Product public opinion analysis method and device, storage medium and computer equipment

Abstract

According to the product public opinion analysis method and device, the storage medium and the computer equipment provided by the invention, the multi-modal features in the public opinion data are extracted and feature fusion is carried out to obtain the fused feature data. In this way, information in different forms can be effectively integrated, and the public opinion situation can be analyzed more comprehensively and accurately. And then extracting an entity relationship in the fused feature data to update a preset public opinion knowledge graph. And performing sentiment analysis on the fused feature data in three dimensions of text, video and voice by using a public opinion knowledge graph to obtain a public opinion sentiment score. Therefore, the emotional tendency of the public opinion information can be captured more comprehensively, and the accuracy of public opinion analysis is improved. Meanwhile, an early warning mechanism is triggered when a preset condition is met. And finally, reasoning and plan matching are performed in the public opinion knowledge graph based on the generated public opinion event features so as to determine a public relation plan. According to the process, historical experience and reasoning ability in the knowledge graph can be fully utilized, and the timeliness and accuracy of public opinion management are improved.

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a product public opinion analysis method, device, storage medium and computer equipment. Background Art

[0002] In a market characterized by increasingly fierce brand competition, companies urgently need to accurately monitor and analyze public opinion surrounding their products and brands to maintain their brand image and rapidly respond to public relations crises. Traditional product public opinion analysis relies primarily on superficial processing of text data, using keyword matching or basic sentiment analysis models to assess brand reputation. A typical process involves collecting brand-related text reviews, performing simple sentiment classification, manually annotating key cases, and ultimately generating an offline analysis report.

[0003] However, with the diversification of information formats and the increasing complexity of the public opinion landscape, traditional product public opinion analysis has gradually exposed significant limitations. First, its single data dimension ignores key public opinion clues in multimodal data such as images, videos, and voice, resulting in incomplete analysis results. Second, because the development of public relations plans relies heavily on manual experience, rapid response and scalable application are difficult to achieve. In summary, traditional product public opinion analysis suffers from poor accuracy and responsiveness. Summary of the Invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical deficiencies, especially the technical deficiencies of poor accuracy and responsiveness of traditional product public opinion analysis in the prior art.

[0005] In a first aspect, the present application provides a method for analyzing product public opinion, the method comprising:

[0006] Obtaining public opinion data of the product to be analyzed, extracting multimodal features from the public opinion data, and then performing feature fusion on the extracted multimodal features to obtain fused feature data;

[0007] Extracting entity relationships from the fused feature data, generating corresponding product triples based on each extracted entity relationship, and updating a preset public opinion knowledge graph based on each product triple;

[0008] Using the public opinion knowledge graph, sentiment analysis is performed on the fused feature data in three dimensions: text, video, and voice, to obtain a public opinion sentiment score;

[0009] Obtain the number of times public opinion is spread, trigger an early warning mechanism when the number of times public opinion is spread and the public opinion sentiment score meet preset conditions, and generate public opinion event features based on each product triple;

[0010] Based on the features of the public opinion event, reasoning and plan matching are performed in the public opinion knowledge graph to obtain a public relations plan for the product to be analyzed.

[0011] In one embodiment, after extracting the multimodal features from the public opinion data, the extracted multimodal features are subjected to feature fusion to obtain fused feature data, including:

[0012] Extracting text features from the public opinion data based on the BERT-GPT hybrid model, extracting video features from the public opinion data based on the YOLOv8 model and the CLIP model, and extracting speech features from the public opinion data based on the Wav2Vec2 model;

[0013] forming a multimodal feature based on the text feature, the video feature, and the voice feature;

[0014] A time alignment operation is performed on the features of different modes in the multimodal features, and a multi-head attention mechanism is used to fuse multiple features in the multimodal features after the time alignment operation to obtain fused feature data.

[0015] In one embodiment, extracting entity relationships from the fused feature data, generating corresponding product triples according to each extracted entity relationship, and updating a preset public opinion knowledge graph according to each product triple includes:

[0016] performing sequence labeling on the fused feature data to identify key information in the fused feature data, and forming at least one entity relationship based on the identified key information;

[0017] Structurally organize each entity relationship and determine the product triples corresponding to each entity relationship;

[0018] Compare each product triple with the preset public opinion knowledge graph to identify product triples that contain new entities or new entity relationships;

[0019] The identified product triples are added to corresponding positions in the public opinion knowledge graph to update the public opinion knowledge graph.

[0020] In one embodiment, the use of the public opinion knowledge graph to perform sentiment analysis on the fused feature data in three dimensions: text, video, and voice to obtain a public opinion sentiment score includes:

[0021] Combined with the public opinion knowledge graph, sentiment tendency analysis is performed on the fused feature data in three dimensions: text, video, and voice, to obtain a text feature vector, a video feature vector, and a voice feature vector of the fused feature data;

[0022] The text feature vector, the video feature vector and the speech feature vector are input into a preset fusion model for cross-modal fusion to explore the association between feature vectors of different modalities under the action of the self-attention mechanism, and obtain the public opinion sentiment score output by the fusion model.

[0023] In one embodiment, the early warning mechanism is triggered when the number of public opinion dissemination times and the public opinion sentiment score meet preset conditions, and public opinion event features are generated based on each product triple, including:

[0024] When the number of times the public opinion spreads reaches a first preset threshold and the public opinion sentiment score reaches a second preset threshold, a warning prompt is issued, and the type of public opinion event is determined based on the entities and relationships in each product triple;

[0025] The fermentation degree of the public opinion event is evaluated in combination with the number of public opinion dissemination times and the public opinion sentiment score, and the public opinion event characteristics are generated according to the public opinion event type and the fermentation degree.

[0026] In one embodiment, the process of performing reasoning and plan matching in the public opinion knowledge graph based on the public opinion event characteristics to obtain a public relations plan for the product to be analyzed includes:

[0027] Based on the graph search, at least one path similar to the characteristics of the public opinion event is found in the public opinion knowledge graph, and the public opinion event corresponding to each path and its public relations plan and public relations effect are obtained;

[0028] According to the public opinion events corresponding to each path and their public relations plans and public relations effects, the public relations plan with the best public relations effect is selected, and the selected public relations plan is determined as the public relations plan for the product to be analyzed.

[0029] In one embodiment, the method further comprises:

[0030] Predicting the public opinion trend of the product to be analyzed in the future based on the public opinion knowledge graph, the public opinion sentiment score, and the number of public opinion disseminations;

[0031] Anomaly detection is performed on the public opinion trend of the product to be analyzed based on the isolation forest algorithm. If an anomaly point is detected, the anomaly information and anomaly time corresponding to the anomaly point are obtained;

[0032] Generate alarm information according to the abnormal information and the abnormal time, and issue the alarm information.

[0033] In a second aspect, the present application provides a product public opinion analysis device, the device comprising:

[0034] A feature fusion module is used to obtain public opinion data of the product to be analyzed, extract multimodal features from the public opinion data, and then perform feature fusion on the extracted multimodal features to obtain fused feature data;

[0035] A graph updating module is used to extract entity relationships from the fused feature data, generate corresponding product triples based on each extracted entity relationship, and update a preset public opinion knowledge graph based on each product triple;

[0036] A sentiment analysis module is used to use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in three dimensions: text, video, and voice, to obtain a public opinion sentiment score;

[0037] An early warning trigger module is used to obtain the number of public opinion propagation times, trigger the early warning mechanism when the number of public opinion propagation times and the public opinion sentiment score meet preset conditions, and generate public opinion event features based on each product triple;

[0038] The plan determination module is used to perform reasoning and plan matching in the public opinion knowledge graph based on the characteristics of the public opinion event to obtain the public relations plan of the product to be analyzed.

[0039] In a third aspect, the present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the product public opinion analysis method described in any of the above embodiments.

[0040] In a fourth aspect, the present application provides a computer device, comprising: one or more processors, and a memory;

[0041] The memory stores computer-readable instructions, and when the one or more processors execute the computer-readable instructions, they perform the steps of the product public opinion analysis method described in any of the above embodiments.

[0042] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0043] The product public opinion analysis method, apparatus, storage medium, and computer device provided in this application acquire public opinion data on the product to be analyzed, extract multimodal features from the public opinion data, and then perform feature fusion on the extracted multimodal features to obtain fused feature data. This effectively integrates information in different forms, avoiding the limitations of a single data source, and thus enabling a more comprehensive and accurate analysis of public opinion trends. Entity relationships are then extracted from the fused feature data, and corresponding product triples are generated based on each extracted entity relationship. A preset public opinion knowledge graph is then updated based on each product triple, thereby providing rich semantic associations for subsequent analysis. The public opinion knowledge graph is then used to perform sentiment analysis on the fused feature data across three dimensions: text, video, and voice, to obtain a public opinion sentiment score. This allows for a more comprehensive capture of the sentimental tendencies of public opinion information, avoids the one-sidedness of single-dimensional analysis, and improves the accuracy of public opinion analysis. Simultaneously, the number of public opinion transmissions is obtained. When the number of public opinion transmissions and the public opinion sentiment score meet preset conditions, an early warning mechanism is triggered to promptly identify potential public opinion risks and enable rapid response through the early warning mechanism. Finally, public opinion event features are generated based on each product triple. Based on these features, reasoning and contingency matching are performed within the public opinion knowledge graph to produce a public relations plan for the product being analyzed. This process fully leverages the historical experience and reasoning capabilities of the knowledge graph to effectively curb the spread of negative public opinion and improve the timeliness and accuracy of public opinion management. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 A flowchart of a product public opinion analysis method provided in an embodiment of the present application;

[0046] Figure 2 A flowchart of a process for determining fusion feature data provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of a process for using a public opinion knowledge graph to perform sentiment analysis on fused feature data in three dimensions: text, video, and voice, provided in an embodiment of the present application;

[0048] Figure 4 A schematic diagram of the process of reasoning and matching plans in a public opinion knowledge graph based on public opinion event features provided in an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the structure of a product public opinion analysis device provided in an embodiment of the present application;

[0050] Figure 6 This is a diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0052] In one embodiment, this application provides a method for analyzing product public opinion. The following embodiments illustrate this method as applied to a server. It is understood that the method can be executed on a single server or a server cluster consisting of multiple servers, and this application does not impose specific limitations on this.

[0053] like Figure 1 As shown, this application provides a product public opinion analysis method, which includes:

[0054] S101: Obtain public opinion data of the product to be analyzed, extract multimodal features from the public opinion data, and then perform feature fusion on the extracted multimodal features to obtain fused feature data.

[0055] Public opinion data refers to all publicly available information related to the product being analyzed, including but not limited to social media posts, news reports, user comments, forum discussions, video content, images, and so on. Multimodal features refer to the various types of information features extracted from public opinion data, including text features (such as keywords and sentiment), visual features (such as color, texture, and object recognition), and speech features (such as intonation, speaking speed, and emotional expression). Fused feature data refers to the comprehensive feature data obtained by integrating multimodal features extracted from public opinion data. Fused feature data can more comprehensively reflect the overall picture of public opinion information.

[0056] In this step, when users need to understand public sentiment toward a particular product, they can conduct a public opinion analysis on that product (i.e., the product to be analyzed). This first requires obtaining public opinion data on the product to be analyzed. Specifically, web crawler technology can be used to capture text, images, videos, and voice data related to the product to be analyzed from social media platforms, news websites, forums, and other channels to generate public opinion data for the product to be analyzed. After obtaining the public opinion data, multimodal features can be extracted from the data. Specifically, feature extraction can be performed using appropriate methods tailored to the data modality. For example, for text data, natural language processing (NLP) techniques can be used to extract features such as keywords and sentiment. The extracted multimodal features are then fused to produce fused feature data.

[0057] Specifically, feature fusion can be achieved through a variety of methods, such as weighted averaging, feature concatenation, or deep learning models (such as multimodal fusion networks). For example, text sentiment scores, image recognition results, and speech sentiment analysis results can be combined to generate a comprehensive feature vector for subsequent public opinion analysis. This application does not limit the specific method of feature fusion.

[0058] S102: Extract entity relationships from the fused feature data, generate corresponding product triples based on each extracted entity relationship, and update the preset public opinion knowledge graph based on each product triple.

[0059] Entity relationships include two entities and the relationship between them. For example, if user A complains about product B's quality issues, user A and product B are considered entities, and the complaint about product quality issues is considered the relationship between the two entities. A product triple is structured information represented in the form of entity-relationship-entity. A public opinion knowledge graph is a knowledge base that uses a graph structure to represent entities and their relationships within public opinion data. The public opinion knowledge graph can reflect the latest public opinion information and entity relationships.

[0060] In this step, after determining the fused feature data, entity relationships can be extracted from the fused feature data based on sequence annotation. Each extracted entity relationship is then structured and organized to obtain the product triples corresponding to each entity relationship. Finally, each product triple is updated to the preset public opinion knowledge graph. It can be understood that updating the public opinion knowledge graph can integrate new public opinion information into the existing knowledge system, enrich the content of the knowledge graph, and enhance its ability to reflect public opinion dynamics. For example, when a new public opinion event occurs, the knowledge graph can be updated in a timely manner to provide the latest background knowledge and related information for public opinion analysis.

[0061] Specifically, when extracting entity relationships from fused feature data, natural language processing techniques can be used to extract these relationships. For example, named entity recognition and dependency parsing tools can be used to identify entities and their relationships within text. For example, when analyzing a user review stating "Product A's battery life is too short," the entity "Product A," the entity "user," and the relationship "battery life is short" can be extracted.

[0062] S103: Use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in the three dimensions of text, video, and voice to obtain a public opinion sentiment score.

[0063] The public opinion sentiment score is the result of a quantitative assessment of the emotional tendencies in public opinion data using sentiment analysis techniques. The public opinion sentiment score can be expressed as a numerical value ranging from -1 to 1, where -1 represents strong negative sentiment, 1 represents strong positive sentiment, and 0 represents neutral sentiment.

[0064] In this step, when using the public opinion knowledge graph to perform sentiment analysis on the fused feature data in the three dimensions of text, video, and voice, the main contents include the following aspects:

[0065] (1) Use the public opinion knowledge graph to perform sentiment analysis on the text dimension of the fused feature data. Specifically, natural language processing (NLP) technology can be used in combination with the associated information in the public opinion knowledge graph to perform sentiment analysis on the text data.

[0066] (2) Use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in the visual dimension. Specifically, for image and video data, computer vision technology can be used to extract emotion-related features. For example, by identifying the user's facial expressions (such as frowning and smiling) in the video, the emotional tendency can be judged.

[0067] (3) Use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in the voice dimension. Specifically, speech recognition technology can be used to convert speech into text, and then sentiment analysis can be performed on the text. At the same time, features such as voice intonation and speaking speed can also be combined to assist in sentiment judgment.

[0068] After performing sentiment analysis on the fused feature data across the three dimensions of text, video, and voice, a public opinion sentiment score can be calculated using methods such as weighted averaging. It's understandable that by analyzing sentiment across the three dimensions of text, video, and voice, the emotional tendencies within public opinion information can be more comprehensively captured. For example, text data can reflect a user's specific opinions and evaluations, video data can provide intuitive emotional clues through their expressions and behaviors, and voice data can reveal a user's emotional state through features such as intonation and speaking speed. This multi-dimensional sentiment analysis avoids the one-sidedness of single-dimensional analysis and provides a more accurate and comprehensive sentiment assessment. Furthermore, the introduction of a public opinion knowledge graph can combine the sentiment information in multimodal data with existing knowledge systems, further enhancing the accuracy and interpretability of sentiment analysis.

[0069] S104: Obtain the number of times public opinion is spread, trigger an early warning mechanism when the number of times public opinion is spread and the public opinion sentiment score meet preset conditions, and generate public opinion event features based on each product triple.

[0070] The number of public opinion transmissions refers to the number of times public opinion information is disseminated on social media platforms or other online channels, including the total number of forwarding, likes, and comments. This number reflects the reach and influence of public opinion information. Preset conditions refer to pre-set thresholds for early warning mechanisms. For example, when the number of public opinion transmissions exceeds 1,000 and the sentiment score is below -0.5, the early warning mechanism is triggered. Public opinion event features refer to key information extracted from product triples to describe public opinion events, including event type, severity, scope of impact, and sentiment.

[0071] In this step, the number of public opinion transmissions can be calculated based on the number of reposts, likes, and comments on social media platforms. After determining the number of transmissions, the number of transmissions and sentiment scores are determined to meet pre-set conditions. If these conditions are met, an early warning mechanism is triggered, notifying relevant personnel to take appropriate measures. Furthermore, public opinion event features can be generated based on the information contained in each product triplet to facilitate subsequent emergency response matching.

[0072] Furthermore, when the early warning mechanism is triggered, relevant personnel may be notified via one or more combinations of methods such as email, text message, phone call or system notification.

[0073] S105: Based on the characteristics of the public opinion event, reasoning and plan matching are performed in the public opinion knowledge graph to obtain the public relations plan for the product to be analyzed.

[0074] Among them, the public relations plan refers to the response strategy and action plan formulated in advance for specific public opinion events.

[0075] In this step, the extracted public opinion event features are input into the public opinion knowledge graph. Using the inference rules and historical event data in the knowledge graph, inference is performed on the current public opinion event. Based on the inference results, the corresponding public relations plan is then matched in the public opinion knowledge graph to obtain a public relations plan for the product being analyzed.

[0076] Furthermore, when a corresponding public relations plan is matched, the public relations plan can be adaptively adjusted according to the characteristics of the public opinion event or the public opinion data, and then the adjusted public relations plan can be used as the final public relations plan for the product to be analyzed.

[0077] It's understandable that by inputting the characteristics of public opinion events into the public opinion knowledge graph for inference, we can leverage the historical data and inference rules within the public opinion knowledge graph to quickly deduce the likely development trends and response strategies for the current event, thus avoiding misjudgments caused by human inexperience. Secondly, plan matching can quickly identify the most appropriate response plan based on the characteristics of the current event, improving the efficiency of public opinion response and reducing the time and cost of crisis management.

[0078] In the above embodiment, public opinion data for the product to be analyzed is obtained and multimodal features are extracted from the public opinion data. The extracted multimodal features are then fused to produce fused feature data. This effectively integrates information from different forms, avoiding the limitations of a single data source, and enabling a more comprehensive and accurate analysis of public opinion trends. Entity relationships are then extracted from the fused feature data. Corresponding product triples are generated based on each extracted entity relationship. A pre-set public opinion knowledge graph is then updated based on each product triple, providing rich semantic associations for subsequent analysis. The public opinion knowledge graph is then used to perform sentiment analysis on the fused feature data across three dimensions: text, video, and voice, to obtain a public opinion sentiment score. This allows for a more comprehensive capture of the sentimental tendencies of public opinion information, avoids the bias inherent in single-dimensional analysis, and improves the accuracy of public opinion analysis. Simultaneously, the number of public opinion transmissions is obtained. When both the number of transmissions and the public opinion sentiment score meet pre-set conditions, an early warning mechanism is triggered, enabling timely identification of potential public opinion risks and rapid response through the early warning mechanism. Finally, public opinion event features are generated based on each product triple. Based on these features, reasoning and contingency matching are performed within the public opinion knowledge graph to produce a public relations plan for the product being analyzed. This process fully leverages the historical experience and reasoning capabilities of the knowledge graph to effectively curb the spread of negative public opinion and improve the timeliness and accuracy of public opinion management.

[0079] like Figure 2 As shown, in one embodiment, after extracting multimodal features from public opinion data, feature fusion is performed on the extracted multimodal features to obtain fused feature data, including:

[0080] S201: Extract text features from public opinion data based on the BERT-GPT hybrid model, extract video features from public opinion data based on the YOLOv8 model and CLIP model, and extract voice features from public opinion data based on the Wav2Vec2 model.

[0081] S202: forming multimodal features based on text features, video features, and speech features.

[0082] S203: Performing a time alignment operation on features of different modes in the multimodal features, and using a multi-head attention mechanism to fuse multiple features in the multimodal features after the time alignment operation to obtain fused feature data.

[0083] The BERT-GPT hybrid model combines BERT (Bidirectional Encoder Representations from Transformers) and GPT (Generative Pre-trained Transformer). Combining BERT's bidirectional encoding capabilities with GPT's generative capabilities, the BERT-GPT hybrid model is used to extract semantic features from text data, enabling better understanding of text content and generating relevant feature representations. The YOLOv8 model is an object detection model that can quickly and accurately detect objects in images and provide bounding boxes and category labels for each detected object. The CLIP (Contrastive Language-Image Pre-training) model is a multimodal model that maps images and text into the same feature space to extract semantic features from image content. The Wav2Vec2 model is a Transformer-based self-supervised speech representation learning model that can extract high-quality speech features from raw speech signals.

[0084] In this embodiment, when the BERT-GPT hybrid model is used to process the text data in the public opinion data, the BERT module in the hybrid model is responsible for extracting the semantic features of the text, while the GPT module is used to generate relevant feature representations, namely text features. Next, the YOLOv8 model is used to perform target detection on the image or video data in the public opinion data to extract the objects in the image and their position information. The CLIP model is then used to map the extracted image content to the semantic space to extract the semantic features of the image, namely the visual features. Finally, the Wav2Vec2 model is used to process the voice data in the public opinion data to extract the emotional and content features in the voice. In this way, the integration of text features, visual features and voice features can generate multimodal features.

[0085] After determining the multimodal features, they are temporally aligned to ensure temporal consistency across modalities. For example, if a user's expression in a video is grim and their voice also expresses anger, these features need to be aligned to the same point in time. A multi-head attention mechanism is then used to fuse the temporally aligned multimodal features. This mechanism dynamically focuses on the relationships between features across modalities, generating fused feature data.

[0086] Furthermore, when performing a time alignment operation on multimodal features, the multimodal features can be mapped to a unified time axis based on the timestamp and key information of the data to complete the time alignment operation.

[0087] It can be understood that the above process can not only fully utilize the advantages of multimodal information, but also improve the accuracy and reliability of public opinion analysis, and provide more powerful support for public opinion management and crisis response of enterprises and organizations.

[0088] In one embodiment, entity relationships are extracted from the fused feature data, corresponding product triples are generated based on each extracted entity relationship, and a preset public opinion knowledge graph is updated based on each product triple, including:

[0089] S1: Sequence labeling is performed on the fused feature data to identify key information in the fused feature data, and at least one entity relationship is formed based on the identified key information.

[0090] S2: Structurally organize each entity relationship and determine the product triples corresponding to each entity relationship.

[0091] S3: Compare each product triple with the preset public opinion knowledge graph to identify product triples with new entities or new entity relationships in each product triple.

[0092] S4: Add the identified product triples to the corresponding positions in the public opinion knowledge graph to update the public opinion knowledge graph.

[0093] Among them, key information includes but is not limited to the event subject, time, location, relationship, etc.

[0094] In this embodiment, named entity recognition and dependency syntax analysis tools are used to perform sequence annotation on the fused feature data. This is to identify key information in the fused feature data, such as entities, relationships, time and place, etc., and then form at least one entity relationship based on this information. Each entity relationship is then structured and sorted to obtain the product triples corresponding to each entity relationship. The generated product triples are compared with the preset public opinion knowledge graph to identify new entities or new entity relationships. For example, if there is already a "Product A" node in the knowledge graph, but there is no "short battery life" relationship, it is identified as a new entity relationship. Finally, the identified new entity or new entity relationship is added to the corresponding position in the public opinion knowledge graph to obtain a new public opinion knowledge graph.

[0095] Specifically, adding new entities or entity relationships to the public opinion knowledge graph can enrich the content of the knowledge graph, making it more comprehensive in reflecting public opinion dynamics. For example, when new product issues or user feedback arise, the knowledge graph can be updated in a timely manner, providing the latest information support for public opinion monitoring and response.

[0096] In one embodiment, the public opinion knowledge graph can also be used in multi-round dialogue interaction scenarios. Specifically, a computer device can receive a user's question, extract key information from the question, use this key information to search the current public opinion knowledge graph, generate a response based on the search results, and return the generated response to the user. For example, if a user asks, "Analyze the reasons for the negative reviews of product A last week," the public opinion knowledge graph can be used to search for information related to product A, such as the subject of the event and causal relationships, and present it to the user in a visual manner. The user can then ask further questions, such as, "What specific models are involved?" The computer device can then provide further answers based on the associated information in the knowledge graph.

[0097] like Figure 3 As shown, in one embodiment, the public opinion knowledge graph is used to perform sentiment analysis on the fused feature data in the three dimensions of text, video, and voice to obtain the public opinion sentiment score, including:

[0098] S301: Combined with the public opinion knowledge graph, sentiment tendency analysis is performed on the fused feature data in the three dimensions of text, video, and voice to obtain the text feature vector, video feature vector, and voice feature vector of the fused feature data.

[0099] In the text dimension, the BERT-GPT hybrid model can be used to analyze the fused feature data in the text dimension to generate a contextual semantic vector. This contextual semantic vector is then combined with a preset sentiment lexicon to perform fine-grained sentiment classification on the fused feature data in the text dimension to obtain a text feature vector. Furthermore, adversarial training can be used to improve the hybrid model's ability to handle complex semantics. This allows the BERT-GPT hybrid model to better learn sentiment features and maintain stable performance even in the face of data perturbations.

[0100] In the visual dimension, the YOLOv8 model and the CLIP model are combined to extract visual features from images and identify key elements such as people, scenes, and objects. The FER (Facial Expression Recognition) model and the CLIP model are then combined to analyze facial expressions and scene emotions in the image, determining the emotional tendency conveyed by the image to obtain a visual feature vector.

[0101] In the speech dimension, the Wav2Vec2 model is used to convert speech data into text and add sentiment labels derived from emotional intonation analysis. The converted text is then subjected to fine-grained sentiment classification using the same methods used for text sentiment analysis, based on sentiment lexicon augmentation and adversarial training, to generate speech feature vectors.

[0102] S302: Input the text feature vector, video feature vector and speech feature vector into the preset fusion model for cross-modal fusion, so as to explore the correlation between the feature vectors of different modalities under the action of the self-attention mechanism, and obtain the public opinion sentiment score output by the fusion model.

[0103] The fusion model is used to fuse features from multiple modalities and, during the fusion process, to mine the correlations between feature vectors from different modalities under the influence of a self-attention mechanism to output a public opinion sentiment score. In one example, the fusion model can employ the MMBT (MultiModal BiTransformers) model, a supervised multimodal bitransformer model for image and text classification.

[0104] In this step, the data formats of text feature vectors, video feature vectors, and speech feature vectors can be unified first, and then these feature vectors are input into the fusion model. The fusion model includes multiple self-attention mechanism models. These modules learn the association weights between feature vectors of different modalities and explore the potential relationship between text, video, and speech. Then, feature mapping is performed through the fully connected layer to obtain a comprehensive public opinion sentiment score. For example, when analyzing a news report with pictures, if the scene in the image and the text description are highly correlated, the fusion model will assign a higher weight to both, and vice versa. In addition, the fusion model will learn the consistency or complementarity of emotional expression of different modalities. For example, if the positive description in the text and the cheerful scene in the image confirm each other, the fusion model will enhance the judgment of positive emotions.

[0105] Specifically, at the output of the fully connected layer, an activation function is used to convert the numerical value into a comprehensive sentiment score. Common activation functions, such as Sigmoid or Softmax, map the output value to a range between -1 and 1, or to a probability distribution range for various sentiment categories, thereby generating a public opinion sentiment score that comprehensively reflects the sentiment embedded in the public opinion data for the product being analyzed.

[0106] It is understood that using the public opinion knowledge graph to perform sentiment analysis on the fused feature data across the three dimensions of text, video, and voice, and employing multiple methods to improve the judgment of sentiment trends during the sentiment analysis process, can more comprehensively capture the sentiment trends of public opinion information, avoid the one-sidedness caused by single-dimensional analysis, and maximize the accuracy of public opinion analysis.

[0107] In one embodiment, an early warning mechanism is triggered when the number of public opinion transmissions and the public opinion sentiment score meet preset conditions, and public opinion event features are generated based on each product triple, including:

[0108] S1: When the number of public opinion dissemination reaches the first preset threshold and the public opinion sentiment score reaches the second preset threshold, an early warning prompt is issued, and the type of public opinion event is determined based on the entities and relationships in each product triplet.

[0109] S2: Combine the number of public opinion dissemination and public opinion sentiment score to evaluate the fermentation degree of public opinion events, and generate public opinion event characteristics based on the type and fermentation degree of public opinion events.

[0110] The first and second preset thresholds are empirical values and can be adjusted based on actual needs. The public opinion event type indicates the category of the public opinion event, such as a failure issue or a quality issue. The fermentation level measures the impact and harm of the public opinion event.

[0111] In this embodiment, when the number of times a public opinion spreads reaches a first preset threshold and the public opinion sentiment score reaches a second preset threshold, a warning alert can be issued, the type of public opinion event can be determined, and the degree of fermentation of the public opinion event can be measured based on the number of times a public opinion spreads and the public opinion sentiment score. Finally, a public opinion event feature is generated based on the type and degree of fermentation of the public opinion event.

[0112] Specifically, an early warning will be automatically triggered when the preset conditions are met, which can promptly remind relevant personnel to take action. In addition, public opinion event characteristics will be generated based on the current information to facilitate the subsequent determination of public relations plans.

[0113] In one embodiment, the thresholds in the preset conditions can also be set to be adaptive, dynamically adjusting the thresholds based on the number of times public opinion spreads and the volatility of the public opinion sentiment score. Furthermore, in addition to the above implementations, the number of times public opinion spreads and the public opinion sentiment score can be weighted separately to calculate a comprehensive trigger index, which can then be compared with the threshold to determine whether to trigger an early warning. This can make the early warning mechanism more flexible and accurate.

[0114] like Figure 4 As shown, in one embodiment, reasoning and plan matching are performed in the public opinion knowledge graph based on the characteristics of public opinion events to obtain a public relations plan for the product to be analyzed, including:

[0115] S401: Based on the graph search, at least one path similar to the characteristics of the public opinion event is searched in the public opinion knowledge graph, and the public opinion events corresponding to each path and their public relations plans and public relations effects are obtained.

[0116] S402: Based on the public opinion events corresponding to each path and their public relations plans and public relations effects, a public relations plan with the best public relations effect is selected, and the selected public relations plan is determined as the public relations plan for the product to be analyzed.

[0117] Among them, public relations effect refers to the evaluation of the actual impact and results of public opinion events after the implementation of the public relations plan.

[0118] In this embodiment, a graph search algorithm (such as breadth-first search or the shortest path algorithm) is used within the public opinion knowledge graph to locate paths with similar characteristics to the current public opinion event. Similarity can be determined by quantifying the path based on a similarity assessment algorithm and then comparing it against a set threshold to determine similarity. For each path found, the corresponding public opinion event, public relations plan, and public relations effect are extracted. Based on the obtained public relations effect data, a comprehensive evaluation is conducted on each public relations plan in terms of the magnitude of the decrease in the number of public opinion transmissions and the degree of improvement in sentiment scores. The plan with the best public relations effect is then selected as the public relations plan for the product to be analyzed.

[0119] It's understandable that by searching for similar paths through graph search, we can quickly locate historical events with similar characteristics to the current event, providing a wealth of reference cases for decision-making. This knowledge graph-based search method fully leverages existing public opinion data and experience. Finally, selecting the plan with the best public relations effect as the public relations plan for the product being analyzed ensures that companies can take the most effective measures when faced with a public opinion crisis, minimizing any negative impact.

[0120] In one embodiment, the product public opinion analysis method further includes:

[0121] S1: Predict the public opinion trend of the product to be analyzed in the future based on the public opinion knowledge graph, public opinion sentiment score and the number of public opinion dissemination.

[0122] S2: Anomaly detection is performed on the public opinion trend of the product to be analyzed based on the isolation forest algorithm. If an anomaly point is detected, the anomaly information and anomaly time corresponding to the anomaly point are obtained.

[0123] S3: Generate alarm information based on the abnormal information and abnormal time, and issue the alarm information.

[0124] In this embodiment, a Transformer-TCN hybrid model is used to predict public opinion trends over a period of time. This model combines the Transformer model with a TCN model. TCN (Temporal Convolutional Network) is a sequence modeling architecture based on convolutional neural networks. The isolation forest algorithm is then used to detect anomalies in the public opinion trends of the product being analyzed, obtaining anomaly information and the time of the anomaly. This information is used to generate and issue an alert. This allows for the rapid detection of sudden changes or abnormal fluctuations in public opinion data, such as an unusual increase in the number of times a public opinion is disseminated or a sudden change in sentiment scores. This allows relevant personnel to prepare for response measures in advance.

[0125] In one embodiment, the public opinion analysis and public opinion trend prediction in this application can automatically generate relevant reports. Specifically, relevant reports can be generated based on relevant data and corresponding templates, and sent to the user end for user viewing.

[0126] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0127] The following describes the product public opinion analysis device provided in an embodiment of the present application. The product public opinion analysis device described below and the product public opinion analysis method described above can be referenced to each other.

[0128] like Figure 5 As shown, the present application provides a product public opinion analysis device 500, which includes:

[0129] The feature fusion module 501 is used to obtain public opinion data of the product to be analyzed, extract multimodal features from the public opinion data, and then perform feature fusion on the extracted multimodal features to obtain fused feature data;

[0130] Graph update module 502, used to extract entity relationships from the fused feature data, generate corresponding product triples based on each extracted entity relationship, and update the preset public opinion knowledge graph based on each product triple;

[0131] Sentiment analysis module 503, used to use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in the three dimensions of text, video, and voice to obtain a public opinion sentiment score;

[0132] The early warning trigger module 504 is used to obtain the number of public opinion transmission times, trigger the early warning mechanism when the number of public opinion transmission times and the public opinion sentiment score meet the preset conditions, and generate public opinion event features based on each product triple;

[0133] The plan determination module 505 is used to perform reasoning and plan matching in the public opinion knowledge graph based on the characteristics of public opinion events to obtain the public relations plan for the product to be analyzed.

[0134] In the above embodiment, public opinion data for the product to be analyzed is obtained and multimodal features are extracted from the public opinion data. The extracted multimodal features are then fused to produce fused feature data. This effectively integrates information from different forms, avoiding the limitations of a single data source, and enabling a more comprehensive and accurate analysis of public opinion trends. Entity relationships are then extracted from the fused feature data. Corresponding product triples are generated based on each extracted entity relationship. A pre-set public opinion knowledge graph is then updated based on each product triple, providing rich semantic associations for subsequent analysis. The public opinion knowledge graph is then used to perform sentiment analysis on the fused feature data across three dimensions: text, video, and voice, to obtain a public opinion sentiment score. This allows for a more comprehensive capture of the sentimental tendencies of public opinion information, avoids the bias inherent in single-dimensional analysis, and improves the accuracy of public opinion analysis. Simultaneously, the number of public opinion transmissions is obtained. When both the number of transmissions and the public opinion sentiment score meet pre-set conditions, an early warning mechanism is triggered, enabling timely identification of potential public opinion risks and rapid response through the early warning mechanism. Finally, public opinion event features are generated based on each product triple. Based on these features, reasoning and contingency matching are performed within the public opinion knowledge graph to produce a public relations plan for the product being analyzed. This process fully leverages the historical experience and reasoning capabilities of the knowledge graph to effectively curb the spread of negative public opinion and improve the timeliness and accuracy of public opinion management.

[0135] In one embodiment, the feature fusion module includes:

[0136] The feature extraction submodule is used to extract text features from public opinion data based on the BERT-GPT hybrid model, visual features based on the YOLOv8 model and CLIP model, and speech features based on the Wav2Vec2 model.

[0137] A feature generation submodule is used to generate multimodal features based on text features, video features, and speech features;

[0138] The feature fusion submodule is used to perform time alignment operations on features of different modes in the multimodal features, and adopts a multi-head attention mechanism to fuse multiple features in the multimodal features after the time alignment operation to obtain fused feature data.

[0139] In one embodiment, the map update module includes:

[0140] An information identification submodule, configured to perform sequence annotation on the fused feature data to identify key information in the fused feature data and form at least one entity relationship based on the identified key information;

[0141] The triple determination submodule is used to perform structured sorting on each entity relationship and determine the product triple corresponding to each entity relationship;

[0142] The information comparison submodule is used to compare each product triple with the preset public opinion knowledge graph to identify product triples with new entities or new entity relationships in each product triple;

[0143] The graph update submodule is used to add the identified product triples to the corresponding positions in the public opinion knowledge graph to update the public opinion knowledge graph.

[0144] In one embodiment, the sentiment analysis module includes:

[0145] The sentiment analysis submodule is used to analyze the sentiment tendency of the fused feature data in the three dimensions of text, video, and voice in combination with the public opinion knowledge graph, and obtain the text feature vector, video feature vector, and voice feature vector of the fused feature data;

[0146] The sentiment fusion submodule is used to input text feature vectors, video feature vectors and speech feature vectors into the preset fusion model for cross-modal fusion, so as to explore the correlation between feature vectors of different modalities under the action of the self-attention mechanism and obtain the public opinion sentiment score output by the fusion model.

[0147] In one embodiment, the early warning triggering module includes:

[0148] The type determination submodule is used to issue a warning when the number of public opinion propagation times reaches a first preset threshold and the public opinion sentiment score reaches a second preset threshold, and to determine the type of public opinion event based on the entities and relationships in each product triple;

[0149] The degree determination submodule is used to evaluate the fermentation degree of public opinion events by combining the number of public opinion dissemination times and the public opinion sentiment score, and to generate public opinion event characteristics based on the type and fermentation degree of public opinion events.

[0150] In one embodiment, the plan determination module includes:

[0151] The path determination submodule is used to find at least one path with similar characteristics to the public opinion event in the public opinion knowledge graph based on graph search, and obtain the public opinion event corresponding to each path and its public relations plan and public relations effect;

[0152] The plan determination submodule is used to select the public relations plan with the best public relations effect based on the public opinion events corresponding to each path and its public relations plan and public relations effect, and determine the selected public relations plan as the public relations plan for the product to be analyzed.

[0153] In one embodiment, the product public opinion analysis device further includes:

[0154] The trend prediction module is used to predict the public opinion trend of the product to be analyzed in the future based on the public opinion knowledge graph, public opinion sentiment score and the number of public opinion transmissions;

[0155] The anomaly detection module is used to detect anomalies in the public opinion trends of the products being analyzed based on the isolation forest algorithm. If an anomaly point is detected, the anomaly information and anomaly time corresponding to the anomaly point are obtained;

[0156] The alarm module is used to generate alarm information based on abnormal information and abnormal time, and issue alarm information.

[0157] The division of the various modules in the above-mentioned product public opinion analysis device is only for illustration. In other embodiments, the product public opinion analysis device can be divided into different modules as needed to complete all or part of the functions of the above-mentioned product public opinion analysis device. The various modules in the above-mentioned product public opinion analysis device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0158] In one embodiment, the present application also provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the product public opinion analysis method as described in any of the above embodiments.

[0159] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the product public opinion analysis method as described in any one of the above embodiments.

[0160] Schematically, as Figure 6 As shown, Figure 6 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 600 can be provided as a server. Figure 6 Computer device 600 includes a processing component 602, which further includes one or more processors, and memory resources represented by memory 601 for storing instructions executable by processing component 602, such as applications. The applications stored in memory 601 may include one or more modules, each corresponding to a set of instructions. In addition, processing component 602 is configured to execute the instructions to perform the product public opinion analysis method of any of the above-mentioned embodiments.

[0161] The computer device 600 may further include a power supply component 603 configured to perform power management of the computer device 600, a wired or wireless network interface 604 configured to connect the computer device 600 to a network, and an input / output (I / O) interface 605. The computer device 600 may operate based on an operating system stored in the memory 601, such as Windows Server™, Mac OS X™, Unix™, Linux™, Free BSD™, or the like.

[0162] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0163] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment comprising a series of elements not only include those elements, but also include other elements not clearly listed, or also include elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements limited by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or equipment comprising the elements. Herein, the singular forms "one", "an" and "said / the" may also include plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "include / comprise" or "have" etc. specify the existence of stated features, wholes, steps, operations, components, parts or combinations thereof, but do not exclude the possibility of the existence or addition of one or more other features, wholes, steps, operations, components, parts or combinations thereof. At the same time, the term "and / or" used in this specification includes any and all combinations of the relevant listed items.

[0164] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can be referenced to each other.

[0165] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A product public opinion analysis method, characterized in that: The method comprises: Obtaining public opinion data of the product to be analyzed, extracting multimodal features from the public opinion data, and then performing feature fusion on the extracted multimodal features to obtain fused feature data; Extracting entity relationships from the fused feature data, generating corresponding product triples based on each extracted entity relationship, and updating a preset public opinion knowledge graph based on each product triple; Using the public opinion knowledge graph, sentiment analysis is performed on the fused feature data in three dimensions: text, video, and voice, to obtain a public opinion sentiment score; Obtain the number of times public opinion is spread, trigger an early warning mechanism when the number of times public opinion is spread and the public opinion sentiment score meet preset conditions, and generate public opinion event features based on each product triple; Based on the features of the public opinion event, reasoning and plan matching are performed in the public opinion knowledge graph to obtain a public relations plan for the product to be analyzed.

2. The product public opinion analysis method according to claim 1, characterized in that: After extracting the multimodal features from the public opinion data, performing feature fusion on the extracted multimodal features to obtain fused feature data includes: Extracting text features from the public opinion data based on the BERT-GPT hybrid model, extracting video features from the public opinion data based on the YOLOv8 model and the CLIP model, and extracting speech features from the public opinion data based on the Wav2Vec2 model; forming a multimodal feature based on the text feature, the video feature, and the voice feature; A time alignment operation is performed on the features of different modes in the multimodal features, and a multi-head attention mechanism is used to fuse multiple features in the multimodal features after the time alignment operation to obtain fused feature data.

3. The product public opinion analysis method according to claim 1, characterized in that: The extracting entity relationships from the fused feature data, generating corresponding product triples according to each extracted entity relationship, and updating a preset public opinion knowledge graph according to each product triple, includes: performing sequence labeling on the fused feature data to identify key information in the fused feature data, and forming at least one entity relationship based on the identified key information; Structurally organize each entity relationship and determine the product triples corresponding to each entity relationship; Compare each product triple with the preset public opinion knowledge graph to identify product triples that contain new entities or new entity relationships; The identified product triples are added to corresponding positions in the public opinion knowledge graph to update the public opinion knowledge graph.

4. The product public opinion analysis method according to claim 1, characterized in that: The method of using the public opinion knowledge graph to perform sentiment analysis on the fused feature data in three dimensions: text, video, and voice, to obtain a public opinion sentiment score includes: Combined with the public opinion knowledge graph, sentiment tendency analysis is performed on the fused feature data in three dimensions: text, video, and voice, to obtain a text feature vector, a video feature vector, and a voice feature vector of the fused feature data; The text feature vector, the video feature vector and the speech feature vector are input into a preset fusion model for cross-modal fusion to explore the association between feature vectors of different modalities under the action of the self-attention mechanism, and obtain the public opinion sentiment score output by the fusion model.

5. The product public opinion analysis method according to claim 1, characterized in that: The warning mechanism is triggered when the number of public opinion propagation times and the public opinion sentiment score meet preset conditions, and public opinion event features are generated according to each product triple, including: When the number of times the public opinion spreads reaches a first preset threshold and the public opinion sentiment score reaches a second preset threshold, a warning prompt is issued, and the type of public opinion event is determined based on the entities and relationships in each product triple; The fermentation degree of the public opinion event is evaluated in combination with the number of public opinion dissemination times and the public opinion sentiment score, and the public opinion event characteristics are generated according to the public opinion event type and the fermentation degree.

6. The product public opinion analysis method according to claim 1, characterized in that: The reasoning and plan matching based on the public opinion event characteristics in the public opinion knowledge graph to obtain the public relations plan for the product to be analyzed includes: Based on the graph search, at least one path similar to the characteristics of the public opinion event is found in the public opinion knowledge graph, and the public opinion event corresponding to each path and its public relations plan and public relations effect are obtained; According to the public opinion events corresponding to each path and their public relations plans and public relations effects, the public relations plan with the best public relations effect is selected, and the selected public relations plan is determined as the public relations plan for the product to be analyzed.

7. The product public opinion analysis method according to any one of claims 1 to 6, characterized in that: The method further comprises: Predicting the public opinion trend of the product to be analyzed in the future based on the public opinion knowledge graph, the public opinion sentiment score, and the number of public opinion disseminations; Anomaly detection is performed on the public opinion trend of the product to be analyzed based on the isolation forest algorithm. If an anomaly point is detected, the anomaly information and anomaly time corresponding to the anomaly point are obtained; Generate alarm information according to the abnormal information and the abnormal time, and issue the alarm information.

8. A product public opinion analysis device, characterized in that: The device comprises: A feature fusion module is used to obtain public opinion data of the product to be analyzed, extract multimodal features from the public opinion data, and then perform feature fusion on the extracted multimodal features to obtain fused feature data; A graph updating module is used to extract entity relationships from the fused feature data, generate corresponding product triples based on each extracted entity relationship, and update a preset public opinion knowledge graph based on each product triple; A sentiment analysis module is used to use the public opinion knowledge graph to perform sentiment analysis on the fused feature data in three dimensions: text, video, and voice, to obtain a public opinion sentiment score; An early warning trigger module is used to obtain the number of public opinion propagation times, trigger the early warning mechanism when the number of public opinion propagation times and the public opinion sentiment score meet preset conditions, and generate public opinion event features based on each product triple; The plan determination module is used to perform reasoning and plan matching in the public opinion knowledge graph based on the characteristics of the public opinion event to obtain the public relations plan of the product to be analyzed.

9. A storage medium, characterized in that: The storage medium stores computer-readable instructions, which, when executed by one or more processors, enable the one or more processors to execute the steps of the product public opinion analysis method as described in any one of claims 1 to 7.

10. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, which, when executed by the one or more processors, execute the steps of the product public opinion analysis method according to any one of claims 1 to 7.

Images