances the model's generalization ability but also improves the accuracy of negative sentiment recognition. Especially when dealing with social media texts with complex semantics and implicit emotions, the generative adversarial network can effectively simulate and recognize these complex data patterns.

[0076] S3. Train a negative sentiment recognition model using the preprocessed text data. The negative sentiment recognition model is a deep neural network, which includes convolutional neural networks, recurrent neural networks, and long short-term memory networks. The convolutional neural network is used to extract local features from the text data. The convolutional neural network processes the text data through a series of convolutional layers and pooling layers. Each convolutional layer contains multiple convolutional kernels, which can extract different features of the text, as expressed by the following formula:

[0077]

[0078] in, These are elements in the feature map after the l-th convolutional layer, where σ is the activation function and b (l) It is a bias term. The weights of the convolution kernel, x i+m,j+n These are elements of the input feature map, such as... Figure 1 As shown, the process of a convolutional neural network is as follows: First, after initializing the weights, given an input vector and a target output, the input data is passed through hidden layers and output layers, i.e., the convolutional layer, downsampling layer, and fully connected layer propagates forward to obtain the output value. Then, the error between the network's output value and the target value is calculated. When the error is greater than our expected value, the error is propagated back into the network, and the errors of the fully connected layer, downsampling layer, and convolutional layer are calculated in sequence (the error of each layer can be understood as how much of the total error of the network should be borne). Then, the corresponding error gradient is calculated, and the weights are updated according to the error gradient before being re-inputted. When the error is equal to or less than our expected value, the process ends, and the weights are updated according to the calculated error.

[0079] The recurrent neural network and long short-term memory network are used to process time-series information in text data. The recurrent neural network processes information at each time step of the sequence through its recurrent structure, while the long short-term memory network solves the long-term dependency problem in traditional recurrent neural networks by introducing forget gates, input gates, and output gates. Figure 2 The diagram shows a basic recurrent neural network. X is the input vector, O is the output vector, and S represents the values ​​of the hidden layer. U is the weight matrix from the input layer to the hidden layer, V is the weight matrix from the hidden layer to the output layer, and the weight matrix W represents the weights of the previous hidden layer values ​​used as the current input. The diagram is expanded chronologically as follows: Figure 3 As shown, specifically, a recurrent neural network can be represented by the following formula:

[0080] s t =f(U·x t +W·s t-1 )

[0081] o t =g(V·s) t )

[0082] Among them, s t In the hidden state at time t, o t It is the output layer, x t The input is at time t, where U, V, and W are weight matrices, and f and g are activation functions.

[0083] The Long Short-Term Memory network is represented by the following formula:

[0084] f t =σ(W f ·[h t-1 ,x t ]+b f ),

[0085] i t =σ(W i ·[h t-1 ,x t ]+b i ),

[0086]

[0087]

[0088] o t =σ(W o ·[h t-1 ,x t ]+b o ), h t =o t *tanh(C t ),

[0089] Among them, f t i t o t These are the activation values ​​for the forget gate, input gate, and output gate, respectively. It is a candidate cell state, C t It represents the cell state at the current time step, h. t It represents the hidden state, where W and b are the weight matrix and bias term, respectively, and * denotes the Hadamard product.

[0090] In step S3, deep neural networks, by combining the advantages of CNNs and RNNs / LSTMs, are able to extract negative sentiment features from social media texts more comprehensively and accurately, and perform effective recognition. Furthermore, the multi-layered structure and non-linear activation functions of deep neural networks give the model stronger feature representation capabilities and higher recognition accuracy.

[0091] Furthermore, when processing social media data, there may be an imbalance in the number of emotion samples of different categories. To address this issue, this invention employs a weighted loss function, assigning higher weights to emotion categories with fewer samples and improving the model's generalization ability. Therefore, to balance the uneven distribution of emotion samples in the dataset, in step S3, a weighted cross-entropy loss function is applied when training the negative emotion recognition model. This weighted loss function is expressed by the following formula:

[0092]

[0093] in, Here, y is the loss function, and y is the true label. These are predicted labels, where C is the total number of categories, and w is the total number of categories. c It is the weight corresponding to category c, y c and These are the values ​​of category c in the true label and the predicted label, respectively. Also, the category weight w... c The weight of less frequent emotion categories can be determined based on their distribution frequency in the training dataset, with higher weights assigned to these categories to improve the model's ability to recognize them.

[0094] Furthermore, in step S3, a negative sentiment recognition model is trained using support vector machines, decision trees, and random forest algorithms, wherein:

[0095] (1) The support vector machine is used to find the optimal decision boundary in a high-dimensional space to distinguish different sentiment categories. The specific training process can be represented by the following optimization problem:

[0096]

[0097] sty i (w·φ(x i )+b)≥1-ξ i ,ξ i ≥0;

[0098] Where w is the weight vector, b is the bias term, and φ(x) is the weight vector. i ξ is a function that maps input data to a high-dimensional space, C is the regularization parameter, and ξ is the regularization parameter. i It is a slack variable.

[0099] (2) The decision tree classifies data by splitting the dataset layer by layer to form a tree-like decision path. The splitting of each node is based on the criteria of information gain or Gini impurity.

[0100] (3) The random forest consists of multiple decision trees, and the final classification result is determined through a voting mechanism. Training the random forest involves randomly selecting features and samples to construct multiple decision trees and integrating the predictions from these trees. This invention improves the model's recognition accuracy and robustness through model testing and optimization, as well as precise network training and feature selection.

[0101] S4. Use the trained negative sentiment recognition model to identify negative sentiment in text data from social media. This step involves applying the model by deploying the trained model to actual social media platforms to monitor and identify users' negative sentiments in real time. Specifically, step S4 includes:

[0102] (1) Real-time data stream processing: A real-time data stream processing framework is built to capture text data streams from social media platforms in real time. This includes real-time data capture, buffering and processing mechanisms to ensure that the data is up-to-date before being processed by the negative emotion recognition model, thereby enabling real-time monitoring of social media and effectively improving the real-time performance of negative emotion recognition.

[0103] (2) Negative sentiment recognition: The trained negative sentiment recognition model is used to analyze the real-time captured text data. The model determines the sentiment attributes of the text by analyzing keywords, semantic structure and sentiment tendency in the text, which can be expressed by the following formula:

[0104]

[0105] Where E(x) represents the sentiment score of text x, w i It is word x i The weights, S(x) i ) is the word x i Emotional polarity score;

[0106] (3) Based on the results of negative emotion recognition, provide real-time feedback and intervention mechanisms. For example, when high-risk negative emotion text is detected, a preset emergency response process can be automatically triggered, such as sending a warning, providing links to mental health resources, or notifying relevant personnel.

[0107] (4) Data update and model optimization: The negative sentiment recognition model is optimized by using new data collected during real-time data stream processing. The optimization includes adjusting model parameters and optimizing the training set to improve the accuracy and response speed of the model, while adapting to the dynamic changes in social media data.

[0108] Based on step S4 above, the present invention can monitor the data flow of social media platforms in real time, identify and respond to negative emotions in a timely manner, and effectively reduce the spread and impact of harmful information.

[0109] Furthermore, to achieve multimodal negative emotion recognition, i.e., to improve the comprehensiveness of negative emotion recognition, the method also includes comprehensively analyzing emotional information in text, image, audio, and video data based on multimodal data fusion technology, as follows:

[0110] (1) Multimodal data processing: Processing data of different modalities in social media and extracting emotional features from the data, wherein the data of different modalities includes text, image, audio and video data;

[0111] (2) Data fusion: Data fusion technology is used to combine data features from different modalities, which can be expressed by the following formula:

[0112] F = γ1·F text +γ2·F image +γ3·F audio +γ4·F video ,

[0113] Where F is the fused feature vector, F text F image F audio and F video These are the feature vectors for text, image, audio, and video, respectively, and γ1, γ2, γ3, and γ4 are the fusion weights for each modality of data;

[0114] (3) Comprehensive Sentiment Analysis: Based on the fused feature vectors, a negative emotion recognition model is used for comprehensive sentiment analysis. This invention utilizes multimodal data fusion technology to comprehensively analyze different types of data to provide more comprehensive emotion recognition.

[0115] Based on the above multimodal optimization, step S4 further includes negative emotion recognition of image, audio, and video data in social media, specifically as follows: For image data, a convolutional neural network is used to extract features from the image data. The convolutional neural network automatically learns image features through multi-layer filters and can identify negative emotional expressions in the image, such as sad expressions or somber tones, which can be expressed by the following formula:

[0116] f(x) = σ(W*x + b),

[0117] Where f(x) is the feature extraction function, W is the weight matrix of the convolution kernel, * denotes the convolution operation, b is the bias term, and σ is the activation function; for audio data processing, a recurrent neural network is used to analyze the tone and rhythm of speech in the audio data to identify the emotional coloring in the speech. For example, a fast and sharp speech rhythm may express anger or anxiety. For video data, the aforementioned image and audio data processing methods are combined to analyze the visual and auditory information in the video and comprehensively identify negative emotions in the video. Specifically, video data processing involves multimodal data analysis, requiring the simultaneous processing of visual and auditory information and their combination for comprehensive judgment. This invention not only processes text data but also multimodal data such as images, audio, and video. By comprehensively analyzing the emotional information in these data, it provides a more comprehensive recognition of negative emotions.

[0118] Furthermore, the method includes analyzing user behavior data in social media using a negative emotion recognition model, and determining whether a user has negative emotions based on the analysis results, as follows:

[0119] (1) Collection of user behavior data: Collect user behavior data on social media, including posting frequency, interaction patterns (likes, comments, reposts) and online time. This behavior data provides indirect indicators of the user's psychological state.

[0120] (2) Behavioral Pattern Analysis: Machine learning algorithms are applied to perform pattern analysis on user behavior data to identify behavioral patterns that indicate negative emotions. The machine learning algorithms include cluster analysis and sequence pattern mining techniques. One embodiment of the present invention uses cluster analysis, which can be expressed by the following formula:

[0121]

[0122] Where S is the population internal variance, C i It is the i-th cluster, x is a data point, and μ i It is clustering C i The center.

[0123] (3) Negative Sentiment Prediction: Combining user behavior patterns with user-related data analysis results, negative sentiment is predicted comprehensively. User-related data includes text, images, audio, and video. Predictive models such as Support Vector Machine (SVM) or Random Forest algorithms are used to predict negative sentiment based on the comprehensive characteristics of behavioral patterns and content analysis.

[0124] (4) Application of prediction results: Based on the prediction results, corresponding support and intervention measures are provided to users with negative emotions. These measures include pushing positive content to users, recommending professional psychological counseling services to users, managing and supervising social media platforms and promptly deleting harmful information based on the negative emotion identification results, sending mental health reminders, and providing links to psychological counseling resources. The specific implementation of the measures adopted in one embodiment of the present invention can be as follows:

[0125] Measure 1: Push positive content to users based on negative sentiment recognition results. First, establish a content recommendation system based on users' negative sentiment recognition results. This system can recommend positive articles, images, videos, etc. This recommendation mechanism can be implemented through collaborative filtering or content-based recommendation algorithms. Then, based on users' historical behavior data and negative sentiment recognition results, personalized positive content recommendations are made for users. For example, for users who like music, recommend relaxing and cheerful music; for users who like reading, recommend inspirational and positive articles. Secondly, a user feedback mechanism can be established, allowing users to evaluate and provide feedback on the recommended content to optimize the accuracy of content recommendations. For example, the following formula can be used to calculate feedback evaluation:

[0126] R(u,i)=α·C(u,i)+(1-α)·F(u,i)

[0127] Where R(u,i) is the total score of recommended item i for user u, C(u,i) is the content-based recommendation score, F(u,i) is the user feedback score, and α is a parameter that adjusts the importance of the two.

[0128] Measure 2: Recommend professional psychological counseling services to users based on negative emotion recognition results to help them solve psychological problems. First, build an intelligent recommendation system that recommends corresponding psychological counseling services based on the user's negative emotion recognition results. The system can recommend personalized psychological counseling resources based on the user's emotion intensity, emotion type, and historical interaction data. Then, use algorithms to evaluate the user's psychological state, providing decision support for the recommendation system. The psychological state evaluation can be expressed by the following formula:

[0129] P(u)=β1·E(u)+β2·B(u),

[0130] Wherein, P(u) represents the psychological state score of user u, E(u) is the emotion score obtained based on negative emotion recognition, B(u) is the score of user behavior data, and β1 and β2 are weighting coefficients; secondly, based on the evaluation results, the intelligent recommendation system matches psychological counseling services suitable for users, such as online counseling, telephone counseling or face-to-face counseling, to ensure that users can receive timely professional psychological support and help.

[0131] Measure 3: Manage and supervise social media platforms, and promptly remove harmful information based on negative sentiment identification results. Specific implementation methods include: real-time monitoring of public content on social media platforms, including various forms of data such as text, images, audio, and video; analyzing the monitored data using the aforementioned negative sentiment identification model to identify content containing harmful information. This process can be combined with multimodal analysis to ensure effective processing of different types of data; classifying the identified harmful information and processing it accordingly according to the policies and guidelines of the social media platform. For example, the system can automatically mark and remove statements involving negative information or other content that violates platform regulations; establishing a feedback mechanism that allows users to report misjudgments, and the system continuously learns and optimizes based on this feedback to improve the accuracy and efficiency of identification.

[0132] Compared with existing technologies, this invention can more accurately and efficiently identify negative emotions in social media, possessing higher real-time monitoring capabilities and providing users with better mental health protection. Furthermore, the technical solution of this invention has broad application prospects and can be applied to fields such as search engines, intelligent customer service systems, online comment filtering, and news sentiment analysis, thereby improving the overall quality of social media and user experience.

[0133] This invention has broad application prospects. It not only provides social media platforms with an effective tool for emotion monitoring and management, but also opens up new application possibilities for many other fields. Specific application prospects include, but are not limited to, the following:

[0134] 1. Social Media Platform Management: This invention helps social media platforms monitor and manage user content in real time, effectively identify and process negative emotional information, and improve the platform's information quality and user experience.

[0135] 2. Mental Health Support: By identifying users' negative emotions, this invention can assist mental health professionals in providing more accurate psychological assessments and interventions, thereby improving the effectiveness of mental health services.

[0136] 3. Online Education and Training: In the field of online education and training, this invention can help identify learners' emotional states and provide a more personalized learning experience and support.

[0137] 4. Intelligent Customer Service System: In the intelligent customer service system, this invention can improve the customer service representative's ability to recognize user emotions and provide a more humanized service experience.

[0138] 5. Public safety and surveillance: In the field of public safety, this invention can assist in monitoring social media and other public forums to prevent and reduce cyberbullying and other harmful behaviors.

[0139] 6. Business Intelligence and Market Analysis: This invention can also be applied to the field of business intelligence and market analysis to help companies understand consumer sentiment and optimize products and services.

[0140] The present invention has been described above by way of example with reference to the accompanying drawings. Obviously, the specific implementation of the present invention is not limited to the above-described manner. Any non-substantial improvements made using the inventive concept and technical solution; or the direct application of the inventive concept and technical solution to other situations without modification, are all within the protection scope of the present invention.

Claims

1. A method for identifying negative sentiment on social media based on generative artificial intelligence, characterized in that: The method includes the following steps: S1. Collect text data from social media, including posts, comments, and messages from social media; S2. The collected text data is preprocessed using a generative artificial intelligence model. The preprocessing steps include text cleaning, word segmentation, stop word removal, and semantic annotation. The generative artificial intelligence model is a generative adversarial network (GAN), which includes a generator and a discriminator. The generator is used to generate synthetic text data, and the discriminator is used to distinguish between the generated synthetic text data and real text data. The generator G and the discriminator D are trained using the following formula: ， ， in, It is the input noise vector of the generator. It is the text data input to the discriminator; S3. Train a negative sentiment recognition model using the preprocessed text data; the negative sentiment recognition model is a deep neural network, which includes a convolutional neural network, a recurrent neural network, and a long short-term memory network. The convolutional neural network is used to extract local features from the text data, as expressed by the following formula: ， in, It is the first Elements in the feature map after convolution. It is an activation function. It is a bias term. These are the weights of the convolution kernel. These are elements of the input feature map; the recurrent neural network and long short-term memory network are used to process time series information in text data, and the recurrent neural network is expressed by the following formula: in, It is in time The hidden layer state, It is the output layer. It is in time Input, , , and It is a weight matrix. and It is the activation function; the Long Short-Term Memory network is represented by the following formula: ， ， ， ， ， ， in, , , These are the activation values ​​for the forget gate, input gate, and output gate, respectively. It is a candidate cell state. It represents the cell state at the current time step. It is in a hidden state. and These are the weight matrix and the bias term, respectively. It represents the Hadamardi (or Hadama) stack; A weighted loss function is applied when training the negative emotion recognition model to balance the uneven distribution of various emotion samples in the dataset. The weighted loss function is expressed by the following formula: ； in, It is a loss function. It's a real label. It is a predicted label. It is the total number of categories. It corresponds to the category The weight, and These are the categories in the true labels and the predicted labels. The value; S4. Use a trained negative sentiment recognition model to identify negative sentiment in text data from social media.

2. The method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 1, characterized in that: The generative adversarial network is trained using the following loss function: ； in, It is the distribution of real data. It is the distribution of the generator input noise. Indicates the desired operation.

3. The method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 1, characterized in that: In step S3, a negative sentiment recognition model is trained using support vector machines, decision trees, and random forest algorithms, wherein: (1) The support vector machine is used to find the optimal decision boundary in a high-dimensional space to distinguish different sentiment categories, as expressed by the following formula: ； ； in, It is a weight vector. It is a bias term. It is a function that maps input data to a high-dimensional space. It is a regularization parameter. It is a slack variable; (2) The decision tree classifies the data by splitting the dataset layer by layer to form a tree-like decision path; (3) The random forest consists of multiple decision trees, and the final classification result is determined by a voting mechanism.

4. The method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 1, characterized in that: Step S4 includes: (1) Real-time data stream processing: Construct a real-time data stream processing framework to capture text data streams from social media platforms in real time; (2) Negative sentiment recognition: The trained negative sentiment recognition model is used to analyze the real-time captured text data to determine the sentiment attributes of the text, expressed by the following formula: ； in, Representing text Emotional score It is a word The weight, It is a word Emotional polarity score; (3) Provide real-time feedback and intervention mechanisms based on the results of negative emotion identification; (4) Data update and model optimization: The negative emotion recognition model is optimized by new data collected during the real-time data stream processing. The optimization includes adjusting model parameters and optimizing the training set.

5. The method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 1, characterized in that: The method includes comprehensively analyzing emotional information in text, image, audio, and video data based on multimodal data fusion technology, and the steps are as follows: (1) Multimodal data processing: Processing data of different modalities in social media and extracting emotional features from the data, wherein the data of different modalities includes text, image, audio and video data; (2) Data fusion: Data fusion technology is used to combine data features from different modalities, as expressed by the following formula: ， in, It is the fused feature vector. , , and These are feature vectors for text, images, audio, and video, respectively. , , and These are the fusion weights of the data from each modality; (3) Comprehensive sentiment analysis: Based on the fused feature vector, a negative sentiment recognition model is used to perform comprehensive sentiment analysis.

6. The method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 5, characterized in that: Step S4 includes identifying negative emotions from image, audio, and video data on social media. For image data, a convolutional neural network is used to extract features from the image data and identify negative emotional expressions in the image, expressed by the following formula: ,in, It is a feature extraction function. It is the weight matrix of the convolution kernel. This represents the convolution operation. It is a bias term. It is an activation function; for audio data processing, a recurrent neural network is used to analyze the tone and rhythm of speech in the audio data and identify the emotional color in the speech; for video data, the image and audio data processing methods are combined to analyze the visual and auditory information in the video and comprehensively identify the negative emotions in the video.

7. A method for identifying negative sentiment on social media based on generative artificial intelligence according to claim 1 or 5, characterized in that: The method includes analyzing user behavior data in social media using a negative sentiment recognition model, and determining whether a user has negative emotions based on the analysis results. The steps are as follows: (1) Collection of user behavior data: Collect user behavior data on social media, including posting frequency, interaction patterns and online duration; (2) Behavioral pattern analysis: Apply machine learning algorithms to perform pattern analysis on user behavior data to identify behavioral patterns that indicate negative emotions. The machine learning algorithms include cluster analysis and sequence pattern mining techniques. (3) Negative emotion prediction: Combine user behavior patterns and data analysis results related to users to comprehensively predict users' negative emotions. Data related to users includes text, pictures, audio and video. (4) Application of prediction results: Based on the prediction results, provide corresponding support and intervention measures for users with negative emotions. These measures include pushing positive content to users, recommending professional psychological counseling services to users, managing and supervising social media platforms, and promptly deleting harmful information based on the negative emotion identification results.

Images