Method for identifying personnel intention through guiding type communication

Abstract

The invention discloses a method for identifying personnel intentions through guided communication, belongs to the technical field of man-machine interaction and artificial intelligence, and aims to solve the problems of low communication efficiency and poor user experience caused by difficulty in processing fuzzy and incomplete user input and neglecting a user interaction state in an intention identification system in the prior art. The method comprises the following steps: analyzing initial input of a user to obtain an initial intention; analyzing and generating an information gap list based on a preset intention-information element mapping library; interaction states such as emotion and uncertainty of the user are perceived and quantified in real time; the core lies in that the information gap and the user state are fused, 'what is asked '(the information gap) and'what is asked' (the user state) are dynamically combined, more intelligent and humanized guidance can be carried out, the accuracy and the communication efficiency of intentional D graph recognition in a complex or emergency scene are remarkably improved, and the man-machine interaction experience is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of human-computer interaction and artificial intelligence, in particular to a method for recognizing personnel intention through guided communication. BACKGROUND

[0002] With the rapid development of human-computer interaction technology, intention recognition has become a core technology in application scenarios such as intelligent customer service, virtual assistants, and task-oriented robots. Existing technologies usually rely on natural language processing (NLP) models, such as deep learning networks or pre-trained language models, to perform semantic analysis on user input text by extracting keywords, entities, and syntactic structures to determine the user's purpose. These technologies can achieve high accuracy when dealing with structured and complete information queries.

[0003] However, these technologies have significant limitations in practical applications. In real-world scenarios, users, especially individuals in informal communication or emergency situations, often input vague, scattered, and incomplete information. When faced with such input, existing systems often interrupt the conversation or give meaningless general replies due to the inability to extract all pre-set key information, and cannot effectively drive the conversation towards problem-solving. The root of this limitation lies in the passive information reception mode, and the system lacks the ability to actively identify what "information gaps" exist between the current conversation and the goal, and to build efficient follow-up questions and guidance around the gap.

[0004] Furthermore, traditional intention recognition methods almost completely ignore the dynamic context information in the interaction process. A successful communication is not only about understanding the literal meaning, but also about perceiving the state of the other party. Existing technologies usually treat all user inputs as homogeneous text data, and fail to incorporate the user's real-time interaction state, such as emotional fluctuations, expression certainty, or hesitation level, into their decision-making model. This "deafness" to the user's state makes the system's follow-up or guidance rhetoric appear mechanical and rigid, unable to adaptively adjust according to the user's real-time feedback. When the user shows confusion or impatience, the system's blunt feedback may even exacerbate the situation, not only reducing communication efficiency, but also seriously affecting user experience.

[0005] Therefore, how to transform the intention recognition system from a passive information parser to an intelligent conversation partner that can actively guide, perceive the state, and adaptively communicate is a technical problem that needs to be solved in the field. SUMMARY

[0006] The technical problem to be solved by the present application is that existing personnel intention recognition technology is mostly passive, and when the user's input information is vague, incomplete, or has deep intentions, it is difficult to accurately identify, and lacks a mechanism for personalized communication combined with the user's real-time state.

[0007] To solve the above technical problems, the present application provides a method for identifying personnel intent through guided communication, which realizes accurate exploration of user intent by actively guiding, real-time sensing of user state and dynamic adjustment of strategy.

[0008] The technical solutions provided by the present application are as follows:

[0009] A method for identifying personnel intent through guided communication, comprising the following steps:

[0010] Step s1: Analyzing the initial input of the user to obtain the preliminary intent. In a specific embodiment, this step is based on a pre-set domain knowledge graph to construct a semantic analysis model. The domain knowledge graph contains structured knowledge such as professional terms, business processes, entity relationships, etc. in a specific industry or scenario. After receiving the initial input of the user in the form of text or voice, etc., the semantic analysis model processes it to identify the core entity in the input content and one or more most likely preliminary intents.

[0011] Step s2: Analyzing the information required to confirm the final intent and the information already obtained to generate an information gap. In a specific embodiment, this step is realized through a pre-set "intent-information element" mapping library. The mapping library clearly defines a complete set of information elements required to confirm each possible intent. The system compares the information obtained in step s1, such as the core entity, with the complete set of information elements corresponding to the preliminary intent in the mapping library, and through set difference operation, a dynamic "information gap" list is generated, which clearly indicates what information is currently missing.

[0012] Step s3: Real-time sensing of the user's interactive state to obtain the user's interactive state vector. In a specific embodiment, the system analyzes the user's input behavior to sense their state at each round of interaction with the user. The user's interactive state can include at least one of the following features: the sentiment polarity of the user's input text, which reflects whether the user's current emotion is positive, negative or neutral; the content modification index of the user during the input process, which measures the degree of hesitation or uncertainty; and the input rate index of the user, whose abnormal change may be associated with a specific emotional state.

[0013] Further, the user interactive state vector S t At the tth round of interaction, it can be constructed as follows:

[0014] S t =[s emo,t ,c mod,t ,v input,t ];

[0015] where s emo,t is the sentiment polarity score calculated by the sentiment analysis model, c mod,t is the content modification index quantifying the input content modification behavior, v input,t is the normalized input rate index. This vector provides a quantitative basis for the adaptive adjustment of the subsequent guidance strategy.

[0016] Step s4: generating and executing a guidance strategy according to the information gap and the user interaction state vector.

[0017] In a specific embodiment, the system predefines a guidance strategy library, which at least includes: an open guidance strategy for guiding the user to make an open description; a closed guidance strategy for guiding the user to make a clear choice; and a scenario-based guidance strategy for asking questions in combination with domain specifications and knowledge.

[0018] The process of generating a guidance strategy by the system is to evaluate multiple candidate guidance responses R through a comprehensive scoring function Score(R), and select the response with the highest score to execute. A specific implementation of the scoring function is:

[0019] Score(R) = a · f match (R, e target,t ) + β · f adapt (R, S t );

[0020] where e target,t is the current round guidance target determined according to the information gap; f match is a content matching degree function for evaluating the effectiveness of the candidate response R in guiding the user to provide the target information e target,t ; f adapt is a state adaptability function for evaluating the fit of the communication manner of the candidate response R with the current user interaction state vector S t ; a and β are preset weight coefficients for balancing task orientation and user experience. This mechanism enables the method to dynamically select soothing, encouraging or direct communication manners according to the user's state of anxiety, confusion, etc.

[0021] Step s5: iteratively through dynamic interaction with the user until the final intent is confirmed. In a specific embodiment, this step constitutes a closed-loop interaction process. After the system executes the guidance strategy and receives the user's feedback information, it updates the acquired information using the feedback information and returns to execute steps s2, s3 and s4 to regenerate the information gap and the next round of guidance strategy.

[0022] To enable the iterative process to terminate in time, the system calculates the current information completeness at each iteration, which is usually the ratio of the number of necessary information elements acquired to the total number of information elements required. When the information completeness reaches a preset confirmation threshold, the system will pause the guidance and trigger the confirmation process of the final intent.

[0023] In the confirmation process, if the user denies the summary intent expressed by the system, the system can automatically backtrack to step s2 and modify the information gap based on the explicit denial information, such as adjusting the candidate intent or reducing the confidence of some acquired information, and then restart a new round of guidance.

[0024] The present application provides a method for identifying the intent of a person through guided communication. It has the following advantages:

[0025] 1. The present application can overcome the problem of traditional intent recognition, which often gets stuck when encountering incomplete information from the user. By introducing an information gap analysis module, the system is no longer a passive receiver of information, but can actively diagnose which key information is missing between the current conversation and the final intent. This "diagnostic" analysis enables the system to ask targeted and focused questions, effectively guiding the conversation forward, even with the most ambiguous initial input.

[0026] 2. Another outstanding contribution of the present application is to give the machine preliminary perception ability. Through the interaction state perception module, the system can go beyond the literal meaning of the text and capture the emotional state of the user, such as uncertainty, anxiety or decisiveness, revealed in tone and word choice. This enables the subsequent generated guidance strategy to "tailor the dish to the person", using a more soothing tone for anxious users and a more direct confirmation for decisive users, thus establishing a smoother and more trustworthy interaction relationship, greatly improving the user experience.

[0027] 3. The core originality of the present application lies in its generation mechanism of guidance strategy. Instead of simply choosing the next information point to ask, it dynamically weighs the two dimensions of "what to ask to complete the task" (content matching degree) and "what to ask in the current atmosphere" (state adaptability) through a comprehensive scoring function. This mechanism ensures that the guidance response selected by the system at any time is the best balance between "advancing the task" and "maintaining user experience", making the decision-making process both intelligent and reasonable.

[0028] 4. The dialogue of the present application is not a "one-time" judgment, but a gradual approach and continuous clarification process. The closed-loop interaction process designed by the present application allows the system to return to the initial step after receiving each feedback from the user, and to refresh the judgment of "information gap" and "user state" with the latest information. This continuous self-correction and adaptation ability makes the whole dialogue process have strong fault tolerance and flexibility, which can calmly cope with the change of user's intention or the correction of previous information, and ensures the robustness of communication.

[0029] 5. Through the above-mentioned features, the present application has great application value in practical application, especially in emergency response, safety production scheduling, complex customer service and other scenes with strict requirements on communication efficiency and accuracy. By intelligently reducing unnecessary exploration and invalid interaction rounds, the present application can quickly lock the real intention of the personnel, thereby saving valuable time for downstream decision-making and action, and effectively reducing the economic loss or safety risk caused by communication misunderstanding. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 a structural block diagram of the intention recognition device of the present application;

[0031] Figure 2 a flowchart of the method of the present application;

[0032] Figure 3 a data flow diagram of the present application;

[0033] Figure 4 a working principle diagram of the comprehensive scoring function of the present application. DETAILED DESCRIPTION

[0034] To make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0035] Referring to the drawings, Figure 1 the present application provides a method for personnel intention recognition through guided communication, which establishes a complete technical closed loop from information gap analysis to fusion user state guided strategy generation, and then to dynamic interaction iteration, aiming to solve the problems of insufficient recognition ability when facing user's ambiguous and incomplete expression, and rigid human-computer interaction experience in the prior art.

[0036] The intent recognition device can be a server deployed in the cloud, a local computer or any hardware entity with corresponding computing and storage capabilities. Specifically, the device includes a processing unit a, a storage unit b, a communication interface c, and an input / output interface d. The processing unit a, such as a central processing unit (CPU) or a graphics processing unit (GPU), is responsible for executing program instructions and processing data. The storage unit b, such as memory or hard disk, is used to store the computer program instructions corresponding to the method of the present invention and various types of data required during execution, such as domain knowledge graph, "intent-information element" mapping library, etc. The communication interface c is used for data exchange with external devices or user terminals, and the input / output interface d is used for receiving user instruction input and presenting system guidance output.

[0037] The computer program instructions stored in the storage unit b, when executed by the processing unit a, collectively constitute an intent recognition system e. The intent recognition system e logically includes an analysis module e1, an information gap analysis module e2, an interaction state perception module e3, a guidance strategy generation module e4, and a dynamic interaction control module e5. These modules work together to implement the complete intent recognition process described in the present invention.

[0038] The analysis module e1 is configured to analyze the user's initial input to obtain the preliminary intent. When the user initiates a session through the input / output interface d, the module first performs standardization preprocessing on the received text or voice data, and then calls a semantic analysis model based on domain knowledge graph fine-tuning to identify core entities and one or more candidate preliminary intents, corresponding to step s1 of the method of the present invention.

[0039] The information gap analysis module e2 is configured to analyze the information required to confirm the final intent and the information already obtained to generate the information gap. Based on the preliminary intent output by the analysis module e1, the module queries the pre-set "intent-information element" mapping library from the storage unit b, and by comparing the complete information element set required for the intent with the current obtained entity information, accurately calculates the list of information gaps to be supplemented, corresponding to step s2 of the method of the present invention.

[0040] The interaction state perception module e3 is configured to perceive the user's interaction state in real time and generate a quantitative user interaction state vector accordingly. In each round of interaction, the module analyzes the user's input behavior, extracts one or more state features such as text sentiment polarity, content modification frequency, input rate, etc. These features are integrated and quantified to construct a user interaction state vector S t , which at the tth round of interaction can be represented as:

[0041] S t = [semo,t ,c mod,t ,v input,t ]

[0042] where s emo,t represents sentiment polarity score, c mod,t represents content modification index, and v input,t represents input rate index. This vector provides the system with insights beyond literal semantics, about the user's current psychological state, corresponding to step s3 of the invented method.

[0043] The guidance policy generation module e4 is one of the cores of the invention, configured to generate and execute guidance policy according to the information gap and the user interaction state vector. The decision process of this module not only depends on "what information is missing", but more importantly, incorporates "what state the user is in now". It selects the optimal guidance reply through a comprehensive scoring function, one specific implementation of which is:

[0044] Score(R) = a · f match (R, e target,t ) + β · f adapt (R, S t )

[0045] where R is a candidate guidance reply, Score(R) is its comprehensive score; f match is the content matching degree function, evaluating the relevance of the reply R to the current guidance target e target,t ; f adapt is the state adaptability function, evaluating the fit of the communication style of the reply R to the user interaction state vector S t ; a and β are preset weight coefficients balancing task orientation and user experience. Through this mechanism, the module can dynamically generate the most suitable guidance rhetoric, corresponding to step s4 of the invented method.

[0046] The dynamic interaction control module e5 is configured to iterate through dynamic interaction with the user until the final intent is confirmed. This module is responsible for managing the flow of the entire conversation, driving other modules to update information and generate new strategies after each round of guidance and feedback. At the same time, this module also calculates the current information completeness, when it reaches the preset confirmation threshold, it triggers the final intent confirmation process. If the user denies the confirmation result, this module is also responsible for executing backtracking logic, guiding the system to self-correct, thus ensuring the robustness and accuracy of the entire recognition process, corresponding to step s5 of the invented method.

[0047] In a complete intent recognition session, the aforementioned modules work collaboratively under the scheduling of processing unit a, forming a complete closed loop from initial parsing to dynamic, adaptive guidance, and finally confirmation, thereby effectively achieving accurate recognition of the person's intent.

[0048] Reference Figures 2 to 3 In this embodiment, a detailed process implementation description will be provided for each step of the method for identifying human intent through guided communication executed by the aforementioned intent recognition system e.

[0049] In step s1, the process of parsing the user's initial input to obtain preliminary intent begins with the parsing module e1 receiving the user input. This input can be text or text data converted using speech-to-text technology. This invention employs a pre-trained language model, such as the BERT model, fine-tuned based on domain knowledge to simultaneously perform intent classification and entity extraction. This model is trained on text data containing entities and relationships specific to a particular domain (such as gas safety), giving it high domain adaptability. For an initial input text U... text,1 The goal of intent classification is to output a set of intents within a predefined intent set I = {i1, i2, ..., i...} N The probability distribution on} is used to select the intent with the highest probability as the initial candidate intent.

[0050]

[0051] Where, P(i k |U text,1 ) is the input text U for model judgment. text,1 Belongs to intention i k The probability of [something]. Simultaneously, the model extracts domain-related entity information from the text, forming an entity set E. extracted,1 Each entity consists of its type and value.

[0052] In step s2, the information gap analysis module e2 is responsible for generating information gaps. This embodiment constructs an "intent-information element" mapping library M. req And store it in storage unit b. This mapping library is in a structured form for each intent i. k It precisely defines a complete set of information elements necessary to confirm the intention. For example, for the intention to report a gas leak, the required set of information elements can be defined as M. r eq(gas leak) = {specific location, odor description, presence of open flame, contact number}. In the t-th round of interaction, this module will express the initial intent... Required set of elements With the collection of all entity types E collected so fartypes,collected,t A set difference operation is performed to dynamically compute the information gap list L gap,t :

[0053]

[0054] where E types,collected,t is the union of all extracted entity types from the first round to the t-th round. In addition, each information element in the mapping library is also assigned a priority weight, with safety-related elements having the highest weight. The items in the information gap list L gap,t are accordingly sorted to ensure that the user is prioritized to fill the most critical information.

[0055] In step s3, the interaction state perception module e3 perceives the user's interaction state in real time in parallel. In this embodiment, the user's interaction state is quantified as a multi-dimensional user interaction state vector S t . The construction of this vector integrates multiple dimensions of features: one is the sentiment polarity score s emo,t , which is calculated by an independent sentiment analysis model and has a value range between -1 (extremely negative) and +1 (extremely positive); the second is the content modification index c mod,t , which quantifies the user's uncertainty by monitoring the frequency and amplitude of text deletion and modification in a single input round; the third is the input rate index v input,t , which is the normalized user typing speed or speaking speed. These features together constitute the state vector S t = s emo,t , c mod,t , v input,t , providing the system with quantitative input on the user's real-time psychological state beyond the literal semantics.

[0056] In step s4, the guidance strategy generation module e4 fuses the information gap with the user state to generate and execute the guidance strategy. This module first selects the highest priority item from the sorted information gap list L gap,t as the guidance target e target,t for this round. Subsequently, the module selects the optimal guidance response R from a pre-set guidance strategy library containing open, closed, and scenario-based guidance strategies through a comprehensive scoring function Score(R). The core of this scoring function lies in balancing the content target and the user's feelings:

[0057] Score(R) = a · f match (R, e target,t ) + β · f adapt (R, S t )

[0058] In this function, fmatch (R,e target,t ) is a content fitness function that evaluates the direct effectiveness of candidate reply R in guiding user to provide target information e target,t in aspect e. adapt (R,S t ) is a state fitness function that evaluates the fitness of stylistic attributes of reply R, such as tone and structure, to current interaction state vector S t . For example, when S t indicates that user is highly anxious, a soothing and closed-ended reply will get a higher f adapt score. α and β are preset weight coefficients that regulate the relative importance of task orientation and humanistic care. The system finally selects and executes the guiding reply R guide,t with the highest score.

[0059] In step s5, dynamic interaction control module e5 is responsible for managing the entire dynamic interaction iteration process until the final intent is confirmed. After the system presents the guiding reply and receives the next round of input from the user, a complete interaction iteration closed loop is completed. The control module will drive the system to re-execute steps s2 to s4 to update the information gap and user state, and generate new guidance. The termination of this iteration process is controlled by the information completeness , which is defined as the ratio of the number of required information elements collected to the total required amount:

[0060]

[0061] When the confirmation threshold θ confirm , for example 0.9, is reached or exceeded, the iteration is paused, and the system generates a summary statement to initiate the final intent confirmation to the user. If the user expresses affirmation, the process ends. If the user denies, this denial information will be considered as high-confidence negative feedback, and the control module will execute backtracking correction logic, for example, instructing information gap analysis module e2 to re-evaluate and correct the information gap list, or instructing resolution module e1 to replace a candidate intent with the next highest confidence, and based on this, start a new round of more targeted guiding iteration.

[0062] To further illustrate the present application, the following takes an application scenario of gas industry third-party construction safety communication as an example to demonstrate the end-to-end process of the method of identifying personnel intent through guided communication described in the present application.

[0063] In this embodiment, it is assumed that personnel of a construction unit initiates a conversation with the intelligent customer service system (i.e. the intent recognition device of the present application) of a gas company through an instant messaging application.

[0064] First round of interaction:

[0065] User initial input: "Hello, our unit wants to dig a ditch in the Happiness Community, and we want to inform you."

[0066] After the system receives this input, the parsing module e1 starts working. Based on the semantic parsing model fine-tuned on the gas industry knowledge graph, the core entities are identified: {Location: "Happiness Community"} and {Action: "Digging a ditch"}. At the same time, the model classifies the input and gets the highest probability of the preliminary intent "Third-party construction notification".

[0067] Subsequently, the information gap analysis module e2 starts. The module queries the "intent-information element" mapping library and learns that the complete information element set required for the "third-party construction notification" intent includes {construction range, construction period, whether aware of pipeline location, contact information}. By comparing with the obtained entity information, the information gap list L gap,1 is generated: {construction range, construction period, whether aware of pipeline location, contact information}. The list is sorted according to the preset priority, and "whether aware of pipeline location" has the highest priority because it involves safety.

[0068] At the same time, the interaction state perception module e3 analyzes the user's input, which is short and direct, without obvious emotional words, no modification traces, and normal input speed. The user interaction state vector S1 generated accordingly indicates that the user is calm and cooperative.

[0069] The guidance strategy generation module e4 receives the information gap and user state. The guidance target e target,1 is "whether aware of pipeline location". Since the user is calm and this information gap requires a clear answer, the system selects the "closed guidance" strategy from the guidance strategy library and generates the guidance language based on domain knowledge. The final output is: "Hello, we have received your construction notification. To ensure construction safety, please tell us whether the construction area is clearly located at the specific position of our gas pipeline?"

[0070] Second round of interaction:

[0071] After receiving the guidance, the user replies: "I'm not sure, where is your pipeline located?"

[0072] The parsing module e1 parses this reply and does not extract new core entities, but identifies that the user expresses "unclear" state.

[0073] The dynamic interaction control module e5 updates the information and marks the status of the "whether aware of pipeline location" element as "unknown". The information gap list L gap,2 still exists, and this item still has the highest priority.

[0074] The interaction state perception module e3 analyzes again that the user's reply contains a question, expressing doubt. The state vector S2 is updated to reflect that the user has a demand for information acquisition and a certain degree of confusion.

[0075] The guiding strategy generation module e4 makes a decision based on this state. Since the user expresses confusion and needs more specific information to assist in answering, simple closed questioning is not suitable. The module selects the "scenario-based guidance" strategy, aiming to guide the user by providing more specific information. The generated guiding language is: "Understand. To avoid safety risks caused by damaging underground pipelines, we need accurate construction scope. Please provide the specific construction building number or circle the construction area on the community map. We can also provide you with the underground pipeline distribution map of the area as a reference."

[0076] Third round of interaction:

[0077] The user replies: "OK, we are between building 3 and building 4, and the construction period is about a week. Please send me the map, and the contact number is 138xxxxxxxx."

[0078] The parsing module e1 successfully extracts multiple new entity information from the reply: {construction scope: "between building 3 and building 4"}, {construction period: "a week"}, {contact information: "138xxxxxxxx"}.

[0079] The dynamic interaction control module e5 adds these new information to the acquired information list. The information gap analysis module e2 compares again and finds that all the necessary information elements in the "intention information element" mapping library have been filled. At this time, the information completeness C3 reaches 100%, exceeding the preset confirmation threshold θ confirm·

[0080] The iteration process is paused, and the system triggers the final intention confirmation process. The dynamic interaction control module e5 generates a summary statement and outputs: "OK, the information has been confirmed. Your needs are summarized as: construction work between building 3 and building 4 in the Happiness Community for about a week, and we need to provide a pipeline location map to ensure safety. Is this information correct?"

[0081] The user replies: "Correct."

[0082] At this point, a complete process of accurately identifying and confirming the intention of the personnel through multiple rounds of guided communication is completed. This embodiment clearly demonstrates how the invention can transform a vague initial notification into a complete, element-specific, and risk-controllable business event through active guidance, state perception, and closed-loop iteration, demonstrating its technical value in solving practical problems.

[0083] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A method for identifying human intent through guided communication, characterized in that, Includes the following steps: s1: Parse the user's initial input to obtain preliminary intent; s2: Analyze the information needed to confirm the final intent and the information already obtained, generating information gaps; s3: Real-time perception of user interaction state to obtain user interaction state vector; s4: Generate and execute a guidance strategy based on the information gap and the user interaction state vector; s5: Iterate through dynamic interaction with the user until the final intent is confirmed.

2. The method for identifying human intent through guided communication according to claim 1, characterized in that, Step s1 specifically includes: A semantic parsing model is constructed based on a pre-defined domain knowledge graph to process the user's initial input in order to identify core entities and the initial intent.

3. The method for identifying human intent through guided communication according to claim 1, characterized in that, Step s2 specifically includes: A preset "intent-information element" mapping library is provided, which defines the set of information elements required for different intents; The information gap is generated by comparing the acquired information with the set of information elements corresponding to the initial intent in the mapping library.

4. The method for identifying human intent through guided communication according to claim 1, characterized in that, In step s3, the user's interaction state includes at least one of the following: the emotional polarity of the text input, the content modification index, or the input rate index.

5. The method for identifying human intent through guided communication according to claim 4, characterized in that, The user interaction state vector S t In the t-th round of interaction, it is constructed as follows: S t =[s emo,t ,c mod,t ,v input,t ] Among them, s emo,t For the emotional polarity score, c mod,t Content modification index, v input,t This is the input rate index.

6. The method for identifying human intent through guided communication according to claim 1, characterized in that, The guidance strategy is selected from a preset guidance strategy library, which includes at least: open guidance for ambiguous information gaps, closed guidance for explicit information gaps, and scenario-based guidance that combines domain specifications.

7. The method for identifying human intent through guided communication according to claim 6, characterized in that, The generation and execution of the guidance strategy specifically includes: scoring the candidate guidance responses R using a comprehensive scoring function Score(R), and selecting the response with the highest score as the guidance strategy; the scoring function is: Score(R)=α·f match (R,e target,t )+β·f adapt (R,S t ); Among them, f match f is the content matching function. adapt Let e ​​be the state fitness function. target,t As the guiding objective for the current round, S t Let α and β be the user interaction state vector, and let α and β be preset weight coefficients.

8. The method for identifying human intent through guided communication according to claim 1, characterized in that, The dynamic interactive iteration specifically includes: Receive user feedback regarding the guidance strategy; Update the acquired information based on the feedback information, and re-execute steps s2, s3 and s4 to update the information gap and generate the next round of guidance strategy.

9. A method for identifying human intent through guided communication according to claim 8, characterized in that, The dynamic interactive iteration also includes: Calculate information completeness at each iteration; When the completeness of the information reaches a preset confirmation threshold, the confirmation process for the final intent is triggered.

10. A method for identifying human intent through guided communication according to claim 9, characterized in that, The process for confirming the final intent includes: When the user denies the final intent proposed by the system, the system automatically backtracks to step s2, re-analyzes and corrects the information gap based on the denial information.

Images