A robot customer service problem intelligent processing method and system

Abstract

The application provides a robot customer service problem intelligent processing method and system, the method comprises the following steps: obtaining a user problem and performing feature extraction and integration on the user problem; constructing a user dynamic causal reasoning model to generate a multi-level problem solving strategy according to the user problem; constructing an adaptive response strategy generation model according to the generated multi-level problem solving strategy and the integrated feature vector to generate a response strategy most suitable for the current corresponding context; executing the response strategy and judging whether to trigger learning according to the user feedback; performing real-time adaptive learning and optimization on the adaptive response strategy generation model through user feedback and historical user feedback; analyzing the long-term trend of the model parameters of the adaptive response strategy generation model after multiple optimizations and historical user feedback, performing global optimization combined with the long-term trend, and generating a final response strategy. The application can perform well in a complex user service scene.

Description

Technical Field

[0001] This invention belongs to the technical field of artificial intelligence, and in particular relates to a method and system for intelligent handling of robot customer service questions. Background Technology

[0002] In modern business-to-user interactions, chatbot systems have become a crucial tool. These systems automate the processing of numerous user requests, providing real-time service and reducing operating costs. Current chatbot systems primarily rely on predefined rules, knowledge bases, or machine learning models trained on large amounts of labeled data. These systems perform well when handling simple, structured queries, such as account balance inquiries, order status checks, or operation guides. However, with the diversification and complexity of user needs, existing chatbot systems face numerous challenges and limitations.

[0003] First, existing systems lack a deep understanding of complex problems. Many user questions are not simply information queries, but complex issues involving multiple factors. For example, a user's problem may arise from the combined effect of several independent factors. Existing systems struggle to analyze the root causes of these complex problems, often providing only superficial and incomplete answers, leading to a decline in user experience and satisfaction. Second, existing systems are insufficiently adaptable to new problems. Traditional chatbot systems rely on static rules or models that require frequent updates and maintenance to address constantly changing user needs and question types. This manual updating and maintenance is time-consuming and labor-intensive, and cannot respond promptly to new problems, resulting in poor system performance when faced with unseen issues. Furthermore, many chatbot systems are significantly inadequate in handling user emotions and understanding context. Emotion recognition and context understanding are crucial for improving user experience, but most current systems lack accurate identification of user emotions and effective utilization of dialogue context, resulting in stiff and impersonal responses.

[0004] Existing technologies also suffer from insufficient utilization of user feedback and interaction data. While many systems can collect vast amounts of interaction data, effectively using this data for system optimization and self-improvement remains a challenge. Most systems cannot achieve real-time learning and optimization based on user feedback, thus hindering continuous improvement in their problem-solving capabilities and service levels. Especially when facing complex and ever-changing user needs and market environments, such systems lacking dynamic adaptability struggle to provide a high-quality service experience. Summary of the Invention

[0005] The purpose of this invention is to design an intelligent problem-solving method and system for robot customer service. It achieves a deep understanding of complex problems through causal reasoning and multi-level analysis, enhances contextual understanding and decision-making capabilities by utilizing biological neural network simulation and recurrent neural networks, improves the system's rapid adaptability through few-shot learning, and achieves self-optimization capabilities by combining reinforcement learning. These innovative points are organically combined to overcome the technical shortcomings of existing robot customer service systems, enabling this invention to perform exceptionally well in complex user service scenarios.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for intelligent handling of robot customer service questions, the method comprising the following steps:

[0007] S1. Obtain user questions and extract and integrate their features, specifically including:

[0008] S101. Obtain the user's complete input dialogue and preprocess the dialogue. Convert the preprocessed dialogue into a text representation and embed each word in the text as a semantic feature. At the same time, design an emotion embedding model to extract emotion features from the input dialogue and design a context feature extractor to capture the context relationship of the user in multiple rounds of dialogue to obtain context features.

[0009] S102. Generate a complexity score based on semantic features, sentiment features, and contextual features, and generate a comprehensive feature vector from the semantic features, sentiment features, contextual features, and complexity score;

[0010] S2. Construct a user dynamic causal reasoning model based on the integrated feature vectors. Generate multi-level problem-solving strategies based on user questions, specifically including:

[0011] S201. Construct a dynamic causal reasoning model, take the comprehensive feature vector as input, reason about the causal dependencies behind the user's question, and establish a causal function for each feature by analyzing the dependencies of contextual features and semantic features.

[0012] S202. Based on the complexity of user problems, user problems are divided into three levels, including: surface feature layer, potential cause layer, and root cause layer.

[0013] S203. Generate three problem-solving strategies for the three levels respectively, including: surface level strategy for problems at the surface feature level, potential level strategy for problems at the potential cause level, and root level strategy for problems at the root cause level.

[0014] S204. If a certain strategy fails to solve the problem effectively, the causal relationship between features is re-evaluated through real-time user feedback signals, and the causal dependency weights in the causal function are dynamically adjusted.

[0015] S3. Based on the generated multi-level problem-solving strategies and the integrated feature vectors, construct an adaptive response strategy generation model to generate the response strategy best suited to the current context, including:

[0016] S301. Construct an adaptive response strategy generation model to generate response strategies at different levels based on the level of the user's question and its corresponding comprehensive feature vector.

[0017] S302. Add regularization terms based on sentiment characteristics and question complexity to the response strategy to constrain the generation of the response strategy;

[0018] S303. Generate an initial response strategy and dynamically adjust the response strategy based on real-time user feedback to obtain the adjusted response strategy;

[0019] S304. Merge the multi-level response strategies generated by the user to generate the final response strategy, and update the model parameters of the adaptive response strategy generation model in real time based on the feedback from user interaction.

[0020] S4. Execute the agreement strategy and determine whether to trigger learning based on user feedback, including:

[0021] Execute the final response strategy and listen for user feedback. Extract feedback signals from the user feedback and determine whether to trigger learning based on the feedback signals:

[0022] If the feedback signal indicates a poor response, the learning module is triggered to update the response model;

[0023] If the feedback signal indicates a good response, then learning is not triggered;

[0024] S5. Real-time adaptive learning and optimization of the adaptive response strategy generation model based on user feedback and historical user feedback;

[0025] S6. Analyze the long-term trend of model parameters of the adaptive response strategy generation model based on historical user feedback and multiple optimizations, perform global optimization based on the long-term trend, optimize the response strategy that best suits the current context, and generate the final response strategy.

[0026] Furthermore, each word in the text is embedded as a vector to introduce a context-dependent feature representation, and the semantic information of each word is captured using the context information of the entire dialogue; the sentiment embedding model maps the dialogue content to the sentiment feature space, captures the user's emotional expression in specific sentences, and combines the dialogue context to obtain the dynamic process of sentiment changes over time; the context feature extractor is used to capture the contextual relationships of the user in multi-turn dialogues.

[0027] Furthermore, the dynamic causal reasoning model is expressed as follows:

[0028] Define each feature in the cause-effect graph as a node v i The edges between all nodes have dependency weights w. i Indicates the causal relationship between features;

[0029] By analyzing historical data, a causal relationship model was derived, as follows:

[0030]

[0031] in, Pa(v) represents the loss function of the causal relationship model. i ) represents feature v i The set of causal parent nodes, f i Let i represent the causal function for estimating the features, and let i represent the number of features.

[0032] The surface symptom layer is used to process semantic and emotional features, directly reflecting the problems currently expressed by the user;

[0033] The potential cause layer is used to handle complex problems, and identifies the potential causes behind surface problems based on contextual features and problem complexity, and through a causal relationship graph.

[0034] The root cause layer is used to trace the root cause of a problem by combining emotional features, contextual features, and historical data from multi-turn dialogues.

[0035] The surface hierarchy strategy directly generates standardized responses to problems at the surface feature layer;

[0036] The potential hierarchy strategy generates a dialogue strategy to guide the user to the next step of describing the problem for the potential cause layer.

[0037] The fundamental level strategy addresses the root cause of the problem by generating complex response schemes based on emotional fluctuations and contextual information from long-term dialogue history.

[0038] Further, S301 specifically includes: for the surface symptom layer, the response strategy is based on semantic features and emotional features; for the potential cause layer and the root cause layer, a detailed response is generated based on semantic features and emotional features, as well as in combination with context features and complexity scores.

[0039] Meanwhile, hierarchical adaptive coefficients are introduced in the surface symptom layer, potential cause layer, and root cause layer to control the complexity of the response strategy at different problem levels. The largest hierarchical adaptive coefficient indicates the most complex problem.

[0040] Furthermore, the real-time user feedback signals include emotional fluctuations and contextual changes; the merging of the multi-level response strategies generated by the user is performed by weighted summation.

[0041] Furthermore, the feedback signal includes sentiment change and semantic matching degree, wherein the sentiment change represents the emotional fluctuation from the current dialogue round to the previous dialogue round, and the semantic matching degree represents the semantic matching between the robot's response and the user's question.

[0042] Furthermore, once the learning trigger is activated, the adaptive response strategy generation model is updated based on the feedback signal. The adaptive response strategy generation model is updated using the gradient descent method, which optimizes the adaptive response strategy generation model by minimizing the loss function.

[0043] Furthermore, S5 specifically includes:

[0044] S501. Design a regularization term based on semantic and sentiment features and construct a loss function. (F,R final ), which is represented as follows:

[0045]

[0046] Where F represents the real-time feedback signal, R final This represents the final response strategy, where α and β represent the weights of the balancing feedback signal and the deviation from the response strategy; F optimal Represents the preset ideal feedback signal; R optimal γ represents the ideal response strategy; γ represents the weight coefficient of the semantic and sentiment-based regularization term. The regularization term, based on semantics and sentiment, is used to ensure that the adaptive response strategy generation model balances semantic and sentiment factors during adjustment, and is represented as follows:

[0047]

[0048] Where e represents the sentiment feature, X represents the semantic feature, and e optimal X represents the ideal emotional characteristic. optimal λ1 and λ2 represent the ideal semantic features, and λ1 and λ2 represent the weights that balance sentiment and semantic bias.

[0049] S502. An adaptive learning rate is introduced, which is dynamically adjusted according to the strength and complexity of the feedback signal. This ensures that the adaptive response strategy generation model can adjust the fastest when the feedback signal is at its maximum, and can update the adaptive response strategy generation model when the feedback signal is at its weakest. The adaptive learning rate η is expressed as follows:

[0050]

[0051] Where η0 represents the initial learning rate; κ represents the adjustment coefficient; and q represents the problem complexity.

[0052] S503. Construct a historical feedback accumulation term to combine feedback signals from multi-turn dialogues and guide the adaptive response strategy generation model for long-term optimization.

[0053] S504. After the adaptive response strategy generation model is updated, a dynamic verification mechanism is designed to verify whether the adaptive response strategy generation model has been updated, as shown below:

[0054]

[0055] if A negative value indicates successful model optimization; otherwise, adjustments will be made to ensure that the model gradually converges to the ideal response.

[0056] Furthermore, S6 specifically includes:

[0057] S601. Construct a long-term feedback trend function based on the problem classification characteristics, analyze the long-term changing trend of the feedback signal, and identify long-term deviation points and optimization space; wherein, the problem classification characteristics are extracted from the long-term feedback signal trend of the response strategy of multi-level problems.

[0058] S602. Optimize the model parameters θ of the adaptive response strategy generation model globally based on the long-term trend of the feedback signal.

[0059] S603. A strategy evolution factor is introduced during the long-term optimization of the adaptive response strategy generation model, and the model is dynamically adjusted according to the rules of the adaptive response strategy generation model based on the long-term feedback signal.

[0060] S604. An adaptive architecture extension mechanism was designed to automatically extend different model modules or update the functionality of existing modules based on the volatility of feedback signals, dialogue complexity, and the diversity of user needs.

[0061] S605. After model updates and policy optimizations, the effectiveness of long-term optimization is verified through testing.

[0062] A second aspect of the present invention provides a robot customer service problem intelligent processing system, the system comprising:

[0063] The user issue acquisition subsystem is used to acquire user issues, extract and integrate their features, and specifically includes:

[0064] S101. Obtain the user's complete input dialogue and preprocess the dialogue. Convert the preprocessed dialogue into a text representation and embed each word in the text as a semantic feature. At the same time, design an emotion embedding model to extract emotion features from the input dialogue and design a context feature extractor to capture the context relationship of the user in multiple rounds of dialogue to obtain context features.

[0065] S102. Generate a complexity score based on semantic features, sentiment features, and contextual features, and generate a comprehensive feature vector from the semantic features, sentiment features, contextual features, and complexity score;

[0066] The user problem analysis subsystem is used to construct a dynamic causal reasoning model for users based on the integrated feature vectors and to generate multi-level problem-solving strategies based on user problems, specifically including:

[0067] S201. Construct a dynamic causal reasoning model, take the comprehensive feature vector as input, reason about the causal dependencies behind the user's question, and establish a causal function for each feature by analyzing the dependencies of contextual features and semantic features.

[0068] S202. Based on the complexity of user problems, user problems are divided into three levels, including: surface feature layer, potential cause layer, and root cause layer.

[0069] S203. Generate three problem-solving strategies for the three levels respectively, including: surface level strategy for problems at the surface feature level, potential level strategy for problems at the potential cause level, and root level strategy for problems at the root cause level.

[0070] S204. If a certain strategy fails to solve the problem effectively, the causal relationship between features is re-evaluated through real-time user feedback signals, and the causal dependency weights in the causal function are dynamically adjusted.

[0071] The user question response subsystem is used to construct an adaptive response strategy generation model based on the generated multi-level question-solving strategies and the integrated feature vectors to generate a response strategy that best suits the current context, including:

[0072] S301. Construct an adaptive response strategy generation model to generate response strategies at different levels based on the level of the user's question and its corresponding comprehensive feature vector.

[0073] S302. Add regularization terms based on sentiment characteristics and question complexity to the response strategy to constrain the generation of the response strategy;

[0074] S303. Generate an initial response strategy and dynamically adjust the response strategy based on real-time user feedback to obtain the adjusted response strategy;

[0075] S304. Merge the multi-level response strategies generated by the user to generate the final response strategy, and update the model parameters of the adaptive response strategy generation model in real time based on the feedback from user interaction.

[0076] The response learning subsystem is used to execute the response strategy and determine whether to trigger learning based on user feedback, including:

[0077] Execute the final response strategy and listen for user feedback. Extract feedback signals from the user feedback and determine whether to trigger learning based on the feedback signals:

[0078] If the feedback signal indicates a poor response, the learning module is triggered to update the response model;

[0079] If the feedback signal indicates a good response, then learning is not triggered;

[0080] The optimization subsystem is used to perform real-time adaptive learning and optimization of the adaptive response strategy generation model based on user feedback and historical user feedback. It analyzes the long-term trends of the model parameters of the adaptive response strategy generation model after multiple optimizations based on historical user feedback, performs global optimization based on the long-term trends, optimizes the response strategy that is most suitable for the current context, and generates the final response strategy.

[0081] The beneficial technical effects of the present invention are at least as follows:

[0082] This invention proposes an adaptive chatbot customer service system based on multi-level intelligent analysis, aiming to solve the aforementioned problems in existing technologies and improve the performance of chatbot customer service systems in complex scenarios. First, this invention employs causal reasoning and multi-level problem analysis techniques, overcoming the limitations of existing systems in deeply understanding complex problems. Through a causal reasoning model, the system can analyze the root causes of user problems, rather than simply addressing surface symptoms. This innovation enables the system to provide more accurate and effective solutions, thereby improving user satisfaction. Through a multi-level problem analysis framework, the system can distinguish different levels of problems, including surface symptoms, underlying causes, and root causes, making solutions more targeted.

[0083] Secondly, this invention introduces biological neural network simulation and recurrent neural network technology to enhance the system's decision-making and contextual understanding capabilities. By mimicking the information processing methods of biological neural networks, the system can better understand the user's intent and the problem's background when facing complex decision-making scenarios. The introduction of recurrent neural networks allows the system to capture the dynamic evolution of the problem and changes in the user's intent when handling dialogues spanning long periods, thereby providing more coherent and human-like responses during the dialogue process.

[0084] Furthermore, this invention enhances the system's adaptability to new problems through few-shot learning technology. Traditional chatbot systems require large amounts of labeled data for training, while this invention, through few-shot learning, enables the system to quickly learn new problem types and solutions with minimal data. This rapid adaptability not only significantly reduces the frequency of manual intervention and system updates but also allows the system to respond more promptly to users' new needs and questions.

[0085] Finally, the adaptive learning module of this invention combines user feedback and reinforcement learning mechanisms, enabling it to continuously self-optimize. The system can dynamically adjust its strategies and model parameters based on real-time user feedback, performing self-calibration and optimization. Through this continuous learning and improvement capability, the system can continuously enhance its service level and maintain its competitiveness in a complex and ever-changing market environment.

[0086] In summary, the main innovations of this invention include: achieving a deep understanding of complex problems through causal reasoning and multi-level analysis; enhancing contextual understanding and decision-making capabilities by utilizing biological neural network simulation and recurrent neural networks; improving the system's rapid adaptability through few-shot learning; and achieving self-optimization capabilities by combining reinforcement learning. These innovations are organically combined to overcome the technical shortcomings of existing robot customer service systems, enabling this invention to perform exceptionally well in complex user service scenarios. Attached Figure Description

[0087] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0088] Figure 1 This is a flowchart of a robot customer service problem intelligent processing method according to an embodiment of the present invention.

[0089] Figure 2 This is a framework diagram of a robot customer service problem intelligent processing system according to an embodiment of the present invention. Detailed Implementation

[0090] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0091] In one or more embodiments, such as Figure 1 As shown, the present invention discloses an intelligent problem-solving method for robot customer service, the method comprising steps 1 to 6, including:

[0092] S1. Obtain user questions and extract and integrate their features.

[0093] In intelligent customer service systems, the diversity and complexity of user input data necessitate targeted preprocessing steps. In particular, to accurately understand user needs, the system must not only process the input text but also capture its sentiment, contextual information, and the complexity of the question. The key to this step lies in extracting multi-dimensional features from the user's natural language input data, laying the foundation for subsequent causal reasoning and problem-solving strategy generation.

[0094] Specifically, firstly, when a user interacts with the customer service system via text or voice, the input data is T = {t1, t2, ..., t...} n} represents the user's complete input dialogue.

[0095] For voice input, it is first converted into text using a speech recognition model. This text then needs to be processed to ensure it is understandable and usable by subsequent models. Text preprocessing includes common steps such as stop word removal and lexical morphology standardization, but more importantly, it involves generating contextual embedding representations of the words, i.e., capturing the semantic information of each word through a context-aware embedding model. Unlike traditional methods, this invention introduces context-related feature representations that fully utilize the context of the entire dialogue. Each word t... i Embedded as vector x i Its generation method relies on contextual information within the dialogue, rather than performing word embeddings in isolation. The embedding method is defined by the following formula:

[0096] x i =ContextEmbedding(t i ,T)

[0097] ContextEmbedding here is not simply word vector generation, but rather semantic representation of words based on the entire dialogue history. This context-aware embedding method ensures that the system not only understands the meaning of a single sentence, but also accurately grasps the core information in longer dialogues, thus providing sufficient semantic basis for subsequent analysis.

[0098] Furthermore, beyond semantic understanding, the user's emotional state plays a crucial role in customer service systems. Traditional systems rely solely on the semantic structure of sentences, easily overlooking user emotions, which often affects the system's response strategy. Therefore, this invention designs an emotion embedding model in the text preprocessing stage to extract emotional features from the input dialogue.

[0099] This invention uses a specially trained emotion recognition module to map dialogue content to an emotion vector space. Emotion embedding not only captures the user's emotional expression in specific statements but also incorporates the dialogue context, considering the dynamic process of emotion changes over time. The emotion feature is represented by e, and expressed by the following formula:

[0100] e = EmotionEmbedding(T)

[0101] Unlike conventional sentiment classification models, this invention incorporates subtle changes in user emotions during sentiment recognition, particularly the emotional fluctuations during multi-turn conversations. This allows the system to not only discern whether a user is expressing confusion or anger, but also to capture trends in emotional changes, helping the system predict the user's subsequent emotional responses. This directly impacts the generation of subsequent response strategies, as the system can more flexibly adjust its strategies to ensure appropriate responses to the user's emotional state.

[0102] Furthermore, in customer service scenarios, a single dialogue statement often cannot independently express a user's problem. Contextual understanding is a crucial step in ensuring that the system can provide accurate responses. Therefore, based on recurrent neural network (RNN) or Transformer architecture, this invention designs a context feature extractor specifically designed to capture the contextual relationships between users in multi-turn dialogues.

[0103] The context feature extractor takes the entire dialogue history as input and outputs a vector c representing the current dialogue state. This vector considers not only the current sentence but also previous dialogue history, ensuring that the system can comprehensively utilize all dialogue information to make decisions.

[0104] c = ContextModel(T)

[0105] This method allows the system to understand the continuity of user questions; even if the user's expression is unclear, the system can still infer the complete intent from the context. Traditional methods typically only process single-sentence input, ignoring the contextual relationships in multi-turn dialogues, while the contextual feature extraction of this invention ensures that the system can handle complex dialogues more intelligently.

[0106] Furthermore, in customer service systems, different questions have varying degrees of complexity. Some simple questions can be resolved with standardized responses, while complex questions require more detailed explanations and processing. Therefore, question complexity assessment becomes a crucial part of the preprocessing stage.

[0107] This invention designs a problem complexity evaluation module that generates a complexity score q by analyzing multiple dimensions such as dialogue length, syntactic complexity, and keyword frequency. This complexity score not only considers the surface features of the dialogue but also incorporates the context vector c to ensure its accuracy.

[0108] q = ComplexityScore(T,X,c)

[0109] This evaluation method allows the system to differentiate between problems of varying complexity and provides crucial reference for generating response strategies for subsequent steps. For more complex problems, the system may need to introduce multi-step dialogues to guide the user step-by-step to a solution; while simple problems can be resolved quickly with a direct response.

[0110] Furthermore, this invention integrates all extracted features to generate a comprehensive feature vector f. This feature vector combines semantic features X, sentiment features e, contextual features c, and problem complexity q, ensuring that subsequent causal reasoning and strategy generation receive sufficient information input.

[0111] f = [X, e, c, q]

[0112] This feature vector is not merely a simple accumulation of semantic information; it simultaneously considers multi-dimensional information such as sentiment, context, and complexity. It incorporates every aspect of user input, providing a complete perspective for the system's subsequent decision-making.

[0113] Through the preprocessing steps described above, this invention not only performs standard text processing on user input but also innovatively combines sentiment analysis, contextual understanding, and question complexity assessment to provide a comprehensive multi-dimensional feature vector for subsequent steps. This feature vector fully reflects the user's intent and the difficulty of the question, ensuring that the system can perform more accurate causal reasoning and response strategy generation in subsequent steps.

[0114] S2. Construct a user dynamic causal reasoning model based on the integrated feature vectors and generate multi-level problem-solving strategies based on user questions.

[0115] In customer service systems, handling complex user issues requires going beyond traditional simple question-and-answer models, especially when the issues have multiple layers of causes. The system must be able to deeply understand the underlying logic of the user input. This process relies not only on surface semantics but also on causal reasoning to identify hidden causal relationships and gradually solve the problem through multi-layered problem analysis. This invention utilizes the multi-dimensional features f = [X, e, c, q] (semantic feature X, sentiment feature e, contextual feature c, and problem complexity q) extracted from step 1 to perform causal reasoning, identify the relationships between features, analyze the problem layer by layer, and generate solution strategies.

[0116] Specifically, the feature vector f obtained in step 1 is used as input, and a causal relationship graph is constructed. This invention can infer the causal dependencies behind user questions. For each user question, there are potential causal relationships between features; for example, the question complexity q may be influenced by both contextual information c and semantic features X. To address this, this invention establishes a dynamic causal inference model that determines the dependencies between features by minimizing residuals.

[0117] This invention defines each feature in a causal graph as a node v. i The edges w between these nodes i This represents the causal relationship between features. Through analysis of historical data, the following causal relationship model is derived:

[0118]

[0119] Where Pa(v) i ) represents feature v i The set of causal parent nodes, and f i To estimate the causal function of the features, for each feature, such as problem complexity q, this invention establishes a causal function f by analyzing the dependency between contextual feature c and semantic feature X. q :

[0120] q≈f q (X,c)

[0121] This causal reasoning not only reveals the dynamic dependencies between features, but also helps the system identify the potential causes of complex user problems by interpreting the features. In this way, the system can update the causal relationships between features in real time and adjust its response strategies for different user problems.

[0122] Furthermore, after constructing the causal relationship diagram, the system needs to perform hierarchical analysis of user problems. This is a key innovation in the patented solution, because user problems are usually not isolated but have different levels of causes. This invention divides problems into three levels: surface symptoms, underlying causes, and root causes.

[0123] Surface Symptom Layer: This layer primarily processes semantic features X and sentiment features e, which directly reflect the problem the user is currently expressing. For example, when a user asks "I can't log in to my account," this is a surface symptom, and the system can generate a direct response based on semantic features and the existing knowledge base.

[0124] Potential Cause Layer: For complex problems, the system identifies underlying causes behind the surface problem using a causal relationship graph. Contextual features *c* and problem complexity *q* play a crucial role at this layer. For example, if the user has mentioned other account-related issues in previous conversations, this could be a potential cause, and the system can further confirm these potential issues by asking the user.

[0125] Root Cause Layer: The final layer traces the root cause of the problem. This requires integrating emotional features (e), context (c), and historical data from multiple rounds of dialogue. For long-standing unresolved problems, changes in user emotions are often one of the root causes; the system can infer the deeper reasons for the problem by progressively analyzing changes in dialogue emotions.

[0126] Through this hierarchical analysis, the system can address problems layer by layer from the surface to the root cause, avoiding the limitation of simple problem-solving systems that only deal with surface symptoms and ignore the underlying causes.

[0127] Furthermore, after identifying the different levels of the problem, the system needs to generate corresponding response strategies for each level. This part is closely related to the causal reasoning model, because the system's response strategies must match the identified causal dependencies and problem levels.

[0128] Surface-level strategy: For surface symptoms, the system can directly generate standardized responses. In this case, response generation primarily relies on semantic features X, for example:

[0129] Response1 = f r (X)

[0130] f in the formula r This represents a standard semantic response function, primarily used to generate fast responses based on semantic features.

[0131] Latent Hierarchy Strategy: For potential causes, the system generates a dialogue strategy to guide the user to further describe the problem. In this case, the contextual feature *c* and the problem complexity *q* are used as key parameters for strategy generation. For example:

[0132] Response2 = f r (X,c,q)

[0133] This response strategy guides users through multiple rounds of dialogue, allowing them to gradually describe more details of the potential problem.

[0134] Root Cause Strategy: For the root cause, the system needs to generate more complex response schemes based on emotional fluctuations and contextual information from long-term dialogue history. At this point, the system can combine emotional changes from multiple rounds of dialogue to generate responses tailored to the user's emotions, ensuring the user feels understood and supported.

[0135] Furthermore, after generating a response strategy, the system dynamically adjusts the weights in the causal relationship graph based on user feedback. Specifically, if a strategy fails to effectively solve the problem, the system will re-evaluate the causal relationships between features using user feedback signals and dynamically adjust the causal dependency weights w. i The updated formula is:

[0136]

[0137] Through this feedback-driven adjustment, the system can continuously optimize the causal reasoning model, thereby making more accurate inferences and responses when dealing with similar problems in the future.

[0138] S3. Based on the generated multi-level problem-solving strategies and the integrated feature vectors, construct an adaptive response strategy generation model to generate the response strategy that best suits the current context.

[0139] In step 2, the system uses a causal relationship graph. User questions were analyzed hierarchically, generating questions at various levels, including surface symptoms, underlying causes, and root causes. These outputs (hierarchical question L, semantic features X, sentiment features e, contextual features c, and question complexity q) will serve as input to generate an adaptive response strategy. The system needs to generate flexible response strategies based on the question's hierarchy, sentiment state, and contextual features, and dynamically adjust these strategies during actual interactions.

[0140] Specifically, firstly, based on the hierarchical problem L and its features (semantic X, context c, problem complexity q) output from step 2, this invention defines an adaptive response strategy generation function f. s This function combines these inputs to generate response strategies R at different levels. i .

[0141] The generation of response strategies depends on different levels of the problem. For the surface symptom level, the response strategy is mainly based on semantic features X and sentiment features e; for the underlying cause level and the root cause level, the complexity of the strategy increases, requiring the combination of context c and problem complexity q to generate a more detailed response.

[0142] This invention introduces a hierarchical adaptive coefficient λ i It controls the complexity of the response strategy at different problem levels. The generated response strategy is represented as:

[0143] R i =f s (λ i X,c,q)

[0144] Among them, R i Let λ represent the response strategy generated at level i. i It is the adaptive complexity coefficient of the i-th level, which depends on the problem level L. i . λ i The larger the value, the more complex the problem, and the more context and historical information needs to be considered in strategy generation.

[0145] For example, for the root cause layer, λi It may be taken as a large value to ensure that the response can cover multiple aspects of the user's problem; for the surface symptom layer, λ i A smaller value allows for a simpler response strategy.

[0146] Furthermore, to avoid overly complex or overly simple response strategies, this invention introduces a regularization term Ω(R) when generating responses. This regularization term constrains the generation of the response strategy based on the user's emotional characteristics e and the question complexity q, ensuring that the strategy can adapt to the user's current emotional state and the complexity of the question. The improved response strategy generation formula is as follows:

[0147]

[0148] Where Ω(R) is the regularization term, defined as:

[0149] Ω(R)=α·∥Re∥ 2 +β·∥Rq∥ 2

[0150] In this regularization term, α and β control the degree of matching between the response strategy and the emotional feature e and the question complexity q. This allows the system to dynamically adjust the response generation based on the user's current emotional state and the question's complexity, preventing simple questions from generating complex responses or complex questions from generating simplified responses.

[0151] Furthermore, after generating an initial response strategy, the system dynamically adjusts the strategy based on the user's real-time feedback F. The feedback signal F includes emotional fluctuation Δe and contextual change Δc, which are used to correct the initial response strategy.

[0152] The adjusted response strategy formula is as follows:

[0153]

[0154] Among them, f a (F) is the feedback adjustment function, which includes the immediate emotional change Δe and the contextual change Δc. Through feedback-driven mechanisms, the system can adjust its response strategy based on the user's emotional fluctuations, avoiding the generation of stiff or inappropriate responses.

[0155] Furthermore, user problems often involve multiple levels (surface symptoms, underlying causes, and root causes), therefore the system needs to generate response strategies R for each level. i The solutions are then merged. This invention uses a weighted summation method to synthesize the response strategies at each level into a final response:

[0156]

[0157] Where, α iThe weights are determined based on the problem complexity q. In this way, the system can comprehensively consider the problems at all levels in the final response, ensuring that the response strategy is comprehensive and accurate.

[0158] Furthermore, to improve the system's long-term responsiveness, the system records user feedback data after each user interaction and updates the response strategy generation model f based on this historical feedback. s and adjustment function f a This optimization uses gradient descent to continuously improve the response strategy. The optimization formula is:

[0159]

[0160] Where γ is the learning rate. This represents the loss gradient based on historical feedback. Through this mechanism, the system can gradually accumulate experience in dealing with complex problems and optimize the quality of future responses.

[0161] S4. Execute the agreement strategy and determine whether to trigger learning based on user feedback.

[0162] Specifically, in step 3, the final response strategy R has been generated. fi na l This strategy is generated through multi-level strategy merging and optimization, combining the user's semantic features X, sentiment features e, contextual features c, and question complexity q. At this stage, the system will execute the generated response strategy R. final And interact with users through actual dialogue.

[0163] The system will respond according to the response strategy R final The system sends appropriate responses and listens for the user's subsequent reactions to ensure that each response matches the user's question and emotional state. The system monitors the effectiveness of responses by processing user replies and reactions in real time. This process is based on the previously established adaptive response strategy generation model (see step 3), ensuring that each response is context-aware.

[0164] For example, user feedback f after execution feedback This feedback will be monitored, including the user's reaction time to the response, emotional fluctuations, and problem-solving progress. By analyzing this feedback, the effectiveness of the current response strategy can be determined.

[0165] Furthermore, the system needs to extract feedback signals from the user's real-time reactions to provide input for subsequent learning and optimization. The core variables of the feedback include sentiment change Δe and semantic matching degree ΔX. This invention defines a feedback signal function f. feedback Its output is the feedback signal F. The specific definition of the feedback signal F is as follows:

[0166] F = f feedback (Δe,ΔX)

[0167] Here, Δe represents the change in user emotional characteristics, indicating the emotional fluctuation from the current dialogue turn to the previous dialogue turn; ΔX represents the change in semantic matching degree, indicating the semantic matching between the system's response and the user's question. The system quantifies these signals using natural language processing techniques, thus providing them to the learning trigger module for further processing.

[0168] For example, Δe can be quantified by analyzing the user's tone, word choice, and speech features; while ΔX uses text similarity algorithms (such as cosine similarity) to determine the semantic deviation between the system's response and the user's question.

[0169] Furthermore, based on the real-time feedback signal F, the system needs to decide whether to trigger the learning module to further optimize the model. This invention defines a learning trigger function T(F), whose output determines whether learning should proceed.

[0170] The core mechanism of learning triggering relies on the strength of the feedback signal. If the feedback signal indicates a poor response (e.g., significant emotional fluctuations or low semantic matching), the learning module is triggered to update the response model; if the feedback signal indicates a good response, learning is not triggered. The triggering condition formula is:

[0171]

[0172] Here, θ is the threshold for triggering learning, which depends on the system design. When F exceeds the threshold, it indicates that there is a deviation in the response, and learning needs to be triggered; when it is below the threshold, it indicates that the response is relatively successful, and no further learning needs to be triggered.

[0173] Furthermore, once the learning trigger is activated, the system will generate a model f based on the feedback signal F to determine the response strategy. s The model is updated using gradient descent, which optimizes the response strategy generation model by minimizing the loss function. Defined as:

[0174]

[0175] Among them, F optimal It is the ideal feedback signal set by the system, R optimal The system represents the ideal response strategy, where α and β are weighting parameters that control the relative importance of the feedback signal and the response strategy in optimization. By minimizing the loss, the system can continuously adjust the generated response strategy R, thereby improving the response quality in subsequent dialogues.

[0176] Through learning triggers and model updates, the system can gradually improve its responsiveness in different dialogue rounds. This learning mechanism enables the system to adapt to the needs of different users and diverse dialogue scenarios, providing more intelligent services.

[0177] S5. The adaptive response strategy generation model is adaptively learned and optimized in real time through user feedback and historical user feedback.

[0178] In step 4, the system executes the generated response strategy R. final It also collects real-time feedback signals F. These signals, quantified by the user's emotional change Δe and semantic matching degree ΔX, represent the degree to which the current response strategy adapts to the user's needs.

[0179] This step will use F as the core input to trigger an adaptive learning mechanism, adjusting the response model f. s The parameters are configured to enable more accurate and personalized responses in subsequent interactions. Simultaneously, the system will integrate historical feedback from multiple rounds of dialogue to optimize the model's long-term performance. Based on the feedback signals, the core task of this step is to minimize the loss function. This is used to update the parameters θ of the response strategy generation model. The loss function considers not only the current feedback but also the balance between emotional and semantic feedback.

[0180] Specifically, firstly, this invention defines an innovative loss function. A regularization term based on semantics and sentiment was added, enabling the model to consider the balance between semantic and sentiment features during updates. The loss function is as follows:

[0181]

[0182] Where α and β are the weights balancing the feedback signal and the response strategy bias; F optimal It is the ideal feedback signal preset by the system; Ro p tim a l It is an ideal response strategy; regularization term This is an innovative design targeting user sentiment and semantic features, ensuring that the model balances semantic and sentiment factors during adjustment. This regularization term is defined as:

[0183]

[0184] The innovation of this regularization term lies in its requirement that the system, during the optimization process, not only pursues a match between the response strategy and the feedback signal, but also maintains consistency with the user's emotions (e) and semantics (X). This design is well-suited for customer service, as changes in user emotions often have a significant impact on the success of response strategies.

[0185] To ensure a more efficient and dynamic model update process, this invention introduces an adaptive learning rate η. This rate is dynamically adjusted based on the strength and complexity of the feedback signal F, ensuring that the model can quickly adjust when the feedback signal is large (indicating user dissatisfaction with the current response strategy) and update more smoothly when the feedback signal is weak (indicating the response strategy is basically successful). The innovative formula for the adaptive learning rate η is designed as follows:

[0186]

[0187] Where η0 is the initial learning rate; κ is the adjustment coefficient; and q is the problem complexity. Adding ∥q∥ aims to further slow down the learning rate when the problem complexity is high, thus avoiding over-adjustment.

[0188] When the feedback signal F is strong (indicating that the response strategy is not ideal), the learning rate will increase; while when the problem complexity q is high, the system tends to make more cautious adjustments to avoid over-reliance on single feedback for adjustments in complex scenarios.

[0189] In customer service systems, a single feedback instance cannot fully represent a user's long-term needs. Therefore, this invention designs a historical feedback accumulation term H to combine feedback signals from multiple rounds of dialogue and guide the model in long-term optimization. By accumulating historical feedback, the model can better cope with emotional fluctuations and semantic changes in complex scenarios. The calculation formula for the historical feedback accumulation term is:

[0190] H new =ρH old +(1-ρ)F

[0191] Among them, H new It is the new historical feedback cumulative value; H old It represents the accumulated historical feedback; ρ is the feedback attenuation factor, controlling the weight of old and new feedback. Through this design, the system can combine new feedback signals with past historical feedback after each interaction, gradually accumulating understanding of different users' emotional and semantic needs, and optimizing the long-term response strategy generation model f. s .

[0192] After the model is updated, the system will re-verify the model's performance using feedback signals. If the updated model performs well, the feedback signal F will significantly approach the ideal feedback F. optimal Otherwise, the system will readjust the learning rate η and the regularization term. The weight parameters λ1 and λ2.

[0193] This dynamic validation mechanism makes the model update process more flexible, avoiding instability caused by adjusting too many parameters at once. This invention measures the update effect using the following formula:

[0194]

[0195] if A negative value indicates successful model optimization; otherwise, the system will make further adjustments to ensure that the model gradually converges to the ideal response.

[0196] S6. Analyze the long-term trend of model parameters of the adaptive response strategy generation model based on historical user feedback and multiple optimizations, perform global optimization based on the long-term trend, optimize the response strategy that best suits the current context, and generate the final response strategy.

[0197] In step 5, the system has optimized the response strategy generation model f through real-time adaptive learning and historical feedback accumulation mechanisms. s At this point, the system has accumulated a large amount of real-time feedback (F) and historical feedback (H), and the model has been updated multiple times. However, long-term optimization requires not only feedback from individual conversations but also system-level upgrades and long-term strategy optimization to adapt to future complex dialogue scenarios and diverse user needs.

[0198] The task of this step is to further optimize the response strategy and gradually update the entire system based on the data accumulated from long-term system interactions, so as to ensure that the service remains efficient and intelligent in the long run.

[0199] Specifically, the main inputs to this step are the accumulated historical feedback H and the model parameters θ after multiple optimizations. new These parameters are derived from the optimization results in step 5. The system needs to analyze the long-term trends of these historical data to identify structural problems or potential optimization spaces in the response strategy. The long-term feedback trend function T(H) is used to analyze the long-term changing trend of the feedback signal, identifying long-term deviations and optimization spaces. This trend function is defined as:

[0200]

[0201] Among them, H i Let represent the historical feedback in the i-th dialogue, and n be the cumulative number of dialogues. By calculating the average feedback signal, the system can identify potential problems in long-term responses, such as insufficient responses to certain question types or biases in emotional processing.

[0202] In this process, the present invention introduces a problem classification feature P. The system analyzes long-term feedback by category based on feedback signals from different problem categories P, ensuring targeted optimization. For example, for technical support problems P... tech The system can focus on optimizing strategies related to problem complexity q; while for emotion support problems P... emotional The system, on the other hand, pays more attention to the matching degree of the emotional feature e.

[0203] Furthermore, the core of long-term optimization lies in continuously improving the overall responsiveness of the system through the accumulation of a large amount of dialogue data. To achieve this goal, the system needs to gradually update the core parameters of the model, while further improving system performance through the generation of new strategies. Based on long-term feedback trends, this invention performs global optimization of the model parameters θ. This optimization minimizes the long-term loss function. accomplish:

[0204]

[0205] in, It is the ideal historical feedback signal in each round of interaction; This represents the ideal strategy in each round of response; α and β represent the importance weights of historical feedback and response strategies, respectively.

[0206] Through this optimization, the system can continuously converge to the ideal response strategy in the long term, ensuring that the system maintains high response accuracy and flexibility even in complex and ever-changing scenarios.

[0207] Furthermore, long-term optimization involves not only tuning existing model parameters, but also "evolving" the response strategy itself based on long-term trends. This evolution is reflected in the structural updates of the response strategy and the improvement of the generation rules. To this end, this invention introduces a strategy evolution factor μ, which dynamically adjusts the rules of the strategy generation model based on long-term feedback signals. The strategy evolution formula is:

[0208] R evolved =R new +μ·T(H)

[0209] Among them, R new The response strategy is updated through long-term feedback. T(H) is the result of the feedback trend function, and μ is the strategy evolution factor used to control the rate of strategy evolution. Through this evolutionary process, the system can gradually adapt to different types of problems and emotional needs, and make more creative and adaptive improvements to its response strategy.

[0210] Furthermore, during long-term operation, with the increase in data volume and the complexity of scenarios, the system also requires architectural-level optimization and expansion. This involves not only adjusting the model parameters but also optimizing the architecture of the entire response system to adapt to higher complexity. An adaptive architecture expansion mechanism is designed, which automatically expands different model modules or updates the functionality of existing modules based on the volatility of feedback signals, dialogue complexity, and the diversity of user needs. The triggering condition for adaptive architecture expansion is as follows:

[0211]

[0212] Where, τ q and τ F These are thresholds for problem complexity and feedback bias, respectively. When the complexity or feedback bias exceeds the threshold, the system will trigger an architecture upgrade. For example, the system might add more deep learning model modules for emotion recognition, or introduce more complex causal reasoning modules to handle highly complex problems.

[0213] Furthermore, after model updates and policy evolution, the system needs to undergo a series of tests to verify the effectiveness of long-term optimization. System-level verification involves simulating multi-turn dialogue scenarios and using historical data for automated testing. Key verification metrics include:

[0214] Response accuracy: The degree to which the system's response matches the ideal answer;

[0215] Sentiment matching degree: The effectiveness of the system's processing of sentiment feature e in its response;

[0216] Dialogue complexity handling capability: The system's ability to cope with highly complex scenarios, mainly measured by ||q||.

[0217] The system evaluates the optimization effect of the model through dynamic testing of these indicators, and further adjusts the strategy and model parameters based on the test results to ensure that the system maintains good performance in long-term operation.

[0218] Step S6 aims to comprehensively optimize and upgrade the response strategy and system architecture through the accumulation and trend analysis of long-term feedback signals. This is achieved by designing an innovative long-term optimization loss function. By combining a strategy evolution factor μ and an adaptive architecture expansion mechanism, the system can continuously improve its intelligence and responsiveness while dealing with complex scenarios. Furthermore, after each optimization, the system undergoes rigorous verification testing to ensure the effectiveness of continuous optimization, providing an efficient and stable intelligent customer service solution for long-term use in patented scenarios.

[0219] In one or more embodiments, such as Figure 2As shown, this invention discloses an intelligent problem-solving system for robot customer service, the system comprising:

[0220] User Question Acquisition Subsystem 101 is used to acquire user questions, extract and integrate their features, and specifically includes:

[0221] S101. Obtain the user's complete input dialogue and preprocess the dialogue. Convert the preprocessed dialogue into a text representation and embed each word in the text as a semantic feature. At the same time, design an emotion embedding model to extract emotion features from the input dialogue and design a context feature extractor to capture the context relationship of the user in multiple rounds of dialogue to obtain context features.

[0222] S102. Generate a complexity score based on semantic features, sentiment features, and contextual features, and generate a comprehensive feature vector from the semantic features, sentiment features, contextual features, and complexity score;

[0223] User problem analysis subsystem 102 is used to construct a dynamic causal reasoning model for users based on the integrated feature vectors and to generate multi-level problem-solving strategies based on user problems, specifically including:

[0224] S201. Construct a dynamic causal reasoning model, take the comprehensive feature vector as input, reason about the causal dependencies behind the user's question, and establish a causal function for each feature by analyzing the dependencies of contextual features and semantic features.

[0225] S202. Based on the complexity of user problems, user problems are divided into three levels, including: surface feature layer, potential cause layer, and root cause layer.

[0226] S203. Generate three problem-solving strategies for the three levels respectively, including: surface level strategy for problems at the surface feature level, potential level strategy for problems at the potential cause level, and root level strategy for problems at the root cause level.

[0227] S204. If a certain strategy fails to solve the problem effectively, the causal relationship between features is re-evaluated through real-time user feedback signals, and the causal dependency weights in the causal function are dynamically adjusted.

[0228] User question response subsystem 103 is used to construct an adaptive response strategy generation model based on the generated multi-level question-solving strategy and the integrated feature vector to generate a response strategy that best suits the current context, including:

[0229] S301. Construct an adaptive response strategy generation model to generate response strategies at different levels based on the level of the user's question and its corresponding comprehensive feature vector.

[0230] S302. Add regularization terms based on sentiment characteristics and question complexity to the response strategy to constrain the generation of the response strategy;

[0231] S303. Generate an initial response strategy and dynamically adjust the response strategy based on real-time user feedback to obtain the adjusted response strategy;

[0232] S304. Merge the multi-level response strategies generated by the user to generate the final response strategy, and update the model parameters of the adaptive response strategy generation model in real time based on the feedback from user interaction.

[0233] The response learning subsystem 104 is used to execute the response strategy and determine whether to trigger learning based on user feedback, including:

[0234] Execute the final response strategy and listen for user feedback. Extract feedback signals from the user feedback and determine whether to trigger learning based on the feedback signals:

[0235] If the feedback signal indicates a poor response, the learning module is triggered to update the response model;

[0236] If the feedback signal indicates a good response, then learning is not triggered;

[0237] The optimization subsystem 105 is used to perform real-time adaptive learning and optimization of the adaptive response strategy generation model based on user feedback and historical user feedback. It analyzes the long-term trends of the model parameters of the adaptive response strategy generation model after multiple optimizations based on historical user feedback, performs global optimization based on the long-term trends, optimizes the response strategy that is most suitable for the current context, and generates the final response strategy.

[0238] These are merely some preferred embodiments of the present invention, and should not be construed as limiting the scope of the invention. Those skilled in the art will understand that implementing all or part of the processes of the above embodiments and making equivalent changes in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A method for intelligent processing of a robot customer service problem, characterized in that, The method comprises the following steps: S1, obtaining a user question and performing feature extraction and integration on the user question, specifically comprising: S101, obtaining a complete input dialogue of a user, pre-processing the dialogue, converting the pre-processed dialogue into a text representation, embedding each word in the text as a semantic feature, designing a sentiment embedding model for extracting sentiment features from the input dialogue, and designing a context feature extractor to capture the context relationship of the user in the multi-round dialogue to obtain context features; S102, generating a complexity score according to the semantic features, the sentiment features and the context features, and generating a comprehensive feature vector from the semantic features, the sentiment features, the context features and the complexity score; S2, constructing a user dynamic causal reasoning model according to the integrated feature vector and generating a multi-level problem solving strategy according to the user question, specifically comprising: S201, constructing a dynamic causal reasoning model, taking the comprehensive feature vector as input, reasoning the causal dependence behind the user question, and for each feature, establishing a causal function by analyzing the dependence of the context features and the semantic features; S202, dividing the user question into three levels according to the complexity of the user question, including: surface feature level, potential reason level and root cause level; S203, generating three kinds of problem solving strategies for the three levels, including: surface level strategy for surface feature level problem, potential level strategy for potential reason level problem and root level strategy for root cause level problem; S204, if a strategy fails to effectively solve the problem, re-evaluate the causal relationship between the features through real-time feedback signals of the user, and dynamically adjust the causal dependence weight in the causal function; S3, generating an adaptive response strategy generation model according to the generated multi-level problem solving strategy and the integrated feature vector to generate a response strategy most suitable for the current corresponding context, comprising: S301, constructing an adaptive response strategy generation model to generate response strategies of different levels according to the levels of the user question and the corresponding comprehensive feature vectors; S302, adding a regular term constraint based on the sentiment features and the problem complexity in the response strategy to generate the response strategy; S303, generating an initial response strategy, and dynamically adjusting the response strategy according to the real-time feedback of the user to obtain an adjusted response strategy; S304, merging the generated multi-level response strategies related to the user to generate a final response strategy, and updating the model parameters of the adaptive response strategy generation model in real time according to the feedback of the interaction with the user; S4, executing the response strategy and judging whether to trigger learning according to the feedback of the user, comprising: execute the final response strategy and listen to the feedback of the user, extract the feedback signal from the feedback of the user, and judge whether to trigger learning according to the feedback signal: if the feedback signal shows that the response effect is poor, trigger the learning module to update the response model; if the feedback signal indicates that the response effect is good, do not trigger learning; S5, performing real-time adaptive learning and optimization of the adaptive response strategy generation model through user feedback and historical user feedback; S6, analyze the long-term trend of the historical user feedback and the model parameters of the adaptive response strategy generation model after multiple optimizations, perform global optimization based on the long-term trend, optimize the response strategy most suitable for the current corresponding context, and generate a final response strategy. 2.The robot intelligent processing method for customer service problems according to claim 1, characterized in that, The embedding of each word in the text into a vector introduces a context-related feature representation, which captures the semantic information of each word using the entire conversation context information; the sentiment embedding model maps the conversation content to a sentiment feature space, capturing the user's emotional expression in a specific sentence while combining the conversation context to obtain the dynamic process of the emotion over time; and the context feature extractor is used to capture the context relationship of the user in multiple rounds of conversation. 3.The robot intelligent processing method for customer service problems according to claim 1, characterized in that, The dynamic causal reasoning model is represented as follows: defining each feature in the causal graph as a node edges between all nodes as dependency weights representing causal relationships between features; Through analysis of historical data, a causal relationship model is derived and represented as follows: ； in, The loss function represents the causal relationship model. Represents a node The set of causal parent nodes, Let i represent the causal function for estimating the features, and let i represent the number of features. The surface feature layer is used to process semantic features and emotional features, directly reflecting the user's current expression of the problem; The latent cause layer is used to process complex problems, identify the underlying causes of the surface problem according to the context features and problem complexity, and through a causal relationship graph; The root cause layer is used to trace the root cause of the problem by combining emotional features, context features, and historical data of multiple rounds of conversation; The surface-level strategy directly generates a standardized response to the problem in the surface feature layer; The latent-level strategy generates a dialogue strategy to guide the user to describe the problem in the next step for the problem in the latent cause layer; The root-level strategy generates a complex response scheme for the problem in the root cause layer based on emotional fluctuations and context information in long-term conversation history. 4.The robot intelligent processing method for customer service problems according to claim 1, characterized in that, S301, specifically including: for the surface feature layer, the response strategy is based on semantic features and emotional features; for the latent cause layer and the root cause layer, detailed responses are generated based on semantic features and emotional features in combination with context features and complexity scores; Meanwhile, a hierarchical adaptive coefficient is introduced in the surface feature layer, the latent cause layer, and the root cause layer to control the complexity of the response strategy at different problem levels, and the hierarchical adaptive coefficient is the largest, indicating the most complex problem. 5.The robot intelligent processing method for customer service problems according to claim 4, characterized in that, The user real-time feedback signal includes emotional fluctuations and context changes; and the multi-level response strategies generated by the user are combined by using a weighted summation method. 6.The robot intelligent processing method for customer service problems according to claim 1, characterized in that, The feedback signal includes emotional changes and semantic matching degrees, wherein the emotional changes represent emotional fluctuations from the current conversation round to the last conversation round, and the semantic matching degrees represent the semantic matching of the robot response and the user problem.

7. The method of claim 6, wherein, When the learning trigger is activated, the adaptive response strategy generation model is updated according to the feedback signal, and the adaptive response strategy generation model update uses a gradient descent method to optimize the adaptive response strategy generation model by minimizing the loss function. 8.The method of claim 6, wherein, S5, specifically including: S501、designing a semantic and emotion based regularization term according to the semantic feature and the emotion feature and constructing a loss function is represented as follows: ； wherein, represents a real-time feedback signal, represents a final response strategy, , represents a weight balancing the feedback signal and the deviation of the response strategy; represents a preset ideal feedback signal; represents an ideal response strategy; represents a weight coefficient of a regularization term based on semantics and emotion; represents a regularization term based on semantics and emotion, used to ensure that the adaptive response strategy generation model balances semantic and emotional factors when adjusting, and is represented as follows: ； wherein, represents an emotional feature, represents a semantic feature, represents an ideal emotional feature, represents an ideal semantic feature, and represents a weight balancing the emotional and semantic bias. S502, the adaptive learning rate is introduced to dynamically adjust according to the intensity and complexity of the feedback signal, to ensure that in the case of the maximum feedback signal, the adaptive response strategy generation model can be adjusted the fastest; in the case of the weakest feedback signal, the adaptive response strategy generation model can be updated; the adaptive learning rate is represented as follows: ； wherein, denotes an initial learning rate; denotes an adjustment coefficient; denotes a problem complexity; S503, construct a historical feedback accumulation item for combining the feedback signals of multiple rounds of conversation to guide the adaptive response strategy generation model to perform long-term optimization; S504, after the adaptive response strategy generation model is updated, a dynamic verification mechanism is designed to verify whether the adaptive response strategy generation model is updated, and is indicated as follows: ； If is negative, it means that the model optimization is successful; otherwise, adjustments will be made to ensure that the model converges gradually to the desired response effect. 9.The robot intelligent processing method for customer service problems according to claim 6, characterized in that, The S6 specifically comprises: S601, a long-term feedback trend function is constructed according to the problem classification characteristics, the long-term change trend of the feedback signal is analyzed, and the long-term deviation point and optimization space are found out; wherein the problem classification characteristics represent the long-term feedback signal trend extracted from the multi-level problem response strategy; S602、According to the long-term change trend of the feedback signal, the model parameters of the adaptive response strategy generation model are generated Global optimization is performed; S603, a strategy evolution factor is introduced in the long-term optimization process of the adaptive response strategy generation model, and the rules of the adaptive response strategy generation model are dynamically adjusted according to the long-term feedback signal; S604, an adaptive architecture expansion mechanism is designed, and different model modules or the functions of the existing modules are automatically expanded according to the volatility of the feedback signal, the dialogue complexity and the diversity of user demand; S605, after the model is updated and the strategy is optimized, the effect of long-term optimization is verified through testing. 10.A robot intelligent handling system for customer service problems, characterized in that, The system comprises: A user question acquisition subsystem for acquiring user questions and performing feature extraction and integration on the user questions, specifically comprising: S101, obtaining the complete input dialogue of the user, pre-processing the dialogue, converting the pre-processed dialogue into a text representation, embedding each word in the text into a semantic feature, designing a sentiment embedding model to extract sentiment features from the input dialogue, and designing a context feature extractor to capture the context relationship of the user in the multi-round dialogue to obtain context features; S102, generating a complexity score according to the semantic features, the sentiment features and the context features, and generating a comprehensive feature vector from the semantic features, the sentiment features, the context features and the complexity score; A user question analysis subsystem for constructing a user dynamic causal reasoning model according to the integrated feature vector and generating a multi-level problem solving strategy according to the user question, specifically comprising: S201, constructing a dynamic causal reasoning model, taking the comprehensive feature vector as input, reasoning the causal dependence behind the user question, and for each feature, establishing a causal function by analyzing the dependence of the context features and the semantic features; S202, dividing the user question into three levels according to the complexity of the user question, including: surface feature level, potential reason level and root cause level; S203, generating three kinds of problem solving strategies for the three levels, including: surface level strategy for surface feature level problem, potential level strategy for potential reason level problem and root level strategy for root cause level problem; S204, if a certain strategy fails to effectively solve the problem, the causal relationship between the features is re-evaluated through real-time user feedback signals, and the causal dependence weight in the causal function is dynamically adjusted; A user question response subsystem for constructing an adaptive response strategy generation model according to the generated multi-level problem solving strategy and the integrated feature vector to generate a response strategy most suitable for the current corresponding context, comprising: S301, constructing an adaptive response strategy generation model to generate response strategies of different levels according to the levels of the user question and the corresponding comprehensive feature vectors; S302, adding a regular term constraint based on emotional features and question complexity in the response strategy, and generating the response strategy; S303, generating an initial response strategy, and dynamically adjusting the response strategy according to real-time feedback of the user to obtain an adjusted response strategy; S304, merging the multi-level response strategies generated by the user to generate a final response strategy, and updating model parameters of the adaptive response strategy generation model in real time according to feedback of interaction with the user; The response learning subsystem is used to execute the response strategy and judge whether to trigger learning according to feedback of the user, and includes: executing the final response strategy and listening to feedback of the user, extracting feedback signals from the feedback of the user, and judging whether to trigger learning according to the feedback signals: if the feedback signals show that the response effect is poor, the learning module is triggered to update the response model; if the feedback signals show that the response effect is good, learning is not triggered; The optimization subsystem is used to perform real-time adaptive learning and optimization of the adaptive response strategy generation model through user feedback and historical user feedback, analyze long-term trends of historical user feedback and model parameters of the adaptive response strategy generation model after multiple optimizations, perform global optimization combined with the long-term trends, optimize the response strategy most suitable for the current context, and generate a final response strategy.

Images