Intelligent Evaluation Method for Dialogue Content Context Adaptability Based on Text-Based Sentiment Computation

Abstract

This invention discloses an intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing, belonging to the field of dialogue sentiment detection technology. The method includes: a corpus construction and context labeling stage: an expert system uses a multi-layered labeling system to label dialogues in the corpus with contextual tags; a sentiment computing model construction stage: using KL divergence and cosine similarity as dual indicators to evaluate the matching degree between the output sentiment and the expected medical scenario and optimize the dialogue strategy; an sentiment adaptability evaluation mechanism establishment stage: generating a comprehensive contextual adaptability score to evaluate the contextual adaptability of the output content of the intelligent simulated patient ISP; and a real-time optimization and adjustment stage: optimizing the dialogue strategy, language expression, and sentiment adaptability in real time through reinforcement learning and expert system feedback, dynamically adjusting the dialogue style according to case requirements. This invention overcomes the limitations of single text sentiment analysis, achieving accurate identification and dynamic adaptation of sentiment expression, making the dialogue more emotionally natural and realistic.

Description

Technical Field

[0001] This invention relates to the field of dialogue sentiment detection technology, and in particular to an intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing. Background Technology

[0002] The concept of Affective Computing was proposed by Professor Rosalind Picard of MIT in 1997. This technology integrates knowledge from multiple disciplines, including computer science, psychology, cognitive science, and neuroscience, with the core goal of endowing computers with more natural and human-emotional interaction capabilities. Currently, affective computing technology shows broad application potential in various fields such as intelligent customer service, education, healthcare, administration, and entertainment. Through emotion recognition technology, computers can gain a deeper understanding of human emotional states and dynamically adjust their behavior and feedback based on changes in user emotions. This allows intelligent systems to consider users' emotional factors in the decision-making process, providing more personalized content and services. Furthermore, affective computing technology can also enable intelligent systems to exhibit a certain degree of empathy, thereby more effectively understanding and supporting user needs, improving the system's intelligence level, enhancing its ability to interact with human society, and ultimately increasing its influence in society.

[0003] Text sentiment analysis, also known as sentiment technology, is a key branch of Natural Language Processing (NLP). It primarily focuses on collecting and analyzing public opinions, thoughts, and feelings regarding different topics, products, issues, and services, aiming to automatically identify and classify the emotional tendencies contained in text using computer technology. The core of text sentiment analysis lies in the processing and analysis of text data, and its methods can be broadly categorized into three main types: those based on sentiment dictionaries, those based on traditional machine learning, and those based on deep learning. This technology has broad application prospects in numerous fields, covering business analytics, healthcare, reputation management, public opinion monitoring, and customer feedback. With the help of text sentiment analysis, it is possible to automate the processing and analysis of large amounts of text data, perform multimodal sentiment analysis, and build personalized user sentiment models to more accurately grasp users' emotional expression patterns and sentiment tendencies.

[0004] The Intelligent Simulation Patient (ISP) is a virtual patient simulator that integrates natural language processing (NLP) and Intelligent Tutoring System (ITS) technologies. Currently, ISP systems are developing towards a deeper integration of NLP and ITS technologies. By simulating real clinical scenarios, this system allows students to train in clinical diagnostic reasoning within a virtual environment, gradually replacing some traditional clinical teaching activities in medical education. Faced with increasingly strained educational resources and increasingly complex clinical environments, ISP effectively alleviates the teaching difficulties of traditional clinical teaching models in meeting student needs, providing students with more realistic clinical scenarios and a more natural language interaction experience without increasing additional human resources costs. Simultaneously, the system can track students' learning progress and performance, providing real-time feedback and thus offering high-quality clinical training. Although NLP technology has made significant progress, in some cases, ISP may still fail to accurately understand students' output and lacks a comprehensive evaluation mechanism for the output.

[0005] Contextual adaptability assessment is a method used to determine whether an object or system can operate or function effectively and reasonably in a specific context. This assessment method mainly encompasses various types, including dimension-based, test-type-based, assessment subject-based, and data-driven approaches. It is applicable to assessment needs in fields such as scenario-based teaching, medical device usability evaluation, and artificial intelligence, providing a comprehensive assessment of contextual adaptability from multiple dimensions, avoiding the limitations of a single assessment indicator. Appropriate assessment methods and indicators can be flexibly selected based on different assessment objects and objectives to improve the accuracy and effectiveness of the assessment. This method is applicable to different operational forms (individual or team participation) and different simulation scenarios, flexibly adapting to various needs. Furthermore, its assessment results are instructive, providing clear directions and basis for improvement and optimization, contributing to enhanced system performance and user experience. In medical education, contextual adaptability assessment can reveal problems in the teaching process, providing teachers with clear feedback and helping them adjust teaching content, methods, and strategies to better meet students' learning needs.

[0006] The following problems exist in the existing technology: 1. Most existing intelligent simulated patient systems lack emotional computing capabilities. Some systems are not flexible enough in speech recognition and dialogue generation, have simple emotional expression, unnatural communication, and cannot dynamically adapt to clinical situations.

[0007] 2. Traditional intelligent simulated patient evaluation systems use relatively limited data sources and often rely on prior knowledge to set a fixed classification framework during model design. They do not adequately consider the differences between different scenarios, lack a systematic scenario-adaptive evaluation mechanism, and have a single evaluation dimension, making it difficult to comprehensively assess the quality of the output content.

[0008] 3. Existing systems often rely on simple decision trees or scripts for dialogue management, which makes it difficult to handle complex dialogue processes. When responding to users, they usually follow fixed patterns and express themselves mechanically, failing to achieve a flexible and natural interactive experience and lacking adaptive learning capabilities.

[0009] 4. Existing intelligent patient simulation systems are weak in processing real-time data and making dynamic adjustments. Limited by plain text interfaces and a lack of multimodal data integration, they are difficult to adapt to new medical scenarios or changing medical needs, and their application scenarios are often limited to traditional medical practice teaching. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the prior art and provide an intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing.

[0011] The objective of this invention is achieved through the following technical solution: an intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing, comprising the following steps: Corpus construction and context annotation stage: Collect clinical dialogue texts and voices from multiple data sources and different medical scenarios. The expert system uses a multi-layer annotation system to annotate the dialogues in the corpus with context labels. In the emotion computing model construction phase: synchronously collect dialogue text and voice data from intelligent simulated patient ISPs to construct a multimodal dataset; use the multimodal Transformer architecture to fuse text and voice features to generate multimodal emotion vectors, and use KL divergence and cosine similarity as dual indicators to evaluate the matching degree between the output emotion and the expected medical scenario and optimize the dialogue strategy; The stage of establishing the emotional adaptability evaluation mechanism: establish an evaluation index system covering emotional consistency score, semantic relevance score, and interaction naturalness score, use the analytic hierarchy process and fuzzy comprehensive evaluation to calculate multiple indicators by weighting, generate a comprehensive situational adaptability score, which is used to evaluate the situational adaptability of the output content of the intelligent simulated patient ISP. Real-time optimization and adjustment phase: Through reinforcement learning and expert system feedback, dialogue strategies, language expression and emotional adaptation are optimized in real time, and the dialogue style is dynamically adjusted according to the needs of the case.

[0012] Preferably, the corpus construction and context annotation stage further includes the following steps: Dialogue data covering different disease types, patient age groups, medical institutions, and medical scenarios were selected from medical training cases, patient interview records, and standardized patient simulation dialogues, with increased weighting for dialogue data of rare diseases and complex cases. Establish a multi-layered annotation system that includes disease type, patient emotional state, and medical communication strategies; the expert system uses multi-person annotation and cross-validation to label each dialogue data based on the multi-layered annotation system.

[0013] Preferably, the emotion computing model construction stage further includes the following steps: The dialogue text of the intelligent simulated patient ISP is recorded. Natural speech processing methods are used to perform word segmentation, noise reduction, and standardization preprocessing on the dialogue text, and speech data corresponding to the content of the dialogue text is collected. The speech data includes pitch, speech rate, and pitch variation. The pre-trained BERT model is used to contextually encode the dialogue text, extract semantic vectors, and capture the sentiment tendency and fine-grained emotion categories in the dialogue text; at the same time, TF-IDF and word embedding methods are combined to improve the representation ability of text features. Short-time spectral features of speech signals are extracted by deep convolutional neural networks, and temporal dependencies are captured by BiLSTM networks to generate speech emotion vectors. An end-to-end speech recognition method is used to extract semantic information from the speech data and enhance the feature fusion effect. By using a cross-modal attention layer, the association weights between text features and speech features are calculated, and dynamic fusion is performed through an attention mechanism to generate a fused multimodal sentiment vector. The KL divergence is obtained by calculating the difference in probability distribution between the multimodal sentiment vector and the sentiment label of the scene annotation; By combining vector normalization methods, cosine similarity is obtained by measuring the directional consistency between multimodal sentiment vectors and scene-expected sentiment vectors in semantic space. The KL divergence and cosine similarity are weighted to generate an emotional consistency score, and a dynamic weight adjustment strategy is adopted to optimize the scoring model according to the needs of different scenarios.

[0014] Preferably, the establishment phase of the emotional adaptability evaluation mechanism further includes the following steps: The matching degree between the patient's output emotion in the intelligent simulation and the expected patient emotion is measured by two indicators: the accuracy of emotion classification and the matching degree of emotion intensity. The BERT model was used to calculate the semantic matching degree between the responses of the intelligent simulated patient's ISP and the clinical context. The semantic relevance score was obtained by combining semantic role labeling and dependency parsing. A combination of language model scoring and human evaluation was used to score the fluency, logic, and contextual coherence of the dialogue to obtain an interaction naturalness score. The weights of each evaluation indicator were determined using AHP, and the weight allocation was optimized by combining expert scoring and statistical analysis. The scores of each evaluation indicator are fuzzified and combined with weights to generate a comprehensive context suitability score. The stability of the evaluation results is improved by using fuzzy membership functions and fuzzy comprehensive evaluation matrices.

[0015] Preferably, the real-time optimization and adjustment stage further includes the following steps: A reinforcement learning environment was established, using the dialogue strategy of the intelligent simulated patient ISP as the learning object, defining the state space, action space, and reward function; combined with feedback from the expert system, the speech expression and emotional adaptability of the intelligent simulated patient ISP were continuously optimized; and a combination of online and offline learning methods was adopted to improve the adaptability of the model. The system analyzes the characteristics and needs of the current medical context, including disease type, patient emotions, and healthcare roles; it uses context-aware methods to capture context changes in real time; based on the context analysis results, it dynamically adjusts the dialogue strategy of the intelligent simulated patient ISP to ensure that its output content is highly consistent with real doctor-patient communication in terms of style, tone, and word choice; it uses a multimodal generation model to generate dialogue content that meets contextual needs; and it continuously optimizes the dialogue strategy of the intelligent simulated patient ISP by monitoring the dialogue effect in real time to improve its context adaptability.

[0016] The beneficial effects of this invention are: 1) This invention uses a multimodal text sentiment computing model to accurately identify, analyze, and match the emotional expressions in intelligent simulated patients' dialogues. By combining text and voice dual-modal data collection, it breaks through the limitations of single text sentiment analysis, achieving accurate identification and dynamic adaptation of emotional expression, making the dialogue more emotionally natural and realistic.

[0017] 2) This invention constructs a multi-dimensional evaluation system, including multiple indicators such as emotional consistency, semantic relevance, and interaction naturalness. It also uses the analytic hierarchy process and fuzzy comprehensive evaluation algorithm to generate a comprehensive contextual suitability score, which can more comprehensively and accurately evaluate the suitability of the intelligent simulated patient's output content.

[0018] 3) This invention introduces a reinforcement learning mechanism, which, combined with feedback from professionals, dynamically adjusts the dialogue style, tone, and word choice of the intelligent simulated patient, enabling it to adapt to different medical situations, optimize language expression and emotional adaptability, improve the intelligence level of the system, and promote the realism of the interaction.

[0019] 4) This invention is not only applicable to clinical communication training for medical students, but can also be widely applied to various medical scenarios such as general practitioner consultation simulation and psychological treatment scenarios. It has stronger versatility and scalability, and can meet the needs of different medical education and clinical simulation.

[0020] 5) Enhancing the context adaptability of intelligent simulated patients: This invention utilizes diverse data sources of clinical dialogue text and speech to construct a corpus of simulated patients and develops a context adaptability evaluation method. By optimizing the dialogue through a reinforcement learning-based optimization mechanism, the context adaptability of intelligent simulated patients is effectively enhanced.

[0021] 6) Enhancing the emotional naturalness of interactive training: This invention employs a multimodal text sentiment computing model, breaking through the limitations of traditional single-text sentiment analysis and accurately capturing the emotional expression characteristics of users. This multimodal sentiment recognition and analysis capability enables the intelligent simulated patient to exhibit more natural and realistic emotional responses when interacting with users, significantly improving the emotional naturalness of interactive training.

[0022] 7) Constructing an intelligent evaluation system to improve the quality of medical training: The evaluation system of this invention comprehensively considers multiple dimensions such as emotional consistency, semantic relevance, and interaction naturalness. Through the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation algorithm, multiple evaluation indicators are weighted and calculated to generate a comprehensive situational adaptability score, providing a scientific and accurate evaluation basis for medical training.

[0023] 8) Optimize the adaptive learning ability of intelligent simulated patients: Based on the reinforcement learning optimization mechanism and combined with feedback from professionals, this invention can dynamically adjust the dialogue strategy of intelligent simulated patients, enabling them to adapt to different medical situations.

[0024] 9) Applicable to a variety of medical scenarios: This invention can flexibly adapt to different medical needs. It is not only suitable for clinical communication training for medical students, but can also be widely used in diverse medical scenarios such as general practitioner consultation simulation and psychological diagnosis and treatment. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for intelligent evaluation of the contextual adaptability of dialogue content based on text sentiment computing. Detailed Implementation

[0026] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This invention addresses the context-appropriateness evaluation of output content from intelligent simulated patient systems. It proposes a method based on multimodal text sentiment computing and deep learning for evaluating the context-appropriateness of output content in intelligent simulated patient systems. This method focuses on solving key problems in existing intelligent simulated patient systems, such as limited emotional expression, insufficient dynamic adaptability, and the lack of a systematic evaluation mechanism in clinical context interactions. Utilizing text sentiment computing technology, this method identifies, analyzes, and matches the emotions in the dialogue content of intelligent simulated patients, constructing a cross-modal emotion consistency model. This overcomes the limitations of single-text emotion recognition, accurately capturing the emotional expression characteristics of intelligent simulated patients and establishing multi-dimensional indicators (emotional consistency, semantic relevance, and interaction naturalness). It achieves a quantitative score for the adaptability of output content to clinical scenarios, and can be widely applied to clinical communication training for medical students, simulated consultations for general practitioners, and psychological treatment scenarios, providing a novel intelligent solution for the field of medical education technology. This method effectively improves the interaction quality of intelligent simulated patients, making their dialogue content more consistent with clinical context needs, thereby enhancing the authenticity and effectiveness of doctor-patient communication training in medical education.

[0028] See Figure 1 This invention provides a technical solution: an intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing, comprising the following steps: Corpus construction and context annotation stage: Collect clinical dialogue texts and voices from multiple data sources and different medical scenarios. The expert system uses a multi-layer annotation system to annotate the dialogues in the corpus with context labels. In the emotion computing model construction phase: synchronously collect dialogue text and voice data from intelligent simulated patient ISPs to construct a multimodal dataset; use the multimodal Transformer architecture to fuse text and voice features to generate multimodal emotion vectors, and use KL divergence and cosine similarity as dual indicators to evaluate the matching degree between the output emotion and the expected medical scenario and optimize the dialogue strategy; The stage of establishing the emotional adaptability evaluation mechanism: establish an evaluation index system covering emotional consistency score, semantic relevance score, and interaction naturalness score, use the analytic hierarchy process and fuzzy comprehensive evaluation to calculate multiple indicators by weighting, generate a comprehensive situational adaptability score, which is used to evaluate the situational adaptability of the output content of the intelligent simulated patient ISP. Real-time optimization and adjustment phase: Through reinforcement learning and expert system feedback, dialogue strategies, language expression and emotional adaptation are optimized in real time, and the dialogue style is dynamically adjusted according to the needs of the case.

[0029] In this embodiment, affective computing technology, multimodal data fusion, and context-adaptability evaluation methods are employed, combined with reinforcement learning optimization techniques, to effectively overcome the limitations of existing intelligent patient simulation systems in terms of expression, emotion recognition and analysis, context adaptability, and adaptive learning capabilities. Compared with existing technologies, this invention exhibits significant advantages in context adaptability, intelligent evaluation system, and adaptive learning capabilities, providing a more realistic, natural, and intelligent solution for medical education and clinical simulation.

[0030] This invention utilizes technologies such as natural language processing, affective computing, and multimodal affective analysis to construct a comprehensive assessment system for the contextual adaptability of intelligent simulated content. This method can effectively evaluate the appropriateness of dialogue content between intelligent simulated patients in diverse medical scenarios, thereby enhancing the realism, coherence, and adaptability of their interactions.

[0031] The main contents include the following: 1. Corpus construction and context labeling: In order to ensure the context adaptability of the output content of the intelligent simulated patient, this method first needs to construct a standardized medical dialogue corpus and perform clinical context labeling. (1) Data collection: Collect clinical dialogue text and voice from multiple data sources such as medical training cases, patient interview records, and standardized patient simulation dialogues. It covers different medical scenarios including initial diagnosis, follow-up, emergency, preoperative, and postoperative. (2) Manual annotation of context labels: Form a medical and nursing professional team and use a multi-layer annotation system to label the dialogue content in the corpus with scenarios such as disease type (chronic disease, acute disease, mental illness, etc.), patient emotional state (anxiety, anger, sadness, calmness, etc.), and medical communication strategies (comfort, explanation, guidance, empathy, etc.).

[0032] 2. Intelligent simulated patient emotion computing model: In order to evaluate whether the emotional expression of the intelligent simulated patient dialogue matches the medical scenario, break through the limitations of single text emotion analysis, and realize the accurate identification and dynamic adaptation of the emotional expression of the intelligent simulated patient, this method designs a multimodal text emotion computing model. The specific steps are as follows: (1) Simultaneous acquisition of dual-modal data: Simultaneously collect the text dialogue content and speech output data (such as pitch, speech rate, tone changes, etc.) of the intelligent simulated patient, and construct a time-aligned multimodal dataset. (2) Cross-modal feature fusion model: Adopt the multimodal Transformer architecture to dynamically fuse text features (BERT encoding) and speech features (acoustic features extracted based on CNN+BiLSTM) through the attention mechanism. Use the pre-trained BERT model to perform context encoding on the dialogue text and extract semantic vectors; capture the emotional tendency (positive, neutral, negative) and fine-grained emotion categories (such as anxiety, sadness, calmness) in the text through the BiLSTM+Attention network. Using deep convolutional neural networks (CNNs), short-time spectral features (MFCC, fundamental frequency, energy, etc.) of speech signals are extracted, and BiLSTM is used to capture temporal dependencies to generate speech emotion vectors (such as anger, calmness, tension). A cross-modal attention layer is designed to calculate the association weights of text and speech features and generate fused multimodal emotion vectors. (3) Emotional consistency analysis: Quantitatively evaluate the matching degree between the emotions output by the intelligent simulated patient and the expected emotions in the medical scenario, and optimize the dialogue strategy in real time. The KL divergence and cosine similarity are used for joint evaluation: KL divergence is used to calculate the difference between the probability distribution of the emotional output of the intelligent simulated patient (multimodal fusion result) and the emotional label of the scenario. The smaller the difference, the higher the fit. Cosine similarity is used to measure the directional consistency between the output emotion vector and the expected emotion vector in the semantic space. Finally, the emotional consistency is calculated by weighting the two indicators.

[0033] 3. Context Adaptability Evaluation Mechanism: In order to conduct a comprehensive context adaptability evaluation of the output content of the intelligent simulated patient, this part designs a multi-dimensional evaluation mechanism, including: (1) Evaluation index system: The emotional consistency score is used to measure whether the output emotion of the intelligent simulated patient is consistent with the expected patient emotion; the semantic relevance score is used to calculate whether the response of the intelligent simulated patient is consistent with the clinical context (the BERT model is used for semantic matching); the interaction naturalness score is used to score the fluency, logic and context coherence of the dialogue. (2) Evaluation algorithm: The Analytic Hierarchy Process (AHP) and Fuzzy Comprehensive Evaluation are used to perform weighted calculations on multiple evaluation indicators to generate a comprehensive context adaptability score; the score range is set to 0-100, and the higher the score, the better the context adaptability of the output content of the intelligent simulated patient.

[0034] 4. Optimization and Real-time Adjustment of Intelligent Simulated Patient Dialogue: Based on the above evaluation results, this invention also proposes an optimization mechanism based on reinforcement learning to optimize the text generation strategy of intelligent simulated patients, specifically including: (1) Feedback mechanism: Adjust the dialogue strategy of intelligent simulated patients through reinforcement learning environment so that it can adapt to different medical situations; Combined with feedback from professionals, continuously optimize the language expression and emotional adaptability of intelligent simulated patients. (2) Adaptive situation adjustment: According to the needs of the case, dynamically adjust the dialogue style, tone, word choice, etc. of intelligent simulated patients to make them more in line with the needs of real doctor-patient communication.

[0035] In some embodiments, the corpus construction and context annotation stage further includes the following steps: Dialogue data covering different disease types, patient age groups, medical institutions, and medical scenarios were selected from medical training cases, patient interview records, and standardized patient simulation dialogues, with increased weighting for dialogue data of rare diseases and complex cases. Establish a multi-layered annotation system that includes disease type, patient emotional state, and medical communication strategies; the expert system uses multi-person annotation and cross-validation to label each dialogue data based on the multi-layered annotation system.

[0036] In this embodiment, clinical dialogue text and voice data are selected from various data sources, including medical training cases, patient interview records, and standardized patient simulation dialogues. The data covers dialogue data from different disease types, patient age groups, and medical institutions. The collected data covers different medical scenarios such as initial diagnosis, follow-up visits, emergency care, pre-operative care, and post-operative care, with particular attention paid to dialogue data from rare diseases and complex cases.

[0037] Manual annotation of contextual tags: A multi-layered annotation system was designed, including disease type (e.g., chronic disease, acute disease, mental illness), patient emotional state (e.g., anxiety, anger, sadness, calmness), and healthcare communication strategies (e.g., comforting, explaining, guiding, empathizing). The annotation team meticulously tagged each dialogue entry according to this system. For example, "I've been feeling anxious lately, and I can't sleep well at night" could be tagged as "Disease type: mental illness, patient emotional state: anxiety, healthcare communication strategy: comforting, dialogue stage: description of symptoms." Multiple annotations and cross-validation were used to ensure consistency in the annotations.

[0038] In some embodiments, the emotion computing model construction phase further includes the following steps: The dialogue text of the intelligent simulated patient ISP is recorded. Natural speech processing methods are used to perform word segmentation, noise reduction, and standardization preprocessing on the dialogue text, and speech data corresponding to the content of the dialogue text is collected. The speech data includes pitch, speech rate, and pitch variation. The pre-trained BERT model is used to contextually encode the dialogue text, extract semantic vectors, and capture the sentiment tendency and fine-grained emotion categories in the dialogue text; at the same time, TF-IDF and word embedding methods are combined to improve the representation ability of text features. Short-time spectral features of speech signals are extracted by deep convolutional neural networks, and temporal dependencies are captured by BiLSTM networks to generate speech emotion vectors. An end-to-end speech recognition method is used to extract semantic information from the speech data and enhance the feature fusion effect. By using a cross-modal attention layer, the association weights between text features and speech features are calculated, and dynamic fusion is performed through an attention mechanism to generate a fused multimodal sentiment vector. The KL divergence is obtained by calculating the difference in probability distribution between the multimodal sentiment vector and the sentiment label of the scene annotation; By combining vector normalization methods, cosine similarity is obtained by measuring the directional consistency between multimodal sentiment vectors and scene-expected sentiment vectors in semantic space. The KL divergence and cosine similarity are weighted to generate an emotional consistency score, and a dynamic weight adjustment strategy is adopted to optimize the scoring model according to the needs of different scenarios.

[0039] In this embodiment, dual-modal data synchronous acquisition utilizes a high-precision microphone and professional recording equipment to ensure the accuracy of the speech data. The dialogue text generated by the intelligent simulated patient is recorded, and natural language processing techniques are used to preprocess the text, including word segmentation, noise reduction, and standardization. Speech output data corresponding to the text dialogue content is acquired, including acoustic features such as pitch, speech rate, and tone variations.

[0040] Cross-modal feature fusion model: A pre-trained BERT model is used to contextually encode the dialogue text, extract semantic vectors, and capture the sentiment tendency and fine-grained emotion categories in the text. TF-IDF and word embedding techniques are combined to further enhance the representational ability of text features. A deep convolutional neural network (CNN) is used to extract short-time spectral features of the speech signal (such as MFCC, fundamental frequency, energy, etc.), and a BiLSTM network is combined to capture temporal dependencies, generating speech sentiment vectors (such as anger, calm, tension, etc.). End-to-end speech recognition technology is employed to extract semantic information from the speech, enhancing the effect of feature fusion. A cross-modal attention layer is designed to calculate the association weights between text and speech features, and dynamic fusion is performed through an attention mechanism to generate a fused multimodal sentiment vector.

[0041] Emotional Consistency Analysis: KL divergence calculation measures the difference between the probability distribution of the emotional output from the intelligent simulated patient (multimodal fusion effect) and the probability distribution of the emotional tags labeled in the scene. A smaller difference indicates higher suitability. Smoothing techniques are used to avoid cases where the probability distribution is 0. Cosine similarity calculation measures the directional consistency between the output emotional vector and the expected emotional vector in the scene within the semantic space. Vector normalization is combined to improve the stability of the similarity calculation. Emotional Consistency Score: The KL divergence and cosine similarity are weighted and calculated to generate the final emotional consistency score. A dynamic weight adjustment strategy is used to optimize the scoring model according to the needs of different scenarios.

[0042] In some embodiments, the establishment phase of the emotional adaptability evaluation mechanism further includes the following steps: The matching degree between the patient's output emotion in the intelligent simulation and the expected patient emotion is measured by two indicators: the accuracy of emotion classification and the matching degree of emotion intensity. The BERT model was used to calculate the semantic matching degree between the responses of the intelligent simulated patient's ISP and the clinical context. The semantic relevance score was obtained by combining semantic role labeling and dependency parsing. A combination of language model scoring and human evaluation was used to score the fluency, logic, and contextual coherence of the dialogue to obtain an interaction naturalness score. The weights of each evaluation indicator were determined using AHP, and the weight allocation was optimized by combining expert scoring and statistical analysis. The scores of each evaluation indicator are fuzzified and combined with weights to generate a comprehensive context suitability score. The stability of the evaluation results is improved by using fuzzy membership functions and fuzzy comprehensive evaluation matrices.

[0043] In this embodiment, the emotional consistency score uses both emotional classification accuracy and emotional intensity matching degree to measure the matching degree between the emotions output by the intelligent simulated patient and the expected patient emotions. The semantic relevance score uses a BERT model to calculate the semantic matching degree between the intelligent simulated patient's response and the clinical context, combined with semantic role labeling and dependency parsing to improve the accuracy of semantic matching. The interaction naturalness score uses a combination of language model scoring and human evaluation to score the fluency, logic, and contextual coherence of the dialogue.

[0044] The evaluation algorithm uses Analytic Hierarchy Process (AHP) to determine the weights of each evaluation indicator and optimizes the weight allocation by combining expert scoring and statistical analysis. The scores of each evaluation indicator are fuzzified and combined with the weights to generate a comprehensive situational suitability score, ranging from 0 to 100, with higher scores indicating better situational suitability. Fuzzy membership functions and a fuzzy comprehensive evaluation matrix are employed to improve the stability of the evaluation results.

[0045] In some embodiments, the real-time optimization and adjustment phase further includes the following steps: A reinforcement learning environment was established, using the dialogue strategy of the intelligent simulated patient ISP as the learning object, defining the state space, action space, and reward function; combined with feedback from the expert system, the speech expression and emotional adaptability of the intelligent simulated patient ISP were continuously optimized; and a combination of online and offline learning methods was adopted to improve the adaptability of the model. The system analyzes the characteristics and needs of the current medical context, including disease type, patient emotions, and healthcare roles; it uses context-aware methods to capture context changes in real time; based on the context analysis results, it dynamically adjusts the dialogue strategy of the intelligent simulated patient ISP to ensure that its output content is highly consistent with real doctor-patient communication in terms of style, tone, and word choice; it uses a multimodal generation model to generate dialogue content that meets contextual needs; and it continuously optimizes the dialogue strategy of the intelligent simulated patient ISP by monitoring the dialogue effect in real time to improve its context adaptability.

[0046] In this embodiment, a reinforcement learning environment is designed, using the dialogue strategies of the intelligent simulated patient as the learning object. A state space, action space, and reward function are defined to ensure the effectiveness of reinforcement learning. Feedback from professionals is incorporated to continuously optimize the language expression and emotional adaptability of the intelligent simulated patient. A combination of online and offline learning methods is employed to enhance the model's adaptability.

[0047] Adaptive context adjustment analyzes the characteristics and needs of the current medical situation, including disease type, patient emotions, and healthcare roles. Context-aware technology captures real-time changes in the context. Based on the context analysis results, the dialogue strategy of the intelligent simulated patient is dynamically adjusted to ensure that its output content closely matches real doctor-patient communication in terms of style, tone, and vocabulary. A multimodal generation model is used to generate dialogue content that meets the context's requirements. By monitoring the dialogue effect in real time, the dialogue strategy of the intelligent simulated patient is continuously optimized to improve its context adaptability.

[0048] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. An intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing, characterized by: Includes the following steps: Corpus construction and context annotation stage: Collect clinical dialogue texts and voices from multiple data sources and different medical scenarios. The expert system uses a multi-layer annotation system to annotate the dialogues in the corpus with context labels. The affective computing model construction phase involves: synchronously collecting dialogue text and speech data from an intelligent simulated patient ISP to construct a multimodal dataset; using a multimodal Transformer architecture to fuse text and speech features to generate multimodal affective vectors; and employing KL divergence and cosine similarity as dual indicators to evaluate the matching degree between the output affective vectors and the expected medical scenario, thereby optimizing the dialogue strategy. The affective computing model construction phase includes: calculating the probability distribution difference between the multimodal affective vectors and the scene-annotated affective labels to obtain the KL divergence. By combining vector normalization methods, cosine similarity is obtained by measuring the directional consistency between multimodal sentiment vectors and scene-expected sentiment vectors in semantic space. The KL divergence and cosine similarity are weighted to generate an emotional consistency score, and a dynamic weight adjustment strategy is adopted to optimize the scoring model according to the needs of different scenarios. The stage of establishing the emotional adaptability evaluation mechanism: establish an evaluation index system covering emotional consistency score, semantic relevance score, and interaction naturalness score, use the analytic hierarchy process and fuzzy comprehensive evaluation to calculate multiple indicators by weighting, generate a comprehensive situational adaptability score, which is used to evaluate the situational adaptability of the output content of the intelligent simulated patient ISP. Real-time optimization and adjustment phase: Through reinforcement learning and expert system feedback, dialogue strategies, language expression and emotional adaptation are optimized in real time, and the dialogue style is dynamically adjusted according to the needs of the case.

2. The intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing according to claim 1, characterized in that: The corpus construction and context annotation stage also includes the following steps: Dialogue data covering different disease types, patient age groups, medical institutions, and medical scenarios were selected from medical training cases, patient interview records, and standardized patient simulation dialogues, with increased weighting for dialogue data of rare diseases and complex cases. Establish a multi-layered annotation system that includes disease type, patient emotional state, and medical communication strategies; the expert system uses multi-person annotation and cross-validation to label each dialogue data based on the multi-layered annotation system.

3. The intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing according to claim 1, characterized in that: The aforementioned emotion computing model construction phase also includes the following steps: The dialogue text of the intelligent simulated patient ISP is recorded. Natural language processing methods are used to perform word segmentation, noise reduction, and standardization preprocessing on the dialogue text, and voice data corresponding to the content of the dialogue text is collected. The voice data includes pitch, speech rate, and tone variation. The pre-trained BERT model is used to contextually encode the dialogue text, extract semantic vectors, and capture the sentiment tendency and fine-grained emotion categories in the dialogue text; at the same time, TF-IDF and word embedding methods are combined to improve the representation ability of text features. Short-time spectral features of speech signals are extracted by deep convolutional neural networks, and temporal dependencies are captured by BiLSTM networks to generate speech emotion vectors. An end-to-end speech recognition method is used to extract semantic information from the speech data and enhance the feature fusion effect. By using a cross-modal attention layer, the association weights between text features and speech features are calculated, and dynamic fusion is performed through an attention mechanism to generate a fused multimodal sentiment vector.

4. The intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing according to claim 1, characterized in that: The establishment phase of the aforementioned emotional adaptability evaluation mechanism also includes the following steps: The matching degree between the emotions output by the intelligent simulated patient and the emotions expected by the patient was measured by two indicators: the accuracy of emotion classification and the matching degree of emotion intensity. The BERT model was used to calculate the semantic matching degree between the responses of the intelligent simulated patient's ISP and the clinical context. The semantic relevance score was obtained by combining semantic role labeling and dependency parsing. A combination of language model scoring and human evaluation was used to score the fluency, logic, and contextual coherence of the dialogue to obtain an interaction naturalness score. The weights of each evaluation indicator were determined using AHP, and the weight allocation was optimized by combining expert scoring and statistical analysis. The scores of each evaluation indicator are fuzzified and combined with weights to generate a comprehensive context suitability score. The stability of the evaluation results is improved by using fuzzy membership functions and fuzzy comprehensive evaluation matrices.

5. The intelligent evaluation method for the contextual adaptability of dialogue content based on text sentiment computing according to claim 1, characterized in that: The real-time optimization and adjustment phase also includes the following steps: A reinforcement learning environment was established, using the dialogue strategy of the intelligent simulated patient ISP as the learning object, defining the state space, action space, and reward function; combined with feedback from the expert system, the speech expression and emotional adaptability of the intelligent simulated patient ISP were continuously optimized; and a combination of online and offline learning methods was adopted to improve the adaptability of the model. The system analyzes the characteristics and needs of the current medical context, including disease type, patient emotions, and healthcare roles; it uses context-aware methods to capture context changes in real time; based on the context analysis results, it dynamically adjusts the dialogue strategy of the intelligent simulated patient ISP to ensure that its output content is highly consistent with real doctor-patient communication in terms of style, tone, and word choice; it uses a multimodal generation model to generate dialogue content that meets contextual needs; and it continuously optimizes the dialogue strategy of the intelligent simulated patient ISP by monitoring the dialogue effect in real time to improve its context adaptability.

Images