Artificial intelligence interaction method and device based on multiple analysis models

Through the artificial intelligence interaction method of multi-analysis models, the problem of insufficient context understanding and common sense reasoning capabilities in complex dialogues is solved. Through word embedding and generative models, intelligent dialogue strategies are generated, and the accuracy and emotion of human-computer interaction are improved. Identification ability.

CN118312593BActive Publication Date: 2024-11-19TIDEMEDIA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410456651.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-16
Publication Date
2024-11-19
Estimated Expiration
2044-04-16

AI Technical Summary

Technical Problem

Existing dialogue systems lack the ability to handle contextual understanding and common sense reasoning in complex dialogues, especially poor performance in long dialogues, and overfitting and lack of diversity data lead to performance degradation.

Method used

Using an artificial intelligence interaction method based on multi-analysis models, natural language text is mapped to continuous vector space through word embedding rules, context encoding is used using preset recurrent neural networks, and intelligent dialogue strategies are generated by combining generative adversarial networks and automatic regression models. And improve the accuracy of the reply content through iterative adjustments.

Benefits of technology

It improves the accuracy of human-computer interaction and emotional recognition capabilities, enhances the contextual understanding and common sense reasoning capabilities of the dialogue system, reduces the risk of overfitting, and adapts to diverse emotional scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118312593B_ABST
    Figure CN118312593B_ABST
Patent Text Reader

Abstract

The embodiment of the present application provides an artificial intelligence interaction method and device based on a multi-analysis model, the method comprising: mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to a word embedding rule, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic emotion; training a generative adversarial network and an auto-regression model respectively according to the reply type and reply style of the historical natural language text, combining the model according to the output of the generator of the generative adversarial network and the auto-regression model, iteratively adjusting the combined model according to the reply content, and obtaining an intelligent dialogue strategy; the present application can effectively improve the accuracy of human-computer interaction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically to an artificial intelligence interaction method and device based on multiple analysis models. Background Art

[0002] Natural Language Processing (NLP) is one of the core technologies of intelligent dialogue. By performing word segmentation, part-of-speech tagging, named entity recognition and other processing on the input text, the system can understand the text content input by the user and extract key information.

[0003] Intelligent dialogue technology is one of the research hotspots in the field of natural language processing (NLP) and artificial intelligence (AI), and has made remarkable progress. However, there are still some challenges and shortcomings.

[0004] With the development of deep learning technology, natural language processing (NLP) has made significant progress, and the introduction of pre-trained models (such as BERT and GPT) has improved the language understanding ability of dialogue systems. The model can understand the context, process complex sentence structures, and has a certain semantic understanding ability.

[0005] However, the inventors found that despite the great success of pre-trained models, there are still challenges in dealing with contextual understanding in complex conversations, especially in long conversations. At the same time, the performance of most dialogue systems is highly dependent on large-scale annotated datasets. The lack of sufficiently diverse data may lead to overfitting of the model. In addition, most dialogue systems lack common sense reasoning capabilities, resulting in poor performance when dealing with some dialogue scenarios that require common sense. Summary of the invention

[0006] In response to the problems in the prior art, the present application provides an artificial intelligence interaction method and device based on multiple analysis models, which can effectively improve the accuracy of human-computer interaction.

[0007] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides an artificial intelligence interaction method based on a multi-analysis model, comprising:

[0009] According to the word embedding rules, each word in the segmented natural language text is mapped to a continuous vector space to form a sentence, the sentence is context-encoded according to a preset recurrent neural network to obtain a corresponding semantic representation, the semantic representation is judged, and the corresponding semantic sentiment is determined, wherein the recurrent neural network is trained by historical natural language text;

[0010] According to the reply type and reply style of the historical natural language text, a generative adversarial network and an auto-regression model are trained respectively, and a model combination is performed according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, and the combined model is iteratively adjusted according to the reply content to obtain an intelligent dialogue strategy;

[0011] According to the specific type of the semantic emotion, the matching degree of each intelligent dialogue strategy in the set intelligent dialogue library is determined, and the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library is determined as the intelligent dialogue strategy corresponding to the semantic emotion, and the corresponding reply content is determined according to the intelligent dialogue strategy and the natural language text.

[0012] Furthermore, mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to the word embedding rule includes:

[0013] Receive a natural language text sent by a user and perform word segmentation on the natural language text to obtain individual words;

[0014] By setting the semantic relationship between words in the word embedding model, each word is mapped to a high-dimensional real number vector in a continuous vector space, and the vector representation of the entire sentence is determined by the vector combination of the high-dimensional real number vectors of each word.

[0015] Furthermore, the context encoding of the sentence is performed according to a preset recurrent neural network to obtain a corresponding semantic representation, the semantic representation is judged, and the corresponding semantic emotion is determined, wherein the recurrent neural network is trained by historical natural language texts, and includes:

[0016] Performing context encoding on the vector representation of the entire sentence through a preset recurrent neural network to obtain a sequential relationship between the words in the sentence, and determining a semantic representation of the sentence according to the sequential relationship between the words;

[0017] The supervised learning model obtained by training the training data with emotion labels is used to perform emotion judgment on the semantic representation of the sentence to determine the corresponding semantic emotion.

[0018] Furthermore, the generative adversarial network and the auto-regression model are trained respectively according to the reply type and reply style of the historical natural language text, including:

[0019] According to each response type and each response style in the historical natural language text, a corresponding generator network and a discriminator network are trained, and a generative adversarial network is obtained by iteratively tuning a generator loss function of the generator network and a discriminator loss function of the discriminator network;

[0020] The preset automatic regression model is trained according to each response type and each response style in the historical natural language text, and the regression loss function of the automatic regression model is iteratively tuned according to the result of the model training to obtain the automatic regression model.

[0021] Furthermore, the model combination is performed based on the output of the generator of the generative adversarial network and the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, including:

[0022] Using the output of the generator of the generative adversarial network as the input of the autoregressive model;

[0023] According to the output of the auto-regression model and the data feedback of the discriminator of the generative adversarial network, the reply content output by the generative adversarial network is obtained.

[0024] Furthermore, the iterative adjustment of the combined model according to the reply content to obtain an intelligent dialogue strategy includes:

[0025] Adjusting the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model according to user feedback on the reply content;

[0026] According to the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model after the adjustment, the combined model and the intelligent dialogue strategy with the specific response type and specific response style output by the model are obtained.

[0027] Further, the determining of the matching degree of each intelligent dialogue strategy in the set intelligent dialogue library according to the specific type of the semantic emotion, and determining the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, includes:

[0028] Calculating the similarity matching degree between the specific type of the semantic emotion and each of the intelligent dialogue strategies in the set intelligent dialogue library;

[0029] The intelligent dialogue strategy with the greatest similarity matching degree is determined as the intelligent dialogue strategy corresponding to the semantic emotion.

[0030] In a second aspect, the present application provides an artificial intelligence interaction device based on a multi-analysis model, comprising:

[0031] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the artificial intelligence interaction method based on multiple analysis models when executing the program.

[0032] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the artificial intelligence interaction method based on multiple analysis models.

[0033] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the artificial intelligence interaction method based on multiple analysis models.

[0034] It can be seen from the above technical scheme that the present application provides an artificial intelligence interaction method and device based on a multi-analysis model, by mapping each word in the segmented natural language text into a continuous vector space according to the word embedding rule and forming a sentence, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic emotion; training a generative adversarial network and an automatic regression model respectively according to the reply type and reply style of the historical natural language text, and combining the model according to the output of the generator of the generative adversarial network and the automatic regression model, iteratively adjusting the combined model according to the reply content, and obtaining an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 This is one of the flowcharts of the artificial intelligence interaction method based on multiple analysis models in the embodiment of the present application;

[0037] Figure 2 The second flowchart of the artificial intelligence interaction method based on multiple analysis models in the embodiment of the present application;

[0038] Figure 3The third flowchart of the artificial intelligence interaction method based on multiple analysis models in the embodiment of the present application;

[0039] Figure 4 This is a fourth flow chart of the artificial intelligence interaction method based on multiple analysis models in an embodiment of the present application;

[0040] Figure 5 This is a fifth flow chart of the artificial intelligence interaction method based on multiple analysis models in the embodiment of the present application;

[0041] Figure 6 This is the sixth flow chart of the artificial intelligence interaction method based on multiple analysis models in the embodiment of the present application;

[0042] Figure 7 FIG7 is a flowchart of an artificial intelligence interaction method based on multiple analysis models in an embodiment of the present application;

[0043] Figure 8 It is a structural diagram of an artificial intelligence interaction device based on multiple analysis models in an embodiment of the present application;

[0044] Fig. 9 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0047] Considering that although the pre-training model has achieved great success in the prior art, there are still challenges in dealing with the context understanding problem in complex dialogues, especially the context understanding in long dialogues. At the same time, the performance of most dialogue systems is highly dependent on large-scale annotated data sets. The lack of sufficiently diverse data may lead to overfitting of the model. In addition, most dialogue systems lack common sense reasoning capabilities, resulting in poor performance when dealing with some dialogue scenarios that require common sense. The present application provides an artificial intelligence interaction method and device based on a multi-analysis model, which maps each word in the segmented natural language text to a continuous vector space and forms a sentence according to the word embedding rule, context-encodes the sentence according to a preset recurrent neural network, obtains the corresponding semantic representation, judges the semantic representation, and determines the corresponding semantic emotion; trains a generative adversarial network and an auto-regression model according to the reply type and reply style of the historical natural language text, and combines the model according to the output of the generator of the generative adversarial network and the auto-regression model, and iteratively adjusts the combined model according to the reply content to obtain an intelligent dialogue strategy, thereby the present application can effectively improve the accuracy of human-computer interaction.

[0048] It is understandable that in order to make the intelligent dialogue system more humane, human-computer interaction and emotion recognition technology are crucial. By analyzing the user's voice, text, expression and other information, the system can identify the user's emotional state, such as happiness, anger, sadness, etc., and adjust the dialogue strategy according to the emotional state. This helps to improve the user's comfort and satisfaction when communicating with the intelligent dialogue system. In recent years, human-computer interaction and emotion recognition technology based on deep learning has made significant progress, such as deep convolutional neural network (DCNN), recurrent neural network (RNN), etc., which brings a richer emotional communication experience to the intelligent dialogue system.

[0049] Natural Language Inference (NLI) is a task in the field of natural language processing. Its goal is to understand and determine the logical relationships between texts, including entailment, contradiction, and neutral relationships. Semantic analysis plays a key role in natural language inference models, helping the model understand the semantic information in the text.

[0050] In order to effectively improve the accuracy of human-computer interaction, the present application provides an embodiment of an artificial intelligence interaction method based on a multi-analysis model, see Figure 1 The artificial intelligence interaction method based on multiple analysis models specifically includes the following contents:

[0051] Step S101: Mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to a word embedding rule, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text;

[0052] Optionally, in this embodiment, first, the natural language text after segmentation is processed. Based on the word segmentation library, open source tools such as jieba are used to perform Chinese word segmentation, or spaces are used to perform English word segmentation. A word list after word segmentation is obtained. Then, each word is mapped to a continuous vector space using a pre-trained word embedding model, such as Word2Vec or BERT. This mapping can be obtained by querying the weight matrix of the embedding model. These word vectors are combined in order to form a representation of the entire sentence. Thus, this embodiment uses word embedding rules to map vocabulary in natural language to dense vectors, retaining semantic information. The continuous vector representation better captures the semantic relationship between words, making the overall representation of the sentence richer.

[0053] For example, suppose the input text is: "This movie is really wonderful!", after word segmentation, we get ["this", "movie", "really", "too", "wonderful", "!"]. Through the Word2Vec model, "movie" may be mapped to [0.2, 0.5, -0.3, ...].

[0054] Optionally, in this embodiment, a recurrent neural network (RNN) can be used to contextually encode the mapped sentence vector. The recurrent neural network processes the sentence word by word through a recurrent unit, gradually updates the implicit state, and forms an abstract representation of the context of the entire sentence. This context encoding process can be implemented through models such as LSTM (Long Short-Term Memory Network) or GRU (Gated Recurrent Unit). As a result, the recurrent neural network can capture long-range dependencies in the text and better understand the context of the sentence. Context encoding provides an abstract representation of the overall semantics of the sentence, providing strong support for subsequent sentiment judgment.

[0055] For example, suppose context encoding is performed through LSTM, and the context of the sentence is represented as a vector of fixed length, such as [0.1, -0.2, 0.5, ...].

[0056] Optionally, in this embodiment, a pre-trained sentiment classification model is used for sentiment determination. This model can be a binary classification (positive / negative) or multi-classification (positive / neutral / negative) model. The semantic representation after context encoding is input, and the model outputs the corresponding sentiment label. As a result, the sentiment determination model can accurately identify the sentiment polarity of the text and improve the system's ability to understand the user's emotions.

[0057] For example, if the sentiment classification model determines that the above sentence has a positive sentiment, then the determined semantic sentiment is "Positive".

[0058] Through the above steps, this embodiment realizes the whole process from natural language text input by the user to semantic representation in continuous vector space and then to emotion judgment. This provides a solid foundation for the subsequent intelligent dialogue system, enabling the system to more accurately understand the semantic emotions input by the user and provide more appropriate support for intelligent replies.

[0059] Step S102: training a generative adversarial network and an auto-regression model respectively according to the reply type and reply style of the historical natural language text, and combining the models according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, and iteratively adjusting the combined model according to the reply content to obtain an intelligent dialogue strategy;

[0060] Optionally, in this embodiment, according to the different reply types and reply styles in the historical natural language text, the generative adversarial network (GAN) and the auto-regression model are trained respectively. For the reply type, it may include declarative reply, question reply, etc. For the reply style, it may include humorous style, formal style, etc. There is a corresponding training set for each type and style of reply. This can make the generated replies more suitable for different application scenarios, improving the adaptability and usability of the model.

[0061] For example, for declarative responses, the model may learn in the training set to focus more on factual expressions, while in humorous responses, the model may be more inclined to learn witty and humorous expressions.

[0062] Optionally, the present embodiment can also combine the generator of the generative adversarial network with the output of the autoregressive model. The function of the generator is to generate the initial reply, while the discriminator of the autoregressive model is responsible for evaluating the rationality of the reply. The output of the generator is fed back by the discriminator so that the generated reply is more consistent with the grammatical structure and contextual logic. Thus, by combining the generator and the discriminator, the system can generate a more logical and grammatically correct reply. If the reply output by the generator is too outrageous, the discriminator can point out the unreasonableness in time, so as to make iterative adjustments.

[0063] Then, this embodiment can also make the system's intelligent dialogue strategy gradually converge by continuously iteratively adjusting the parameters of the generator and the discriminator. This process may involve fine-tuning the model parameters of the generator and the feedback mechanism of the discriminator to improve the quality and logic of the response. The continuous iterative adjustment of the intelligent dialogue strategy enables the system to better adapt to user input in practical applications and improves the intelligence level of the system. Through iterative adjustment, the system can more accurately capture contextual information when the user asks a question and generate a more reasonable response that meets the user's expectations.

[0064] This series of steps, through the training of response types and styles, the combination of generators and discriminators, and iterative adjustments, ultimately resulted in a system with intelligent dialogue strategies. The system can generate logical and reasonable responses based on historical texts, improving the quality of human-computer interaction experience.

[0065] Step S103: Determine the degree of match between the intelligent dialogue strategies in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest degree of match in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

[0066] Optionally, in this embodiment, according to the specific type of the semantic emotion, the system sets a series of intelligent dialogue strategies for each emotion type in the intelligent dialogue library. Each intelligent dialogue strategy has a certain degree of matching, indicating that it is suitable for different semantic emotions. By setting a different intelligent dialogue strategy for each semantic emotion, the adaptability of the system to diverse emotions is improved.

[0067] For example, if a user expresses a happy emotion, the system may select the most matching "encouraging" intelligent dialogue strategy to better respond to the user's emotion.

[0068] Optionally, in this embodiment, the system can also determine the corresponding reply content according to the intelligent dialogue strategy with the highest matching degree. This step takes into account the characteristics of the intelligent dialogue strategy, so that the generated reply is more in line with the emotions and context expressed by the user. By matching the intelligent dialogue strategy, the system can generate a reply that is closer to the user's emotions, thereby improving the emotional intelligence level of the system.

[0069] For example, if a user expresses anxiety, the system may select the most matching "comforting" intelligent dialogue strategy and generate a response that can relieve the user's anxiety.

[0070] Through the above steps, this embodiment enables the system to better understand and reflect the user's semantic emotions when replying to the user through the matching of intelligent dialogue strategies and the determination of reply content. Such an intelligent dialogue system can better meet the user's personalized communication needs and improve the intelligence level of the system in emotional interaction.

[0071] From the above description, it can be seen that the artificial intelligence interaction method based on a multi-analysis model provided in the embodiment of the present application can map each word in the segmented natural language text into a continuous vector space and form a sentence according to the word embedding rule, context-encode the sentence according to the preset recurrent neural network, obtain the corresponding semantic representation, judge the semantic representation, and determine the corresponding semantic emotion; train the generative adversarial network and the automatic regression model according to the reply type and reply style of the historical natural language text, respectively, and combine the model according to the output of the generator of the generative adversarial network and the automatic regression model, iteratively adjust the combined model according to the reply content, and obtain an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction.

[0072] In order to accurately identify the user's emotional state, in one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 2 , and can also include the following:

[0073] Step S201: receiving a natural language text sent by a user and performing word segmentation on the natural language text to obtain individual words;

[0074] Step S202: Mapping each word to a high-dimensional real number vector in a continuous vector space by setting the semantic relationship between words in the word embedding model, and determining the vector representation of the entire sentence by the vector combination of the high-dimensional real number vectors of each word.

[0075] Optionally, in this embodiment, in this step, the system receives the natural language text sent by the user, and then uses advanced natural language processing technology to perform word segmentation. This process is intended to split the text input by the user into independent words, providing a basis for subsequent processing. A preliminary analysis of the text input by the user is achieved, providing basic data for subsequent processing.

[0076] In this step, the system uses a pre-set word embedding model to map each word obtained through word segmentation to a continuous vector space. The word embedding model can capture the semantic relationship between words and express it as a high-dimensional real number vector. It improves the ability to understand the semantics of words, converts discrete words into continuous vectors, and better expresses the semantic relationship between words.

[0077] For example, after mapping the words “happy” and “delight” to a continuous vector space, their vector representations become closer in the space, reflecting their semantic similarity.

[0078] These two steps enable the system to understand the language of the user input through the analysis of the user input text and the application of word embedding technology, and convert it into a continuous vector representation, providing input data for subsequent context encoding and semantic sentiment judgment. This helps the system better understand the user's intention and improves the accuracy of natural language processing.

[0079] In one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 3 , and can also include the following:

[0080] Step S301: context encoding the vector representation of the entire sentence through a preset recurrent neural network to obtain the order relationship of each word in the sentence, and determine the semantic representation of the sentence according to the order relationship of the words;

[0081] Step S302: Using a supervised learning model obtained by training the training data with emotion labels added thereto, an emotion judgment is performed on the semantic representation of the sentence to determine the corresponding semantic emotion.

[0082] Optionally, in this embodiment, in this step, we use a preset recurrent neural network (RNN) to contextually encode the vector representation of the sentence. First, we perform word segmentation to decompose the natural language text into individual words. Then, through the set word embedding model, we map each word to a high-dimensional real vector in a continuous vector space. These word vectors are combined to determine the vector representation of the entire sentence.

[0083] This preset RNN can effectively capture the semantic information of the sentence in the vector representation by learning the order relationship of each word in the entire sentence. Through iterative learning, RNN can encode the context of the sentence, making the semantics of the sentence richer and more accurate. This improves the understanding of the relationship between each word in the sentence. The semantic information of the sentence is effectively expressed, allowing the system to better understand the natural language text entered by the user.

[0084] For example, consider a sentence: "This movie is so wonderful!" After being processed by the preset RNN, the system can understand the relationship between "wonderful" and "movie" and better grasp the user's evaluation of the movie.

[0085] Optionally, this embodiment can also use a supervised learning model to perform sentiment judgment on the semantic representation of a sentence. First, we train the model by adding training data with sentiment labels. These sentiment-labeled data enable the model to learn to map the semantic representation of a sentence to the corresponding sentiment category.

[0086] Through supervised learning, we can make emotional judgments on the semantic representation of sentences, thereby determining the semantic emotions corresponding to the sentences. This step provides a basis for subsequent intelligent dialogue, enabling the system to more comprehensively grasp the emotional tendencies of user input sentences. Through supervised learning, the model can accurately determine the emotions of sentences. It provides important information about user emotions for subsequent intelligent dialogue strategies.

[0087] For example, consider the sentence: "I feel very satisfied with this product." Through the supervised learning model, the system can accurately determine that the sentiment of the sentence is "satisfied."

[0088] These two steps enable the system to understand the user's input sentences more deeply by using the preset recurrent neural network and supervised learning model, and provide a basis for the subsequent determination of semantic sentiment. This helps the system to more comprehensively grasp the user's emotional tendencies and make more accurate intelligent replies. The entire process ensures that the system conducts a full and detailed semantic analysis of the user's input.

[0089] In one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 4 , and can also include the following:

[0090] Step S401: training a corresponding generator network and a discriminator network according to each response type and each response style in the historical natural language text, and obtaining a generative adversarial network by iteratively tuning the generator loss function of the generator network and the discriminator loss function of the discriminator network;

[0091] Step S402: Perform model training on a preset automatic regression model according to each response type and each response style in the historical natural language text, and iteratively tune the regression loss function of the automatic regression model according to the result of the model training to obtain an automatic regression model.

[0092] Optionally, in this embodiment, first, a corresponding generator network and a discriminator network are constructed according to each response type and each response style in the historical natural language text. The generator network is responsible for generating text of a specific type and style, while the discriminator network is responsible for evaluating the difference between the generated text and the real text. The iterative tuning process of the loss function includes training the generator to deceive the discriminator, and training the discriminator to better identify the difference between the real text and the generated text. This process is repeated through iterations, so that the generator gradually generates more realistic text that conforms to a specific type and style.

[0093] Optionally, this embodiment can also perform model training on a preset automatic regression model according to each response type and each response style in the historical natural language text. The goal of model training is to enable the automatic regression model to generate text that matches the historical response type and style based on the input semantic representation. During the training process, by iteratively tuning the regression loss function, the model gradually learns to generate text that is consistent with the target response type and style. The key to this step is to train the model through supervised learning so that it can accurately generate response content that meets the requirements.

[0094] The combination of these two steps constructs a generative adversarial network and an autoregressive model as a whole. Through iterative tuning of the loss function, these two models gradually improve the quality and consistency of their generated texts, enabling them to better meet specific response types and style requirements.

[0095] Through supervised learning of historical natural language texts, generative adversarial networks and autoregressive models gradually learn the characteristics of different response types and styles during training, improving the quality and diversity of text generated by the system.

[0096] For example, in historical conversations, the system has learned the style of actively responding to users and the type of responses required when answering questions. The generative adversarial network generates more realistic active responses by deceiving the discriminator, while the autoregressive model learns through supervised learning how to generate text that matches the question answer type based on the semantic representation of the input. This enables the system to respond to users more flexibly and naturally in interactions.

[0097] In one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 5 , and can also include the following:

[0098] Step S501: using the output of the generator of the generative adversarial network as the input of the auto-regression model;

[0099] Step S502: Obtaining the response content output by the generative adversarial network according to the output of the auto-regression model and the data feedback of the discriminator of the generative adversarial network.

[0100] Optionally, in this embodiment, the output of the generator of the generative adversarial network is used as the input of the autoregressive model. This means that the text generated by the generative adversarial network is introduced into the autoregressive model for further processing and improvement. The text generated by the generative adversarial network may contain features of a specific response type and style, and the autoregressive model will process these features through the knowledge learned from its training.

[0101] This embodiment can also obtain the reply content output by the generative adversarial network based on the output of the auto-regressive model and the data feedback of the discriminator of the generative adversarial network. In this process, the auto-regressive model continuously iterates through the feedback information of the generated text, adjusts the output of the generative adversarial network to better meet the expectations of the discriminator, and thus generates reply content that is more in line with the characteristics of the real text.

[0102] Through the combination of these two steps, the generative adversarial network and the autoregressive model form a synergistic mechanism that leverages their respective strengths to improve the quality and consistency of the generated response content.

[0103] By combining generative adversarial networks and autoregressive models, the system can better maintain semantic consistency and conform to preset response types and styles when generating response content.

[0104] Example: The generative adversarial network generates a positive response. By introducing the auto-regression model, the system can further adjust the text to make it more consistent with the context and style of the historical conversation while maintaining a positive sentiment tendency. This combination mechanism makes the responses generated by the system more natural and accurate.

[0105] In one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 6 , and can also include the following:

[0106] Step S601: adjusting the creativity weight parameter of the generator in the generative adversarial network and the grammatical rule weight of the auto-regressive model according to the user feedback of the reply content;

[0107] Step S602: According to the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model after the adjustment, the combined model and the intelligent dialogue strategy of the specific response type and specific response style output by the model are obtained.

[0108] Optionally, in this embodiment, the system uses user feedback information to adjust the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the auto-regressive model. User feedback can include evaluation, correction or other forms of guidance information for the generated response. Based on these feedbacks, the system dynamically adjusts the creative weight parameters of the generator to increase or decrease the innovation of the generated response, and also adjusts the grammatical rule weights of the auto-regressive model to ensure that the generated response is more reasonable and natural in grammatical structure.

[0109] Optionally, this embodiment can obtain a combined model based on the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model after adjustment. This model outputs an intelligent dialogue strategy with a specific response type and a specific response style. By dynamically adjusting the creative weight parameters and grammatical rule weights, the system can better adapt to the user's preferences and the specific requirements of the dialogue scene, making the generated dialogue more personalized and close to practical applications.

[0110] By dynamically adjusting the creativity weight of the generator and the grammatical rule weight of the auto-regression model through user feedback, the system can better control the rationality of the grammatical structure while maintaining the diversity of generated content. This helps the generated dialogue content better meet user expectations and application scenario requirements.

[0111] For example, if a user provides feedback that the generated response is too innovative and wants to pay more attention to the accuracy of the grammatical structure, the system will reduce the creativity weight of the generator and increase the grammatical rule weight based on this feedback to produce conversation content that is more in line with user expectations. This feedback loop mechanism enables the system to continuously optimize the intelligent conversation strategy.

[0112] In one embodiment of the artificial intelligence interaction method based on multiple analysis models of the present application, see Figure 7 , and can also include the following:

[0113] Step S701: Calculate the similarity matching degree between the specific type of the semantic emotion and each of the intelligent dialogue strategies in the set intelligent dialogue library;

[0114] Step S702: Determine the intelligent dialogue strategy with the greatest similarity matching degree as the intelligent dialogue strategy corresponding to the semantic emotion.

[0115] Optionally, in this embodiment, the system calculates the similarity matching degree between the specific type of semantic emotion and each intelligent dialogue strategy in the intelligent dialogue library. The similarity matching degree can be calculated by various methods, such as cosine similarity, Euclidean distance, etc. The system maps the specific type of semantic emotion to a vector or feature representation, and compares it with the representation of each dialogue strategy in the intelligent dialogue library to calculate the similarity matching degree.

[0116] Optionally, the system of this embodiment determines the intelligent dialogue strategy with the maximum matching degree as the intelligent dialogue strategy corresponding to the semantic emotion according to the calculated similarity matching degree. This means that the system selects the dialogue strategy that best matches the user input in terms of semantic emotion to provide a more personalized and emotionally reasonable response.

[0117] By calculating the similarity matching degree, the system can more accurately select a dialogue strategy that matches the semantic sentiment of the user input, thereby improving the emotional coherence of the dialogue response and user experience.

[0118] For example, if the user's input expresses positive emotions, the system will calculate the similarity matching between each intelligent dialogue strategy and positive emotions, and finally select the intelligent dialogue strategy with the greatest similarity matching as the response. This helps the system better understand and respond to user emotional tendencies.

[0119] In order to effectively improve the accuracy of human-computer interaction, the present application provides an embodiment of an artificial intelligence interaction device based on a multi-analysis model for implementing all or part of the contents of the artificial intelligence interaction method based on a multi-analysis model, see Figure 8 The artificial intelligence interaction device based on multiple analysis models specifically includes the following contents:

[0120] A semantic sentiment determination module 10 is used to map each word in the segmented natural language text into a continuous vector space according to a word embedding rule to form a sentence, perform context encoding on the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judge the semantic representation, and determine the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text;

[0121] A dialogue strategy determination module 20 is used to train a generative adversarial network and an auto-regression model according to the response type and response style of the historical natural language text, and to combine the models according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate a response content, and to iteratively adjust the combined model according to the response content to obtain an intelligent dialogue strategy;

[0122] The intelligent matching module 30 is used to determine the matching degree of each intelligent dialogue strategy in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

[0123] From the above description, it can be seen that the artificial intelligence interaction device based on the multi-analysis model provided in the embodiment of the present application can map each word in the segmented natural language text into a continuous vector space according to the word embedding rules and form sentences, and context-encode the sentences according to the preset recurrent neural network to obtain the corresponding semantic representation, judge the semantic representation, and determine the corresponding semantic emotion; train the generative adversarial network and the automatic regression model according to the reply type and reply style of the historical natural language text, and combine the models according to the output of the generator of the generative adversarial network and the automatic regression model, and iteratively adjust the combined model according to the reply content to obtain an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction.

[0124] From the hardware level, in order to effectively improve the accuracy of human-computer interaction, the present application provides an embodiment of an electronic device for implementing all or part of the contents of the artificial intelligence interaction method based on multiple analysis models, and the electronic device specifically includes the following contents:

[0125] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the artificial intelligence interaction device based on multiple analysis models and related devices such as core business systems, user terminals and related databases; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., but the present embodiment is not limited thereto. In the present embodiment, the logic controller can be implemented with reference to the embodiment of the artificial intelligence interaction method based on multiple analysis models and the embodiment of the artificial intelligence interaction device based on multiple analysis models in the embodiment, and the contents thereof are incorporated herein, and the repeated parts are not repeated.

[0126] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0127] In practical applications, part of the artificial intelligence interaction method based on multiple analysis models can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0128] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0129] Fig. 9 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Fig. 9 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Fig. 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0130] In one embodiment, the artificial intelligence interaction method function based on multiple analysis models can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0131] Step S101: Mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to a word embedding rule, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text;

[0132] Step S102: training a generative adversarial network and an auto-regression model respectively according to the reply type and reply style of the historical natural language text, and combining the models according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, and iteratively adjusting the combined model according to the reply content to obtain an intelligent dialogue strategy;

[0133] Step S103: Determine the degree of match between the intelligent dialogue strategies in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest degree of match in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

[0134] From the above description, it can be seen that the electronic device provided by the embodiment of the present application maps each word in the segmented natural language text into a continuous vector space and forms a sentence according to the word embedding rule, context-encodes the sentence according to a preset recurrent neural network, obtains a corresponding semantic representation, judges the semantic representation, and determines the corresponding semantic emotion; trains a generative adversarial network and an automatic regression model according to the reply type and reply style of the historical natural language text, and combines the models according to the output of the generator of the generative adversarial network and the automatic regression model, iteratively adjusts the combined model according to the reply content, and obtains an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction.

[0135] In another embodiment, the artificial intelligence interaction device based on multiple analysis models can be configured separately from the central processing unit 9100. For example, the artificial intelligence interaction device based on multiple analysis models can be configured as a chip connected to the central processing unit 9100, and the function of the artificial intelligence interaction method based on multiple analysis models can be realized through the control of the central processing unit.

[0136] like Fig. 9 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Fig. 9 In addition, the electronic device 9600 may also include Fig. 9 For components not shown, reference may be made to the prior art.

[0137] like Fig. 9 As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0138] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0139] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0140] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0141] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0142] The communication module 9110 is a transmitter / receiver 9110 that sends and receives signals via an antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0143] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module, etc. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0144] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps of the artificial intelligence interaction method based on multiple analysis models in the above embodiments, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all the steps of the artificial intelligence interaction method based on multiple analysis models in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0145] Step S101: Mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to a word embedding rule, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text;

[0146] Step S102: training a generative adversarial network and an auto-regression model respectively according to the reply type and reply style of the historical natural language text, and combining the models according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, and iteratively adjusting the combined model according to the reply content to obtain an intelligent dialogue strategy;

[0147] Step S103: Determine the degree of match between the intelligent dialogue strategies in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest degree of match in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

[0148] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application maps each word in the segmented natural language text into a continuous vector space and forms a sentence according to the word embedding rule, context-encodes the sentence according to a preset recurrent neural network, obtains a corresponding semantic representation, judges the semantic representation, and determines the corresponding semantic emotion; trains a generative adversarial network and an automatic regression model according to the reply type and reply style of the historical natural language text, respectively, and combines the models according to the output of the generator of the generative adversarial network and the automatic regression model, and iteratively adjusts the combined model according to the reply content to obtain an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction.

[0149] The embodiments of the present application also provide a computer program product capable of implementing all the steps of the artificial intelligence interaction method based on multiple analysis models in the above embodiments, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the artificial intelligence interaction method based on multiple analysis models are implemented. For example, the computer program / instruction implements the following steps:

[0150] Step S101: Mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to a word embedding rule, context encoding the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judging the semantic representation, and determining the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text;

[0151] Step S102: training a generative adversarial network and an auto-regression model respectively according to the reply type and reply style of the historical natural language text, and combining the models according to the output of the generator of the generative adversarial network and the output of the auto-regression model, so that the output of the generator is fed back by the discriminator of the auto-regression model to generate reply content, and iteratively adjusting the combined model according to the reply content to obtain an intelligent dialogue strategy;

[0152] Step S103: Determine the degree of match between the intelligent dialogue strategies in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest degree of match in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

[0153] From the above description, it can be seen that the computer program product provided by the embodiment of the present application maps each word in the segmented natural language text into a continuous vector space and forms a sentence according to the word embedding rule, context-encodes the sentence according to a preset recurrent neural network, obtains the corresponding semantic representation, judges the semantic representation, and determines the corresponding semantic emotion; trains the generative adversarial network and the automatic regression model according to the reply type and reply style of the historical natural language text, and combines the models according to the output of the generator of the generative adversarial network and the automatic regression model, and iteratively adjusts the combined model according to the reply content to obtain an intelligent dialogue strategy, so that the present application can effectively improve the accuracy of human-computer interaction.

[0154] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0155] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0156] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0158] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. An artificial intelligence interaction method based on multiple analysis models, characterized in that: The method comprises: According to the word embedding rules, each word in the segmented natural language text is mapped to a continuous vector space to form a sentence, the sentence is context-encoded according to a preset recurrent neural network to obtain a corresponding semantic representation, the semantic representation is judged, and the corresponding semantic sentiment is determined, wherein the recurrent neural network is trained by historical natural language text; According to each reply type and each reply style in the historical natural language text, the corresponding generator network and discriminator network are trained, and the generative adversarial network is obtained by iteratively tuning the generator loss function of the generator network and the discriminator loss function of the discriminator network; according to each reply type and each reply style in the historical natural language text, the preset auto-regression model is trained, and the regression loss function of the auto-regression model is iteratively tuned according to the result of the model training to obtain the auto-regression model, and the output of the generator of the generative adversarial network is used as the input of the auto-regression model, and the reply content output by the generative adversarial network is obtained according to the output of the auto-regression model and the data feedback of the discriminator of the generative adversarial network, and the combined model is iteratively adjusted according to the reply content to obtain the intelligent dialogue strategy; According to the specific type of the semantic emotion, the matching degree of each intelligent dialogue strategy in the set intelligent dialogue library is determined, and the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library is determined as the intelligent dialogue strategy corresponding to the semantic emotion, and the corresponding reply content is determined according to the intelligent dialogue strategy and the natural language text.

2. The artificial intelligence interaction method based on multiple analysis models according to claim 1 is characterized in that: The method of mapping each word in the segmented natural language text into a continuous vector space and forming a sentence according to the word embedding rule includes: Receive a natural language text sent by a user and perform word segmentation on the natural language text to obtain individual words; By setting the semantic relationship between words in the word embedding model, each word is mapped to a high-dimensional real number vector in a continuous vector space, and the vector representation of the entire sentence is determined by the vector combination of the high-dimensional real number vectors of each word.

3. The artificial intelligence interaction method based on multiple analysis models according to claim 1 is characterized in that: The context encoding of the sentence is performed according to a preset recurrent neural network to obtain a corresponding semantic representation, the semantic representation is judged, and the corresponding semantic emotion is determined, wherein the recurrent neural network is trained by historical natural language texts, and includes: Performing context encoding on the vector representation of the entire sentence through a preset recurrent neural network to obtain the order relationship of each word in the sentence, and determining the semantic representation of the sentence according to the order relationship of the words; The supervised learning model obtained by training the training data with emotion labels is used to perform emotion judgment on the semantic representation of the sentence to determine the corresponding semantic emotion.

4. The artificial intelligence interaction method based on multiple analysis models according to claim 1 is characterized in that: The iterative adjustment of the combined model according to the reply content to obtain an intelligent dialogue strategy includes: Adjusting the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model according to user feedback on the reply content; According to the creative weight parameters of the generator in the generative adversarial network and the grammatical rule weights of the autoregressive model after the adjustment, a combined model and an intelligent dialogue strategy with a specific response type and a specific response style output by the model are obtained.

5. The artificial intelligence interaction method based on multiple analysis models according to claim 1 is characterized in that: The step of determining the matching degree of each intelligent dialogue strategy in the intelligent dialogue library according to the specific type of the semantic emotion, and determining the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, includes: Calculating the similarity matching degree between the specific type of the semantic emotion and each of the intelligent dialogue strategies in the set intelligent dialogue library; The intelligent dialogue strategy with the greatest similarity matching degree is determined as the intelligent dialogue strategy corresponding to the semantic emotion.

6. An artificial intelligence interactive device based on multiple analysis models, characterized in that: include: A semantic sentiment determination module is used to map each word in the segmented natural language text into a continuous vector space according to a word embedding rule to form a sentence, perform context encoding on the sentence according to a preset recurrent neural network to obtain a corresponding semantic representation, judge the semantic representation, and determine the corresponding semantic sentiment, wherein the recurrent neural network is trained by historical natural language text; A dialogue strategy determination module is used to train the corresponding generator network and discriminator network according to each response type and each response style in the historical natural language text, and obtain a generative adversarial network by iteratively tuning the generator loss function of the generator network and the discriminator loss function of the discriminator network; perform model training on a preset autoregressive model according to each response type and each response style in the historical natural language text, iteratively tune the regression loss function of the autoregressive model according to the result of the model training, and obtain an autoregressive model; use the output of the generator of the generative adversarial network as the input of the autoregressive model, obtain the response content output by the generative adversarial network according to the output of the autoregressive model and the data feedback of the discriminator of the generative adversarial network, and iteratively adjust the combined model according to the response content to obtain an intelligent dialogue strategy; An intelligent matching module is used to determine the matching degree of each intelligent dialogue strategy in the set intelligent dialogue library according to the specific type of the semantic emotion, determine the intelligent dialogue strategy with the greatest matching degree in the intelligent dialogue library as the intelligent dialogue strategy corresponding to the semantic emotion, and determine the corresponding reply content according to the intelligent dialogue strategy and the natural language text.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the artificial intelligence interaction method based on multiple analysis models described in any one of claims 1 to 5 are implemented.

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 steps of the artificial intelligence interaction method based on multiple analysis models described in any one of claims 1 to 5 are implemented.