Implicit fine-grained sentiment analysis method based on knowledge driving and intelligent reasoning

Abstract

The invention discloses an implicit fine-grained sentiment analysis method based on knowledge driving and intelligent reasoning, and relates to the technical field of sentiment analysis, and the method comprises the following steps: S1, carrying out data preprocessing and text depth representation modeling on original text data; s2, multi-dimensional semantic embedding is adopted; s3, recessive emotion reasoning based on a causal chain; s4, performing model training of the adaptive deep learning framework; s5, optimizing an emotion reasoning path driven by reinforcement learning; s6, introducing logical constraint and causal reasoning path fine regulation and control; according to the implicit fine-grained sentiment analysis method based on knowledge driving and intelligent reasoning, the domain knowledge graph, logical reasoning and reinforcement learning are combined, the recognition ability of implicit sentiment is remarkably improved, the method effectively makes up for the defects in the prior art, and the method is suitable for popularization and application. The method is suitable for various complex text analysis tasks and has high application value and commercial potential.

Description

Technical Field

[0001] The present invention relates to the field of sentiment analysis technology, and in particular to an implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning. Background Art

[0002] Sentiment analysis is a key task in the field of natural language processing (NLP) and is widely used in scenarios such as public opinion monitoring and consumer feedback analysis. Traditional sentiment analysis methods typically rely on explicit sentiment words (such as "good" and "bad") to determine the emotional tendency of text. However, the sentiment expression of many texts is implicit and does not directly contain explicit sentiment words. For example, "The room has a sea view, but it is 30 kilometers away from the city." This sentence does not contain direct sentiment words, but implicitly expresses the inconvenience that may be caused by the long distance, thus carrying negative sentiment.

[0003] Traditional sentiment analysis methods mainly include dictionary-based methods, machine learning-based methods, and deep learning-based methods:

[0004] Dictionary-based methods: They rely on predefined sentiment vocabulary and rule matching to judge sentiment tendencies, cannot handle implicit sentiment expressions, and have poor generalization capabilities.

[0005] Machine learning-based methods, such as support vector machines (SVMs), convolutional neural networks (CNNs), and long-term time-travel (LSTMs), can perform sentiment classification through feature extraction, but they still rely on explicit sentiment words and cannot perform causal reasoning.

[0006] Deep learning-based methods, such as pre-trained models like BERT, can improve the accuracy of sentiment analysis. However, they lack the use of domain knowledge and have difficulty understanding implicit sentiment relationships, resulting in limited generalization capabilities.

[0007] In summary, existing technologies still have significant limitations in implicit sentiment analysis and are unable to cope with complex contexts and cross-domain sentiment reasoning needs. To this end, an implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning is proposed. Summary of the Invention

[0008] The purpose of this invention is to solve the problems in the prior art and to propose an implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning.

[0009] The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning includes the following steps:

[0010] S1. Perform data preprocessing and text deep representation modeling on the original text data;

[0011] S2, adopts multi-dimensional semantic embedding;

[0012] S3, implicit emotional reasoning based on causal chains;

[0013] S4, Model training of the adaptive deep learning framework;

[0014] S5, Optimization of the emotion reasoning path driven by reinforcement learning;

[0015] S6, Introduction of logical constraints and fine-tuning of causal reasoning paths;

[0016] S7, Output of reasoning results and interpretability analysis.

[0017] Preferably, in the step S1, the data preprocessing and text deep representation modeling of the original text data include the following steps:

[0018] (1) Cleaning the original text data, including removing stop words, spelling correction, and syntactic standardization;

[0019] (2) Performing word segmentation, part-of-speech tagging, syntactic analysis, and dependency parsing on the cleaned text data to construct the basic structure of the text;

[0020] (3) Using named entity recognition technology to extract key emotion-related entities in the text;

[0021] (4) Obtaining context semantic information from the text data through the BERT pre-trained model to generate high-dimensional word vector embeddings.

[0022] Preferably, in the step S2, the multi-dimensional semantic embedding includes the following steps:

[0023] (1) Collecting and integrating entity and relationship data from external knowledge bases;

[0024] (2) Using a deep learning-based relationship extraction model to automatically extract emotion causal relationships from unstructured text, and the relationship extraction model is a model combination of BERT+BiLSTM+CRF;

[0025] (3) Constructing a knowledge graph and using a graph neural network for embedding to fuse the knowledge graph with the text semantic vector;

[0026] (4) Through an adaptive knowledge enhancement mechanism, dynamically adjusting the weight of knowledge graph information in the reasoning process.

[0027] Preferably, in the step S3, the implicit emotion reasoning based on the causal chain includes the following steps:

[0028] (1) Establishing an emotion causal chain according to the reasoning logic of event-emotion reaction-emotion polarity;

[0029] (2) Using a causal reasoning algorithm based on an event graph to construct an emotion causal graph and optimizing the reasoning path using reinforcement learning;

[0030] (3) Combine the Transformer model to calculate the sentiment intensity and dynamically adjust the inference direction.

[0031] Preferably, in the step S4, the model training of the adaptive deep learning framework includes the following steps:

[0032] (1) Adopt a multi-modal information fusion strategy, combine text, graph, and audio data to enrich the sentiment features;

[0033] (2) Based on the graph attention network, introduce an attention mechanism to optimize the weight assignment of entity relationships;

[0034] (3) Through the meta-learning method, construct a model that can adapt to different scenarios and improve the cross-domain generalization ability;

[0035] (4) Adopt a combination of supervised learning and unsupervised learning to enhance the model's understanding ability of implicit sentiment.

[0036] Preferably, in the step S5, the optimization of the sentiment inference path driven by reinforcement learning includes the following steps:

[0037] (1) Design a multi-level reward mechanism, and optimize the inference path based on the policy gradient optimization method of reinforcement learning;

[0038] (2) Adopt the DQN or PPO algorithm to dynamically adjust the inference strategy and improve the accuracy of sentiment analysis in complex situations;

[0039] (3) Through hierarchical reinforcement learning, establish different levels of sentiment causal chains to improve the inference accuracy.

[0040] Preferably, in the step S6, the introduction of logical constraints and fine-tuning of the causal inference path include the following steps:

[0041] (1) Introduce a logical reasoning algorithm based on the knowledge graph during the inference process to ensure the reasonableness of the inference results;

[0042] (2) Combine rule matching and deep learning techniques to restrict unreasonable sentiment causal relationships;

[0043] (3) Adopt an adversarial constraint mechanism to enhance the model's robustness through the GAN training method.

[0044] Preferably, in the step S7, the output of the inference result and the interpretability analysis include the following steps:

[0045] (1) Adopt visualization technology to display the sentiment inference path and provide a transparent inference process;

[0046] (2) Generate an emotional analysis report, including key information such as emotional categories, emotional intensities, and causal paths;

[0047] (3) Adopt techniques based on explainable AI to provide interpretability analysis for the inference results and improve user trust.

[0048] Compared with the existing technologies, the advantages of the present invention are as follows:

[0049] 1. The present invention combines a domain knowledge graph and causal reasoning to achieve precise recognition of texts without explicit emotional words. By constructing multi-dimensional semantic embeddings and causal reasoning chains, the emotional understanding ability of complex texts is improved. Through reinforcement learning, the inference direction is dynamically adjusted to achieve automatic attribution and inference of implicit emotions.

[0050] 2. The present invention adopts graph neural networks and cross-modal information fusion to improve the adaptability of the model in multiple fields. Through meta-learning methods, adaptive learning in different scenarios is achieved, the accuracy of cross-domain generalization analysis is optimized. By using reinforcement learning-driven policy gradient optimization, the inference path is dynamically optimized to improve the accuracy of emotional prediction. By establishing different levels of emotional causal chains through hierarchical reinforcement learning, the inference depth and accuracy are improved.

[0051] 3. The present invention adopts knowledge-graph-based logical reasoning, combines rule matching and deep learning to ensure the rationality of the inference results. Through an adversarial constraint mechanism, the GAN training method is introduced to improve the robustness of the model. Brief Description of the Drawings

[0052] Figure 1 It is a flowchart of the emotional analysis method in the present invention. Detailed Embodiments

[0053] To make the technical means, creative features, achieved purposes, and functions of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0054] Refer to Figure 1 As shown, the implicit fine-grained emotional analysis method based on knowledge-driven and intelligent reasoning includes the following steps:

[0055] S1. Perform data preprocessing and text deep representation modeling on the original text data;

[0056] S2. Adopt multi-dimensional semantic embedding;

[0057] S3. Implicit emotional reasoning based on the causal chain;

[0058] S4. Model training of the adaptive deep learning framework;

[0059] S5. Optimization of the emotional reasoning path driven by reinforcement learning;

[0060] S6. Introduce logical constraints and finely regulate causal reasoning paths;

[0061] S7. Output the reasoning results and perform interpretability analysis.

[0062] In this embodiment, in the step S1, the data preprocessing and text deep representation modeling of the original text data include the following steps:

[0063] (1) Clean the original text data, including removing stop words, spelling correction, and syntactic standardization;

[0064] (2) Perform word segmentation, part-of-speech tagging, syntactic analysis, and dependency parsing on the cleaned text data to construct the basic structure of the text;

[0065] (3) Use the named entity recognition (NER) technology to extract key sentiment-related entities in the text;

[0066] (4) Obtain context semantic information from the text data through the BERT pre-trained model to generate high-dimensional word vector embeddings.

[0067] In this embodiment, in the step S2, the multi-dimensional semantic embedding includes the following steps:

[0068] (1) Collect and integrate entity and relationship data from external knowledge bases (such as WordNet, DBpedia);

[0069] (2) Use a deep learning-based relation extraction model to automatically extract sentiment causal relationships from unstructured text, and the relation extraction model is a model combination of BERT+BiLSTM+CRF;

[0070] (3) Construct a knowledge graph and use a graph neural network (GNN) for embedding to fuse the knowledge graph with the text semantic vector to improve the text understanding ability and achieve more accurate sentiment analysis;

[0071] (4) Through an adaptive knowledge enhancement mechanism, dynamically adjust the weight of the knowledge graph information in the reasoning process to make it adapt to different scenarios.

[0072] In this embodiment, in the step S3, the implicit sentiment reasoning based on the causal chain includes the following steps:

[0073] (1) Establish a sentiment causal chain according to the reasoning logic of event - emotional reaction - emotional polarity;

[0074] (2) Adopt the event graph-based causal reasoning algorithm (ECG-RL) to construct an Emotional Causal Graph, and use reinforcement learning to optimize the reasoning path;

[0075] (3) Combine the Transformer model to calculate the emotional intensity and dynamically adjust the reasoning direction.

[0076] In this embodiment, in step S4, the model training of the adaptive deep learning framework includes the following steps:

[0077] (1) Adopt a multi-modal fusion strategy, combine text, graph, and audio data to enrich emotional features;

[0078] (2) Based on the Graph Attention Network (GAT), introduce an attention mechanism to optimize the weight assignment of entity relationships;

[0079] (3) Through the meta-learning method, construct a model that can adapt to different scenarios, improve cross-domain generalization ability, and adapt to different situations;

[0080] (4) Adopt a combination of supervised learning and unsupervised learning to enhance the model's understanding ability of implicit emotions.

[0081] In this embodiment, in step S5, the optimization of the emotional reasoning path driven by reinforcement learning includes the following steps:

[0082] (1) Design a multi-level reward mechanism, and optimize the reasoning path based on the policy gradient optimization method of reinforcement learning;

[0083] (2) Adopt the DQN or PPO algorithm to dynamically adjust the reasoning strategy to improve the accuracy of emotional analysis in complex situations;

[0084] (3) Through hierarchical reinforcement learning (Hierarchical RL), establish emotional causal chains at different levels to improve the reasoning accuracy.

[0085] In this embodiment, in step S6, the introduction of logical constraints and fine-tuning of the causal reasoning path includes the following steps:

[0086] (1) Introduce a knowledge graph-based logical reasoning algorithm (KG-Logic Inference) during the reasoning process to ensure the reasonableness of the reasoning results;

[0087] (2) Combine rule matching and deep learning techniques to restrict unreasonable emotional causal relationships, such as "positive emotion → negative consequence", to ensure that the reasoning conforms to common sense in reality;

[0088] (3) Adopt an adversarial constraint mechanism to enhance the robustness of the model through GAN training.

[0089] In this embodiment, in step S7, the inference result output and interpretability analysis include the following steps:

[0090] (1) Use visualization techniques to display the emotional inference path and provide a transparent inference process;

[0091] (2) Generate an emotional analysis report, including key information such as emotional category, emotional intensity, and causal path;

[0092] (3) Adopt techniques based on explainable AI to provide interpretability analysis for the inference results and improve user trust.

[0093] In summary, the implicit fine-grained emotional analysis method based on knowledge-driven and intelligent inference proposed by the present invention combines a domain knowledge graph, logical reasoning, and reinforcement learning, significantly improving the ability to identify implicit emotions. This method effectively makes up for the deficiencies of the prior art, is applicable to various complex text analysis tasks, and has strong application value and commercial potential.

[0094] As is known by technical common sense, the present invention can be implemented by other embodiments that do not depart from its spiritual essence or essential features. Therefore, the above-disclosed embodiments are illustrative in all aspects and not exclusive. All changes within the scope of the present invention or equivalent to the present invention are encompassed by the present invention.

Claims

1. An implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning, characterized in that: It includes the following steps: S1. Perform data preprocessing and text deep representation modeling on the original text data; S2. Adopt multi-dimensional semantic embedding; S3. Perform implicit emotion reasoning based on the causal chain; S4. Train the model of the adaptive deep learning framework; S5. Optimize the emotion reasoning path driven by reinforcement learning; S6. Introduce logical constraints and fine-tune the causal reasoning path; S7. Output the reasoning result and perform interpretability analysis.

2. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, characterized in that: In the step S1, performing data preprocessing and text deep representation modeling on the original text data includes the following steps: (1) Clean the original text data, including removing stop words, spelling correction, and syntactic standardization; (2) Perform word segmentation, part-of-speech tagging, syntactic analysis, and dependency parsing on the cleaned text data to construct the basic structure of the text; (3) Use named entity recognition technology to extract key emotion-related entities in the text; (4) Obtain context semantic information from the text data through the BERT pre-trained model to generate high-dimensional word vector embeddings.

3. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, wherein: In the step S2, adopting multi-dimensional semantic embedding includes the following steps: (1) Collect and integrate entity and relationship data from external knowledge bases; (2) Use a relationship extraction model based on deep learning to automatically extract emotion causal relationships from unstructured text, and the relationship extraction model is a model combination of BERT+BiLSTM+CRF; (3) Construct a knowledge graph and use a graph neural network for embedding to fuse the knowledge graph with the text semantic vector; (4) Through an adaptive knowledge enhancement mechanism, dynamically adjust the weight of the knowledge graph information in the reasoning process.

4. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, wherein: In the step S3, performing implicit emotion reasoning based on the causal chain includes the following steps: (1) Establish an emotion causal chain according to the reasoning logic of event-emotion response-emotion polarity; (2) Use a causal reasoning algorithm based on an event graph to construct an emotion causal graph and optimize the reasoning path using reinforcement learning; (3) Combine the Transformer model to calculate the emotion intensity and dynamically adjust the reasoning direction.

5. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, wherein: In the step S4, training the model of the adaptive deep learning framework includes the following steps: (1) Adopt a multi-modal information fusion strategy, combine text, graph, and audio data to enrich emotion features; (2) Based on the graph attention network, introduce an attention mechanism to optimize the weight assignment of entity relationships; (3) Through meta-learning methods, construct a model that can adapt to different scenarios and improve cross-domain generalization ability; (4) Adopt a combination of supervised learning and unsupervised learning to enhance the model's understanding ability of implicit emotions.

6. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, characterized in that: In the step S5, optimizing the emotion reasoning path driven by reinforcement learning includes the following steps: (1) Design a multi-level reward mechanism and optimize the reasoning path based on the policy gradient optimization method of reinforcement learning; (2) Use the DQN or PPO algorithm to dynamically adjust the reasoning strategy to improve the accuracy of emotion analysis in complex situations; (3) Through hierarchical reinforcement learning, establish emotion causal chains at different levels to improve the reasoning accuracy.

7. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, characterized in that: In the step S6, introducing logical constraints and fine-tuning the causal reasoning path includes the following steps: (1) Introduce a logical reasoning algorithm based on a knowledge graph during the reasoning process to ensure reasonable reasoning results; (2) Combine rule matching and deep learning techniques to restrict unreasonable emotional causal relationships; (3) Adopt an adversarial constraint mechanism to enhance the robustness of the model through GAN training.

8. The implicit fine-grained sentiment analysis method based on knowledge-driven and intelligent reasoning according to claim 1, wherein: In step S7, the inference result output and interpretability analysis include the following steps: (1) Use visualization techniques to display the emotional reasoning path and provide a transparent reasoning process; (2) Generate an emotional analysis report, including key information such as emotional categories, emotional intensities, and causal paths; (3) Adopt techniques based on explainable AI to provide interpretability analysis for the inference results and improve user trust.

Images