Auxiliary reply method and system based on psychological strategy

Abstract

According to the auxiliary reply method and system based on the psychological strategy, replies are not directly generated, the most suitable psychological communication strategy is dynamically selected through real-time evaluation of the state of the user, and reply suggestions capable of effectively relieving the negative emotion of the user and improving the psychological comfort degree are generated based on the strategy. The invention aims to solve the problem of negative psychological states such as anxiety, pressure and hesitation which are generally generated due to the fact that a user does not determine how to properly reply in modern digital communication (especially instant messaging and e-mail). The prior art mainly focuses on improvement of communication efficiency and task completion degree, but neglects subjective psychological feelings and emotional demands of a user in a communication process. The invention aims to provide a brand new technical solution from the perspective of relieving the negative psychological state of the user and improving the communication happiness.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to an auxiliary response method and system based on psychological strategies. Background Technology

[0002] With the development of artificial intelligence technology, a collaborative technical system has been established in related fields, encompassing multimodal data collection, behavioral pattern analysis, and natural language generation. Specifically, this technological ecosystem includes key components such as semantic parsing based on chat logs, user behavior feature extraction, and pre-defined template matching. The integration of reinforcement learning algorithms and knowledge graph technology has enabled a systematic evolution from quick input method replies to professional, scenario-based templates. However, existing technical methods directly employ efficiency-oriented generation models without establishing a dynamic mapping relationship between user psychological states and reply strategies. This may lead to a disconnect between reply suggestions and users' actual psychological needs, or cause cognitive overload in multi-option scenarios, thereby affecting the improvement of communication satisfaction.

[0003] Existing technologies mainly focus on the following directions: 1. Improving the relevance of reply content: Generating more relevant auxiliary prompts by analyzing chat element information (such as language style, emotion, and topic), with the core being to improve the level of intelligence and content matching. 2. Providing social simulation and practice: Helping users practice social interaction in a virtual environment by constructing AI virtual avatars and dialogue simulations to improve their social skills in real life, with the core being "training" rather than "real-time assistance". 3. Improving reply operation efficiency: Providing auxiliary reply items directly based on conversation messages at the input method level, with the core being to simplify operations and improve efficiency. 4. Template-based professional replies: For email scenarios, generating professional reply content by matching email type and emotional templates, with the core being to improve the accuracy and professionalism of replies.

[0004] Existing technological solutions all share a common and fundamental drawback: a singular goal orientation that ignores human feelings. All existing technologies take "generating a high-quality response" as the ultimate objective, with evaluation systems centered around task indicators such as efficiency, accuracy, and relevance. They treat users as efficiency-driven "task executors," completely ignoring their intrinsic needs as "emotional experiencers." Therefore, they fail to address the most fundamental psychological issues users face in communication, such as anxiety, fear, and lack of confidence. Sometimes, they may even exacerbate users' decision-making difficulties and anxiety because the provided options "all seem good." Summary of the Invention

[0005] The present invention aims to at least partially solve one of the technical problems in the related art.

[0006] Therefore, the first objective of this invention is to propose an auxiliary response method based on psychological strategies.

[0007] The second objective of this invention is to provide an auxiliary response device based on psychological strategies.

[0008] To achieve the above objectives, a first aspect of the present invention proposes an auxiliary response method based on psychological strategies, comprising: S1, collect multimodal data of users in digital communication scenarios, including dialogue context data, real-time user behavior data, user profile data and application scenario data; S2, input the real-time user behavior data into the pre-trained psychological state assessment model, and output the quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure and self-confidence level. S3, based on the integration result of the psychological state vector and the multimodal data, the core guidance strategy that is compatible with the user's current psychological state is selected by matching the strategy cards in the configurable psychological theory knowledge base through the psychological strategy decision network; S4, invoke a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidance of the core guidance strategy and combined with the user profile data to achieve personalized expression. S5, present the suggested response and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principles of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

[0009] In one embodiment of the present invention, S2 includes: S21 calculates anxiety levels using a weighted average algorithm, with the weighting coefficients dynamically adjusted based on users' historical behavior data. S22 employs multimodal data fusion technology to fuse typing speed, backspace rate, and speech intonation features at the feature level.

[0010] In one embodiment of the present invention, S3 includes: S31, when a keyword with ambiguous boundaries is detected in the dialogue context, the activation rule of the topic separation strategy is triggered. The specific judgment condition is as follows:

[0011] in The number of times the boundary keyword appears. The preset trigger threshold; S32 employs a dynamic weight adjustment mechanism to adjust the matching score of strategy cards based on the communication style type in the user profile data.

[0012] In one embodiment of the present invention, S4 further includes: S41, the generation guidance constraint includes the four-element structure of the Nonviolent Communication (NVC) framework, specifically requiring that the response content must include four parts: observation, feeling, need, and request; S42, using language style parameters from user profile data To personalize the expression, style constraints are added to the LLM Prompt during generation.

[0013] In one embodiment of the present invention, S5 includes: S51 generates explanatory feedback by translating a pre-set strategy principle template, which includes a popular description of communication skills and a quantitative prediction of expected results. S52 uses a hierarchical structure to present explanatory feedback. The first layer shows the core strategy name, the second layer shows the strategy application scenarios, and the third layer shows the specific implementation steps.

[0014] To achieve the above objectives, a second aspect of the present invention provides an auxiliary response device based on psychological strategies, comprising: The multimodal data acquisition module is used to collect multimodal data of users in digital communication scenarios. The multimodal data includes dialogue context data, real-time user behavior data, user profile data, and application scenario data. The psychological state assessment module is used to input the user's real-time behavior data into a pre-trained psychological state assessment model and output a quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure, and self-confidence level. The psychology strategy matching module is used to select core guidance strategies that are suitable for the user's current psychological state by matching strategy cards in the configurable psychology theory knowledge base through the psychology strategy decision network, based on the integration results of the psychological state vector and the multimodal data. The response generation module is used to call a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidelines of the core guidance strategy and combined with the user profile data to achieve personalized expression. The feedback presentation module is used to present the suggested response and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principle of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

[0015] The beneficial effects that this technical solution can bring are: Significantly reduces user communication anxiety: Unlike traditional auxiliary tools, this invention takes "alleviating users' negative psychological state" as its primary goal. By introducing psychological strategies, it provides users with emotional support and psychological "scaffolding," directly and effectively reducing their anxiety, stress, and uncertainty in communication.

[0016] Enhancing the intrinsic quality of communication: By applying established psychological theories such as "nonviolent communication," the system-generated responses can not only soothe users but also more effectively promote mutual understanding and positive interaction with the communication partner, thereby improving the quality of the communication relationship.

[0017] Empowering and enhancing users' capabilities: Through the innovative design of "explanatory feedback," this invention not only provides immediate assistance but also acts as a "communication coach." Long-term use helps users learn and internalize healthy communication patterns, fundamentally improving their communication confidence and skills.

[0018] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an auxiliary response method based on psychological strategies according to an embodiment of the present invention; Figure 2 This is an architecture diagram of an auxiliary response method based on psychological strategies according to an embodiment of the present invention; Figure 3 This is a structural diagram of an auxiliary response device based on psychological strategies according to an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] The following description, with reference to the accompanying drawings, describes an auxiliary response method and system based on psychological strategies according to an embodiment of the present invention.

[0023] Example 1 Figure 1 This is a flowchart of an auxiliary response method based on psychological strategies according to an embodiment of the present invention, such as... Figure 1 As shown, it includes: S1, collect multimodal data of users in digital communication scenarios, including dialogue context data, real-time user behavior data, user profile data and application scenario data.

[0024] In some implementations, the system monitors the user's input behavior in instant messaging or email clients and detects the time the user spends on the reply screen. and input-deletion frequency .when (in This is a preset threshold, typically set between 10 and 30 seconds, depending on the urgency of the application scenario. ( When the maximum allowed hesitation frequency (e.g., more than 5 input-delete operations per minute) is reached, the system triggers the data collection process. The collection module will simultaneously acquire four types of data: dialogue context data (such as chat logs and original email text), real-time user behavior data (including typing speed), and more. pause frequency Rejection rate User profile data (such as communication style tags, such as "highly ingratiating" or "avoidant", obtained through historical behavior modeling) and application scenario data (such as identifying whether the current email is a work email, private chat, or customer communication, usually based on a natural language classification model).

[0025] Specifically, the system typically collects behavioral data every 2 seconds to ensure real-time performance and data continuity. Typing speed Backspace rate (in characters per second) Defined as the ratio of the number of backspaces to the total number of inputs, i.e. ,in The number of backspaces. This represents the total number of inputs. These parameters will serve as input features for subsequent mental state assessment models.

[0026] In practical applications, this step is widely applicable to various digital communication platforms, including but not limited to enterprise email systems, instant messaging software (such as WeChat and Slack), and online customer service systems. The system can be automatically activated when users are under high pressure, experiencing high emotional fluctuations, or facing significant communication uncertainty, providing a data foundation for subsequent psychological strategy selection.

[0027] S2, input the real-time user behavior data into the pre-trained psychological state assessment model, and output the quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure and self-confidence level.

[0028] The system inputs real-time user behavior data into a pre-trained psychological state assessment model to output a quantified psychological state vector. This vector contains three key dimensions: anxiety level, decision-making pressure, and self-confidence level, each represented by a normalized value (0-1 range) to describe the user's psychological state in the current communication context.

[0029] In some implementations, mental state assessment models are based on deep learning architectures, such as multilayer perceptrons (MLPs) or Transformers. Their input layer receives multidimensional behavioral feature vectors, including but not limited to typing speed (characters / second), pause frequency (number of pauses / minute), backspace rate (number of backspaces / total inputs), and cursor movement trajectory (pixels / second). These features are standardized by a feature encoder, typically using Z-score normalization to eliminate scale differences between different user behavior patterns. The model output layer consists of three independent regression units, each corresponding to an anxiety level (…). ), decision-making pressure ( ) and self-confidence level ( Its output value satisfies The higher the value, the more significant the corresponding psychological state.

[0030] Specifically, the evaluation accuracy of a model can be measured using mean squared error (MSE) or mean absolute error (MAE). For example, during the training phase, the model's loss function can be defined as... To ensure consistent predictions across the three psychological dimensions, the model's inference delay should be controlled within a certain range. This is to meet the needs of real-time interaction.

[0031] In practical applications, this step typically runs on mobile or desktop instant messaging or email clients, and occurs when the user spends more than [a certain amount of time] on the reply screen. Or appear The system is activated when the backspace rate is reached. It is deployed locally or in the cloud using a lightweight model, and leverages edge computing technology to achieve low-latency response.

[0032] Furthermore, S2 includes: S21 calculates anxiety levels using a weighted average algorithm, with weighting coefficients dynamically adjusted based on users' historical behavior data. The specific formula is:

[0033] in For the first The weight of each behavioral feature, For the first A normalization function for each behavioral feature.

[0034] In some implementations, psychological strategy decision networks may employ graph neural networks (GNNs) or attention-enhanced multilayer perceptrons (MLPs) to process heterogeneous input data. Input features include, but are not limited to, the user's current anxiety level. Decision-making pressure Confidence level The features include semantic features of the dialogue context, communication style tags in user profiles (such as "highly ingratiating"), and application scenario classifications (such as "work emails" or "personal chats"). These features are encoded into vector representations of a uniform dimension and input into a decision network for policy matching.

[0035] Specifically, anxiety level The range of values ​​is ,in Indicates no anxiety. This indicates extreme anxiety. This indicator is output by a psychological state assessment model, calculated based on a weighted fusion of user behavioral data (such as typing speed, backspace frequency, pause duration, etc.) and historical behavioral patterns. Weighting coefficients... It can be dynamically adjusted based on users' historical behavior data. For example, it can use a sliding window mechanism to calculate users' anxiety sensitivity in different situations, thereby optimizing the accuracy of the current strategy selection.

[0036] In practical applications, this step can be deployed in instant messaging software, email clients, or intelligent assistant systems. When a user exhibits hesitation on the reply screen (e.g., remaining on the screen for more than [time limit]), [this step is implemented]. Seconds, or backspace rate The system will automatically activate the psychological strategy decision-making process. Through real-time assessment and strategy matching, the system can provide guidance on communication strategies that match the user's psychological state even before the user has clearly expressed their intentions.

[0037] S22 employs multimodal data fusion technology to fuse typing speed, backspace rate, and speech intonation features at the feature level. The fusion formula is as follows:

[0038] in For typing speed, For the rate of rejection, The value represents the fluctuation of speech intonation, and the denominator is the maximum value of each feature.

[0039] The system uses a pre-trained psychological state assessment model to quantitatively analyze real-time user behavior data, outputting a vector describing the user's current psychological state. This model is based on multimodal input, including but not limited to behavioral indicators such as typing speed, pause frequency, and backspace rate, as well as speech and intonation features (such as speech rate, pitch, and pause duration), to achieve a high-precision assessment of the user's immediate psychological state.

[0040] In some implementations, mental state assessment models can employ deep neural network architectures, such as Transformer or LSTM, to capture time-series features of user behavior. The model input consists of standardized multimodal data, such as typing speed normalized to characters per second. The backspace rate is defined as the ratio of the number of backspaces to the total number of inputs. ,in Indicates the number of backspaces. This represents the total number of inputs. Speech intonation features include pitch, energy, and speech rate, which can be extracted using speech signal processing techniques such as MFCC or ProsodyNet.

[0041] Furthermore, the model output is a mental state vector. ,in This indicates the number of psychological dimensions, such as anxiety level, decision-making stress, and self-confidence level. The value range for each dimension is typically [value range missing]. This indicates the intensity of the psychological state. For example, when a user frequently backspaces and their typing speed significantly decreases in the reply interface, the model might output... (High anxiety level) (High decision-making pressure) (Low self-confidence level).

[0042] In practical applications, this step can be deployed in a local or cloud module of an instant messaging or email client. It automatically triggers the assessment process when the user exhibits signs of hesitation or anxiety by real-time monitoring of user input behavior and voice input (such as voicemail or voice reply functions). The system must ensure real-time data collection with low latency, typically requiring the processing latency of behavioral data to be no more than [a certain value]. To maintain a smooth user experience.

[0043] S3. Based on the integration result of the psychological state vector and the multimodal data, the core guidance strategy that matches the user's current psychological state is selected by matching strategy cards in the configurable psychological theory knowledge base through the psychological strategy decision network.

[0044] In some implementations, the mental state vector is output by a pre-trained mental state assessment model, which typically includes quantitative indicators across multiple dimensions, such as anxiety level. Decision-making pressure Confidence level The range of values ​​for each dimension is as follows: ,in This indicates that the psychological state has reached its highest intensity. Multimodal data includes dialogue context, user behavior features (such as typing speed and backspace rate), user profiles (such as communication style tags), and application scenario classifications (such as work emails and private chats). After feature encoding, this data is concatenated with the psychological state vector to form a unified input representation.

[0045] The psychological strategy decision-making network employs a multilayer perceptron (MLP) or graph neural network (GNN) structure, with an input layer dimension of [missing information]. ,in This represents the dimension of the mental state vector. The network uses an attention mechanism to weightedly fuse features from various modalities, outputting a policy matching score vector. ,in Indicates the first The strategy card is selected based on its match with the current input. The strategy card with the highest score is then chosen as the core guiding strategy.

[0046] Each strategy card in the configurable psychological theory knowledge base contains standardized activation rules and generation guidelines, such as in... and Under certain conditions, the "cognitive reappraisal strategy" is activated, and specific language structure templates and emotion regulation suggestions are provided. This knowledge base supports dynamic updates and user-defined configurations, meeting the needs of personalized mental health interventions.

[0047] Furthermore, in practical applications, this step can be deployed in the backend services of instant messaging or email clients and invoked in real time via API interfaces. Its technical value lies in transforming psychological theories into computable and configurable strategy modules, realizing a shift from "task-oriented" to "psychological-oriented," thereby embedding therapeutic functions into communication assistance and enhancing users' psychological comfort and communication well-being.

[0048] Furthermore, S3 includes: S31, when a keyword with ambiguous boundaries is detected in the dialogue context, the activation rule of the topic separation strategy is triggered. The specific judgment condition is as follows:

[0049] in The number of times the boundary keyword appears. This is the preset trigger threshold.

[0050] Specifically, in some implementations, when the system detects ambiguous keywords (such as "must" or "should") in the dialogue context, it triggers the activation rule of the topic separation strategy. This step is technically implemented based on a combination of keyword recognition and contextual semantic analysis in Natural Language Processing (NLP). Specifically, the system first scans the current dialogue context using a predefined ambiguous keyword library, which includes, but is not limited to, words with mandatory or obligatory semantics such as "must," "should," "have to," and "always." The keyword recognition module employs rule-based matching algorithms and part-of-speech tagging (POS) technology to ensure accuracy and contextual relevance.

[0051] Furthermore, the system combines contextual semantic analysis to determine whether keywords exhibit blurred psychological boundaries in their expression. For example, if a user uses "I should explain clearly" in a reply, the system uses semantic role labeling (SRL) and intent recognition models to determine whether the expression implies excessive self-responsibility or increased emotional burden. The judgment criteria can be set as: keyword frequency. Furthermore, the contextual semantics contain semantic features related to attribution of responsibility or emotional stress. ,in Indicates the number of times the keyword appears. This represents the semantic stress score, which is output by a pre-trained sentiment analysis model.

[0052] This step is primarily applicable to communication scenarios where users face high emotional loads or a strong sense of responsibility, such as critical emails in the workplace or conflicting conversations in private chats. When the system identifies keywords with ambiguous boundaries, it will automatically activate the issue separation strategy, guiding users to distinguish their own responsibilities from the expectations of others, thereby reducing psychological burden.

[0053] S32 employs a dynamic weight adjustment mechanism to adjust the matching score of strategy cards based on the communication style type in the user profile data. The adjustment formula is:

[0054] in This is a style correction factor. The degree of strategy adaptation to the communication style.

[0055] Specifically, in some implementations, the psychological strategy decision-making network employs a dynamic weight adjustment mechanism. Its core purpose is to adjust the matching score of strategy cards based on the communication style type (e.g., highly ingratiating) in user profile data, thereby improving the personalized adaptability of strategy selection and the effectiveness of psychological intervention. This mechanism dynamically adjusts the weight parameters in the strategy matching process based on a communication style feature vector constructed from long-term user behavior data, combined with the current psychological state assessment results, to ensure that the output communication strategies better meet the user's psychological needs in terms of emotional support and behavioral guidance.

[0056] Specifically, the system first extracts communication style features from user profile data, such as "appeasing," "adversarial," and "avoidant." These features are typically modeled using methods such as sentiment analysis, semantic role labeling (SRL), and behavioral pattern clustering. In this invention, communication style features are quantized into a feature vector. ,in Indicates the first The intensity value of each communication style dimension, with a value range of [value range missing]. For example, for "highly pleasing" users, It may be close to 1, while The result may be lower.

[0057] During the strategy matching process, the system presets a basic matching score for each strategy card. The score is calculated from multiple dimensions, including contextual semantic similarity, strategy applicability, and sentiment consistency. Subsequently, the system introduces a correction coefficient based on communication style characteristics in the user profile. The calculation method is as follows:

[0058] in, The parameter for adjusting sensitivity (usually set to) ), For the threshold of this style dimension (e.g.) (This represents the threshold for identifying highly pleasing users). This function maps continuous style intensity values ​​to policy adjustment weights, thereby enabling dynamic adjustment of the matching score.

[0059] Finally, the revised strategy score for:

[0060] In practical applications, this correction mechanism is particularly suitable when users are communicating in psychological states such as high anxiety and low self-confidence. By reducing the preference weight for "adversarial" strategies and increasing the matching score for "empathic" or "buffering" strategies, it guides users to choose communication methods that better suit their psychological comfort. This mechanism can be flexibly adjusted in different application scenarios such as work emails and private chats. and The value of is determined to adapt to the psychological needs of users in different communication scenarios, significantly improving the system's adaptability and practicality in psychologically assisted communication.

[0061] S4, invoke a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidance of the core guidance strategy and combined with the user profile data to achieve personalized expression.

[0062] The system invokes a Large Language Model (LLM) to generate specific response suggestions. This process is not simply text generation based on context, but is strictly constrained by selected "core guidance strategies" and their "generation guidelines," and incorporates user profile data to achieve personalized expression. In some implementations, the LLM's input prompt consists of three parts: dialogue context, psychological strategy guidance, and user profile features, thereby ensuring that the generated response is not only semantically coherent but also highly matched in terms of emotional expression and communication style with the user's current psychological state and long-term behavioral patterns.

[0063] In some implementations, upon receiving a Prompt, the LLM first performs contextual understanding to identify the topic, tone, and underlying intent of the current conversation. Subsequently, the system embeds the "core guidance strategy" into the Prompt as structured instructions. For example, in a state of high anxiety, the system might inject an instruction such as, "Please use cognitive reappraisal strategies to express your views in neutral, non-adversarial language." Furthermore, user profile data (such as communication style, emotional inclination, and language preferences) is encoded into semantic feature vectors, which, along with the strategy instructions, serve as input to the LLM, guiding it to generate responses that align with the user's personality traits.

[0064] Specifically, the construction of a Prompt must adhere to certain semantic control specifications. For example, the weight of the policy instruction can be set to 0.6, the weight of the context information to 0.3, and the weight of the user profile to 0.1, to ensure that the policy priority is higher than content matching. Furthermore, the output length of the LLM is typically controlled between 150 and 300 characters to meet the conciseness requirements of instant messaging scenarios. The temperature parameter during the generation process can be set to 0.4-0.7 to achieve a balance between creativity and stability.

[0065] In practical applications, this step is widely applicable to scenarios requiring immediate responses, such as work emails, private chats, and social media comments. Especially when users exhibit high anxiety, low confidence, or decision-making difficulties, the system's strategy-guided response suggestions can effectively alleviate their psychological burden while improving the comfort and effectiveness of communication.

[0066] Furthermore, S4 includes: S41, the generation guidance constraint includes the four-element structure of the Nonviolent Communication (NVC) framework, specifically requiring that the response content must include four parts: observation, feeling, need, and request.

[0067] Specifically, in some implementations, the system first fuses the collected multimodal data with the generated psychological state vector to form a comprehensive input feature vector. ,in Represents dialogue context data, Represents real-time user behavior data. This represents user profile data. This represents application scenario data. The feature vector is then fed into a "psychological strategy decision network," which performs strategy matching calculations based on deep learning models (such as Transformer, graph neural networks, etc.).

[0068] Furthermore, the psychological theory knowledge base contains multiple standardized strategy cards, each corresponding to a psychological communication strategy (such as nonviolent communication, cognitive behavioral therapy, etc.), and includes two parts: "activation rules" and "generation guidelines." The activation rules define under what psychological state vectors the strategy should be prioritized, for example, when anxiety level... And confidence level When prompted, the system will prioritize either the "cognitive reappraisal strategy" or the "nonviolent communication strategy." The generation guidelines provide a structured template for generating subsequent responses. For example, the Nonviolent Communication (NVC) framework requires responses to include four parts: Observation, Feeling, Need, and Request.

[0069] Optionally, the strategy selection module can also introduce a weighted scoring mechanism to quantitatively evaluate the matching degree of each strategy card. For example, cosine similarity can be used. or attention mechanism The matching score between the input feature vector and the policy card is calculated, and the policy with the highest score is selected as the core guiding policy.

[0070] In practical applications, this step is particularly useful when users face communication scenarios with high emotional loads (such as conflicting emails or critical conversations). By dynamically matching psychological strategies, it provides users with structured, emotion-oriented response guidance. Its technological value lies in the deep integration of psychological theories with AI systems, achieving a paradigm shift from "task-oriented" to "psychological empowerment," thereby fundamentally alleviating users' communication anxiety and improving their psychological comfort and communication quality.

[0071] S42, using language style parameters from user profile data To personalize the expression, style constraints are added to the LLM Prompt during generation.

[0072] In some implementations, the system first extracts language style-related feature parameters from the user profile database. This set of parameters typically includes: tone preference (e.g., formal, casual, humorous), emotional tendency (e.g., positive, neutral, negative), expressive complexity (e.g., sentence length, lexical diversity), and communication style type (e.g., highly appeasing, low emotional expression). These parameters can be modeled using users' historical dialogue data, and NLP techniques (e.g., BERT, LSTM, etc.) can be used for feature extraction and cluster analysis to form a quantifiable language style vector.

[0073] Specifically, The specific dimensions and value ranges can be configured according to the actual application scenario. For example, tone preference can be quantified as follows: Continuous values ​​within the interval, where Indicates full formality. This indicates complete arbitrariness; emotional tendency can be expressed using an emotional intensity index (such as one based on the VADER model). Sentiment score); expressive complexity can be assessed through average sentence length (unit: words / sentence) and lexical diversity (such as TTR index, value range). The parameters are described in a structured manner in the Prompt, for example: "Give a reply in a formal tone, neutral sentiment, and moderate sentence length (average sentence length 10-15 words)."

[0074] In practical applications, this step is widely applicable to scenarios requiring users to provide text responses, such as instant messaging, email, and online customer service. Especially when users exhibit high anxiety or low self-confidence, the system guides the LLM to generate responses that better align with the user's psychological comfort level by adjusting the language style parameters in the Prompt, thereby reducing their psychological burden and increasing their willingness to communicate and fluency of expression.

[0075] S5, present the suggested response and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principles of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

[0076] In some implementations, the explanation module can be based on a templated language generation mechanism to match preset explanation statements according to the strategy type (such as cognitive reappraisal, nonviolent communication, topic separation, etc.) and make personalized adjustments based on the user's current psychological state vector (such as anxiety level, decision-making pressure, confidence level, etc.) to enhance the relevance and comprehensibility of the explanation.

[0077] The generation of explanatory feedback needs to meet certain standards of information density and readability. For example, the length of the explanatory text should be controlled between 100 and 200 words to ensure that users can obtain key information in a short time. Simultaneously, the system can set a confidence threshold for feedback generation to ensure that the output explanation has sufficient accuracy and logic. Furthermore, the presentation of explanatory feedback can employ multimodal interaction methods such as highlighting keywords, providing point-by-point explanations, or reading aloud to adapt to different users' cognitive preferences.

[0078] In practical applications, this step is widely applicable to scenarios requiring users to provide text responses, such as instant messaging, email, and online customer service. Especially when users face high-pressure, high-emotional-load communication situations (such as work conflicts, emotional outpourings, and critical feedback), explanatory feedback can effectively reduce users' uncertainty about the response content, enhance their acceptance and trust in the strategy chosen, and thus improve the overall communication experience.

[0079] Furthermore, S5 includes: S51 generates explanatory feedback by translating a pre-set strategy principle template, which includes a simplified description of communication skills and a quantitative prediction of expected results.

[0080] In some implementations, the strategy principle translation templates are jointly developed by psychology experts and NLP engineers. Their content covers simplified explanations of various communication strategies, such as "nonviolent communication," "cognitive reappraisal," and "task separation." Each template contains two core parts: a simplified description of the strategy and a prediction of its potential psychological effects in the current context. The simplified descriptions utilize text summarization and interpretation generation techniques from natural language processing to translate complex psychological theories into language that users can easily understand. For example: "This response uses the task separation technique, which helps you maintain your psychological boundaries and reduce unnecessary psychological burden."

[0081] The quantitative prediction of the expected effect is based on the psychological state vector output in step two, calculated using a pre-defined strategy-effect mapping function. For example, if the current strategy is "cognitive reappraisal," the system can predict the extent to which this strategy reduces anxiety levels based on historical data and psychological models. And the degree of improvement in self-confidence. These predictions can be dynamically adjusted based on user profiles and the current context to improve the personalization of feedback.

[0082] In practical applications, this step can be embedded in various instant messaging and email clients as supplementary information to the suggested responses. While viewing the suggested responses, users can also understand the underlying psychological principles and expected effects, thereby enhancing their acceptance of the response content and their willingness to use it.

[0083] S52 uses a hierarchical structure to present explanatory feedback. The first layer shows the core strategy name, the second layer shows the strategy application scenarios, and the third layer shows the specific implementation steps.

[0084] In some implementations, this hierarchical structure includes three levels: the first level is the core strategy name, such as "cognitive reappraisal strategy" or "nonviolent communication strategy," used to quickly convey the type of psychological method employed; the second level is the strategy application scenario, where the system determines the appropriate communication situation based on the current dialogue context and the user's psychological state, such as "when facing critical emails, using the cognitive reappraisal strategy helps reduce emotional reactions"; the third level is the specific implementation steps, describing in concise language how the strategy is applied in the response, such as "by reconstructing the other party's criticism using neutral language, the focus is shifted from emotional reactions to problem-solving."

[0085] Specifically, the presentation of this feedback mechanism needs to strike a balance between information density and readability. For example, strategy names should be limited to 10 characters, application scenario descriptions to no more than 30 characters, and implementation steps to no more than 50 characters, to ensure that users can obtain key information in a short time. In addition, the system can set a feedback presentation frequency threshold, such as forcing the display of explanatory feedback when the user's anxiety level in their psychological state vector exceeds 0.7, in order to enhance the psychological support effect.

[0086] In application scenarios, this step can be widely used in instant messaging, email, online customer service, and other situations that require users to provide text responses. Especially when users exhibit high anxiety, low confidence, or difficulty in decision-making, this feedback mechanism can provide immediate psychological support and behavioral guidance, helping users understand the logic behind the suggested responses, thereby enhancing their sense of psychological security and control over communication.

[0087] An auxiliary response method based on psychological strategies, as described in this invention, can effectively alleviate users' anxiety and stress during communication. By generating personalized response suggestions through real-time psychological assessment and strategy guidance, it can improve psychological comfort and intrinsic quality during communication.

[0088] Example 2 The following describes in detail an auxiliary response method based on psychological strategies according to an embodiment of the present invention, with reference to the accompanying drawings.

[0089] This invention provides an auxiliary response method and system based on psychological strategies. Instead of directly generating a response, it dynamically selects the most suitable psychological communication strategy by real-time assessment of the user's state, and generates response suggestions based on the strategy that can effectively alleviate the user's negative emotions and improve psychological comfort. Figure 2 This paper demonstrates the complete process of an auxiliary response method based on psychological strategies proposed in this invention. The process begins with data collection in step one, proceeds to psychological state assessment in step two, selects core strategies in step three, generates strategy-guided content in step four, and finally presents suggested responses and explanatory feedback to the user in step five, forming a complete technical loop.

[0090] The method includes the following steps: Step 1: Context Awareness and Multimodal Data Acquisition. This method is triggered when a user spends more than a preset threshold time on the reply interface of an instant messaging or email client, or exhibits frequent hesitant behaviors such as typing and deleting. The system collects a set of multimodal data in real time, including: 1) Dialogue context data (chat logs, original email text); 2) Real-time user behavior data (typing speed, pause frequency, backspace rate); 3) User profile data (preset or learned communication styles from historical behavior, such as "highly ingratiating"); 4) Application scenario data (identifying whether the current email is for work or a private chat).

[0091] Step Two: Real-Time User Psychological State Assessment. This step is one of the key innovations of this invention. The system inputs the "real-time user behavior data" collected in Step One into a pre-trained "psychological state assessment model." Based on the behavioral data, this model outputs a quantified "psychological state vector" in real time to describe the user's current inner feelings. For example: {Anxiety level: 0.8, Decision-making pressure: 0.9, Confidence level: 0.2}.

[0092] Step 3: Core Psychological Strategy Selection. This step is the core technology of this invention. The system integrates the various data from Step 1 with the "psychological state vector" from Step 2 as input to the "psychological strategy decision network." Based on the input information, this decision network queries a "configurable psychological theory knowledge base." This knowledge base creates standardized "strategy cards" containing "activation rules" and "generation guidelines" based on various communication theories (such as nonviolent communication, cognitive behavioral therapy, and issue separation). Through matching calculations, the decision network selects the strategy with the highest score in the knowledge base (e.g., selecting the "cognitive reappraisal strategy" in a high-anxiety state) as the "core guiding strategy" for this response.

[0093] Step Four: Strategy-Guided Response Generation. The system invokes a Large Language Model (LLM) to generate specific response suggestions. The prompt received by the LLM is strictly constrained and includes: 1) the dialogue context; 2) the "core guiding strategy" selected in Step Three and its detailed "generation guidelines"; and 3) user profile data to ensure personalized response tone. In this way, the output of the LLM is no longer aimless but rather text that strictly follows psychological strategies and aims to achieve specific psychological reassurance effects.

[0094] Step Five: Explanatory Feedback and Interactive Presentation. The system presents the user with one or more generated response suggestions. Simultaneously, the system includes a "Design Concept" explanation card, explaining in simple language the communication skills or psychological principles behind the suggested response, and its potential benefits. For example: "This response uses the 'topic separation' technique, which helps you maintain your psychological boundaries and reduce unnecessary psychological burden." This step not only provides solutions but also, through empowerment and instruction, enhances the user's communication confidence and abilities over long-term use, forming a positive therapeutic loop.

[0095] The core idea of ​​this invention lies in "real-time psychological assessment + application of psychological strategies," and its specific implementation can have various alternative solutions: Alternatives to psychological state assessment: In addition to analyzing typing behavior, richer modal data can be integrated (with user authorization), such as: Voice intonation analysis: In voice input scenarios, analyzing the user's speech rate, pitch, pauses, etc., can more accurately determine their emotional state. Physiological signal integration: It can be linked with wearable devices such as smartwatches and wristbands, using physiological indicators such as heart rate and galvanic skin response (GSR) as input to the assessment model, greatly improving the accuracy of the assessment.

[0096] Alternatives to strategy selection: In addition to using decision networks for rating selection, a dedicated reinforcement learning model can be employed. This model uses positive feedback (such as decreased anxiety or increased adoption rate) after a user selects a suggestion as a reward signal, and dynamically optimizes its strategy selection ability through continuous learning.

[0097] Alternative to the interaction method: In addition to directly providing response suggestions, the system can also offer a "role-playing dialogue" mode. Users can first practice dialogue with a simulated, psychologically driven "communication coach," and then send out their final response once they feel ready.

[0098] Example 3 To achieve the above embodiments, such as Figure 3 As shown, this embodiment also provides an auxiliary response device 10 based on psychological strategies. The device 10 includes a multimodal data acquisition module 100, a psychological state assessment module 200, a psychological strategy matching module 300, a response generation module 400, and a feedback presentation module 500.

[0099] The multimodal data acquisition module 100 is used to collect multimodal data of users in digital communication scenarios. The multimodal data includes dialogue context data, real-time user behavior data, user profile data, and application scenario data. The psychological state assessment module 200 is used to input the user's real-time behavior data into a pre-trained psychological state assessment model and output a quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure and self-confidence level. The psychology strategy matching module 300 is used to select a core guidance strategy that is compatible with the user's current psychological state by matching strategy cards in the configurable psychology theory knowledge base through the psychology strategy decision network based on the integration result of the psychological state vector and the multimodal data. The response generation module 400 is used to call a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidance of the core guidance strategy and combined with the user profile data to achieve personalized expression. The feedback presentation module 500 is used to present the response suggestions and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principles of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

[0100] Furthermore, the aforementioned psychological state assessment module 200 is also used for: Anxiety levels are calculated using a weighted average algorithm, with the weighting coefficients dynamically adjusted based on users' historical behavior data. Multimodal data fusion technology is used to fuse typing speed, backspace rate and speech intonation features at the feature level.

[0101] Furthermore, the aforementioned psychological strategy matching module 300 is also used for: When a vaguely defined keyword is detected in the dialogue context, the activation rule of the topic separation strategy is triggered. The specific judgment condition is as follows:

[0102] in The number of times the boundary keyword appears. The preset trigger threshold; A dynamic weight adjustment mechanism is adopted to adjust the matching score of strategy cards based on the communication style type in user profile data.

[0103] Furthermore, the aforementioned response generation module 400 is also used for: The generation guidelines and constraints include the four-element structure of the Nonviolent Communication (NVC) framework, specifically requiring that the response content must include four parts: observation, feeling, need, and request. Personalized expression is achieved by using language style parameters from user profile data, and style constraints are added to the LLM Prompt during generation.

[0104] Furthermore, the aforementioned feedback presentation module 500 is also used for: Explanatory feedback is generated by translating pre-set strategy principles into templates. The templates include a simplified description of communication skills and a quantitative prediction of expected results. Explanatory feedback is presented using a hierarchical structure. The first layer displays the core strategy name, the second layer displays the application scenarios of the strategy, and the third layer displays the specific implementation steps.

[0105] An auxiliary response device based on psychological strategies according to an embodiment of the present invention can effectively alleviate users' anxiety and stress in communication. It generates personalized response suggestions through real-time psychological assessment and strategy guidance, thereby improving psychological comfort and intrinsic quality in communication.

[0106] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0107] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A psychological strategy-based auxiliary response method, characterized in that, include: S1, collect multimodal data of users in digital communication scenarios, including dialogue context data, real-time user behavior data, user profile data and application scenario data; S2, input the real-time user behavior data into the pre-trained psychological state assessment model, and output the quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure and self-confidence level. S3, based on the integration result of the psychological state vector and the multimodal data, the core guidance strategy that is compatible with the user's current psychological state is selected by matching the strategy cards in the configurable psychological theory knowledge base through the psychological strategy decision network; S4, invoke a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidance of the core guidance strategy and combined with the user profile data to achieve personalized expression. S5, present the suggested response and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principles of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

2. The method as described in claim 1, characterized in that, S2 includes: S21 calculates anxiety levels using a weighted average algorithm, with the weighting coefficients dynamically adjusted based on users' historical behavior data. S22 employs multimodal data fusion technology to fuse typing speed, backspace rate, and speech intonation features at the feature level.

3. The method as described in claim 1, characterized in that, The S3 includes: S31, when a keyword with ambiguous boundaries is detected in the dialogue context, the activation rule of the topic separation strategy is triggered. The specific judgment condition is as follows: in The number of times the boundary keyword appears. The preset trigger threshold; S32 employs a dynamic weight adjustment mechanism to adjust the matching score of strategy cards based on the communication style type in the user profile data.

4. The method as described in claim 1, characterized in that, The S4 further includes: S41, the generation guidance constraint includes the four-element structure of the Nonviolent Communication (NVC) framework, specifically requiring that the response content must include four parts: observation, feeling, need, and request; S42, using language style parameters from user profile data To personalize the expression, style constraints are added to the LLM Prompt during generation.

5. The method as described in claim 1, characterized in that, The S5 includes: S51 generates explanatory feedback by translating a pre-set strategy principle template, which includes a popular description of communication skills and a quantitative prediction of expected results. S52 uses a hierarchical structure to present explanatory feedback. The first layer shows the core strategy name, the second layer shows the strategy application scenarios, and the third layer shows the specific implementation steps.

6. A psychological strategy-based auxiliary response device, characterized in that, include: The multimodal data acquisition module is used to collect multimodal data of users in digital communication scenarios. The multimodal data includes dialogue context data, real-time user behavior data, user profile data, and application scenario data. The psychological state assessment module is used to input the user's real-time behavior data into a pre-trained psychological state assessment model and output a quantified psychological state vector, which includes numerical representations of anxiety level, decision-making pressure, and self-confidence level. The psychology strategy matching module is used to select core guidance strategies that are suitable for the user's current psychological state by matching strategy cards in the configurable psychology theory knowledge base through the psychology strategy decision network, based on the integration results of the psychological state vector and the multimodal data. The response generation module is used to call a large language model to generate response suggestions. The generation process of the response suggestions is constrained by the generation guidelines of the core guidance strategy and combined with the user profile data to achieve personalized expression. The feedback presentation module is used to present the suggested response and corresponding explanatory feedback to the user. The explanatory feedback explains the communication principle of the selected strategy and its expected effect on alleviating the user's anxiety in plain language.

7. The apparatus as claimed in claim 6, characterized in that, The psychological state assessment module is also used for: Anxiety levels are calculated using a weighted average algorithm, with the weighting coefficients dynamically adjusted based on users' historical behavior data. Multimodal data fusion technology is used to fuse typing speed, backspace rate and speech intonation features at the feature level.

8. The apparatus as claimed in claim 6, characterized in that, The psychological strategy matching module is also used for: When a vaguely defined keyword is detected in the dialogue context, the activation rule of the topic separation strategy is triggered. The specific judgment condition is as follows: in The number of times the boundary keyword appears. The preset trigger threshold; A dynamic weight adjustment mechanism is adopted to adjust the matching score of strategy cards based on the communication style type in user profile data.

9. The apparatus as claimed in claim 6, characterized in that, The response generation module is also used for: The generation guidelines and constraints include the four-element structure of the Nonviolent Communication (NVC) framework, specifically requiring that the response content must include four parts: observation, feeling, need, and request. Personalized expression is achieved by using language style parameters from user profile data, and style constraints are added to the LLM Prompt during generation.

10. The apparatus as claimed in claim 6, characterized in that, The feedback presentation module is also used for: Explanatory feedback is generated by translating pre-set strategy principles into templates. The templates include a simplified description of communication skills and a quantitative prediction of expected results. Explanatory feedback is presented using a hierarchical structure. The first layer displays the core strategy name, the second layer displays the application scenarios of the strategy, and the third layer displays the specific implementation steps.

Images