Interaction processing method, electronic equipment and readable storage medium

Abstract

The invention discloses an interaction processing method, electronic equipment and a readable storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: receiving interaction data of a user; performing feature extraction on the interaction data through a preset small model to obtain interaction features, and performing domain element integrity analysis on the interaction data to generate a domain element missing list; performing semantic definition and integrity evaluation based on the interaction features through a preset large model, and generating a semantic evaluation result; and if the domain element missing list is empty and the semantic assessment node indication semantics is clear and complete, generating a reply for the interaction data through a preset large model. According to the invention, the natural and smooth experience of AI interaction is improved.

Description

Interactive processing methods, electronic devices and readable storage media Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an interactive processing method, an electronic device, and a readable storage medium. Background Technology

[0002] With the rapid development of AI-powered interactive technologies, intelligent dialogue systems are increasingly being used in scenarios such as online customer service and virtual assistants, and interaction modes are constantly evolving towards naturalness and contextualization. Maintaining the fluency and naturalness of the dialogue is crucial in this process, especially when users are engaging in continuous expression or complex task interactions. The system needs to accurately understand the user's intent to avoid undue interruption or premature response, thereby ensuring a consistent and efficient interactive experience.

[0003] Currently, common interactive systems often employ simple time-delay-based strategies to prevent AI (Artificial Intelligence) from "interrupting" or triggering responses during user expression. For example, they may only generate a response after detecting a certain pause in the user's expression; or they may use basic keyword matching and intent classification models to perform real-time shallow analysis of user input to determine whether the user has finished expressing themselves.

[0004] However, these existing methods have significant shortcomings. Simply relying on time thresholds can easily lead to delayed responses or misjudgments, and cannot adapt to the different expression habits of users; while shallow semantic analysis struggles to accurately capture the true intent and completeness of a user's expression in continuous interactions, often resulting in AI prematurely interrupting the user or missing the optimal response opportunity, causing an imbalance in the conversation's rhythm and a fragmented user experience. Therefore, accurately judging user intent to dynamically prevent AI from interrupting has become a key challenge in improving the naturalness of interaction. Summary of the Invention

[0005] The main purpose of this application is to provide an interactive processing method, electronic device, and readable storage medium, which aims to solve the technical problem of how to accurately determine the user's intent to dynamically prevent AI from interrupting and improve the naturalness of the interaction.

[0006] To achieve the above objectives, this application provides an interaction processing method, which includes the following steps: receiving user interaction data; extracting interaction features from the interaction data using a preset small model, and performing domain element completeness analysis on the interaction data to generate a domain element missing list; evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate a semantic evaluation result; if the domain element missing list is empty and the semantic evaluation result indicates semantic clarity and completeness, then generating a response to the interaction data using the preset large model.

[0007] In one embodiment, before the step of generating a response to the interaction data using the preset large model, the method further includes: obtaining a dynamic pause duration threshold corresponding to the user; if no new input from the user is detected within the dynamic pause duration threshold, triggering the generation of a response using the preset large model; if new input from the user is detected within the dynamic pause duration threshold, responding to the new input from the user.

[0008] In one embodiment, the step of obtaining the pause duration threshold corresponding to the user includes: obtaining the user's user behavior feature library, wherein the user behavior feature library includes average speech rate, historical pause duration statistics and / or expression habit tags; determining the dynamic pause duration threshold corresponding to the user based on the user behavior feature library, wherein the user behavior feature library indicates that the shorter the user's historical pause duration, the shorter the dynamic pause duration threshold.

[0009] In one embodiment, the step of responding to a new user input includes: if the new input contains a preset response trigger instruction, then directly executing the step of generating a response to the interaction data using the preset large model; if the new input does not contain the response trigger instruction, then concatenating the new input with the received interaction data to generate updated interaction data; based on the updated interaction data, returning to the step of extracting interaction features from the interaction data using the preset small model to obtain interaction features, and performing domain element integrity analysis on the interaction data to generate a domain element missing list.

[0010] In one embodiment, the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate a semantic evaluation result includes: identifying the current interaction intent and current scene attributes based on the interaction features and the user's historical interaction records; fusing the interaction features, the current interaction intent, the current scene attributes, and historical interaction records to generate context-enhanced features; and evaluating the semantic clarity and completeness of the context-enhanced features using the preset large model to generate the semantic evaluation result.

[0011] In one embodiment, the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model and generating a semantic evaluation result includes: generating semantic clarity scores and semantic completeness scores corresponding to the interaction features using the preset large model; determining the semantic clarity scores and semantic completeness scores, and their respective fusion weights, according to the interaction scenario type; weighting and fusing the semantic clarity scores and semantic completeness scores based on the fusion weights to obtain a comprehensive evaluation score; if the comprehensive evaluation score is greater than a preset threshold, then semantic clarity and completeness are determined as the semantic evaluation result; if the comprehensive evaluation score is less than or equal to the preset threshold, then semantic clarity and completeness are determined as the semantic evaluation result.

[0012] In one embodiment, after the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model and generating a semantic evaluation result, the method further includes: if the domain element missing list is not empty, or the semantic evaluation result does not indicate semantic clarity and completeness, then based on the domain element missing list and the semantic evaluation result, determining a follow-up questioning strategy; constructing follow-up question content to guide the user to supplement or clarify information according to the follow-up questioning strategy; and outputting the follow-up question content to guide the user to conduct subsequent interactions.

[0013] In one embodiment, after the step of generating a response to the interaction data using the preset large model, the method further includes: using the preset large model to perform comprehensibility and completeness verification on the response based on the interaction features and the semantic evaluation results; if the verification fails, the response is rewritten or completed using the preset large model until the verification passes and the final response is output.

[0014] In addition, to achieve the above objectives, this application also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the interactive processing method as described above.

[0015] In addition, to achieve the above objectives, this application also provides a readable storage medium, which is a computer-readable storage medium, on which a computer program is stored, and the computer program is executed by a processor to implement the steps of the interactive processing method described above.

[0016] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the interactive processing method described above.

[0017] The one or more technical solutions proposed in this application have at least the following technical effects: The embodiments of this application effectively improve the problem of AI "interrupting" due to the inability to accurately judge the completeness of user expressions by constructing a two-layer evaluation and decision-making mechanism involving a small model and a large model. Specifically, firstly, a preset small model is used to extract features from the received interaction data and simultaneously perform a completeness analysis of domain elements to quickly identify whether the current input lacks key information in a specific task context. This lays a structured foundation for judging whether the user's expression constitutes a complete request. Then, the extracted interaction features are input into a preset large model, which utilizes its powerful deep semantic understanding capabilities to evaluate the clarity and overall completeness of the expression, surpassing the limitations of relying solely on time thresholds or shallow keyword matching. Finally, "an empty list of missing domain elements" and "clear and complete semantic evaluation results" are used as the dual joint criteria for generating a response. This design allows the system to trigger a response only when it is certain that the user's current expression is complete in the domain context, clear in its intent at the general semantic level, and has ended. This enables the system to intelligently distinguish between natural pauses in the expression process and the actual state of completion, dynamically and precisely controlling the timing of the response, reducing inappropriate interruptions, making the AI's interaction rhythm more in line with human conversation habits, and improving the naturalness and fluency of the interaction experience. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 is a flowchart of the first embodiment of the interactive processing method of this application; Figure 2 is a flowchart of the second embodiment of the interactive processing method of this application; Figure 3 is a flowchart of the dynamic pause duration threshold determination process involved in an embodiment of the interactive processing method of this application; Figure 4 is a flowchart of the interactive processing process involved in an embodiment of the interactive processing method of this application; Figure 5 is a schematic diagram of the hardware operating environment of the interactive processing method device in the embodiment of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] It should be noted that the execution subject of each embodiment of the interactive processing method of this application can be a computing service device with data processing, network communication and program running functions, such as a server, tablet computer, personal computer, mobile phone, etc., or an electronic device that can realize the above functions, such as a VR (Virtual Reality) headset. The embodiments of the interactive processing method of this application do not impose specific limitations on this.

[0024] Based on this, this application proposes an interaction processing method according to a first embodiment. In this embodiment, referring to FIG1, the interaction processing method includes the following steps S10 to S30: Step S10, receiving user interaction data; the interaction data can be received through an integrated front-end interface (such as a microphone array, text input box, camera, etc.), and its form includes but is not limited to voice stream, text characters, touch gestures and visual information.

[0025] Furthermore, the raw interactive data can be preprocessed, including but not limited to speech recognition to text, encoding standardization, removal of irrelevant noise, and possibly associated with the current session ID (Identification), user ID, and context history to form structured data units to be processed, providing a unified format for subsequent analysis.

[0026] Step S20: Interaction features are extracted from the interaction data using a pre-defined small model. A domain element completeness analysis is then performed on the interaction data to generate a list of missing domain elements. A lightweight, low-latency pre-defined small model (such as a small-scale neural network fine-tuned for a specific domain or a rule-based classifier) ​​is used to perform two parallel tasks. First, the model performs real-time feature extraction on the input interaction data, capturing key semantic fragments, intent identifiers, sentiment tendencies, entity information, and contextual vectors, outputting structured "interaction features."

[0027] Simultaneously, the model performs domain-adapted integrity checks: based on the interaction content or session context, it dynamically identifies the domain to which the current dialogue belongs (e.g., "online shopping," "flight booking") and loads a predefined knowledge graph or element verification template bound to that domain. The model then matches and verifies the current interaction data item by item according to the essential elements defined in the template (e.g., "check-in date," "check-out date," "room type," "number of guests," etc., under the "hotel booking" domain). By comparing the identified elements with the template requirements, a clear "domain element missing list" is automatically generated, detailing the necessary information items that have not yet been provided.

[0028] Step S30: Based on the interaction features, a preset large model is used to evaluate the semantic clarity and completeness, and a semantic evaluation result is generated; the structured "interaction features" produced by the small model are input into a preset large model (such as a pre-trained language model) with a larger number of parameters than the small model.

[0029] This large-scale model, leveraging its profound language understanding and logical reasoning capabilities gained through training on massive corpora, performs a higher-dimensional, generative decision-oriented evaluation of interactive features. The evaluation primarily unfolds in two dimensions: first, semantic clarity assessment, which determines whether the user's expressed intent is clear and unambiguous, and whether the core request is prominent and easily understood, such as identifying and excluding cases of ambiguous, polysemous, or contradictory expressions; second, semantic integrity assessment, which, based on natural dialogue logic and common sense, comprehensively judges whether the current input constitutes a semantically self-consistent, independently triggerable, and complete unit of expression within the context, rather than a thought-provoking pause, incomplete turn-taking, or fragmented information. After completing the evaluation, the large-scale model outputs a structured "semantic evaluation result," which typically includes a quantified clarity score, binary or multi-valued integrity judgment labels (such as "complete / incomplete / needs clarification"), or a quantified comprehensive score for semantic clarity and integrity, and may include a brief analysis of any points of ambiguity.

[0030] Step S40: If the list of missing domain elements is empty and the semantic evaluation result indicates that the semantics are clear and complete, then a response to the interaction data is generated through the preset large model.

[0031] The "appropriate response time" is determined when both of the following conditions are met simultaneously: 1) The domain element missing list is empty, indicating that the user has provided all necessary information within the current domain context, with no structured information gaps; 2) The semantic evaluation result clearly indicates that the semantics are clear and complete, indicating that from the perspective of general semantics and dialogue logic, the user's expression has ended and the intent is clear. When both conditions are met, the instruction preset model (or a dedicated model of the same level) is invoked, and based on complete interaction data, refined interaction features, and the entire conversation context, a natural, fluent, accurate, and targeted response is generated.

[0032] Furthermore, if any condition is not met, the generation of an immediate response can be suppressed, and strategies such as waiting for subsequent input, initiating clarifying questions (such as for missing items in a list) or providing confirmation prompts can be adopted instead. This mechanism fundamentally ensures that the AI ​​only responds after fully understanding and confirming that the user has finished expressing themselves, effectively reducing interruptions and accidental triggers, making the rhythm of the conversation more in line with human communication habits, and improving the naturalness of the interaction and the efficiency of collaboration.

[0033] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment can be referred to the above description, and will not be repeated hereafter. Based on this, referring to Figure 2, before the step of generating a response value for the interaction data through the preset large model, the method further includes: step A10, obtaining the dynamic pause duration threshold corresponding to the user; before generating a response for the interaction data, dynamically adjusting the pause duration threshold according to the user's individual characteristics and historical interaction behavior. This threshold reflects the range of pause times that may occur during the user's expression, and is used to determine whether the user has completed the current expression.

[0034] By analyzing a user's speech rate, expression habits, and pause patterns in historical interaction data, a personalized pause duration threshold can be generated for each user. For example, for users who speak quickly and fluently, the pause duration threshold may be shorter; while for users who speak slowly and take longer to think, the pause duration threshold will be correspondingly longer. In this way, it is possible to more accurately adapt to the different expression habits of users and avoid misjudgments caused by fixed time thresholds.

[0035] Step A20: If no new user input is detected within the dynamic pause duration threshold, a response is generated using the preset large model. Once the user's dynamic pause duration threshold is obtained, it is monitored in real time whether the user continues to input new interactive data within that threshold. If no new user input is detected within the dynamic pause duration threshold, the user's current expression is considered complete. At this point, the preset large model is triggered to generate a response based on the previously extracted interactive features and semantic evaluation results. This process ensures a timely response after the user completes their expression, avoiding user experience issues caused by responding too early or too late. By judging the dynamic pause duration threshold, the optimal response time can be more intelligently determined, improving the naturalness and fluency of the interaction.

[0036] Step A30: If a new user input is detected within the dynamic pause duration threshold, then respond to the new user input.

[0037] If new interactive data is detected from the user within the dynamic pause duration threshold, it indicates that the user's expression is not yet complete, or the user is adding new information. At this point, response generation will be paused, and the new input will be processed in real time. For example, the new interactive data will be merged with previous interactive data, feature extraction and semantic evaluation will be performed again, and the domain element missing list and semantic evaluation results will be updated. This process ensures real-time response to new user input, avoiding unintentional interruptions to the user's expression due to premature response generation. In this way, the user's expression progress can be continuously tracked, and the interaction strategy can be dynamically adjusted to ensure the coherence and naturalness of the interaction process.

[0038] In one possible implementation, referring to Figure 3, the step of obtaining the pause duration threshold corresponding to the user includes: Step B10, obtaining the user's user behavior feature library, wherein the user behavior feature library includes average speech rate, historical pause duration statistics, and / or expression habit tags; the "user behavior feature library" associated with the current user can be retrieved from persistent storage or constructed in real time. This feature library is generated and updated by continuously learning and analyzing the user's behavioral patterns in historical interaction sessions. Its key feature data mainly includes: average speech rate, usually measured in words or syllables per minute, reflecting the user's overall expression rhythm; historical pause duration statistics, such as the typical pause time distribution within sentences, between sentences, and during turn-taking; and expression habit tags, which may be labeled by machine learning models to characterize specific patterns of the user, such as "tendency to use long sentences," "habit of pausing briefly before key information," and "frequent use of filler words (such as 'uh,' 'um')," etc. This feature library may be stored locally on the user's device or on a cloud server and incrementally updated with each new interaction, while adhering to privacy protocols.

[0039] Step B20: Determine the dynamic pause duration threshold corresponding to the user based on the user behavior feature library, wherein the user behavior feature library indicates that the shorter the user's historical pause duration, the shorter the dynamic pause duration threshold.

[0040] By invoking a specific algorithm or query logic, and using a user behavior feature database as input, the applicable dynamic pause duration threshold for the current interaction can be calculated. The calculation process follows a principle: the shorter the user's historical pause duration, the shorter the dynamic pause duration threshold will be.

[0041] In a preferred embodiment, the dynamic pause duration threshold is determined based on a structured user behavior feature library and a corresponding weighted calculation model. The feature library contains multi-dimensional personalized data such as average speech rate, historical pause duration statistics, and expression habit tags. These features are processed comprehensively by a weighted calculation model to generate the final dynamic threshold T_dynamic, which can be expressed in its basic form as: T_dynamic = Base_T × (W_speed × F_speed) × (W_habit × F_habit).

[0042] Base_T is the base threshold, and its value is positively correlated with the historical pause duration statistics. This is to use a statistic that better represents the moment when the user "completes their expression." For example, the 70th to 80th percentile of the historical pause duration statistics in the user's historical sessions can be used as the base value. This follows the principle that "the shorter the user's historical pause duration, the shorter the base threshold."

[0043] F_speed is the speech rate factor, and W_speed is the speech rate weight. F_speed is dynamically calculated based on the real-time deviation of the user's current speech rate from its historical average speech rate. Its design logic is: when the user's current speech rate is faster than their average speech rate, this factor is less than 1, aiming to shorten the waiting threshold; conversely, it is greater than 1, to lengthen the waiting threshold. In practice, this factor can be set as the reciprocal of the ratio of the current speech rate to the average speech rate, or it can be obtained by mapping based on this ratio through a preset function. For example, if the user's current speech rate is detected to be 20% faster than their average speech rate, then F_speed can be calculated as 0.8 (i.e., 1 / 1.2), thereby shortening the basic threshold. W_speed, as a weighting coefficient in the model, is used to adjust the overall magnitude of the speech rate feature's influence on the final threshold.

[0044] F_habit is the habit factor, and W_habit is the habit weight. Each predefined expression habit label is associated with a specific adjustment factor. For example, if the expression habit label is "habitual use of filler words (such as 'uh', 'um')," it is determined that there are many thinking pauses, which are easily confused with pauses at the end of the expression. Therefore, an F_habit factor greater than 1 is set, such as 1.3, to appropriately relax the threshold to avoid accidental interruptions. If the expression habit label is "tends to use complex and long sentences," it is determined that there may be long logical pauses within sentences. Similarly, an F_habit factor greater than 1 is set (such as 1.15). If the expression habit label is "expression is usually concise and direct," then an F_habit factor less than or equal to 1 is set, such as 0.9 or 1.0, which does not extend or may even shorten the threshold.

[0045] It should be noted that if a user has multiple habit tags, F_habit can calculate its overall impact value through aggregation methods such as weighted average.

[0046] Through the weighted calculation model that integrates multiple features, the general, fixed pause duration threshold is transformed into a highly personalized, dynamically adaptive T_dynamic. This allows the model to predict and adapt to the actual moment of a user's completion of their expression based on their inherent speaking rhythm (Base_T), real-time changing speech rate (F_speed), and unique language habits (F_habit). Base_T provides a personalized historical baseline, F_speed introduces real-time state compensation, and F_habit specifically corrects judgment biases caused by specific language habits. From three complementary dimensions—historical patterns, real-time state, and inherent habits—it comprehensively models the spatiotemporal characteristics of a user's expression. Compared to a single fixed time threshold, this dynamic model not only fundamentally avoids the response delays (for fast-paced users) or interruptions (for slow-thinking users) caused by a "one-size-fits-all" approach, but also accurately adapts to the rhythm changes of the same user in different scenarios (such as slowing down speech when thinking carefully, or expressing oneself fluently on familiar topics). Thus, the dynamic judgment of this time dimension and the aforementioned semantic integrity assessment form a dual "time-space" guarantee, jointly ensuring that AI responds only at the appropriate time when the user's expression is truly complete, further improving the accuracy and natural fluency of the interaction.

[0047] In one possible implementation, the step of responding to new user input includes: step C10, if the new input contains a preset response trigger instruction, then directly executing the step of generating a response for the interaction data using the preset large model; during the process of responding to new user input, it is detected whether the new input content contains a preset response trigger instruction. These response trigger instructions are predefined specific words, phrases, or symbols used to explicitly indicate to the user that they want the system to generate a response immediately. For example, the user may add instructions such as "Please answer" or "Reply to me now" to the new input. If these trigger instructions are detected, the further analysis and splicing steps of the new input will be skipped, and the preset large model will be directly invoked to generate a response based on the existing interaction data. This mechanism ensures a rapid response when the user explicitly requests a response, avoiding delays caused by additional processing steps, thereby improving the user experience.

[0048] Step C20: If the new input does not contain the response trigger instruction, the new input is concatenated with the received interaction data to generate updated interaction data. If the user's new input does not contain the response trigger instruction, a more detailed processing flow will be entered. In this case, the content of the new input will be concatenated with the previously received interaction data to form updated interaction data. This concatenation process can consider the integration of semantic coherence and contextual logic. For example, the semantic relationship between the new input and existing interaction data can be analyzed to ensure that the concatenated data maintains semantic consistency and coherence. For instance, if the new input is a supplement or correction to previous content, it can be reasonably integrated into the existing data so that the updated interaction data can more accurately reflect the user's complete intent. In this way, the interaction data can be dynamically updated, providing more accurate input for subsequent analysis and response generation.

[0049] Step C30: Based on the updated interaction data, return to the steps of extracting interaction features from the interaction data using a preset small model, performing domain element integrity analysis on the interaction data, and generating a domain element missing list.

[0050] After generating updated interaction data, the analysis process is restarted. Specifically, a pre-defined small model is invoked to perform feature extraction and domain element integrity analysis on the updated interaction data. The small model re-extracts interaction features, including updated semantic units, keywords, and language structures, to reflect the latest state of user expression. Simultaneously, the small model performs domain element integrity analysis on the updated interaction data, checking for any new missing domain elements and generating a new list of missing domain elements. This process ensures that the understanding and analysis of interaction data can be dynamically adjusted based on the latest user input, providing a more accurate foundation for subsequent large-scale model semantic evaluation and response generation. Through this dynamic, iterative analysis mechanism, the user's expression progress can be continuously tracked, ensuring that the most appropriate response is generated at the right time.

[0051] In one possible implementation, the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate semantic evaluation results includes: Step D10, identifying the current interaction intent and current scenario attributes based on the interaction features and the user's historical interaction records; identifying the intent and scenario attributes of the current interaction before performing semantic clarity and completeness evaluation. This process is based on two parts of data: first, interaction features extracted from the interaction data, which reflect the core content and structure of the user's current expression; second, the user's historical interaction records, which contain the user's past interaction patterns, preferences, and contextual information. By analyzing the interaction features and historical records, and using a preset intent recognition algorithm and scenario analysis module, the user's current interaction intent (e.g., inquiry, command, expression of opinion, etc.) and the scenario attributes of the interaction (e.g., shopping consultation, technical support, daily conversation, etc.) are determined. For example, if a user has repeatedly inquired about product prices in historical interactions, and mentions keywords such as "price" and "discount" in the current interaction, the current interaction intent will be identified as "price consultation," and the scenario attribute as "shopping consultation."

[0052] Step D20 involves fusing the interaction features, the current interaction intent, the current scene attributes, and historical interaction records to generate context-enhanced features. After identifying the current interaction intent and scene attributes, this information is fused with the interaction features and historical interaction records to generate context-enhanced features. This process uses a feature fusion algorithm to integrate information from different sources into a comprehensive feature vector. Specifically, interaction features provide the specific content of the user's current expression; interaction intent and scene attributes provide directional guidance for semantic evaluation; and historical interaction records supplement the user's background information and preferences. For example, if the user's current expression mentions "how is the battery life of this phone?", the interaction features will extract keywords such as "phone" and "battery life"; the interaction intent is "product consultation," and the scene attribute is "shopping consultation"; the historical records show that the user previously inquired about the phone's price and features. Fusing this information generates a context-enhanced feature that includes user preferences, intent direction, and specific questions. This feature vector can more comprehensively reflect the user's expression background and intent, providing richer information for subsequent semantic evaluation.

[0053] Step D30: The semantic clarity and completeness of the context enhancement features are evaluated using the preset large model to generate the semantic evaluation result.

[0054] The system utilizes a pre-defined large model to evaluate the semantic clarity and completeness of the generated context-enhanced features. Leveraging its powerful semantic understanding and analysis capabilities, the large model performs in-depth processing on the context-enhanced features. It assesses not only the clarity of the user's expression—whether the user's intent is explicit and the expression coherent—but also the completeness of the semantics—whether the user's expression fully conveys their intended message. For example, if a user expresses "I want to buy a phone with long battery life," the large model will evaluate whether this statement clearly expresses the purchase intent (clarity) and whether it fully describes the need (completeness). If the user expresses "this phone," the large model will identify semantic incompleteness, requiring further inquiry into the user's specific needs. The large model outputs a structured "semantic evaluation result," which may include a clarity level (e.g., high / medium / low), a completeness Boolean value (yes / no), and a specific confidence score or summary of reasons, providing precise semantic-level criteria for the final response generation decision.

[0055] Based on the first and / or second embodiments of this application, in the third embodiment of this application, the content that is the same as or similar to the first and second embodiments described above can be referred to the above description and will not be repeated hereafter. Based on this, the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate semantic evaluation results includes: Step E10, generating semantic clarity scores and semantic completeness scores corresponding to the interaction features using the preset large model; the semantic clarity score measures whether the user's expressed intent is clear, whether the language is coherent, and whether the logic is clear, and is calculated by analyzing keywords, sentence structure, and semantic consistency in the interaction features. For example, if a user expresses "I want to buy a phone with long battery life," the model will give a high semantic clarity score based on its clear intent and clear expression. The semantic completeness score measures whether the user's expression contains enough information and whether it can completely express their needs or opinions. For example, if a user only says "this phone," the model will consider the expression incomplete and give a low semantic completeness score. These two scores provide basic data for subsequent comprehensive evaluation.

[0056] Step E20: Determine the semantic clarity score and semantic completeness score, and their corresponding fusion weights, based on the interaction scenario type. According to the previously identified or inferred interaction scenario type (e.g., "task instruction," "open-ended question answering," "complex problem solving," "simple greeting"), obtain the corresponding fusion weights for the semantic clarity score and semantic completeness score from the pre-configured weight mapping table (the fusion weight W_clarity for semantic clarity corresponds to the fusion weight W_completeness for semantic completeness). Different scenarios have different emphases on these two indicators. For example, in the "task instruction" scenario, the weight of W_completeness is greater than that of W_clarity, emphasizing information completeness; while in the "open-ended question answering" scenario, the weight of W_clarity may be higher, emphasizing question clarity. The weight values ​​are usually between 0 and 1, and W_clarity + W_completeness = 1.

[0057] Step E30: Based on the fusion weights, the semantic clarity score and the semantic completeness score are weighted and fused to obtain a comprehensive evaluation score. After determining the fusion weights for the semantic clarity score and the semantic completeness score, these two scores are weighted and fused to calculate the comprehensive evaluation score. The specific calculation formula is: Comprehensive Evaluation Score = (Semantic Clarity Score × Semantic Clarity Weight) + (Semantic Completeness Score × Semantic Completeness Weight). This score comprehensively reflects the overall degree of compliance of the user's expression in both clarity and completeness in the current specific scenario.

[0058] For example, assuming a semantic clarity score of 80, a semantic completeness score of 60, a semantic clarity weight of 0.7, and a semantic completeness weight of 0.3, the overall evaluation score would be: (80 × 0.7) + (60 × 0.3) = 56 + 18 = 74. This overall evaluation score reflects the overall semantic quality of the user's expression in the current interaction scenario, providing a quantitative basis for subsequent semantic evaluation result judgment.

[0059] Step E40: If the comprehensive evaluation score is greater than a preset threshold, then semantic clarity and completeness are determined as the semantic evaluation result; the comprehensive evaluation score is judged according to the preset threshold. If the comprehensive evaluation score is greater than the preset threshold, it means that the user's expression is both clear and complete in the current interaction scenario, and meets the semantic requirements. For example, if the preset threshold is 70 and the comprehensive evaluation score is 74, then semantic clarity and completeness will be determined as the semantic evaluation result. This indicates that the user's expression is clear and complete enough to generate an accurate response, thereby ensuring the smoothness of the interaction and the user experience.

[0060] Step E50: If the comprehensive evaluation score is less than or equal to a preset threshold, then the semantics are determined to be unclear and incomplete as a semantic evaluation result.

[0061] If the overall evaluation score is less than or equal to a preset threshold, it indicates that the user's expression failed to meet the standards of clarity and completeness in the current interaction scenario. For example, if the overall evaluation score is 65, which is lower than the preset threshold of 70, then the semantic evaluation result will be determined as a lack of clarity and completeness. This suggests that the user's expression may have issues such as unclear intent, incomplete information, or unclear logic. In this case, further measures can be taken, such as asking the user to provide additional information or rephrase their expression, to ensure that subsequent interactions can more accurately meet the user's needs and avoid misunderstandings or incorrect responses caused by unclear or incomplete semantics.

[0062] In one possible implementation, after the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model and generating a semantic evaluation result, the method further includes: step F10, if the domain element missing list is not empty, or the semantic evaluation result does not indicate semantic clarity and completeness, then based on the domain element missing list and the semantic evaluation result, a follow-up questioning strategy is determined; the "domain element missing list" and the "semantic evaluation result" are analyzed simultaneously to generate a structured action guide—the follow-up questioning strategy. Specifically: for domain element missing items: each missing element in the list is parsed, and a corresponding basic follow-up question template is matched according to its predefined type (such as "critical required field", "optional supplement", "conditional constraint"), and the priority of the follow-up question is determined. For example, the absence of "destination" will trigger a high-priority, direct questioning strategy; the absence of "seat preference" will trigger a guidance strategy that provides a list of options.

[0063] For semantically unclear or incomplete issues: The system delves into the specific descriptions in the semantic evaluation results (e.g., "unclear referent: it", "missing benchmark: cheaper", "incomplete logical condition: if...then..."). For each type of semantic problem, the system associates a semantic clarification strategy, such as "request explicit referent", "require comparison object", or "inquire about missing conditions".

[0064] Strategy Integration: Finally, the analysis results from the two approaches are integrated to generate a unified, multi-objective follow-up questioning strategy. This strategy clarifies the list of issues requiring follow-up questions, the questioning methods for each item (direct questioning, selection guidance, example clarification, etc.), and the possible order of follow-up questions, providing clear instructions for generating specific dialogue.

[0065] Step F20 involves constructing follow-up questions to guide the user to supplement or clarify information based on the aforementioned follow-up questioning strategy; this process transforms the strategy into natural, friendly, and executable dialogue content. The system (usually by calling upon or utilizing the natural language generation capabilities of a pre-defined large model) uses the generated "follow-up questioning strategy" as the core instruction, while combining the current "interaction features" and the complete "dialogue history context" as background information to generate the final follow-up question statement.

[0066] The generation process ensures that the content is coherent in the dialogue, for example, by using phrases like "You just mentioned..., so..." to connect the content.

[0067] For complex or multiple follow-up questions, multiple questions may be organized into a well-structured compound sentence or a short sequence of questions to improve interaction efficiency and reduce user confusion.

[0068] The generated dialogue should be natural and avoid a mechanical feel. For example, "Please enter your destination" can be optimized to "Where would you like to go?", and explanatory statements (such as "In order to find the most suitable option for you...") can be incorporated at appropriate times to improve the user experience.

[0069] Step F30: Output the follow-up question to guide the user to conduct subsequent interactions.

[0070] The constructed follow-up questions are presented to the user through speech synthesis, text display, or other appropriate interactive output interfaces. This step is usually accompanied by an update of the dialogue state, such as marking the system status as "waiting for user clarification input," and possibly resetting or adjusting parameters related to response timing (such as dynamic pause duration thresholds). By outputting follow-up questions, the interaction flow is proactively guided towards a state of complete information and semantic clarity, creating conditions for generating accurate responses later. This forms a virtuous cycle of "analysis-follow-up questioning-supplementation-reanalysis," effectively solving the problem of interaction blockage or erroneous responses caused by insufficient information or ambiguous expression.

[0071] In one possible implementation, after the step of generating a response to the interaction data using the preset large model, the method further includes: step G10, performing comprehensibility and completeness verification on the response based on the interaction features and the semantic evaluation results using the preset large model; performing comprehensibility and completeness verification on the response using the preset large model. This process uses the previously extracted interaction features and semantic evaluation results as a reference to ensure the quality of the response. Specifically, the large model checks whether the response can be easily understood by the user, i.e., whether the language is fluent and the expression is clear; at the same time, it also evaluates whether the response completely answers the user's question and whether it covers the key information and needs expressed by the user. For example, if a user asks "How is the battery life of this phone?", the initial generated response might be "The battery life is very good." During the comprehensibility verification, the large model checks whether the response clearly expresses the specific situation of the battery life; during the completeness verification, it checks whether other relevant information that the user may care about has been omitted, such as battery life, charging speed, etc. Through this verification step, problems in the response can be detected in advance, avoiding the direct output of poor-quality responses to the user.

[0072] In step G20, if the verification fails, the response is rewritten or completed using the preset large model until the verification passes and the final response is output.

[0073] If the large model finds that a response fails the understandability or completeness check during the validation process, an optimization process will be initiated. Specifically, the large model will rewrite or complete the response based on the problem description in the validation results. For example, if the validation results show that the response "Battery life is very good" lacks specific information, the large model might rewrite it as "This phone has excellent battery life, lasting a whole day under normal use, and supports fast charging, reaching 50% charge in 30 minutes." After rewriting or completion, understandability and completeness checks will be performed again. This process will be repeated until the response passes the validation. Through this cyclical optimization mechanism, it is ensured that the final output response is both clear and understandable, as well as complete and accurate, thereby increasing user trust and satisfaction with the interaction system and ensuring the quality and effectiveness of the interaction process.

[0074] For example, in order to help understand the technical concept or technical principle of the interaction processing method after combining this embodiment with the first and second embodiments described above, a specific embodiment is listed here. In this specific embodiment, referring to FIG4, the interaction processing flow includes: 1. Multimodal input acquisition: First, multimodal input is acquired, including voice, text, behavioral actions, etc., to obtain the user's interaction data.

[0075] 2. Preprocessing and Initial Analysis: The collected data enters the "input preprocessing" stage, which is handled by the "small model". Two tasks are performed in parallel here: Feature Extraction: Extracting basic features such as "speech rate, pauses, and incomplete sentence structures".

[0076] Confirm feature segmentation: Generate a "List of Missing Domain Elements" by "Loading the Glossary".

[0077] 3. Deep Semantic Understanding and Evaluation: The process then moves to the "Syntactic Reasoning Module," which is led by the "Large Model." This module performs: Context Fusion: integrating "current interaction intent + historical interaction content + current scene attributes."

[0078] Semantic integrity reasoning and scoring: Evaluate “clear intent?” and “semantic integrity?” and give a total score (e.g., out of 100).

[0079] 4. Collaborative Decision Making and Dynamic Adaptation: The system enters the "dual-model collaborative judgment" stage: "two-way verification" is performed to determine whether the conditions of "no missing cores + score > 80" are met.

[0080] If "No", the system proceeds to the "Reply Triggering and Fault Tolerance Module". Here, the system will update the "new segment characteristics" and "wait for user input". The system provides a "fault tolerance" mechanism, allowing "users to trigger replies via keywords, etc."

[0081] If "yes", the system proceeds to the "dynamic threshold adaptation" stage, adjusting the pause threshold by "calling user profile features". Then, the system makes a final "judgment" to confirm whether the "intent is clear and well-expressed". If the judgment fails ("no"), the system may return or wait; if it passes ("yes"), a response is generated.

[0082] 5. Response generation and output: In the "Response Triggering and Fault Tolerance Module", the "large model generates targeted responses", then the "small model optimizes the speech synthesis style", and finally completes the "output".

[0083] 6. Model Iteration and Optimization (Continuous Learning Loop): The entire system includes a "model iteration and optimization module" to form a continuous learning loop: small models are responsible for "uploading the case (set but not triggered / not set but triggered)".

[0084] The large model, on the other hand, "rewards user satisfaction" and "updates its reasoning logic through reinforcement learning".

[0085] In addition, the "small model re-domain vocabulary" is used to ensure the timeliness of the system's knowledge.

[0086] It should be noted that the above examples are only used to help understand this embodiment and do not constitute a limitation on the interactive processing flow of this embodiment. Any simple modifications based on this technical concept are within the protection scope of this application.

[0087] Furthermore, embodiments of this application also propose an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method described above.

[0088] Referring to Figure 5, a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of this application is shown. The electronic device in the embodiments of this application may also include, but is not limited to, mobile terminals such as mobile phones, VR headsets, laptops, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. The electronic device shown in Figure 5 is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0089] As shown in Figure 5, the electronic device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The RAM 1004 also stores various programs and data required for the operation of the electronic device. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. While electronic devices with various systems are shown in the figures, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0090] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0091] The electronic device provided in this application, employing the interaction processing method described in the above embodiments, can solve the technical problem of how to accurately determine user intent to dynamically prevent AI from interrupting and improve the naturalness of interaction. Compared with the prior art, the beneficial effects of the electronic device provided in this application are the same as those of the interaction processing method provided in the above embodiments, and other technical features in this electronic device are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0092] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0093] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0094] In addition, to achieve the above objectives, embodiments of this application also provide a readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the interactive processing method in the above embodiments.

[0095] The computer-readable storage medium provided in this application embodiment may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0096] The aforementioned computer-readable storage medium may be included in an electronic device or may exist independently without being assembled into an electronic device.

[0097] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to perform the process steps of any embodiment of the aforementioned interactive processing method.

[0098] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0099] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0100] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the modules themselves.

[0101] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described interactive processing method. This solves the technical problem of how to accurately determine user intent to dynamically prevent AI from interrupting and improve the naturalness of interaction. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the interactive processing method provided in the above embodiments, and will not be repeated here.

[0102] Furthermore, embodiments of this application also propose a computer program product, including a computer program that, when executed by a processor, implements the steps of the interactive processing method described above.

[0103] The specific implementation of the computer program product in this application is basically the same as the embodiments of the above-described interactive processing method, and will not be repeated here.

[0104] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0105] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0106] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software sensor. This computer software sensor is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause an electronic device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0107] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An interactive processing method, characterized in that, The interaction processing method includes the following steps: receiving user interaction data; extracting interaction features from the interaction data using a preset small model, and performing domain element completeness analysis on the interaction data to generate a domain element missing list; evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate a semantic evaluation result; if the domain element missing list is empty and the semantic evaluation result indicates semantic clarity and completeness, then generating a response to the interaction data using the preset large model.

2. The interactive processing method as described in claim 1, characterized in that, Before the step of generating a response to the interaction data through the preset large model, the method further includes: obtaining a dynamic pause duration threshold corresponding to the user; if no new input from the user is detected within the dynamic pause duration threshold, triggering the generation of a response through the preset large model; if new input from the user is detected within the dynamic pause duration threshold, responding to the new input from the user.

3. The interactive processing method as described in claim 2, characterized in that, The step of obtaining the pause duration threshold corresponding to the user includes: obtaining the user's user behavior feature library, wherein the user behavior feature library includes average speech rate, historical pause duration statistics and / or expression habit tags; determining the dynamic pause duration threshold corresponding to the user based on the user behavior feature library, wherein the user behavior feature library indicates that the shorter the user's historical pause duration, the shorter the dynamic pause duration threshold.

4. The interactive processing method as described in claim 2, characterized in that, The steps for responding to new user input include: if the new input contains a preset response trigger instruction, then directly execute the step of generating a response to the interaction data using the preset large model; if the new input does not contain the response trigger instruction, then concatenate the new input with the received interaction data to generate updated interaction data; based on the updated interaction data, return to execute the steps of extracting interaction features from the interaction data using the preset small model to obtain interaction features, and performing domain element integrity analysis on the interaction data to generate a domain element missing list.

5. The interactive processing method as described in claim 1, characterized in that, The step of evaluating semantic clarity and completeness based on the interaction features using a preset large model to generate a semantic evaluation result includes: identifying the current interaction intent and current scene attributes based on the interaction features and the user's historical interaction records; fusing the interaction features, the current interaction intent, the current scene attributes, and historical interaction records to generate context-enhanced features; and evaluating the semantic clarity and completeness of the context-enhanced features using the preset large model to generate the semantic evaluation result.

6. The interactive processing method according to any one of claims 1 to 5, characterized in that, The step of evaluating semantic clarity and completeness based on the interaction features using a preset large model and generating a semantic evaluation result includes: generating semantic clarity scores and semantic completeness scores corresponding to the interaction features using the preset large model; determining the semantic clarity scores and semantic completeness scores, and their respective fusion weights, according to the interaction scenario type; weighting and fusing the semantic clarity scores and semantic completeness scores based on the fusion weights to obtain a comprehensive evaluation score; if the comprehensive evaluation score is greater than a preset threshold, then semantic clarity and completeness are determined as the semantic evaluation result; if the comprehensive evaluation score is less than or equal to the preset threshold, then semantic clarity and completeness are determined as the semantic evaluation result.

7. The interactive processing method according to any one of claims 1 to 5, characterized in that, After the step of evaluating semantic clarity and completeness based on the interaction features using a preset large model and generating a semantic evaluation result, the method further includes: if the domain element missing list is not empty, or the semantic evaluation result does not indicate semantic clarity and completeness, then based on the domain element missing list and the semantic evaluation result, determining a follow-up questioning strategy; constructing follow-up questioning content to guide the user to supplement or clarify information according to the follow-up questioning strategy; and outputting the follow-up questioning content to guide the user to conduct subsequent interactions.

8. The interactive processing method according to any one of claims 1 to 5, characterized in that, After the step of generating a response to the interaction data using the preset large model, the method further includes: using the preset large model to perform comprehensibility and completeness verification on the response based on the interaction features and the semantic evaluation results; if the verification fails, the response is rewritten or completed using the preset large model until the verification passes and the final response is output.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the interactive processing method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a device control program, which, when executed by a processor, implements the steps of the interactive processing method as described in any one of claims 1 to 8.

Images