Intelligent dialogue method and device, electronic equipment and storage medium

Abstract

The application relates to an intelligent conversation method and device, electronic equipment and a storage medium. The method comprises the following steps: obtaining conversation content input by a target object at present and historical conversation data; inputting the conversation content into a pre-trained emotion recognition model to obtain an emotion label of the target object, and determining a personality portrait of the target object based on the historical conversation data, wherein the emotion recognition model is pre-trained based on a large number of infant conversation samples, and model parameters of the emotion recognition model can be adjusted based on the personality portrait of the target object; determining a candidate interaction strategy based on the emotion label of the target object and a preset mapping relationship, and optimizing and learning the candidate interaction strategy based on the emotion label of the target object and the personality portrait of the target object to determine a target interaction strategy; and generating target reply content based on the emotion label of the target object, the personality portrait of the target object and the target interaction strategy, so that the accuracy of intelligent conversation reply content is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to an intelligent conversation method and device, an electronic device, and a storage medium. BACKGROUND

[0002] With the rapid development of artificial intelligence technology, intelligent conversation systems for young children have also emerged. They are mainly based on auxiliary teaching and emotion management to provide conversation interaction for young children. For example, they are widely used in educational intelligent toys, intelligent speakers, and other devices.

[0003] In related technologies, the voice content or text content input by a young child is usually compared with preset emotional keywords, and a corresponding conversation reply is selected according to built-in rules. For example, when a young child says, "I don't like this", the system may match the young child's "negative" emotion and output a conversation reply such as "That's okay, let's try something else". However, this approach cannot adapt well to the conversation characteristics of young children (such as fuzzy pronunciation or incomplete sentence patterns), and cannot provide personalized reply content for young children with different personalities, resulting in low accuracy of the generated reply content. Therefore, how to improve the intelligent conversation reply content for young children has become a technical problem to be solved. SUMMARY

[0004] The present application provides an intelligent conversation method, device, electronic device, and storage medium to solve the technical problem that existing technologies cannot adapt well to the conversation characteristics of young children and cannot provide personalized reply content for young children with different personalities, resulting in low accuracy of the generated reply content.

[0005] In a first aspect, an embodiment of the present application provides an intelligent conversation method, which comprises:

[0006] obtaining conversation content currently input by a target object and historical conversation data, wherein the target object is any young child;

[0007] inputting the conversation content into a pre-trained emotion recognition model to obtain an emotional label of the target object, and determining a personality portrait of the target object based on the historical conversation data, wherein the emotion recognition model is pre-trained based on a large amount of young child conversation samples, and the model parameters of the emotion recognition model can be adjusted based on the personality portrait of the target object;

[0008] determining a candidate interaction strategy based on the emotional label of the target object and a preset mapping relationship, and optimizing learning the candidate interaction strategy based on the emotional label of the target object and the personality portrait of the target object to determine a target interaction strategy, wherein the preset mapping relationship represents the mapping relationship between the emotional label and the interaction strategy;

[0009] generate target reply content based on the emotion label of the target object, the personality portrait of the target object, and the target interaction strategy.

[0010] Optionally, the emotion recognition model comprises a speech feature extraction sub-model, a text feature extraction sub-model, and an emotion analysis sub-model.

[0011] The inputting of the dialogue content into the pre-trained emotion recognition model to obtain the emotion label of the target object comprises:

[0012] The speech information in the dialogue content is inputted into the speech feature extraction sub-model to obtain the spectrum feature, the pitch change, and the volume intensity corresponding to the speech information, and the speech feature vector corresponding to the target object is determined based on the spectrum feature, the pitch change, and the volume intensity corresponding to the speech information.

[0013] The text information in the dialogue content is inputted into the text feature extraction sub-model to obtain the word embedding, the sentence length, and the context score corresponding to the text information, and the text feature vector corresponding to the target object is determined based on the word embedding, the sentence length, and the context score corresponding to the text information.

[0014] The speech feature vector and the text feature vector are inputted into the emotion analysis sub-model to obtain the emotion label of the target object.

[0015] Optionally, the inputting of the speech feature vector and the text feature vector into the emotion analysis sub-model to obtain the emotion label of the target object comprises:

[0016] The speech feature vector and the text feature vector are inputted into the emotion analysis sub-model, and the emotion analysis sub-model is used to perform feature fusion according to a formula to obtain the emotion label of the target object.

[0017] wherein, represents the emotion label of the target object, represents a dynamic attention weight, represents a normalization function, represents a weight matrix of speech features, represents the speech feature vector, represents a weight matrix of text features, represents the text feature vector.

[0018] Optionally, the determining of the personality portrait of the target object based on the historical dialogue data comprises:

[0019] determine the personality portrait of the target object based on the historical dialogue data.

[0020] determine the personality portrait of the target object based on the historical dialogue data.

[0021] Optionally, after the personality portrait of the target object is determined based on the historical dialogue data, the method further comprises:

[0022] adjust the first parameter or the second parameter in the emotion recognition model based on the personality portrait of the target object, so that the emotion label output by the adjusted emotion recognition model is more close to the real emotion label of the target object and matches the personality portrait of the target object, wherein the first parameter is a related parameter corresponding to a weight matrix of a voice feature, and the second parameter is a related parameter corresponding to a weight matrix of a text feature.

[0023] Optionally, the optimization learning of the candidate interaction strategy based on the emotion label of the target object and the personality portrait of the target object to determine the target interaction strategy comprises:

[0024] use a preset reinforcement learning algorithm to optimize the learning of the candidate interaction strategy, and determine the target interaction strategy from the candidate interaction strategy, wherein the reinforcement learning algorithm is an algorithm that uses the emotion label of the target object and the personality portrait of the target object as a state, an interaction strategy in the candidate interaction strategy as an action, and an emotion relief degree as a reward, selects the target interaction strategy by iteratively calculating the value of different state-action pairs, and the target interaction strategy is an interaction strategy corresponding to the optimal value.

[0025] Optionally, after the target reply content is generated based on the emotion label of the target object, the personality portrait of the target object, and the target interaction strategy, the method further comprises:

[0026] obtain an emotion trend factor and an emotion fluctuation range of the target object;

[0027] in a case where the emotion trend factor of the target object is less than a first preset threshold, trigger a warning notification to warn the emotion change of the target object;

[0028] In a case where the emotional fluctuation range of the target object is greater than a second preset threshold, dynamic suggestion content is generated based on the emotional label of the target object, the personality portrait of the target object, and the dialogue context content of the target object in the current dialogue scenario; and

[0029] A visualization score is generated based on the emotional label of the target object, the personality portrait of the target object, and the emotional trend factor of the target object.

[0030] In a second aspect, the embodiments of the present application further provide an intelligent dialogue device, and the device comprises:

[0031] A first obtaining module is configured to obtain dialogue content input by a target object currently and historical dialogue data, wherein the target object is any infant;

[0032] A first determining module is configured to input the dialogue content into a pre-trained emotion recognition model to obtain an emotional label of the target object, and determine a personality portrait of the target object based on the historical dialogue data, wherein the emotion recognition model is pre-trained based on a large amount of infant dialogue samples, and model parameters of the emotion recognition model can be adjusted based on the personality portrait of the target object;

[0033] A second determining module is configured to determine a candidate interaction strategy based on the emotional label of the target object and a preset mapping relationship, and optimize learning of the candidate interaction strategy based on the emotional label of the target object and the personality portrait of the target object to determine a target interaction strategy, wherein the preset mapping relationship is used to represent a mapping relationship between the emotional label and the interaction strategy;

[0034] A first generating module is configured to generate target reply content based on the emotional label of the target object, the personality portrait of the target object, and the target interaction strategy.

[0035] In a third aspect, the embodiments of the present application further provide an electronic device, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete communication with each other through the communication bus;

[0036] The memory is configured to store a computer program;

[0037] The processor is configured to execute the program stored on the memory, and implement the intelligent dialogue method of any one of the embodiments of the first aspect.

[0038] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the intelligent dialogue method of any one of the embodiments of the first aspect.

[0039] The technical solution provided by the embodiments of the present application has the following advantages compared with the prior art:

[0040] The method provided by the embodiments of the present application, by obtaining the current input dialogue content of the target object and historical dialogue data, wherein the target object is any infant, inputting the dialogue content into a pre-trained emotion recognition model to obtain the emotion label of the target object, and determining the individual portrait of the target object based on the historical dialogue data, wherein the emotion recognition model is pre-trained based on a large number of infant dialogue samples, and the model parameters of the emotion recognition model can be adjusted based on the individual portrait of the target object, determining the candidate interaction strategy based on the emotion label of the target object and a preset mapping relationship, and optimizing learning the candidate interaction strategy based on the emotion label of the target object and the individual portrait of the target object to determine the target interaction strategy, wherein the preset mapping relationship is used to represent the mapping relationship between the emotion label and the interaction strategy, and generating the target reply content based on the emotion label of the target object, the individual portrait of the target object and the target interaction strategy. Through the above manner, the model parameters of the emotion recognition model can be adjusted by using the individual portrait of the target object, so that the emotion recognition model can comprehensively consider the dialogue characteristics and individual portrait of the target object to improve the accuracy of emotion recognition, thereby improving the accuracy of the intelligent dialogue reply content for infants. Moreover, by combining the emotion label of the target object and the individual portrait of the target object, the candidate interaction strategy is optimized and learned, so as to select the target interaction strategy that is most suitable for the target user from the candidate interaction strategy, thereby further improving the accuracy of the target reply content. BRIEF DESCRIPTION OF DRAWINGS

[0041] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and serve to explain the principles of the present application together with the specification.

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained based on these drawings without creative labor.

[0043] Figure 1 A flowchart of an intelligent dialogue method provided by the embodiments of the present application;

[0044] Figure 2 A flowchart of another intelligent dialogue method provided by the embodiments of the present application;

[0045] Figure 3A structural schematic diagram of an intelligent dialogue device provided by an embodiment of the present application is provided.

[0046] Figure 4 A structural schematic diagram of an electronic device provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0047] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0048] Referring to Figure 1 , Figure 1 A flowchart of an intelligent dialogue method provided by an embodiment of the present application is provided. As shown in Figure 1 , the intelligent dialogue method can include the following steps:

[0049] Step S101, obtaining dialogue content currently input by a target object and historical dialogue data, wherein the target object is any infant.

[0050] Specifically, the dialogue content can be voice information currently input by the target object, can be text information currently input by the target object, or can be voice information and text information currently input by the target object, which is not limited in the present application. As an optional implementation, the dialogue content is voice information currently input by the target object, and text information corresponding to the voice information is obtained through voice-to-text processing, and then subsequent analysis and processing are performed. The historical dialogue data refers to dialogue data generated by the target object before the current dialogue content is input.

[0051] Step S102, inputting the dialogue content into a pre-trained emotion recognition model to obtain an emotion label of the target object, and determining a personality portrait of the target object based on the historical dialogue data, wherein the emotion recognition model is pre-trained based on a large amount of infant dialogue samples, and model parameters of the emotion recognition model can be adjusted based on the personality portrait of the target object.

[0052] Specifically, the above-mentioned emotion recognition model is pre-trained based on a large number of infant dialogue samples, and thus can learn the dialogue characteristics such as ambiguous pronunciation or incomplete sentence of the infant during the training process. The model parameters of the emotion recognition model can be adjusted based on the individual portrait of the target object, so that the emotion recognition model can recognize the target object based on the individual portrait of the target object during the process of recognizing the emotion, thereby improving the accuracy of emotion recognition. The emotion label of the target object is the output result of the emotion recognition model, and is used to represent the current emotional state of the target object, such as “happy” and “sad”.

[0053] In step S103, a candidate interaction strategy is determined based on the emotion label of the target object and a preset mapping relationship, and the candidate interaction strategy is optimized and learned based on the emotion label of the target object and the individual portrait of the target object to determine a target interaction strategy. The preset mapping relationship is used to represent the mapping relationship between the emotion label and the interaction strategy.

[0054] Specifically, the above-mentioned preset mapping relationship is used to represent the mapping relationship between the emotion label and the interaction strategy. Generally, each emotion label can correspond to one or more interaction strategies. For example, when the emotion label is “sad”, the interaction strategy can be “comfortable conversation”, “playing music”, etc. The above-mentioned candidate interaction strategy refers to one or more interaction strategies that have a mapping relationship with the emotion label of the target object. When the candidate interaction strategy is optimized and learned based on the emotion label of the target object and the individual portrait of the target object, the optimization can be performed based on a preset optimization learning algorithm, so as to determine the optimal target interaction strategy from the candidate interaction strategy. The optimization learning algorithm can be a Q-learning algorithm, an online reinforcement learning algorithm based on time difference learning (State-Action-Reward-State-Action, referred to as SARSA), a deep deterministic policy gradient algorithm (Deep Deterministic Policy Gradient, referred to as DDPG), a proximal policy optimization (Proximal Policy Optimization, referred to as PPO) algorithm, etc.

[0055] In step S104, the target reply content is generated based on the emotion label of the target object, the individual portrait of the target object, and the target interaction strategy.

[0056] After determining the emotion label of the target object, the individual portrait of the target object, and the target interaction strategy, the context content of the previous rounds of dialogue of the target object can be obtained, and the context relevance score can be calculated. Then, the dialogue generation model is used to learn the dialogue features of the target object, the individual portrait of the target object, the target interaction strategy, and the context relevance score, and the final target reply content is obtained and outputted. The dialogue generation model can be pre-trained based on the training samples labeled with the dialogue features of the children, the individual portrait of the children, the interaction strategy, and the context relevance score, or can be fine-tuned by using an existing large speech model, which is not limited in the present application.

[0057] In the present embodiment, the model parameters of the emotion recognition model can be adjusted by using the individual portrait of the target object, so that the emotion recognition model can comprehensively consider the dialogue characteristics and the individual portrait of the target object to improve the accuracy of emotion recognition, thereby improving the accuracy of the intelligent dialogue reply content for children. In addition, by combining the emotion label of the target object and the individual portrait of the target object, the candidate interaction strategy is optimized and learned, so that the target interaction strategy most suitable for the target user is selected from the candidate interaction strategy, thereby further improving the accuracy of the target reply content.

[0058] In an optional embodiment, the step S102 of inputting the dialogue content into the pre-trained emotion recognition model to obtain the emotion label of the target object comprises:

[0059] The speech information in the dialogue content is input into the speech feature extraction sub-model to obtain the spectrum feature, the pitch change, and the volume intensity corresponding to the speech information, and the speech feature vector corresponding to the target object is determined based on the spectrum feature, the pitch change, and the volume intensity corresponding to the speech information.

[0060] The text information in the dialogue content is input into the text feature extraction sub-model to obtain the word embedding, the sentence length, and the context score corresponding to the text information, and the text feature vector corresponding to the target object is determined based on the word embedding, the sentence length, and the context score corresponding to the text information.

[0061] The speech feature vector and the text feature vector are input into the emotion analysis sub-model to obtain the emotion label of the target object.

[0062] Specifically, the above-mentioned emotion recognition model can include a speech feature extraction sub-model, a text feature extraction sub-model, and an emotion analysis sub-model. Among them, the speech feature extraction sub-model can be implemented by using an improved unsupervised speech pre-training model Wav2Vec 2.0 model (i.e. adding a child speech enhancement layer on the basis of the original Wav2Vec 2.0 model) or other models. The text feature extraction sub-model can be implemented by using a lightweight DistilBERT model (parameter quantity about 66M) or other models. The emotion analysis sub-model can be implemented by using a transfome model or other models.

[0063] When the dialogue content is input into the pre-trained emotion recognition model to obtain the emotion label of the target object, the speech information in the dialogue content can be input into the speech feature extraction sub-model to obtain the spectrum feature, pitch change and volume intensity corresponding to the speech information, and based on the spectrum feature, pitch change and volume intensity corresponding to the speech information, the speech feature vector corresponding to the target object is determined. The speech information here can be the speech dialogue content directly input by the target object, or the speech information converted from the text dialogue content directly input by the target object. At the same time, the text information in the dialogue content can be input into the text feature extraction sub-model to obtain the word embedding, sentence length and context score corresponding to the text information, and based on the word embedding, sentence length and context score corresponding to the text information, the text feature vector corresponding to the target object is determined. The text information here can be the text dialogue content directly input by the target object, or the text information converted from the speech dialogue content directly input by the target object. Then, the speech feature vector and the text feature vector can be input into the emotion analysis sub-model for feature fusion and emotion analysis to obtain the emotion label of the target object.

[0064] In the present embodiment, the speech feature extraction sub-model and the text feature extraction sub-model in the emotion recognition model can be used to extract the speech feature and the text feature of the target object respectively, so that the speech feature and the text feature of the child dialogue can be fully considered in emotion recognition, and the accuracy of emotion recognition is improved.

[0065] In an optional embodiment, the above-mentioned step of inputting the speech feature vector and the text feature vector into the emotion analysis sub-model to obtain the emotion label of the target object comprises:

[0066] inputting the speech feature vector and the text feature vector into the emotion analysis sub-model, and using the emotion analysis sub-model to perform feature fusion according to the formula to obtain the emotion label of the target object;

[0067] wherein, represents the emotion label of the target object, denotes a dynamic attention weight, denotes a normalization function, denotes a weight matrix of speech features, denotes a speech feature vector, denotes a weight matrix of text features, denotes a text feature vector.

[0068] Specifically, when obtaining the emotional label of the target object by using the emotion analysis sub-model, the speech feature vector and the text feature vector can be input into the emotion analysis sub-model, and the emotion analysis sub-model is used to perform feature fusion according to the formula to obtain the emotional label of the target object. Wherein, denotes the emotional label of the target object, that is, the current emotional state of the target object (such as the numerical value corresponding to "sadness"), denotes a dynamic attention weight, which is in the range of 0 to 1, and determines the relative importance of speech features and text features. The initial value can be 0.6, and is dynamically adjusted according to the frequency of the conversation. denotes a normalization function, which ensures that the output value is between 0 and 1, facilitating classification. denotes a weight matrix of speech features, which is used to map speech features to an emotional space, and the size can depend on the number of layers of the improved Wav2Vec 2.0 model (such as 128x64). denotes a speech feature vector, which can include spectral features (represented by MFCC), pitch changes (represented by ΔPitch), and volume intensity (represented by Volume). denotes a weight matrix of text features, which is used to adjust the emotional relevance of text information, and the size can be matched with the output of the DistilBERT model (such as 768x64). denotes a text feature vector, which can include word embeddings (represented by Word_emb), sentence length (represented by Sent_len), and context scores (represented by Context_score).

[0069] In this embodiment, the speech features and text features of the target object are dynamically fused by using the attention mechanism for extraction, so that the speech features and text features of the conversation of the infant can be fully considered in emotion recognition, and the accuracy of emotion recognition is improved.

[0070] In an optional embodiment, the above step S102, based on the historical conversation data, determines the individual portrait of the target object, comprising:

[0071] Based on historical conversation data, the target subject's activity, expression tendency, and emotional stability are calculated. Activity is used to indicate the target subject's active participation in the conversation, expression tendency is used to indicate the target subject's relative reliance on tone and vocabulary in language expression, and emotional stability is used to indicate the target subject's emotional fluctuations.

[0072] Based on the target object's corresponding activity, expression tendency and emotional stability, determine the target object's personality portrait.

[0073] Specifically, based on historical conversation data, the target object's activity, expression tendency, and emotional stability can be calculated. Among them, activity can be used to represent the degree of activeness of the target object in participating in the conversation, which can be calculated using the following formula:

[0074] ;

[0075] in, Indicates activity, Indicates the total interaction time. Indicates that the target object is Total number of interactions within. Activity High means the target person speaks frequently, shows an extroverted and proactive communication tendency, and may like interactive activities. Low means that the target object speaks less, shows introverted, cautious or quiet characteristics, and may be more suitable for low-stimulation interaction strategies. It can guide strategy selection, such as prompting the system to choose non-compulsory and quiet interaction strategies (such as storytelling rather than highly interactive games) when activity is low. It can also reflect the target object's willingness to interact and help the system judge the target object's participation, thereby adjusting the interaction rhythm or strategy intensity.

[0076] Expression tendency can be used to characterize the target subject's relative reliance on tone and vocabulary in language expression, which can be calculated using the following formula:

[0077] ;

[0078] in, Indicates a tendency to express The tone weight is calculated based on the speech features (such as pitch change, speaking speed, volume, etc.) in the historical conversation data, reflecting the intensity of the target object's emotion expressed through tone. Represents the vocabulary weight, which can be calculated based on the complexity of historical conversation data, the frequency of use of emotional vocabulary, etc., reflecting the degree to which the target object expresses emotions through vocabulary. High, indicating that the target person relies more on tone (such as pitch, speaking speed) to express emotions, uses fewer words or emotional vocabulary is not obvious, which may reflect the characteristics of introversion or implicit emotional expression. Low, indicating that the target object relies more on the content of the words to express emotions, with less variation in tone, which may reflect an extroverted or direct expression style. , can optimize emotion recognition results. For example, a high E-value prompts the system to pay more attention to voice features (such as pitch, speaking speed, etc.) rather than text features to improve the accuracy of emotion recognition. In addition, the model parameters of the emotion recognition model can be personalized adjusted. For example, the system lowers the tone detection threshold (Pitchmin=0.05) for child A to capture subtle changes in pitch and adapt to his high expressive tendency.

[0079] Emotional stability can be used to characterize the target's emotional fluctuations, which can be calculated using the following formula:

[0080] ;

[0081] in, Indicates the degree of emotional fluctuation. and They represent the emotion values ​​in two consecutive interactions at time t-1 and time t, respectively (for example, the emotion intensity or category probability estimated by the emotion recognition model). Represents the variance function and calculates the sentiment value and The degree of fluctuation over time. The degree of emotional fluctuation High, indicating that the target person has little emotional change and shows a stable emotional state, which may reflect the characteristics of introversion or good emotional control. Low means that the target object has large emotional fluctuations, is easily affected by external stimuli, and exhibits emotional instability. , the sensitivity of emotion recognition can be adjusted. For example, when the S value is low, the system is prompted to pay less attention to minor emotional fluctuations and focus on major emotional states (such as "sadness") to improve the robustness of recognition; and it can also guide strategy selection. For example, young children with stable emotions may need a gentler interaction strategy, while young children with large emotional fluctuations may need a more dynamic interaction strategy.

[0082] Next, based on the target object's corresponding activity, expression tendency and emotional stability, determine the target object's personality profile .

[0083] In the embodiment, the individual image of the target object can be accurately determined based on the historical dialogue data, and subsequent adjustment of the model parameters of the emotion recognition model based on the individual image of the target object is facilitated, and the target reply content conforming to the individual image of the target object is output.

[0084] In an optional embodiment, after the individual image of the target object is determined based on the historical dialogue data in the step S102, the method further includes:

[0085] The first parameter or the second parameter in the emotion recognition model is adjusted based on the individual image of the target object, so that the emotion label output by the adjusted emotion recognition model is more close to the real emotion label of the target object and matches the individual image of the target object, wherein the first parameter is a related parameter corresponding to a weight matrix of a voice feature, and the second parameter is a related parameter corresponding to a weight matrix of a text feature.

[0086] Specifically, after the individual image of the target object is determined, the first parameter or the second parameter in the emotion recognition model can be adjusted based on the individual image of the target object. For example, for an introverted child, the related parameter (such as a tone threshold ) corresponding to the weight matrix of the voice feature can be reduced; for an extroverted child, the related parameter (such as a word weight ) corresponding to the weight matrix of the text feature can be increased. In this way, the emotion label output by the adjusted emotion recognition model is more close to the real emotion label of the target object and matches the individual image of the target object. When the model parameters of the emotion recognition model are adjusted, the model parameters can be continuously updated in an online learning manner, and the updating manner is as follows:

[0087] ;

[0088] wherein, represents the updated model parameter, which is used for subsequent emotion recognition and dialogue generation. represents the model parameter before updating, which is used as the basis for updating. represents the learning rate, which is in the range of 0 to 1 (such as 0.01), and controls the parameter update step to avoid overfitting. represents the predicted emotion value, which is output by the current emotion recognition model. represents the real emotion value, which is usually obtained by labeling or model secondary inference. represents the individual image of the target object, which is used to adjust the loss weight so that the model pays more attention to the individual features. represents the mean square error loss, which is used to measure the gap between the predicted emotion value and the real emotion value.

[0089] By the above manner, the first parameter or the second parameter in the emotion recognition model can be continuously adjusted based on the personality portrait of the target object each time generated, so that the emotion label output by the adjusted emotion recognition model is more close to the real emotion label of the target object and matches the personality portrait of the target object.

[0090] In an optional embodiment, the step S103 of optimizing and learning the candidate interaction strategy based on the emotion label of the target object and the personality portrait of the target object to determine the target interaction strategy comprises:

[0091] The preset reinforcement learning algorithm is used to optimize and learn the candidate interaction strategy to determine the target interaction strategy from the candidate interaction strategy, wherein the reinforcement learning algorithm is an algorithm that uses the emotion label of the target object and the personality portrait of the target object as a state, an interaction strategy in the candidate interaction strategy as an action, and an emotion relief degree as a reward, selects the target interaction strategy by iteratively calculating the value of different state-action pairs, and the target interaction strategy is an interaction strategy corresponding to an optimal value.

[0092] Specifically, the Q-Learning algorithm can be used as the preset reinforcement learning algorithm to optimize and learn the candidate interaction strategy to determine the target interaction strategy from the candidate interaction strategy. The state is , the action is , and the reward is the emotion relief degree . The value Q of the state-action pair is updated according to the following formula: ;

[0093] wherein is the value of the state-action pair, represents the long-term return of the state under the action . represents the current state, which is composed of the emotion label of the target object and the personality portrait of the target object (for example, ). represents the current action, such as “telling stories” and the like. represents a learning rate, which is in a range from 0 to 1 (for example, 0.1 and the like), and is used to control the Q value update speed. represents an immediate reward, which is calculated from the emotion relief degree (for example, R=0.8 represents that the emotion is improved by 80%). represents a discount factor, which is in a range from 0 to 1 (for example, 0.9 and the like), and is used to balance the immediate and future returns. represents the next state , and represents the optimal Q value of all possible actions of the next state, which is used to predict the return of the best future strategy.

[0094] In the above manner, the preset reinforcement learning algorithm can be used to optimize and learn the candidate interaction strategy, so as to determine the optimal target interaction strategy from the candidate interaction strategy.

[0095] In an optional embodiment, after the target reply content is generated based on the emotional label of the target object, the personality portrait of the target object, and the target interaction strategy in the above step S104, the method further includes:

[0096] obtaining an emotional trend factor and an emotional fluctuation range of the target object;

[0097] In the case where the emotional trend factor of the target object is less than a first preset threshold, triggering a warning notification to warn the emotional change of the target object;

[0098] In the case where the emotional fluctuation range of the target object is greater than a second preset threshold, generating dynamic suggestion content based on the emotional label of the target object, the personality portrait of the target object, and the dialogue context content of the target object in the current dialogue scenario; and,

[0099] generating a visual score based on the emotional label of the target object, the personality portrait of the target object, and the emotional trend factor of the target object.

[0100] Specifically, the first preset threshold and the second preset threshold can be set according to actual needs, which are not limited here. The emotional trend factor refers to the emotional change trend of the target object in a relatively short time, such as changing from “happy” to “anxious” in 5 minutes. The emotional trend factor can be calculated using the exponential smoothing method on the emotional historical data . As an optional implementation, when the emotional trend factor , a warning notification (such as “child's emotion may decline, please pay attention” and the like) can be triggered, and the refresh rate can be 0.5 seconds / time. The emotional fluctuation range refers to the degree of emotional change of the target object. As an optional implementation, when the emotional fluctuation range , dynamic suggestion content (such as “learning after emotional depression, suggest praise for effort” and the like) can be generated based on the emotional label E of the target object, the personality portrait P of the target object, and the dialogue context content of the target object in the current dialogue scenario, through matching matrix . Of course, the visual score can also be generated in real time based on the emotional label E of the target object, the personality portrait P of the target object, and the emotional trend factor of the target object using the following formula:

[0101] ;

[0102] wherein, represents a visualization score, which integrates the emotion label E of the target object, the personality portrait P of the target object, and the emotion trend factor of the target object 3 parameters, with a value range of 0 to 1, for dashboard display. represents a personality weight, with a value range of 0 to 1 (such as 0.5, etc.), for adjusting the influence of personality on the score. represents a trend weight, with a value range of 0 to 1 (such as 0.4, etc.), for highlighting the importance of trend analysis. represents , and the maximum value among the three, for normalization processing. represents an emotion trend factor, with a calculation formula of ). represents the emotion value at time t (such as 0.8). represents the emotion value at time t-1, with an initial value of 0. represents a smoothing coefficient, with a value range of 0 to 1 (such as 0.3, etc.), for balancing the proportion of current and historical data.

[0103] In the above manner, the emotion of the target object can be deeply analyzed, and the warning notification and dynamic suggestion content can be triggered and generated accordingly, providing precise monitoring and intervention support for parents.

[0104] In an optional embodiment, a smart speaker can be used as a smart dialogue device to provide emotion management and education support for children aged 3-6. The smart dialogue device can include an emotion recognition module, a personality adaptive fine-tuning module, an interaction strategy matching module, a context adaptive generation module, and a parent intelligent support module. The specific implementation steps are as follows:

[0105] Step S201, the emotion recognition module collects the voice and text input of the child, extracts the voice feature vector and text feature vector using the improved Wav2Vec 2.0 model and lightweight DistilBERT model respectively, and generates an emotion label through attention mechanism fusion.

[0106] Device preparation: use a smart speaker equipped with a microphone (such as model XYZ-200, sampling rate 16kHz), run an embedded Linux system, and install the system software.

[0107] Specific actions:

[0108] 1. The child says "I don't like kindergarten" in front of the smart speaker, and the microphone collects the voice signal with a frame length of 10ms to generate an original audio file (WAV format).

[0109] 2. The speech feature extraction submodule in the emotion recognition module loads the improved Wav2Vec 2.0 model (with about 50M parameters) and calculates the spectral features (MFCC, 40 dimensions), pitch changes ( , unit Hz / s) and volume intensity (Volume, unit dB), output speech feature vector .

[0110] 3. The text feature extraction submodule in the emotion recognition module converts the audio into the text "I don't like kindergarten" through speech-to-text conversion, loads the lightweight DistilBERT model (with about 66M parameters), and extracts word embeddings ( , 768 dimensions), sentence length ( , 4 words) and context score ( , such as calculating based on the previous sentence "I went to kindergarten today"), output text feature vector .

[0111] Results: Fusion of sentiment analysis submodules in the sentiment recognition module and text feature vector , through the formula , the output emotion label is "sad" with a confidence level of 0.85.

[0112] Step S202: The personality adaptive fine-tuning module analyzes the child's conversation data, constructs a personality profile, and dynamically adjusts the recognition threshold and conversation style based on the personality profile.

[0113] Equipment preparation: The smart speaker is connected to a cloud database that stores historical conversation data of children (e.g., 10 interaction records).

[0114] Specific actions:

[0115] 1. Extract the children's conversation records from the database (e.g. 5 short sentences, slow speaking speed) and calculate their activity level. , expression tendency , emotional stability , generate personality portraits , and judged the child as "introverted".

[0116] 2. Adaptive fine-tuning mechanism adjusts parameters according to P: lowering the tone threshold , reducing the dependence on pitch changes; updating model parameters through the formula ( ), optimize the emotion recognition model.

[0117] Results: After adjusting the model parameters, the system's confidence in identifying "sadness" in young children can be improved from 0.85 to 0.90.

[0118] Step S203, the interactive strategy matching module optimizes the output interactive strategy based on the emotion label and the personality portrait using the Q-Learning optimization algorithm.

[0119] Device preparation: the smart speaker loads the emotion-strategy mapping matrix (5x20 dimensions, pre-training data can be obtained based on Piaget's theory).

[0120] Specific actions:

[0121] 1. Input the emotion label "sadness" and the personality portrait , query the matrix , and get the candidate interactive strategies, such as "comforting conversation" and "playing music".

[0122] 2. Q-Learning optimization: state , action a = "comforting conversation", reward (based on subsequent emotion improvement), update Q value .

[0123] 3. Output strategy: the system selects "comforting conversation" as the target interactive strategy and generates candidate reply content such as "Are you a little sad?".

[0124] Result: after the strategy is executed, the child's emotion improves to "calm", and the adjustment time is 1 minute.

[0125] Step S204, the context-adaptive generation module generates target reply content by combining context-adaptive generation and lightweight Transformer context enhancement.

[0126] Specific actions:

[0127] 1. Input the dialogue features and the personality portrait , dynamically adjust the generation strategy: for short sentence input (such as "don't want to play"), prefer concise replies (such as "OK, change one"); for long sentences (such as "I painted in kindergarten today"), generate detailed responses (such as "Great, what did you paint?"). Combine the child's current multi-round context, such as , use a lightweight Transformer encoder to calculate the context relevance score , and integrate it into the generation process.

[0128] 2. According to the input short sentence "don't want to play", generate the target reply content "OK, change one" through the formula . Among them, represents the final dialogue reply content. represents the candidate reply content generated by the decoder. is the conditional probability, representing the suitability of the reply content under the dialogue feature , context sequence and personality profile P. is the dialogue feature, which integrates the speech feature vector and the text feature vector . is the context sequence, such as the previous 3 rounds of dialogue features and emotion labels (e.g. . is the personality profile, which guides the tone adjustment. is the decoder weight matrix, which matches the size of the vocabulary (e.g. 512 × vocab_size, etc.). is the context relevance weight, which ranges from 0 to 1 (e.g. 0.2), used to adjust the influence of context on the reply. is the context relevance score, which is calculated by the Transformer. is the selected reply with the highest comprehensive probability.

[0129] 3. Tone adjustment is mild (adapted to introverted personality), output audio (speech rate 1.2 times normal, volume -3dB).

[0130] Result: Generate reply content "OK".

[0131] Step S205, the parent intelligent support module analyzes the emotion trend by exponential smoothing method, calculates the trend factor, generates trigger warning and scene suggestion, and generates growth report to support parent intervention.

[0132] Device preparation: smart speaker connects parent mobile application, real-time data synchronization.

[0133] Specific actions:

[0134] 1. The emotion trend analysis module processes historical emotion data, such as , calculates the trend factor (where ), and gets .

[0135] 2. Check trigger conditions: when the emotion fluctuation , according to the emotion label "sadness", the personality profile and the dialogue context content in the current dialogue scene, generate dynamic suggestion content such as "deserve praise", and push it to the parent mobile application.

[0136] 3. Generate weekly report: such as emotion stability , participation .

[0137] Results: Parents received recommendations and implemented them, and the emotional improvement of young children was observed, with an acceptance rate of 85%.

[0138] As can be seen, by improving the Wav2Vec 2.0 model and the lightweight DistilBERT model, extracting voice features and text features, and using attention mechanism fusion, the accuracy of recognition is improved. This adaptability solves the problem of insufficient understanding of children's dialogue features in existing technologies, laying a precise foundation for subsequent educational interventions. Moreover, by constructing a portrait of children's individuality (such as indicators of activity level, expression tendency, and emotional stability), dynamically adjusting the emotional recognition threshold and dialogue style. For example, lowering the tone threshold for introverted children and increasing the vocabulary weight for extroverted children, etc., to improve individual adaptability. This online learning mechanism ensures that model parameters can be optimized with interaction, surpassing traditional unified models, significantly enhancing the personalization of interaction and educational effectiveness. In addition, by introducing emotion trend analysis and scenario triggering mechanism, using exponential smoothing method to predict emotion changes, and triggering warning and suggestions when emotion fluctuation exceeds 0.2 or emotion trend is lower than -0.15, and can combine scene context (such as "after learning"), generate precise intervention strategies (such as "deserve praise"), which is more intelligent than static suggestions, and can better help parents understand the emotional changes of children, effectively improving user experience.

[0139] Reference Figure 3 , Figure 3 A structural schematic diagram of an intelligent dialogue device provided by the embodiments of the present application is shown in FIG. 1. As shown in FIG. 1, the intelligent dialogue device 300 includes: Figure 3

[0140] A first acquisition module 301 is configured to acquire dialogue content currently input by a target object and historical dialogue data, wherein the target object is any young child.

[0141] A first determination module 302 is configured to input the dialogue content into a pre-trained emotion recognition model to obtain an emotional label of the target object, and determine a personal portrait of the target object based on the historical dialogue data, wherein the emotion recognition model is pre-trained based on a large number of young children dialogue samples, and the model parameters of the emotion recognition model can be adjusted based on the personal portrait of the target object.

[0142] A second determination module 303 is configured to determine a candidate interaction strategy based on the emotional label of the target object and a preset mapping relationship, and optimize learning the candidate interaction strategy based on the emotional label of the target object and the personal portrait of the target object to determine a target interaction strategy, wherein the preset mapping relationship is used to represent the mapping relationship between the emotional label and the interaction strategy.

[0143] A first generation module 304 is configured to generate target reply content based on the emotional label of the target object, the personal portrait of the target object, and the target interaction strategy.​

[0144] Further, the emotion recognition model comprises a speech feature extraction sub-model, a text feature extraction sub-model, and an emotion analysis sub-model; the first determination module 302 comprises:

[0145] a first input sub-module, configured to input speech information in the dialogue content into the speech feature extraction sub-model to obtain spectrum features, pitch changes, and volume intensities corresponding to the speech information, and determine a speech feature vector corresponding to the target object based on the spectrum features, the pitch changes, and the volume intensities corresponding to the speech information;

[0146] a second input sub-module, configured to input text information in the dialogue content into the text feature extraction sub-model to obtain word embeddings, sentence lengths, and context scores corresponding to the text information, and determine a text feature vector corresponding to the target object based on the word embeddings, the sentence lengths, and the context scores corresponding to the text information;

[0147] a third input sub-module, configured to input the speech feature vector and the text feature vector into the emotion analysis sub-model to obtain an emotion label of the target object.

[0148] Further, the third input sub-module is specifically configured to:

[0149] input the speech feature vector and the text feature vector into the emotion analysis sub-model, and perform feature fusion according to a formula using the emotion analysis sub-model to obtain the emotion label of the target object;

[0150] wherein, denotes the emotion label of the target object, denotes a dynamic attention weight, denotes a normalization function, denotes a weight matrix of speech features, denotes the speech feature vector, denotes a weight matrix of text features, denotes the text feature vector.

[0151] Further, the first determination module 302 further comprises:

[0152] a calculation sub-module, configured to calculate an activity, an expression tendency, and an emotion stability corresponding to the target object based on historical dialogue data, wherein the activity is used to represent an activity degree of the target object in participating in the dialogue, the expression tendency is used to represent a relative dependence degree of the target object on tone and vocabulary in language expression, and the emotion stability is used to represent an emotion fluctuation degree of the target object;

[0153] a determination sub-module, configured to determine a personal portrait of the target object based on the activity, the expression tendency, and the emotion stability corresponding to the target object.

[0154] Further, the intelligent dialogue device 300 further comprises:

[0155] The adjusting module is configured to adjust the first parameter or the second parameter in the emotion recognition model based on the personality portrait of the target object, so that the adjusted emotion recognition model outputs an emotion label that is closer to the real emotion label of the target object and matches the personality portrait of the target object, wherein the first parameter is a related parameter corresponding to a weight matrix of a speech feature, and the second parameter is a related parameter corresponding to a weight matrix of a text feature.

[0156] Further, the second determining module 303 comprises:

[0157] The optimization learning sub-module is configured to perform optimization learning on the candidate interaction strategy by using a preset reinforcement learning algorithm, and determine the target interaction strategy from the candidate interaction strategy, wherein the reinforcement learning algorithm is an algorithm that uses the emotion label of the target object and the personality portrait of the target object as a state, an interaction strategy in the candidate interaction strategy as an action, and an emotion relief degree as a reward, selects the target interaction strategy by iteratively calculating the value of different state-action pairs, and the target interaction strategy is an interaction strategy corresponding to an optimal value.

[0158] Further, the intelligent dialogue device 300 further comprises:

[0159] The second obtaining module is configured to obtain an emotion trend factor and an emotion fluctuation range of the target object.

[0160] The triggering module is configured to trigger a warning notification to warn the change of the emotion of the target object when the emotion trend factor of the target object is less than a first preset threshold.

[0161] The second generating module is configured to generate dynamic suggestion content based on the emotion label of the target object, the personality portrait of the target object, and the dialogue context content of the target object in the current dialogue scene when the emotion fluctuation range of the target object is greater than a second preset threshold.

[0162] The third generating module is configured to generate a visual score based on the emotion label of the target object, the personality portrait of the target object, and the emotion trend factor of the target object.

[0163] It should be noted that the intelligent dialogue device 300 can implement the steps of the intelligent dialogue method provided by any one of the foregoing method embodiments, and achieve the same technical effects, which will not be repeated here.

[0164] As Figure 4As shown, the embodiment of the present application further provides an electronic device, comprising a processor 411, a communication interface 412, a memory 413 and a communication bus 414, wherein the processor 411, the communication interface 412 and the memory 413 complete mutual communication through the communication bus 414,

[0165] The memory 413 is used for storing a computer program.

[0166] In an embodiment of the present application, the processor 411 is used for executing the program stored in the memory 413, and realizes the intelligent conversation method provided by any one of the foregoing method embodiments.

[0167] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the intelligent conversation method provided by any one of the foregoing method embodiments.

[0168] It should be noted that, in this document, relational terms such as "first" and "second", and the like, are used solely to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. In addition, the terms "includes", "including", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0169] The above descriptions are only specific embodiments of the present application to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent dialogue method, characterized in that: The method comprises: Obtaining the current conversation content and historical conversation data input by the target object, wherein the target object is any child; Inputting the conversation content into a pre-trained emotion recognition model to obtain an emotion label for the target subject, and determining a personality profile of the target subject based on the historical conversation data, wherein the emotion recognition model is pre-trained based on a large number of children's conversation samples, and the model parameters of the emotion recognition model can be adjusted based on the personality profile of the target subject; Determining candidate interaction strategies based on the target object's emotion label and a preset mapping relationship, and optimizing and learning the candidate interaction strategies based on the target object's emotion label and the target object's personality profile to determine a target interaction strategy, wherein the preset mapping relationship is used to characterize the mapping relationship between the emotion label and the interaction strategy; Generate target reply content based on the target object's emotion tag, the target object's personality portrait, and the target interaction strategy; The step of determining the personality profile of the target object based on the historical conversation data includes: Based on the historical conversation data, the activity level, expression tendency, and emotional stability corresponding to the target object are calculated respectively, wherein the activity level is used to represent the degree of activeness of the target object in participating in the conversation, the expression tendency is used to represent the relative reliance of the target object on tone and vocabulary in language expression, and the emotional stability is used to represent the degree of emotional fluctuation of the target object; Determining a personality profile of the target object based on the target object's corresponding activity, expression tendency, and emotional stability; After determining the personality profile of the target object based on the historical conversation data, the method further includes: Based on the personality portrait of the target object, the first parameter or the second parameter in the emotion recognition model is adjusted so that the emotion label output by the adjusted emotion recognition model is closer to the true emotion label of the target object and matches the personality portrait of the target object, wherein the first parameter is the relevant parameter corresponding to the weight matrix of the speech feature, and the second parameter is the relevant parameter corresponding to the weight matrix of the text feature.

2. The intelligent dialogue method according to claim 1, characterized in that: The emotion recognition model includes a speech feature extraction sub-model, a text feature extraction sub-model and an emotion analysis sub-model; Inputting the conversation content into a pre-trained emotion recognition model to obtain the emotion label of the target object includes: Inputting the speech information in the conversation content into the speech feature extraction sub-model to obtain spectral features, pitch changes, and volume intensity corresponding to the speech information, and determining a speech feature vector corresponding to the target object based on the spectral features, pitch changes, and volume intensity corresponding to the speech information; Inputting the text information in the conversation content into the text feature extraction sub-model to obtain word embeddings, sentence lengths, and context scores corresponding to the text information, and determining a text feature vector corresponding to the target object based on the word embeddings, sentence lengths, and context scores corresponding to the text information; The speech feature vector and the text feature vector are input into the emotion analysis sub-model to obtain the emotion label of the target object.

3. The intelligent dialogue method according to claim 2, characterized in that: Inputting the speech feature vector and the text feature vector into the emotion analysis sub-model to obtain the emotion label of the target object includes: The speech feature vector and the text feature vector are input into the emotion analysis sub-model, and the emotion analysis sub-model is used according to the formula Perform feature fusion to obtain the emotion label of the target object; in, represents the emotional label of the target object, represents the dynamic attention weight, represents the normalization function, The weight matrix representing the speech features, represents the speech feature vector, The weight matrix representing the text features, Represents the text feature vector.

4. The intelligent dialogue method according to claim 1, characterized in that: The optimizing and learning the candidate interaction strategies based on the emotion label of the target object and the personality portrait of the target object to determine the target interaction strategy includes: The candidate interaction strategies are optimized and learned using a preset reinforcement learning algorithm, and the target interaction strategy is determined from the candidate interaction strategies. The reinforcement learning algorithm uses the emotional label of the target object and the personality portrait of the target object as states, the interaction strategies in the candidate interaction strategies as actions, and the degree of emotional relief as rewards. The target interaction strategy is selected by continuously iteratively calculating the values ​​of different state-action pairs. The target interaction strategy is the interaction strategy corresponding to the optimal value.

5. The intelligent dialogue method according to claim 1, characterized in that: After generating target reply content based on the target object's emotion tag, the target object's personality portrait, and the target interaction strategy, the method further includes: Obtaining the target object's emotional trend factor and emotional fluctuation range; When the target object's emotion trend factor is less than a first preset threshold, triggering an early warning notification to warn of the target object's emotion change; If the target object's emotional fluctuation range is greater than a second preset threshold, generating dynamic suggestion content based on the target object's emotional label, the target object's personality portrait, and the target object's conversation context in the current conversation scenario; and A visualization score is generated based on the emotion label of the target object, the personality portrait of the target object, and the emotion trend factor of the target object.

6. An intelligent dialogue device, characterized in that: The device comprises: A first acquisition module is used to acquire the conversation content currently input by the target object and historical conversation data, wherein the target object is any child; a first determination module, configured to input the conversation content into a pre-trained emotion recognition model to obtain an emotion label of the target subject, and determine a personality profile of the target subject based on the historical conversation data, wherein the emotion recognition model is pre-trained based on a large number of children's conversation samples, and model parameters of the emotion recognition model can be adjusted based on the personality profile of the target subject; a second determination module, configured to determine candidate interaction strategies based on the target object's emotion label and a preset mapping relationship, and optimize and learn the candidate interaction strategies based on the target object's emotion label and the target object's personality profile to determine a target interaction strategy, wherein the preset mapping relationship is used to represent the mapping relationship between the emotion label and the interaction strategy; A first generating module is configured to generate target reply content based on the target object's emotion label, the target object's personality portrait, and the target interaction strategy; The first determining module includes: a calculation submodule, configured to calculate, based on the historical conversation data, the activity level, expression tendency, and emotional stability corresponding to the target subject, wherein the activity level is used to represent the target subject's activeness in participating in the conversation, the expression tendency is used to represent the target subject's relative reliance on tone and vocabulary in language expression, and the emotional stability is used to represent the target subject's emotional fluctuations; A determination submodule, configured to determine a personality profile of the target object based on the target object's corresponding activity, expression tendency, and emotional stability; Wherein, the intelligent dialogue device further includes: An adjustment module is used to adjust the first parameter or the second parameter in the emotion recognition model based on the personality portrait of the target object, so that the emotion label output by the adjusted emotion recognition model is closer to the real emotion label of the target object and matches the personality portrait of the target object, wherein the first parameter is the relevant parameter corresponding to the weight matrix of the speech feature, and the second parameter is the relevant parameter corresponding to the weight matrix of the text feature.

7. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; The processor is configured to implement the intelligent dialogue method according to any one of claims 1 to 5 when executing a program stored in the memory.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intelligent dialogue method according to any one of claims 1 to 5 is implemented.

Images