A language interaction-based robot interaction method, system, device and medium

Abstract

The application relates to the technical field of robot interaction, and provides a robot interaction method based on language interaction, which comprises the following steps: collecting language information of a robot interaction object and interaction scene information and emotional interaction feedback in a preset time period; collecting language reply information and attached emotional information to be replied by the robot according to the language information; determining causal reply information corresponding to the language information according to a preset causal correlation memory graph library; comprehensively calibrating and analyzing the causal reply information, the language reply information, the emotional interaction feedback and the attached emotional information and adjusting the attached emotional information; and generating language interaction instructions to control the robot interaction according to the adjusted attached emotional information and the language reply information. The application realizes dynamic cooperation of emotional expression and semantic logic, and improves the naturalness and accuracy of interaction.

Description

Technical Field

[0001] This application relates to the technical field of robot interaction, and in particular to a robot interaction method, system, device and medium based on language interaction. Background Technology

[0002] With the rapid development of service robot technology, human-computer language interaction has become one of the core functions of robot applications. Existing robot interaction solutions typically focus on the accuracy of natural language understanding or the fluency of speech synthesis, but often employ static mapping or simple keyword matching strategies at the emotional interaction level. However, in actual interaction, the robot's preset emotional expressions may not match the logical semantics of the current dialogue, or they may fail to dynamically adapt to changes in the user's real-time emotional feedback and the environment. This disconnect between emotional expression and semantic logic, and the user's genuine feelings, leads to robots exhibiting stiff, misaligned, or even offensive emotional responses in complex interaction scenarios, severely impacting the naturalness and empathetic experience of human-computer interaction. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this application provides a robot interaction method, system, device, and medium based on language interaction.

[0004] Firstly, this application provides a robot interaction method based on language interaction, which adopts the following technical solution:

[0005] A language-based robot interaction method, comprising:

[0006] Collect language information of the robot's interaction objects, as well as interaction scenario information and emotional interaction feedback within a preset time period;

[0007] The data collection robot will provide the language response and accompanying emotional information based on the language information.

[0008] Based on a pre-set causal association memory graph database, determine the causal response information corresponding to the linguistic information;

[0009] A comprehensive calibration analysis is performed on the causal response information, the linguistic response information, the emotional interaction feedback, and the ancillary emotional information. Based on the analysis results, the ancillary emotional information is calibrated and adjusted to obtain the adjusted ancillary emotional information.

[0010] Based on the adjusted auxiliary emotional information and the language response information, language interaction instructions are generated to control the robot to perform language interaction with the robot's interaction object.

[0011] In a preferred embodiment, this application can be further configured as follows: a comprehensive calibration analysis is performed on the causal response information, the linguistic response information, the emotional interaction feedback, and the ancillary emotional information, and the ancillary emotional information is calibrated and adjusted based on the analysis results to obtain adjusted ancillary emotional information, including:

[0012] Determine whether the semantics of the causal response information are consistent with the semantics of the linguistic response information. If they are consistent, determine the real emotional information corresponding to the robot interaction object based on the interaction scenario information and the emotional interaction feedback. Input the auxiliary emotional information and the real emotional information into the emotional expression standard for emotional value detection to obtain a first emotional value corresponding to the auxiliary emotional information and a second emotional value corresponding to the real emotional information.

[0013] Determine whether the first emotional value and the second emotional value are consistent. If they are inconsistent, determine the emotional difference between the first emotional value and the second emotional value, and adjust the emotional intensity of the first emotional value based on the emotional difference to obtain the adjusted auxiliary emotional information.

[0014] In a preferred embodiment, this application can be further configured such that: determining whether the semantics of the causal response information is consistent with the semantics of the linguistic response information includes:

[0015] If the semantics of the causal response information are inconsistent with the semantics of the language response information, then the language response information is updated based on the causal response information to obtain the updated language response information;

[0016] The system detects whether there are any abnormal situations in the historical time period where the semantics of the causal response information are inconsistent with the semantics of the language response information. If such situations exist, the number of abnormal situations is determined, and it is judged whether the number of abnormal situations is greater than or equal to a preset maximum number of abnormal situations. If it is greater, maintenance information is generated and sent to a preset terminal for display output.

[0017] In a preferred embodiment, this application can be further configured such that: after generating language interaction instructions based on the adjusted auxiliary emotional information and the language response information, and controlling the robot to perform language interaction with the robot interaction object, the application further includes:

[0018] Acquire emotional calibration data where the first emotional value and the second emotional value are inconsistent during a historical time period, as well as the interaction scenario information corresponding to the emotional calibration data;

[0019] The emotional calibration data and the interaction scene information are analyzed for emotional anomalies according to the timeline to obtain the emotional deviation trend of each scene interference information at different time points.

[0020] Obtain the current emotion definition standard of the robot under the interactive scenario information, and determine the emotion output value of the robot under each scenario interference information according to the emotion definition standard;

[0021] The emotional output value is calibrated and updated based on the emotional deviation trend to obtain the calibrated and updated emotional output value.

[0022] The emotion definition criteria are updated based on the calibrated and updated emotion output values ​​to obtain the updated emotion definition criteria.

[0023] In a preferred embodiment, this application can be further configured as follows: the step of performing emotional anomaly analysis on the emotional calibration data and the interaction scene information according to a timeline to obtain the emotional deviation trend of interference information for each scene at different time points includes:

[0024] Based on the emotion calibration data and the interaction scenario information, determine the first emotion data of the first interaction abnormality and the second emotion data of the non-first interaction abnormality under different interaction scenarios;

[0025] Based on the first emotion data, the primary real emotion value and the primary output emotion value are determined in different interaction scenarios. The primary real emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The primary output emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object.

[0026] The secondary true emotion value and secondary output emotion value are determined based on the second emotion data in different interaction scenarios. The secondary true emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The secondary true emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object.

[0027] The primary real emotion value and the primary output emotion value, and the secondary real emotion value and the secondary output emotion value are respectively processed according to the interaction scenario to obtain the first emotion information corresponding to the primary real emotion value and the second emotion information corresponding to the secondary real emotion value.

[0028] The first emotional information and the second emotional information are integrated, calculated and analyzed according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information.

[0029] The scene interference information and the emotional deviation data are organized and analyzed according to the timeline to obtain the emotional deviation trend of each scene interference information at future time nodes.

[0030] In a preferred embodiment, this application can be further configured as follows: the step of integrating and calculating the first emotional information and the second emotional information according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information includes:

[0031] Construct a first emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0032] The emotional data in the first emotional information and the second emotional information are imported into the first emotional coordinate system according to the interaction scenario and timeline to obtain a first three-dimensional curve corresponding to the first emotional information and a second three-dimensional curve corresponding to the second emotional information.

[0033] Filter out the first non-overlapping abnormal curves in the first three-dimensional curve and the second non-overlapping abnormal curves in the second three-dimensional curve respectively;

[0034] Scene acquisition and analysis are performed on the first abnormal curve and the second abnormal curve to determine the first scene interference information corresponding to the first abnormal curve and the second scene interference information corresponding to the second abnormal curve.

[0035] Calculate the first sentiment deviation data corresponding to the first abnormal curve and the second sentiment deviation data corresponding to the second abnormal curve;

[0036] The first emotional deviation data and the first scene interference information, and the second emotional deviation data and the second scene interference information are respectively processed to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information.

[0037] In a preferred embodiment, this application can be further configured as follows: the step of organizing and analyzing the scene interference information and the sentiment deviation data according to a timeline to obtain the sentiment deviation trend of each scene interference information at future time nodes includes:

[0038] Construct a second emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional deviation data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0039] The scene parameters in the scene interference information and the emotional deviation data are imported into the second emotional coordinate system according to the timeline to obtain the third three-dimensional curve;

[0040] Calculate the dynamic slope of the third three-dimensional curve in different time periods, and perform periodic change analysis on the dynamic slope of the curve to obtain the slope variation period;

[0041] Based on the slope variation period, the trend of the third three-dimensional curve is predicted to obtain the emotional bias trend of interference information in each scene at future time nodes.

[0042] Secondly, this application provides a robot interaction system based on language interaction, which adopts the following technical solution:

[0043] A language-based robot interaction system, comprising:

[0044] The first acquisition module is used to collect language information of the robot's interactive objects, as well as interactive scene information and emotional interaction feedback within a preset time period.

[0045] The second acquisition module is used to acquire the language response information and associated emotional information that the robot will reply to based on the language information.

[0046] The response determination module is used to determine the causal response information corresponding to the language information based on a preset causal association memory graph library.

[0047] The calibration analysis module is used to perform comprehensive calibration analysis on the causal response information, the language response information, the emotional interaction feedback, and the auxiliary emotional information, and to calibrate and adjust the auxiliary emotional information according to the analysis results to obtain the adjusted auxiliary emotional information.

[0048] The interaction control module is used to generate language interaction instructions based on the adjusted auxiliary emotional information and the language response information, and to control the robot to perform language interaction with the robot interaction object.

[0049] In one possible implementation, when the calibration analysis module performs a comprehensive calibration analysis on the causal response information, the linguistic response information, the emotional interaction feedback, and the ancillary emotional information, and adjusts the ancillary emotional information based on the analysis results to obtain the adjusted ancillary emotional information, it is specifically used for:

[0050] Determine whether the semantics of the causal response information are consistent with the semantics of the linguistic response information. If they are consistent, determine the real emotional information corresponding to the robot interaction object based on the interaction scenario information and the emotional interaction feedback. Input the auxiliary emotional information and the real emotional information into the emotional expression standard for emotional value detection to obtain a first emotional value corresponding to the auxiliary emotional information and a second emotional value corresponding to the real emotional information.

[0051] Determine whether the first emotional value and the second emotional value are consistent. If they are inconsistent, determine the emotional difference between the first emotional value and the second emotional value, and adjust the emotional intensity of the first emotional value based on the emotional difference to obtain the adjusted auxiliary emotional information.

[0052] In another possible implementation, when determining whether the semantics of the causal response information are consistent with the semantics of the linguistic response information, the calibration analysis module is specifically used for:

[0053] If the semantics of the causal response information are inconsistent with the semantics of the language response information, then the language response information is updated based on the causal response information to obtain the updated language response information;

[0054] The system detects whether there are any abnormal situations in the historical time period where the semantics of the causal response information are inconsistent with the semantics of the language response information. If such situations exist, the number of abnormal situations is determined, and it is judged whether the number of abnormal situations is greater than or equal to a preset maximum number of abnormal situations. If it is greater, maintenance information is generated and sent to a preset terminal for display output.

[0055] In another possible implementation, the system further includes: an information acquisition module, an anomaly analysis module, an output determination module, an output update module, and a standard update module, wherein,

[0056] The information acquisition module is used to acquire emotional calibration data in which the first emotional value and the second emotional value are inconsistent during a historical time period, as well as the interaction scenario information corresponding to the emotional calibration data;

[0057] The anomaly analysis module is used to perform emotional anomaly analysis on the emotional calibration data and the interaction scene information according to the timeline, and obtain the emotional deviation trend of each scene interference information at different time nodes.

[0058] The output determination module is used to obtain the current robot's emotion definition standard under the interactive scenario information, and determine the robot's emotion output value under each scenario interference information according to the emotion definition standard.

[0059] The output update module is used to calibrate and update the emotional output value based on the emotional deviation trend to obtain the calibrated and updated emotional output value.

[0060] The standard update module is used to update the emotion definition standard according to the calibrated and updated emotion output value to obtain the updated emotion definition standard.

[0061] In another possible implementation, when the anomaly analysis module performs emotional anomaly analysis on the emotional calibration data and the interaction scenario information according to a timeline to obtain the emotional deviation trend of each scenario's interference information at different time points, it is specifically used for:

[0062] Based on the emotion calibration data and the interaction scenario information, determine the first emotion data of the first interaction abnormality and the second emotion data of the non-first interaction abnormality under different interaction scenarios;

[0063] Based on the first emotion data, the primary real emotion value and the primary output emotion value are determined in different interaction scenarios. The primary real emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The primary output emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object.

[0064] The secondary true emotion value and secondary output emotion value are determined based on the second emotion data in different interaction scenarios. The secondary true emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The secondary true emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object.

[0065] The primary real emotion value and the primary output emotion value, and the secondary real emotion value and the secondary output emotion value are respectively processed according to the interaction scenario to obtain the first emotion information corresponding to the primary real emotion value and the second emotion information corresponding to the secondary real emotion value.

[0066] The first emotional information and the second emotional information are integrated, calculated and analyzed according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information.

[0067] The scene interference information and the emotional deviation data are organized and analyzed according to the timeline to obtain the emotional deviation trend of each scene interference information at future time nodes.

[0068] In another possible implementation, when the anomaly analysis module integrates and analyzes the first emotional information and the second emotional information according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information, it is specifically used for:

[0069] Construct a first emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0070] The emotional data in the first emotional information and the second emotional information are imported into the first emotional coordinate system according to the interaction scenario and timeline to obtain a first three-dimensional curve corresponding to the first emotional information and a second three-dimensional curve corresponding to the second emotional information.

[0071] Filter out the first non-overlapping abnormal curves in the first three-dimensional curve and the second non-overlapping abnormal curves in the second three-dimensional curve respectively;

[0072] Scene acquisition and analysis are performed on the first abnormal curve and the second abnormal curve to determine the first scene interference information corresponding to the first abnormal curve and the second scene interference information corresponding to the second abnormal curve.

[0073] Calculate the first sentiment deviation data corresponding to the first abnormal curve and the second sentiment deviation data corresponding to the second abnormal curve;

[0074] The first emotional deviation data and the first scene interference information, and the second emotional deviation data and the second scene interference information are respectively processed to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information.

[0075] In another possible implementation, when the anomaly analysis module organizes and analyzes the scene interference information and the sentiment deviation data according to a timeline to obtain the sentiment deviation trend of each scene interference information at future time points, it is specifically used for:

[0076] Construct a second emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional deviation data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0077] The scene parameters in the scene interference information and the emotional deviation data are imported into the second emotional coordinate system according to the timeline to obtain the third three-dimensional curve;

[0078] Calculate the dynamic slope of the third three-dimensional curve in different time periods, and perform periodic change analysis on the dynamic slope of the curve to obtain the slope variation period;

[0079] Based on the slope variation period, the trend of the third three-dimensional curve is predicted to obtain the emotional bias trend of interference information in each scene at future time nodes.

[0080] Thirdly, this application provides an electronic device that adopts the following technical solution:

[0081] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described language-based robot interaction method.

[0082] Fourthly, this application provides a computer storage medium, as follows:

[0083] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described language-based robot interaction method.

[0084] In summary, this application has the following beneficial technical effects:

[0085] By constructing a comprehensive calibration mechanism that includes a causal association memory graph library, this method deeply integrates and analyzes the robot's language responses with its inherent causal logic, the user's real-time emotional feedback, and environmental scene information. This approach not only solves the problem of the disconnect between emotional expression and semantic logic in traditional interactions but also achieves precise dynamic adjustment of auxiliary emotional information by quantifying emotional differences. Furthermore, this application utilizes historical interaction data to construct a three-dimensional emotional coordinate system, analyzes the trend and periodic patterns of emotional deviations under scene interference, and adaptively updates the emotional definition criteria accordingly. This dual mechanism of "real-time calibration + long-term evolution" significantly improves the accuracy, naturalness, and environmental adaptability of the robot's emotional interaction in complex and ever-changing scenarios, effectively enhancing the empathetic experience and intelligence level of human-computer interaction. Attached Figure Description

[0086] Figure 1 This is a flowchart of a language-based robot interaction method in one embodiment of this application.

[0087] Figure 2 This is a schematic diagram of the structure of a language-based robot interaction system according to one embodiment of this application.

[0088] Figure 3 This is a schematic block diagram of an electronic device according to one embodiment of this application. Detailed Implementation

[0089] The following is in conjunction with the appendix Figure 1 To be continued Figure 3 This application will be described in further detail.

[0090] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0091] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0092] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0093] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0094] This application provides a language-based robot interaction method executed by an electronic device, which can be a server or a terminal device. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smartphone, tablet, laptop, desktop computer, etc., but is not limited to these. The terminal device and the server can be directly or indirectly connected via wired or wireless communication. This application does not impose any limitations on this connection. Figure 1 As shown, the method includes:

[0095] Step S10: Collect language information of the robot's interaction objects, as well as interaction scene information and emotional interaction feedback within a preset time period.

[0096] Specifically, linguistic information refers to the speech or text content emitted by the robot's interactive object, serving as the basic input for triggering interaction. The preset time period is not a fixed, static duration, but rather a dynamically adjusted time window based on the interaction state. For example, in a preferred implementation, this preset time period can employ a sliding window mechanism, i.e., tracing back a specific duration (such as the most recent 30 seconds or 1 minute) from the current moment; alternatively, the preset time period can also be based on an event-triggered conversation cycle, i.e., the time interval between the end of the previous interaction and the start of the current interaction. This dynamic selection logic ensures that the collected scene information and emotional feedback have the strongest temporal relevance to the current linguistic information, avoiding the introduction of irrelevant noise due to an excessively large time span, or the loss of context due to an excessively small time span.

[0097] Interactive scene information includes, but is not limited to, physical environment parameters such as ambient noise level, light intensity, current time, geographical location, and the number of people present. Emotional interaction feedback not only includes voice tone features but can also encompass multimodal data such as facial expressions and body movements. It should be understood that although this embodiment uses voice interaction as an example, in other embodiments, language information can also be collected through text input via a touchscreen or gesture recognition-converted instructions, as long as they can represent the intention of the interactive object.

[0098] Step S11: Collect the language response information and accompanying emotional information that the robot will reply based on the language information.

[0099] Specifically, the language response information is the text content output by the robot's natural language generation module, while the accompanying sentiment information is metadata tags attached to this language content. These tags instruct the robot on the appropriate emotional tone, speech rate, pitch curve, and other expressive parameters to use when delivering the text. It is crucial to emphasize that the accompanying sentiment information exists independently of the semantic content of the language response information; it is a parameterized description of "how to say it," not part of "what to say." For example, for the same sentence "The weather is nice today," the accompanying sentiment information could be "cheerful-high-pitched" or "calm-soothing." Decoupling the two provides a data foundation for subsequently adjusting the sentiment expression independently without altering the semantic content. In practical systems, the accompanying sentiment information is typically initially generated by the dialogue management module according to a preset strategy. However, this initial sentiment is often based solely on keyword matching or simple intent classification and has not undergone deep logical verification, thus potentially leading to discrepancies with the real-world context.

[0100] Step S12: Determine the causal response information corresponding to the language information based on the preset causal association memory map library.

[0101] In this embodiment, the preset causal association memory graph library is the core knowledge base for the present invention to achieve logical consistency calibration. Its storage structure is different from the traditional keyword index or pure semantic vector space, but adopts a structured causal relationship network.

[0102] Specifically, the data units in this knowledge graph library can be represented as knowledge graph triples, such as <user asks about the weather, leading to, obtaining meteorological data>, <meteorological data shows heavy rain, leading to, suggesting carrying rain gear>, <user is depressed, suppressed, cheerfully reporting>, etc. These triples explicitly define the causal chains and logical constraints between events. Upon receiving linguistic information, the system performs subgraph retrieval or path reasoning within the graph to find standard response nodes logically related to the current input, using them as causal response information. The significance of this process lies in providing an objective "logical anchor" or "truth benchmark." Compared to language models that rely solely on probability generation, the causal association memory graph library ensures that the robot's responses conform to common sense logic and established rules, preventing emotional and factual misalignment caused by model illusions. It should be understood that although this embodiment uses knowledge graph triples as an example, in other implementations, the knowledge graph library can also take the form of a production rule base, decision tree, or vector index with causal constraints, as long as it can represent and retrieve standard reference information that has a causal logical relationship with the input information.

[0103] Step S13: Perform a comprehensive calibration analysis on causal response information, linguistic response information, emotional interaction feedback, and affiliative emotional information, and calibrate and adjust the affiliative emotional information based on the analysis results to obtain the adjusted affiliative emotional information.

[0104] In the embodiments of this application, the comprehensive calibration analysis is not a simple weighted average or linear superposition of information from various sources, but a multi-level adaptation process based on logical consistency.

[0105] Specifically, the system first verifies whether the linguistic response and causal response conflict semantically. Based on logical consistency, it further evaluates whether the current auxiliary emotional information matches the user's true state reflected in the emotional interaction feedback and whether it conforms to the atmosphere of the current interaction scenario. If the analysis finds a deviation between the auxiliary emotional information and the user feedback or scenario logic, the system will adjust it according to a preset calibration strategy. For example, if the user feedback shows sadness, while the auxiliary emotional information is marked as "excitement," and the causal response confirms the current topic is a negative event, the calibration analysis will adjust the auxiliary emotional information to "sympathy" or "depression." This calibration mechanism ensures that the final emotional output respects the user's real-time feelings while adhering to objective causal logic, achieving a "fusion of emotion and reason" in the interaction. It should be noted that this step focuses on the overall architecture and functional limitations of the calibration; detailed implementation methods such as specific difference calculations and threshold judgments will be elaborated in subsequent embodiments.

[0106] Step S14: Generate language interaction instructions based on the adjusted auxiliary emotional information and language response information, and control the robot to perform language interaction with the robot interaction object.

[0107] After calibration and adjustment, the system injects the updated auxiliary emotional parameters into the language response information synthesis process, generating a language interaction instruction containing complete semantic content and precise emotional expression parameters. This instruction is sent to the robot's speech synthesis engine or actuator, driving the robot to output a response in a manner consistent with the current context. Through the closed-loop processing of steps S10 to S14 above, this embodiment of the invention establishes a complete interaction link from multi-source perception and logical retrieval to emotional adaptive adjustment. This solution effectively solves the problems of mechanical and context-detached emotional expression in existing technologies, enabling the robot to maintain logical consistency in emotional output and user comfort in complex and ever-changing interaction scenarios, significantly improving the naturalness and intelligence of human-computer interaction.

[0108] This embodiment further refines the specific implementation logic of the comprehensive calibration analysis and constructs a dual processing path based on semantic consistency verification to ensure that the robot can maintain the rationality of the interaction and the stability of the system under different logical states.

[0109] Specifically, the process of comprehensively calibrating and analyzing causal response information, verbal response information, emotional interaction feedback, and auxiliary emotional information, and then calibrating and adjusting the auxiliary emotional information based on the analysis results to obtain the adjusted auxiliary emotional information includes: determining whether the semantics of the causal response information is consistent with the semantics of the verbal response information; if consistent, determining the true emotional information corresponding to the robot's interactive object based on the interaction scenario information and emotional interaction feedback, and inputting the auxiliary emotional information and the true emotional information into the emotional expression standard for emotional value detection to obtain a first emotional value corresponding to the auxiliary emotional information and a second emotional value corresponding to the true emotional information; determining whether the first emotional value and the second emotional value are consistent; if inconsistent, determining the emotional difference between the first emotional value and the second emotional value, and adjusting the emotional intensity of the first emotional value based on the emotional difference to obtain the adjusted auxiliary emotional information.

[0110] This processing method constitutes the system's "normal calibration path." Once the semantic consistency check passes, it means the robot's response is logically correct, and the focus shifts to the accurate adaptation of emotional expression. Emotional expression standards serve as the bridge connecting abstract emotional labels with calculable numerical values. These standards can take the form of a multi-dimensional emotional space model, such as the PAD three-dimensional emotional model (Pleasure, Arousal, Dominance), or discrete emotion scales (such as happiness, sadness, anger, fear, and their corresponding intensities). Taking the PAD model as an example, the system maps auxiliary emotional information (such as "enthusiastic recommendation") to coordinates (P1, A1, D1) in space as the first emotional value, while mapping the true emotional information inferred from the user's voice tone, facial expressions, and current environmental noise to coordinates (P2, A2, D2) as the second emotional value. The so-called emotional difference is the Euclidean distance or weighted Manhattan distance between these two coordinates in multi-dimensional space. Adjustments based on this difference are not simple numerical replacements, but rather a smooth transition or compensation strategy. For example, if the difference exceeds a preset threshold, the system can vector-shift the first emotional value towards the second emotional value according to a preset step size or interpolation algorithm, or dynamically adjust the weight parameters of the emotional synthesis based on the magnitude of the difference. This quantitative adjustment mechanism avoids the abruptness of emotional switching, ensuring that the robot's emotional output retains the original service intent while also resonating with the user's real-time mood, achieving an advancement from "semantic correctness" to "emotional empathy."

[0111] As a complementary and complete branch to the normal calibration path described above, if the semantics of the causal response information are determined to be inconsistent with the semantics of the linguistic response information, an "abnormal correction path" is triggered. Specifically, this includes: updating the linguistic response information based on the causal response information to obtain the updated linguistic response information; detecting whether there are any abnormal situations in the historical time period where the semantics of the causal response information and the linguistic response information are inconsistent; if so, determining the number of abnormal situations and whether the number of abnormal situations is greater than or equal to a preset maximum number of abnormal situations; if greater, generating maintenance information and sending the maintenance information to a preset terminal for display output.

[0112] In this path, the system follows the principle of "logical truth value first." Because the causal association memory graph is typically considered a validated knowledge benchmark, when a generated language response conflicts with it (e.g., a user asks "What medicine should I take for a cold?", the language response generates an incorrect folk remedy, while the causal graph retrieves standard medical advice), the original language response must be forcibly overridden or corrected with causal response information to prevent the spread of misinformation. At this point, sentiment calibration temporarily gives way to content correction, ensuring the safety and accuracy of the interaction. More importantly, this embodiment introduces an anomaly counting and circuit breaker mechanism based on a time window. A single inconsistency may be occasional generation noise, but if semantic conflicts frequently occur within a historical time period (e.g., the past 24 hours or the most recent 100 interactions), it highly likely indicates structural defects in the underlying knowledge base, contaminated data sources, or a pattern collapse in the natural language generation module. When the number of anomalies reaches the preset maximum number of anomalies (this threshold can be flexibly configured according to business tolerance, for example, set to 5 or 10 times), the system no longer attempts automatic repair but instead generates maintenance information containing anomaly logs, conflict samples, and the frequency of occurrence, and pushes it to the preset terminal of the maintenance personnel. This design elevates a simple interaction method into a maintainable system with self-diagnostic capabilities, effectively preventing continuous service degradation caused by algorithmic black boxes and ensuring the long-term reliability of the robot's operation.

[0113] It should be understood that although this embodiment uses the PAD model and Euclidean distance as examples to illustrate the method of emotion quantification and adjustment, in other implementations, the same function can also be achieved using a circular emotion model, vector cosine similarity, or other custom nonlinear mapping functions. Similarly, the statistical window for anomaly counting can be either a fixed duration or a sliding count based on interaction rounds, as long as it can characterize the frequency features of semantic conflicts. These two paths together constitute a robust defense system: pursuing subtle emotional adaptation under normal conditions, and adhering to logical bottom lines and proactively issuing warnings under abnormal conditions, thereby comprehensively improving the intelligence level and security boundaries of human-computer interaction.

[0114] This embodiment further elevates the technical solution from real-time calibration in a single interaction to an adaptive evolutionary dimension of long-term learning. Specifically, after generating language interaction commands based on adjusted auxiliary emotional information and language response information, and controlling the robot to perform language interaction with the robot's interaction object, it also includes:

[0115] Acquire sentiment calibration data where the first sentiment value and the second sentiment value are inconsistent in a historical time period, as well as the corresponding interaction scenario information.

[0116] This application essentially constructs a "mistake set" mechanism for emotional interaction. It's important to note that the emotional calibration data here is not all interaction records, but rather a specially selected set of samples that were deemed "inconsistent" in the comprehensive calibration analysis and underwent adjustment. These samples represent moments where there is a significant deviation between the robot's preset emotional model and the user's actual feelings, possessing extremely high learning value. The corresponding interaction scenario information includes the complete environmental context when this deviation occurred, such as timestamps, ambient noise levels (decibels), illumination lumens, and dialogue topic tags. Through this targeted data mining, the system can avoid ineffective calculations in massive amounts of normal interaction data, focusing instead on the long-tail problems that truly cause a decline in user experience.

[0117] It should be understood that the length of the historical time period can be flexibly configured according to the robot's deployment cycle and learning rate requirements. For example, it can be the past week, a month, or all periods since the last model update. This invention does not limit this.

[0118] Emotional calibration data and interaction scenario information are analyzed for emotional anomalies according to a timeline to obtain the emotional deviation trend of interference information for each scenario at different time points.

[0119] This application aims to extract continuous, regular features from discrete bias samples. Specifically, the system maps collected emotion calibration data onto a timeline and performs cluster analysis in conjunction with interaction scenario information to identify specific scenario factors (i.e., scenario interference information) that are the main variables causing emotion bias. For example, analysis may reveal that when ambient light is below a certain threshold and it is evening, users' acceptance of the robot's "calm" tone decreases significantly, showing a negative emotion bias trend; or in noisy environments, users' tolerance for high-arousal emotions fluctuates periodically over time. This bias trend not only describes the magnitude of the bias but also reveals the dynamic trajectory of the bias changing with time and scenario. Compared to static statistical analysis, trend analysis incorporating the timeline dimension can capture the lag or cumulative effect of environmental factors on emotion perception, providing a more refined basis for subsequent prediction and adaptation.

[0120] Obtain the emotion definition criteria of the robot under the current interactive scenario information, and determine the emotional output value of the robot under each scenario interference information based on the emotion definition criteria.

[0121] Here, it is crucial to clearly distinguish between the "emotion definition standard" and the "emotion expression standard" mentioned above. The emotion expression standard is typically a general, fundamental emotion quantification model (such as the PAD model), while the emotion definition standard is the specific set of parameter configurations used by the robot when applying this general model in a specific scenario. Specifically, the emotion definition standard is a dynamic set of parameters, including but not limited to baseline values ​​for each emotion dimension, allowable fluctuation threshold ranges, and weighting coefficients of different scenario factors on the emotion output. For example, in the "hospital" interaction scenario, the emotion definition standard might stipulate a strict upper limit on "pleasure," and the baseline value for "arousal" might be reduced by 20% by default. Based on this standard, the system can calculate the theoretically target emotion value that the robot should output under specific scenario interference information (such as "hospital + late at night"). This theoretical value serves as the reference baseline for subsequent calibration and updates.

[0122] The sentiment output value is calibrated and updated based on the sentiment deviation trend to obtain the calibrated and updated sentiment output value.

[0123] This is a crucial link connecting historical patterns with future strategies. The system uses the aforementioned deviation trends to correct the determined theoretical emotional output value. If the deviation trend shows that the robot's output consistently exceeds the user's comfort zone in a certain scenario, the emotional output value for that scenario will be reduced accordingly during calibration updates; conversely, the same applies. This update is not a simple numerical addition or subtraction, but a weighted adjustment combining the slope of the trend and confidence level. For example, if the deviation trend shows a rapidly expanding trend, it indicates that the current sense of maladjustment is intensifying, and the calibration update should be appropriately amplified for a rapid response; if the deviation trend tends to stabilize but remains persistent, a small, gradual adjustment is used to avoid overcorrection. In this way, the calibrated and updated emotional output value both incorporates historical lessons and maintains the system's stability.

[0124] The emotion definition criteria are updated based on the calibrated and updated emotion output values ​​to obtain the updated emotion definition criteria.

[0125] Unlike the aforementioned approaches that only adjust the ancillary emotional information of a single interaction, this application directly modifies the underlying rules guiding all future similar interactions—the emotional definition standard. Specifically, the system will deduce and adjust the relevant parameters in the emotional definition standard based on the calibrated and updated emotional output value. For example, if the calibrated output value indicates that a lower pleasure benchmark is needed in a "rainy day" scenario, the system will automatically update the pleasure benchmark parameters for the "rainy day" scenario in the emotional definition standard, or increase the negative weight of the "humidity" factor on the emotional output. The updated emotional definition standard will be persistently stored and will replace the old standard in subsequent interactions. This means that the robot can autonomously adapt to environmental changes and the emotional preferences of user groups during daily operation without manual reprogramming or full retraining.

[0126] It should be understood that, in order to prevent overfitting or individual anomalous data from contaminating the standard, version management, canary releases, or multi-level review mechanisms can be introduced in actual implementation to ensure that standard updates are robust and traceable. Through the closed-loop processing of the above-mentioned technical content, the embodiments of this application have successfully constructed an emotional interaction system with self-evolution capabilities, effectively solving the industry problem that static rules cannot cope with complex dynamic environments.

[0127] This embodiment further refines the specific implementation process of emotion anomaly analysis, and deconstructs the abstract deviation analysis into a computable multidimensional data modeling and decoupling algorithm.

[0128] Specifically, the process of analyzing emotional anomalies by combining emotional calibration data and interaction scenario information along a timeline to obtain the emotional deviation trend of interference information in each scenario at different time points includes: determining the first emotional data of the first interaction anomaly and the second emotional data of non-first interaction anomalies in different interaction scenarios based on emotional calibration data and interaction scenario information.

[0129] This hierarchical data processing mechanism is the prerequisite for achieving precise deviation decoupling in this embodiment. The strict distinction between "first-time" and "non-first-time" interaction anomalies stems from the fundamentally different underlying mechanisms. First-time interaction anomalies typically occur when the robot enters a new environment or initially encounters a new user. Their data often contains significant random noise or a lack of "cold start" cognition in the new scenario, reflecting the system's initial adaptation error to specific scenario parameters. Non-first-time interaction anomalies, on the other hand, are deviations that persist even after multiple interactions. This type of data, after eliminating incidental factors, points more to long-term drift in user emotional preferences, the cumulative effect of environmental factors, or structural defects in the underlying emotional model. By separating the first and second emotional data, the system can avoid using incidental cold start noise to correct a stable long-term model, or using long-term preference trends to mask sudden environmental disturbances, thus providing a purer data foundation for subsequent analysis.

[0130] Based on this, primary genuine emotion values ​​and primary output emotion values ​​are determined according to the first emotion data in different interaction scenarios. The primary genuine emotion value is the interaction abnormality emotion value fed back by the robot interaction object after interacting with the robot, and the primary output emotion value is the interaction emotion value provided by the robot to the robot interaction object. Secondary genuine emotion values ​​and secondary output emotion values ​​are determined according to the second emotion data in different interaction scenarios. The secondary genuine emotion value is the interaction abnormality emotion value fed back by the robot interaction object after interacting with the robot, and the secondary output emotion value is the interaction emotion value provided by the robot to the robot interaction object.

[0131] Here, "primary" and "secondary" do not refer to their relative importance, but rather correspond to the aforementioned first and subsequent data levels. The actual sentiment value originates from the quantification result after multimodal perception fusion, while the output sentiment value is the record of the synthetic parameters actually executed by the robot at that time. These two sets of values ​​constitute the raw material for bias analysis.

[0132] Subsequently, the primary true emotion value and primary output emotion value, and the secondary true emotion value and secondary output emotion value were processed according to the interaction scenario to obtain the first emotion information corresponding to the primary true emotion value and the second emotion information corresponding to the secondary true emotion value. The first emotion information and the second emotion information were integrated and calculated according to the interaction scenario to obtain the scene interference information under different interaction scenarios and the emotion deviation data corresponding to the scene interference information.

[0133] To extract continuous patterns from these discrete data points, this embodiment introduces three-dimensional spatial modeling technology. Specifically, a first emotional coordinate system is constructed, with the X-axis representing different time points, the Y-axis representing emotional data in different units, and the Z-axis representing parameters of different interaction scenarios.

[0134] In this coordinate system, the X-axis represents the time dimension, which can be an absolute timestamp or a relative conversation duration; the Y-axis represents the quantitative intensity of emotional data. Since emotions are multidimensional (e.g., the PAD model), the "emotional data in different units" here can be, in practice, a comprehensive emotional intensity value after dimensionality reduction through principal component analysis (PCA), a normalized value of a key dimension (e.g., pleasure level), or a vector magnitude composed of multiple emotional dimensions; the Z-axis represents the parametric characteristics of the interaction scene, such as continuous variables like ambient noise decibels, illumination lumens, temperature, and the number of people present, or encoded scene type labels. It should be understood that although this embodiment uses a three-dimensional Cartesian coordinate system as an example, in other embodiments, polar coordinates, spherical coordinates, or high-dimensional manifold spaces can also be used, as long as they can simultaneously accommodate the three core variables of time, emotion, and scene.

[0135] Next, the emotional data from the first and second emotional information are imported into the first emotional coordinate system according to the interaction scenario and timeline, resulting in a first three-dimensional curve corresponding to the first emotional information and a second three-dimensional curve corresponding to the second emotional information. This process essentially fits discrete interaction samples into a continuous spatiotemporal trajectory. In specific implementations, algorithms such as B-spline interpolation, Bézier curve fitting, or Gaussian process regression can be used to transform discrete points into smooth three-dimensional curves. The first three-dimensional curve reflects the dynamic game between the user's actual feelings and the robot's output during the adaptation period to the new scenario; while the second three-dimensional curve depicts the inherent, structural deviation between the two during the stable interaction period.

[0136] Then, the first non-overlapping abnormal curves in the first three-dimensional curves and the second non-overlapping abnormal curves in the second three-dimensional curves are screened respectively.

[0137] The "non-overlap" criterion here is the key to identifying valid deviations. It does not refer to complete geometric non-overlap, but rather that the distance between the two curves in a specific interval exceeds a preset significance threshold. Specifically, the system can calculate the Euclidean distance or dynamic time warping (DTW) distance between the first 3D curve and the benchmark ideal curve (or between the second 3D curve and the first 3D curve) at various time points. When the average distance of a certain curve interval is greater than the preset distance threshold, or when the probability of the curve deviating from the confidence interval within that interval exceeds a preset significance level (e.g., p < 0.05), that curve segment is marked as an anomalous curve. This screening criterion based on quantified thresholds effectively eliminates false anomalies caused by sensor jitter or minor random fluctuations, ensuring the reliability of subsequent analysis.

[0138] Scene acquisition and analysis are performed on the first and second abnormal curves to determine the first scene interference information corresponding to the first abnormal curve and the second scene interference information corresponding to the second abnormal curve. The first emotional deviation data corresponding to the first abnormal curve and the second emotional deviation data corresponding to the second abnormal curve are calculated. The first emotional deviation data and the first scene interference information, and the second emotional deviation data and the second scene interference information are sorted out respectively to obtain scene interference information under different interaction scenarios and the emotional deviation data corresponding to the scene interference information.

[0139] Through the aforementioned 3D modeling and curve filtering, the system successfully decouples the mixed emotional biases into "scene-driven" biases. For example, if it is found that whenever the Z-axis value (such as noise) exceeds 60dB, the second 3D curve will show a significant upward shift (i.e., non-coincidence anomaly), and this shift is linearly correlated with the noise value, then "high noise" can be identified as key scene interference information, and the specific bias function relationship can be quantified. Compared with traditional two-dimensional planar analysis, the introduction of Z-axis scene parameters allows the system to intuitively locate the environmental root causes of emotional mismatch while maintaining temporal continuity. This provides a precise, physically meaningful correction basis for updating the emotional definition criteria in Example 3, rather than merely a statistically significant numerical adjustment.

[0140] This embodiment further refines the prediction mechanism for future emotional deviation trends, transforming static deviation analysis into dynamic trend prediction.

[0141] Specifically, the process of organizing and analyzing scene interference information and emotional deviation data according to the timeline to obtain the emotional deviation trend of each scene interference information at future time nodes includes: constructing a second emotional coordinate system, where the X-axis of the second emotional coordinate system represents different time nodes, the Y-axis of the second emotional coordinate system represents emotional deviation data of different units, and the Z-axis of the second emotional coordinate system represents different interaction scene parameters.

[0142] It is important to note that the second emotional coordinate system differs fundamentally from the first emotional coordinate system in terms of functional positioning and data dimensions. The Y-axis of the first emotional coordinate system carries raw emotional data (such as the absolute value of pleasure), and its core purpose is to "identify" the intervals where anomalies occur through curve overlap comparison. In contrast, the Y-axis of the second emotional coordinate system carries decoupled and quantified emotional deviation data (i.e., the difference or distance between the actual emotional value and the output emotional value), and its core purpose is to "analyze" the evolutionary patterns of the anomalies themselves. This dimensional shift signifies a leap in the system's processing logic from "state monitoring" to "trend insight." In this coordinate system, the X-axis still represents the time dimension, but the time granularity can be adjusted according to prediction needs. For example, minute-level sampling can be used for short-term predictions, while hourly or day-level sampling can be used for long-term adaptive learning. The Z-axis continues to represent interaction scenario parameters, ensuring that deviation analysis is always anchored within a specific environmental context, avoiding the vagueness of discussing trends detached from the scenario. It should be understood that although this embodiment uses the description of a three-dimensional rectangular coordinate system, in actual data processing, a higher-dimensional tensor space or manifold structure can also be used depending on the complexity of the scene parameters, as long as the mapping relationship between time, deviation and scene can be maintained.

[0143] Subsequently, the scene parameters and emotional deviation data from the scene interference information are imported into the second emotional coordinate system according to the timeline to obtain the third three-dimensional curve.

[0144] This third-dimensional curve essentially depicts the trajectory of how the intensity of the influence of interference factors in a specific scenario on the robot's emotional expression deviation changes over time. For example, if the Z-axis represents the ambient noise in decibels and the Y-axis represents the negative deviation value of the user's perception of the robot's emotional expression, then the third-dimensional curve may exhibit a hysteretic loop pattern where the deviation gradually accumulates as the noise increases and slowly diminishes as the noise decreases. Compared to discrete deviation data points, the continuous three-dimensional curve retains the inertial characteristics and directional information of the deviation changes, providing a complete topological basis for subsequent dynamic analysis.

[0145] Furthermore, the dynamic slope of the third three-dimensional curve is calculated in different time periods, and the periodic change analysis of the dynamic slope is performed to obtain the slope variation period.

[0146] The dynamic slope of the curve here is not simply the slope of a geometric tangent, but a physical quantity characterizing the rate of change in emotional bias, reflecting the system's sensitivity to specific scenario interference or the dynamic threshold of the user's emotional tolerance. In practice, the instantaneous slope sequence can be obtained by calculating the first derivative or performing discrete difference operations on the third-dimensional curve. More importantly, the periodic change analysis of the slope sequence aims to uncover the biological rhythms or environmental patterns hidden behind seemingly random fluctuations. For example, in a family companionship scenario, a user's negative emotional bias may show a regular peak in the evening (corresponding to a fatigue period), or a periodic increase in anxiety bias may occur on Sunday evenings; or in a factory inspection scenario, the interaction bias caused by machine noise may be highly synchronized with the shift cycle of equipment operation. To accurately extract these periodic features, signal processing algorithms such as Fast Fourier Transform (FFT), wavelet transform, autocorrelation function analysis, or Lomb-Scargle periodograms can be used. Through these algorithms, the system can separate significant fundamental frequency components and their harmonics from noisy slope data, thereby determining one or more statistically significant slope variation periods. This period extraction method based on mathematical transformation has stronger environmental adaptability and scientific basis compared with manually set fixed time windows, effectively proving that the present invention is not a simple time statistics, but a modeling of the deep dynamic characteristics of human-computer interaction.

[0147] Finally, based on the slope variation cycle, the trend of the third three-dimensional curve is predicted to obtain the emotional bias trend of interference information in each scene at future time nodes.

[0148] After obtaining a reliable slope variation cycle, the system can use extrapolation, time series models (such as ARIMA and Prophet), or sequence prediction algorithms based on recurrent neural networks, combined with the current phase state, to infer the possible trend of emotional deviation over a future period. This prediction result directly empowers the robot's proactive emotional strategy. For example, if the prediction shows that due to the superposition of the dual scenes of "rainy day + evening" in the next 30 minutes, the negative deviation of the user's high-arousal emotion will continue to expand at a rate of 0.5 units per hour, the system can initiate defensive adjustments in advance at the current moment: not only adjusting the current emotional output to a mild mode of low arousal, but also proactively planning the rhythm of subsequent interactions, such as reducing the frequency of proactively initiating dialogue, increasing the proportion of soothing non-verbal actions (such as soft lighting breathing effects), and even completing a smooth transition of emotional tone before the user shows obvious dissatisfaction. This proactive adaptation mechanism of "prevention is better than cure" completely changes the passive response mode of traditional robots of "error-calibration-error again", significantly improving the user's emotional comfort and trust in long-term interactions.

[0149] The above embodiments describe a language-based robot interaction method from the perspective of process flow. The following embodiments describe a language-based robot interaction system from the perspective of virtual modules or virtual units. For details, please refer to the following embodiments.

[0150] This application provides a language-based robot interaction system 20, such as... Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the structure of a language-based robot interaction system provided in an embodiment of this application. Specifically, the system 20 may include:

[0151] The first acquisition module 21 is used to acquire language information of the robot's interactive objects, as well as interactive scene information and emotional interaction feedback within a preset time period.

[0152] The second acquisition module 22 is used to acquire the language response information and associated emotional information that the robot will reply based on the language information.

[0153] The response determination module 23 is used to determine the causal response information corresponding to the language information based on the preset causal association memory map library;

[0154] The calibration analysis module 24 is used to perform comprehensive calibration analysis on causal response information, language response information, emotional interaction feedback and auxiliary emotional information, and to calibrate and adjust the auxiliary emotional information based on the analysis results to obtain the adjusted auxiliary emotional information.

[0155] The interaction control module 25 is used to generate language interaction instructions based on the adjusted auxiliary emotional information and language response information, and to control the robot to perform language interaction with the robot interaction object.

[0156] In one possible implementation of this application embodiment, when the calibration analysis module 24 performs comprehensive calibration analysis on causal response information, linguistic response information, emotional interaction feedback, and auxiliary emotional information, and calibrates and adjusts the auxiliary emotional information based on the analysis results to obtain adjusted auxiliary emotional information, it is specifically used for:

[0157] Determine whether the semantics of the causal response information are consistent with the semantics of the linguistic response information. If they are consistent, determine the real emotional information corresponding to the robot's interactive object based on the interaction scenario information and emotional interaction feedback. Input the auxiliary emotional information and the real emotional information into the emotional expression standard for emotional value detection to obtain the first emotional value corresponding to the auxiliary emotional information and the second emotional value corresponding to the real emotional information.

[0158] Determine whether the first emotional value and the second emotional value are consistent. If they are inconsistent, determine the emotional difference between the first emotional value and the second emotional value, and adjust the emotional intensity of the first emotional value based on the emotional difference to obtain the adjusted auxiliary emotional information.

[0159] In another possible implementation of this application embodiment, when determining whether the semantics of the causal response information is consistent with the semantics of the linguistic response information, the calibration analysis module 24 is specifically used for:

[0160] If the semantics of the causal response information are inconsistent with the semantics of the linguistic response information, the linguistic response information is updated based on the causal response information to obtain the updated linguistic response information.

[0161] The system detects whether there are any anomalies in the semantics of causal response information and linguistic response information within a historical time period. If such anomalies exist, the system determines the number of anomalies and checks whether the number of anomalies is greater than or equal to the preset maximum number of anomalies. If it is greater, the system generates maintenance information and sends the maintenance information to a preset terminal for display output.

[0162] In another possible implementation of this application embodiment, system 20 further includes: an information acquisition module, an anomaly analysis module, an output determination module, an output update module, and a standard update module, wherein...

[0163] The information acquisition module is used to acquire emotional calibration data where the first emotional value and the second emotional value are inconsistent during a historical time period, as well as the interaction scenario information corresponding to the emotional calibration data;

[0164] The anomaly analysis module is used to perform emotional anomaly analysis on emotional calibration data and interaction scenario information according to the timeline, and to obtain the emotional deviation trend of interference information for each scenario at different time points.

[0165] The output determination module is used to obtain the emotion definition standard of the robot under the current interactive scenario information, and determine the emotion output value of the robot under each scenario interference information according to the emotion definition standard.

[0166] The output update module is used to calibrate and update the sentiment output value based on the sentiment deviation trend, so as to obtain the calibrated and updated sentiment output value.

[0167] The standard update module is used to update the emotion definition standard based on the calibrated and updated emotion output value, so as to obtain the updated emotion definition standard.

[0168] In another possible implementation of this application embodiment, when the anomaly analysis module performs emotional anomaly analysis on the emotional calibration data and interaction scene information according to the timeline to obtain the emotional deviation trend of interference information for each scene at different time points, it is specifically used for:

[0169] Based on sentiment calibration data and interaction scenario information, determine the first sentiment data of the first interaction abnormality and the second sentiment data of the non-first interaction abnormality in different interaction scenarios;

[0170] Based on the first sentiment data, the primary real sentiment value and the primary output sentiment value are determined in different interaction scenarios. The primary real sentiment value is the abnormal sentiment value of the interaction object after interacting with the robot. The primary output sentiment value is the interaction sentiment value provided by the robot to the robot interaction object.

[0171] The secondary true sentiment value and secondary output sentiment value are determined based on the second sentiment data in different interaction scenarios. The secondary true sentiment value is the interaction abnormal sentiment value fed back by the robot interaction object after interacting with the robot. The secondary true sentiment value is the interaction sentiment value provided by the robot to the robot interaction object.

[0172] The primary real emotion value and primary output emotion value, and the secondary real emotion value and secondary output emotion value are respectively organized according to the interaction scenario to obtain the first emotion information corresponding to the primary real emotion value and the second emotion information corresponding to the secondary real emotion value.

[0173] The first and second emotional information are integrated and analyzed according to the interaction scenario to obtain scene interference information under different interaction scenarios and the emotional deviation data corresponding to the scene interference information.

[0174] By organizing and analyzing scene interference information and emotional bias data according to the timeline, the emotional bias trend of each scene interference information at future time nodes can be obtained.

[0175] In another possible implementation of this application embodiment, when the anomaly analysis module integrates and analyzes the first emotional information and the second emotional information according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information, it is specifically used for:

[0176] Construct a first emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0177] The emotional data in the first emotional information and the second emotional information are imported into the first emotional coordinate system according to the interaction scenario and timeline to obtain the first three-dimensional curve corresponding to the first emotional information and the second three-dimensional curve corresponding to the second emotional information.

[0178] Filter out the first non-overlapping anomalous curves in the first three-dimensional curve and the second non-overlapping anomalous curves in the second three-dimensional curve respectively;

[0179] Scene acquisition and analysis are performed on the first abnormal curve and the second abnormal curve to determine the first scene interference information corresponding to the first abnormal curve and the second scene interference information corresponding to the second abnormal curve.

[0180] Calculate the first sentiment deviation data corresponding to the first abnormal curve and the second sentiment deviation data corresponding to the second abnormal curve;

[0181] The first emotional bias data and the first scene interference information, and the second emotional bias data and the second scene interference information were sorted out to obtain scene interference information under different interaction scenarios and emotional bias data corresponding to the scene interference information.

[0182] Another possible implementation of this application embodiment is that, when the anomaly analysis module organizes and analyzes scene interference information and emotional deviation data according to a timeline to obtain the emotional deviation trend of each scene interference information at future time nodes, it is specifically used for:

[0183] Construct a second emotional coordinate system. The X-axis of the second emotional coordinate system represents different time points, the Y-axis represents emotional deviation data in different units, and the Z-axis represents parameters for different interaction scenarios.

[0184] The scene parameters and emotional deviation data from the scene interference information are imported into the second emotional coordinate system according to the timeline to obtain the third three-dimensional curve;

[0185] Calculate the dynamic slope of the third three-dimensional curve in different time periods, and perform periodic change analysis on the dynamic slope of the curve to obtain the slope variation period.

[0186] Based on the slope variation cycle, the trend of the third three-dimensional curve is predicted to obtain the emotional bias trend of interference information in each scene at future time nodes.

[0187] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the language-based robot interaction system 20 described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0188] This application provides an electronic device, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 3 The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may also include a transceiver 304. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of this electronic device 300 does not constitute a limitation on the embodiments of this application.

[0189] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0190] Bus 302 may include a pathway for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0191] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0192] The memory 303 is used to store application code that executes the solution of this application, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0193] Electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Servers can also be included. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0194] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0195] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0196] The above are only some embodiments of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A robot interaction method based on language interaction, characterized in that, include: Collect language information of the robot's interaction objects, as well as interaction scenario information and emotional interaction feedback within a preset time period; The data collection robot will provide the language response and accompanying emotional information based on the language information. Based on a pre-set causal association memory graph database, determine the causal response information corresponding to the linguistic information; A comprehensive calibration analysis is performed on the causal response information, the linguistic response information, the emotional interaction feedback, and the ancillary emotional information. Based on the analysis results, the ancillary emotional information is calibrated and adjusted to obtain the adjusted ancillary emotional information. Based on the adjusted auxiliary emotional information and the language response information, language interaction instructions are generated to control the robot to perform language interaction with the robot's interaction object.

2. The robot interaction method based on language interaction according to claim 1, characterized in that, The process involves a comprehensive calibration analysis of the causal response information, the linguistic response information, the emotional interaction feedback, and the ancillary emotional information. Based on the analysis results, the ancillary emotional information is calibrated and adjusted to obtain the adjusted ancillary emotional information, including: Determine whether the semantics of the causal response information are consistent with the semantics of the linguistic response information. If they are consistent, determine the real emotional information corresponding to the robot interaction object based on the interaction scenario information and the emotional interaction feedback. Input the auxiliary emotional information and the real emotional information into the emotional expression standard for emotional value detection to obtain a first emotional value corresponding to the auxiliary emotional information and a second emotional value corresponding to the real emotional information. Determine whether the first emotional value and the second emotional value are consistent. If they are inconsistent, determine the emotional difference between the first emotional value and the second emotional value, and adjust the emotional intensity of the first emotional value based on the emotional difference to obtain the adjusted auxiliary emotional information.

3. The robot interaction method based on language interaction according to claim 2, characterized in that, The step of determining whether the semantics of the causal response information are consistent with the semantics of the linguistic response information includes: If the semantics of the causal response information are inconsistent with the semantics of the language response information, then the language response information is updated based on the causal response information to obtain the updated language response information; The system detects whether there are any abnormal situations in the historical time period where the semantics of the causal response information are inconsistent with the semantics of the language response information. If such situations exist, the number of abnormal situations is determined, and it is judged whether the number of abnormal situations is greater than or equal to a preset maximum number of abnormal situations. If it is greater, maintenance information is generated and sent to a preset terminal for display output.

4. The robot interaction method based on language interaction according to claim 1, characterized in that, The step of generating language interaction commands based on the adjusted auxiliary emotional information and the language response information to control the robot to perform language interaction with the robot interaction object further includes: Acquire emotional calibration data where the first emotional value and the second emotional value are inconsistent during a historical time period, as well as the interaction scenario information corresponding to the emotional calibration data; The emotional calibration data and the interaction scene information are analyzed for emotional anomalies according to the timeline to obtain the emotional deviation trend of each scene interference information at different time points. Obtain the current emotion definition standard of the robot under the interactive scenario information, and determine the emotion output value of the robot under each scenario interference information according to the emotion definition standard; The emotional output value is calibrated and updated based on the emotional deviation trend to obtain the calibrated and updated emotional output value. The emotion definition criteria are updated based on the calibrated and updated emotion output values ​​to obtain the updated emotion definition criteria.

5. A robot interaction method based on language interaction according to claim 4, characterized in that, The step of performing emotional anomaly analysis on the emotional calibration data and the interaction scene information according to a timeline to obtain the emotional deviation trend of interference information for each scene at different time points includes: Based on the emotion calibration data and the interaction scenario information, determine the first emotion data of the first interaction abnormality and the second emotion data of the non-first interaction abnormality under different interaction scenarios; Based on the first emotion data, the primary real emotion value and the primary output emotion value are determined in different interaction scenarios. The primary real emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The primary output emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object. The secondary true emotion value and secondary output emotion value are determined based on the second emotion data in different interaction scenarios. The secondary true emotion value is the interaction abnormal emotion value fed back by the robot interaction object after interacting with the interaction robot. The secondary true emotion value is the interaction emotion value provided by the interaction robot to the robot interaction object. The primary real emotion value and the primary output emotion value, and the secondary real emotion value and the secondary output emotion value are respectively processed according to the interaction scenario to obtain the first emotion information corresponding to the primary real emotion value and the second emotion information corresponding to the secondary real emotion value. The first emotional information and the second emotional information are integrated, calculated and analyzed according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information. The scene interference information and the emotional deviation data are organized and analyzed according to the timeline to obtain the emotional deviation trend of each scene interference information at future time nodes.

6. The robot interaction method based on language interaction according to claim 5, characterized in that, The step of integrating and analyzing the first emotional information and the second emotional information according to the interaction scenario to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information includes: Construct a first emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional data in different units, and the Z-axis represents parameters for different interaction scenarios. The emotional data in the first emotional information and the second emotional information are imported into the first emotional coordinate system according to the interaction scenario and timeline to obtain a first three-dimensional curve corresponding to the first emotional information and a second three-dimensional curve corresponding to the second emotional information. Filter out the first non-overlapping abnormal curves in the first three-dimensional curve and the second non-overlapping abnormal curves in the second three-dimensional curve respectively; Scene acquisition and analysis are performed on the first abnormal curve and the second abnormal curve to determine the first scene interference information corresponding to the first abnormal curve and the second scene interference information corresponding to the second abnormal curve. Calculate the first sentiment deviation data corresponding to the first abnormal curve and the second sentiment deviation data corresponding to the second abnormal curve; The first emotional deviation data and the first scene interference information, and the second emotional deviation data and the second scene interference information are respectively processed to obtain scene interference information under different interaction scenarios and emotional deviation data corresponding to the scene interference information.

7. A robot interaction method based on language interaction according to claim 5, characterized in that, The process of organizing and analyzing the scene interference information and the emotional deviation data according to a timeline to obtain the emotional deviation trend of each scene interference information at future time points includes: Construct a second emotional coordinate system, where the X-axis represents different time points, the Y-axis represents emotional deviation data in different units, and the Z-axis represents parameters for different interaction scenarios. The scene parameters in the scene interference information and the emotional deviation data are imported into the second emotional coordinate system according to the timeline to obtain the third three-dimensional curve; Calculate the dynamic slope of the third three-dimensional curve in different time periods, and perform periodic change analysis on the dynamic slope of the curve to obtain the slope variation period; Based on the slope variation period, the trend of the third three-dimensional curve is predicted to obtain the emotional bias trend of interference information for each scene at future time nodes.

8. A robot interaction system based on language interaction, characterized in that, include: The first acquisition module is used to collect language information of the robot's interactive objects, as well as interactive scene information and emotional interaction feedback within a preset time period. The second acquisition module is used to acquire the language response information and associated emotional information that the robot will reply to based on the language information. The response determination module is used to determine the causal response information corresponding to the language information based on a preset causal association memory graph library. The calibration analysis module is used to perform comprehensive calibration analysis on the causal response information, the language response information, the emotional interaction feedback, and the auxiliary emotional information, and to calibrate and adjust the auxiliary emotional information according to the analysis results to obtain the adjusted auxiliary emotional information. The interaction control module is used to generate language interaction instructions based on the adjusted auxiliary emotional information and the language response information, and to control the robot to perform language interaction with the robot interaction object.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as any one of the language-based robot interaction methods as claimed in claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as any one of the language-based robot interaction methods as described in claims 1 to 7.

Images