Hybrid interaction system based on AI large model

Abstract

The invention relates to the technical field of man-machine interaction, in particular to a mixed interaction system based on an AI large model, which comprises a multi-modal question acquisition module, a preprocessing and classification module, a retrieval and analysis module, a mixed processing module, a result integration and output module, an interaction module and an answer display module, the problem preprocessing and classification module performs text cleaning, semantic analysis and classification, and the hybrid processing module distributes tasks to natural language processing or AI large models according to a data complexity threshold value, so that the problem processing flow is optimized, the system performance is improved, and the system performance is improved. According to the system, model parameters are optimized based on a machine learning classification algorithm and transfer learning training, classification accuracy is ensured, the system adapts to problem changes, through cooperation of all the modules, the existing technical problems are effectively solved, the man-machine interaction efficiency and quality are improved, the system has wide application prospects in multiple fields, and the man-machine interaction technology is promoted to be developed to be intelligent and efficient.

Description

Technical Field

[0001] The present invention relates to the field of human-computer interaction technology, and in particular to a hybrid interaction system based on an AI large model. Background Art

[0002] In today's digital age, human-computer interaction technology continues to evolve, aiming to achieve more natural, efficient and intelligent information exchange between people and computers. The traditional human-computer interaction method has undergone a transformation from command line to graphical user interface (GUI), which has significantly improved the convenience and intuitiveness of user operations. However, as people's requirements for information acquisition and processing efficiency continue to increase, natural language interaction has become an important development direction in the field of human-computer interaction.

[0003] In the prior art, traditional natural language processing technology understands natural language based on rules or neural networks with fewer parameters. It has a high understanding accuracy for simple instructions, fast reasoning speed and low deployment cost. However, it is more complicated to process complex instructions and has a low understanding accuracy.

[0004] Therefore, it is necessary to propose a hybrid interactive system based on AI large model to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a hybrid interactive system based on an AI large model to solve the problem that the traditional natural language processing technology understands natural language based on rules or neural networks with fewer parameters, has a high understanding accuracy for simple instructions, and has fast reasoning speed and low deployment cost, but is more complicated to process complex instructions and has a low understanding accuracy.

[0006] To achieve the above object, the present invention provides the following technical solutions: A hybrid interactive system based on an AI large model, including a hybrid interactive system based on an AI large model, characterized by comprising a multimodal question acquisition module, a question preprocessing and classification module, a question retrieval and parsing module, a hybrid processing module, a result integration and output module, an interaction module, and an answer display module; A multimodal question acquisition module, used to receive questions input by a user, wherein the multimodal question acquisition module comprises a voice acquisition unit and a text input unit; The voice acquisition unit is the key part of the system to receive user voice input. Its core function is to accurately obtain the user's voice signal. Based on the transfer learning method, it further improves the recognition accuracy of multiple dialects through real-time feedback and adaptive adjustment functions. The text input unit provides users with a convenient interface for manually inputting text questions. A question preprocessing and classification module, connected to the multimodal question acquisition module signal, is used to perform a preprocessing operation on the acquired question, wherein the preprocessing operation includes text cleaning to remove noise data, semantic analysis based on a semantic understanding model to extract key semantic information, and question classification according to predefined classification rules; Text cleaning mainly removes spelling errors, grammatical errors, extra punctuation marks, and special characters during the input process. The predefined classification rules are formulated based on the analysis and summary of user problems in a large number of actual application scenarios, and the classification algorithm based on machine learning is used to classify problems. A question retrieval and analysis module, connected to the question preprocessing and classification unit signal, is used to retrieve relevant information from the local knowledge base and external data sources based on the classification results, and analyze the retrieved information using natural language analysis technology to extract key content and logical relationships; A hybrid processing module, connected to the question retrieval and analysis module signal, used to process the analyzed question data, the hybrid processing module includes a natural language processing subunit and an AI large model access subunit, and the natural language processing subunit and the AI ​​large model access subunit are both provided with a complexity determination threshold; When the complexity of the parsed data is lower than the complexity determination threshold of the natural language processing subunit, it is processed by the natural language processing subunit; When the complexity of the parsed data exceeds the complexity judgment threshold of the natural language processing sub-unit, the AI ​​big model access sub-unit calls the external AI big model for processing; A result integration and output unit module is connected to the hybrid processing unit module and is used to integrate the processing results. The result integration and output unit module includes a natural language processing result output module and an AI large model result output module, which are used to output the processing results of the natural language processing sub-unit and the AI ​​large model access sub-unit respectively; An interactive module, used to provide an interactive interface between the user and the system, wherein the interactive interface presentation module includes a visual operation screen and a multimodal interactive support component, and the interactive module includes an answer presentation module; The answer display module is connected to the result integration and output unit signal, and includes an answer retrieval submodule, an answer acquisition submodule, an answer summary submodule, an answer integration submodule, and an answer generation submodule.

[0007] Preferably, in the hybrid processing module, as the complexity of the problem gradually approaches the threshold, the performance of the natural language processing sub-unit will change. Within the threshold range, its processing time is relatively stable and short, and the accuracy is high. However, when the complexity of the problem approaches the upper limit of the threshold, the processing time may begin to increase, and more computing resources and time are required to analyze complex grammatical structures and semantic relationships. Since large AI models usually require a lot of computing resources and a long processing time, when the complexity of the problem reaches above the threshold, the large AI model access sub-unit processing can provide more valuable solutions for complex problems. This resource allocation method is based on the reasonable setting of the threshold to balance the processing efficiency and processing quality of the system.

[0008] Preferably, in the multimodal question acquisition module, the voice collection unit interacts with data from other parts of the system, and the input text questions are promptly transmitted to the question preprocessing and classification unit for subsequent processing. During the data transmission process, an efficient data transmission protocol is adopted to ensure the integrity and accuracy of the text data. At the same time, the text input unit is connected to the system's history module, which can automatically record historical questions input by users, making it convenient for users to view and reuse them.

[0009] Preferably, the recognition accuracy of multiple dialects is improved by using a transfer learning method. On the basis of a general language model, a small amount of sample data of a specific dialect is used for fine-tuning training, so that the model can quickly adapt to the characteristics of the new dialect and achieve accurate recognition of multiple dialects.

[0010] Preferably, while the transfer learning method is fine-tuning the sample data, when the real-time feedback and adaptive adjustment function encounters a situation with low recognition accuracy, the system will automatically analyze the cause, such as poor voice signal quality and language model mismatch. If it is a voice signal quality problem, the system will prompt the user to adjust the speaking method or environment; if it is a language model problem, the system will automatically trigger the online learning mechanism, and use the new voice data to update and optimize the model to ensure recognition accuracy.

[0011] Preferably, a classification algorithm based on machine learning is used to classify questions, and the classification model is trained and optimized using training data with labeled classification labels. The sample data is fine-tuned and trained using a transfer learning method. During the training process, the model parameters are optimized so that the model can accurately predict the classification category based on the input question features.

[0012] Preferably, model training and optimization include training data preparation, model training process, and model evaluation and optimization. The training data preparation collects a large amount of labeled problem data as a training set to ensure that each category has a sufficient number of samples and that the samples are diverse, covering different language expressions and problem scenarios. The model training process trains the selected machine learning algorithm using the training set data. During the training process, some hyperparameters are set, and the optimal hyperparameter combination can be found through grid search and random search methods to improve model performance. Model evaluation and optimization evaluates the performance of the trained model using validation set data. Common evaluation indicators include accuracy and precision. Accuracy indicates the proportion of the number of correctly classified problems to the total number of problems. Precision indicates the proportion of the number of problems predicted to be positive and actually positive to the number of problems predicted to be positive.

[0013] Preferably, the answer retrieval optimization submodule is used to further retrieve optimization information in the knowledge base based on user questions and processing results, the answer integration generation submodule is used to integrate processing results and optimization information to generate a final answer, and the multi-format display submodule is used to display the final answer in multiple formats such as text, voice, and charts, and the display format can be automatically adapted according to the user terminal device type and user preferences.

[0014] Technical effects and advantages of the present invention: 1. The multimodal question acquisition module significantly improves the convenience and accuracy of user interaction through the synergy of the voice acquisition unit and the text input unit, while ensuring the fluency of information interaction; 2. The optimized design of the problem preprocessing and classification module and the mixed processing module in the problem handling process has greatly improved the system performance, enhanced the system's ability to handle problems of different complexity, and improved the stability and reliability of the system operation; 3. Adopt a classification algorithm based on machine learning, fine-tune training through transfer learning, and optimize model parameters using a large amount of labeled data, so that the model can accurately predict the classification category of problems. This ensures the accuracy of problem classification. With the continuous addition of new data and continuous training of the model, the system can better adapt to changes in different types of problems, continuously improve classification effects, and provide a solid foundation for the efficient operation of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a schematic diagram of the structure of a hybrid interactive system based on an AI large model in the present invention.

[0016] Figure 2 It is a flow chart of the interaction of the system of the present invention. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The present invention provides Figure 1-Figure 2 A hybrid interactive system based on an AI large model is shown, comprising a multimodal question acquisition module, a question preprocessing and classification module, a question retrieval and parsing module, a hybrid processing module, a result integration and output module, an interactive module, and an answer display module; A multimodal question acquisition module, used to receive questions input by a user, wherein the multimodal question acquisition module comprises a voice acquisition unit and a text input unit; The voice acquisition unit is the key part of the system for receiving user voice input. Its core function is to accurately obtain the user's voice signal. It uses a high-precision microphone array or other advanced audio acquisition equipment to clearly capture sound under different environmental conditions. On the basis of the transfer learning method, it further improves the recognition accuracy of multiple dialects through real-time feedback and adaptive adjustment functions. The text input unit provides users with a convenient interface for manually inputting text questions. Users can input questions through a keyboard or other input devices. During the user input process, the text input unit provides a real-time input prompt function, gives grammar correction prompts through grammar analysis technology, and performs real-time grammar checking on the text input by the user. High-precision microphone arrays or other audio acquisition devices monitor the surrounding sounds in real time. Once a user's voice signal is detected, the acquisition is started immediately. The collected voice signal is first pre-processed, including amplification, filtering and other operations. The amplification operation ensures that the voice signal has sufficient strength for subsequent processing. The filtering removes low-frequency and high-frequency noise, retaining only the voice signal components within the audible frequency range of the human ear to improve the signal quality. The input text issues are formatted and normalized to ensure unified text encoding (such as conversion to UTF-8 encoding), and to remove unnecessary spaces, line breaks and other formatting issues, so that the text format meets the system's subsequent processing requirements.

[0019] In the multimodal question acquisition module, the voice collection unit interacts with data from other parts of the system, and the input text questions are promptly transmitted to the question preprocessing and classification unit for subsequent processing. During the data transmission process, an efficient data transmission protocol is used to ensure the integrity and accuracy of the text data. At the same time, the text input unit is connected to the system's history record module, which can automatically record the historical questions input by the user, making it convenient for users to view and reuse; Improve the recognition accuracy of multiple dialects. Through transfer learning, fine-tune the general language model using a small amount of sample data from a specific dialect, so that the model can quickly adapt to the characteristics of new dialects and achieve accurate recognition of multiple dialects. While the transfer learning method fine-tunes the sample data, the real-time feedback and adaptive adjustment function will automatically analyze the reasons when the recognition accuracy is low, such as poor voice signal quality and language model mismatch. If it is a voice signal quality problem, the system will prompt the user to adjust the speaking method or environment; if it is a language model problem, the system will automatically trigger the online learning mechanism to update and optimize the model with new voice data to ensure recognition accuracy. It provides two input methods, voice and text, to meet the needs of users in different scenarios and is very convenient. The voice collection unit can effectively improve the recognition accuracy of multiple dialects through transfer learning and real-time feedback adaptive adjustment. This enables users from different regions and with different dialect habits to interact with the system using natural language, avoiding the system's inability to understand user intentions due to dialect problems, expanding the system's scope of application, and improving the system's inclusiveness for diverse user groups.

[0020] A question preprocessing and classification module, connected to the multimodal question acquisition module signal, is used to perform a preprocessing operation on the acquired question, wherein the preprocessing operation includes text cleaning to remove noise data, semantic analysis based on a semantic understanding model to extract key semantic information, and question classification according to predefined classification rules; Text cleaning mainly removes spelling errors, grammatical errors, extra punctuation marks, and special characters during the input process. During the semantic analysis process, the model performs word segmentation on the input text, and each word or subword is converted into a vector representation. The model conducts a comprehensive semantic understanding of each part of the text. The predefined classification rules are formulated based on the analysis and summary of user questions in a large number of actual application scenarios, and a classification algorithm based on machine learning is used to classify questions. Use a classification algorithm based on machine learning to classify questions, use training data with labeled classification labels to train and optimize the classification model, and use transfer learning methods to fine-tune the sample data. During the training process, optimize the model parameters so that the model can accurately predict the classification category based on the input question features; Model training and optimization include training data preparation, model training process, and model evaluation and optimization. Training data preparation collects a large amount of labeled problem data as training sets to ensure that each category has a sufficient number of samples and that the samples are diverse, covering different language expressions and problem scenarios. The model training process uses the training set data to train the selected machine learning algorithm. During the training process, some hyperparameters are set, and the grid search and random search methods can be used to find the optimal hyperparameter combination to improve model performance. Model evaluation and optimization evaluates the performance of the trained model using validation set data. Common evaluation indicators include accuracy and precision. Accuracy indicates the proportion of correctly classified problems to the total number of problems. Precision indicates the proportion of problems predicted to be positive and actually positive to the number of problems predicted to be positive. The problem preprocessing and classification module has many important and beneficial effects on the hybrid interactive system based on AI large models. It removes various types of noise data through text cleaning, ensures the standardization and accuracy of the input data, lays a solid foundation for subsequent processing, and reduces the understanding bias caused by erroneous or non-standard text; semantic analysis based on the semantic understanding model converts text into vector representation, deeply mines semantic relationships and information, and enhances the system's ability to handle complex semantic problems; predefined classification rules are based on the analysis and summary of actual application scenarios to make the classification scientific and reasonable and avoid processing confusion; machine learning classification algorithms and labeled data are used to train and optimize the model, combined with transfer learning to fine-tune to adapt to new situations, significantly improve classification accuracy, and accurately route problems to appropriate processing modules or processes; efficient preprocessing accelerates the overall processing flow, reduces the amount of data and processing difficulty, and accurate classification helps precise processing, improves resource utilization efficiency, and ultimately improves system operation results and user satisfaction, greatly enhancing the performance and adaptability of the system, and providing users with a better, more efficient and accurate service experience.

[0021] A question retrieval and analysis module, connected to the question preprocessing and classification unit signal, is used to retrieve relevant information from the local knowledge base and external data sources based on the classification results, and analyze the retrieved information using natural language analysis technology to extract key content and logical relationships; A hybrid processing module, connected to the question retrieval and analysis module signal, used to process the analyzed question data, the hybrid processing module includes a natural language processing subunit and an AI large model access subunit, and the natural language processing subunit and the AI ​​large model access subunit are both provided with a complexity determination threshold; When the complexity of the parsed data is lower than the complexity determination threshold of the natural language processing subunit, it is processed by the natural language processing subunit; When the complexity of the parsed data exceeds the complexity judgment threshold of the natural language processing sub-unit, the AI ​​big model access sub-unit calls the external AI big model for processing; The question retrieval and analysis module retrieves information from the local knowledge base and external data sources based on the classification results, and uses natural language analysis technology to extract key content and logical relationships. This enables the system to quickly locate and obtain rich knowledge resources related to the problem, provides a sufficient information basis for subsequent processing, and enhances the comprehensiveness and accuracy of the system's answers to questions; the hybrid processing module reasonably allocates tasks through the internal complexity judgment threshold. When the complexity of the parsed data is low, the natural language processing subunit quickly gives answers with its high efficiency and ability to handle common problems, meeting the user's needs for quick response to simple questions and improving the system's processing efficiency; when the complexity of the parsed data is high, the AI ​​big model access subunit calls a powerful external AI big model for processing, giving full play to its advantages in complex semantic understanding, logical reasoning and cross-domain knowledge fusion, providing users with in-depth, accurate and insightful answers, improving the system's ability to handle complex problems, and broadening the system's scope of application. The organic combination of the two realizes the optimal allocation of system resources, ensuring that the system can provide high-quality, personalized services when facing various types of problems, greatly improving the user experience and the overall performance of the system.

[0022] A result integration and output unit module is connected to the hybrid processing unit module and is used to integrate the processing results. The result integration and output unit module includes a natural language processing result output module and an AI large model result output module, which are used to output the processing results of the natural language processing sub-unit and the AI ​​large model access sub-unit respectively; The result integration and output unit module plays a vital bridging role in the hybrid interactive system based on the AI ​​large model, and has brought about significant beneficial effects in many aspects. It can effectively integrate the processing results of the natural language processing subunit and the AI ​​large model access subunit in the hybrid processing unit, ensuring that results from different sources and types are gathered in an orderly manner; the natural language processing result output module outputs the processing results of relatively simple problems, ensuring the system's ability to respond quickly to common and basic problems, and can provide users with concise and clear answers in a timely manner, meeting users' needs for efficient information acquisition and improving the timeliness of interaction.

[0023] An interactive module, used to provide an interactive interface between the user and the system, wherein the interactive interface presentation module includes a visual operation screen and a multimodal interactive support component, and the interactive module includes an answer presentation module; An answer display module, connected to the result integration and output unit signal, includes an answer retrieval submodule, an answer acquisition submodule, an answer summary submodule, an answer integration submodule, and an answer generation submodule; The answer display module is used to further retrieve optimization information in the knowledge base based on the user's question and the processing result. The answer integration submodule is used to integrate the processing result and the optimization information to generate the final answer. The multi-format display submodule is used to display the final answer in multiple formats such as text, voice, and charts, and the display format can be automatically adapted according to the user terminal device type and user preference; The visual operation screen and multimodal interaction support components in the interactive module provide users with rich and convenient interaction methods, allowing users to interact with the system in the most natural and comfortable way. Whether through intuitive touch operation, gesture recognition, or clear visual display and voice feedback, it greatly enhances the interactivity and ease of use between users and the system, and improves the user experience; the answer display module further improves the quality and adaptability of the system output; the answer retrieval optimization sub-module deeply mines and optimizes information in the knowledge base based on user questions and processing results, making the final answer more accurate and comprehensive.

[0024] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A hybrid interactive system based on AI large model, characterized in that: It includes multimodal question acquisition module, question preprocessing and classification module, question retrieval and analysis module, hybrid processing module, result integration and output module, interaction module and answer display module; A multimodal question acquisition module, used to receive questions input by a user, wherein the multimodal question acquisition module comprises a voice acquisition unit and a text input unit; The voice acquisition unit is the key part of the system to receive user voice input. Its core function is to accurately obtain the user's voice signal. Based on the transfer learning method, it further improves the recognition accuracy of multiple dialects through real-time feedback and adaptive adjustment functions. The text input unit provides users with a convenient interface for manually inputting text questions. A question preprocessing and classification module, connected to the multimodal question acquisition module signal, is used to perform a preprocessing operation on the acquired question, wherein the preprocessing operation includes text cleaning to remove noise data, semantic analysis based on a semantic understanding model to extract key semantic information, and question classification according to predefined classification rules; Text cleaning mainly removes spelling errors, grammatical errors, extra punctuation marks, and special characters during the input process. The predefined classification rules are formulated based on the analysis and summary of user problems in a large number of actual application scenarios, and the classification algorithm based on machine learning is used to classify problems. A question retrieval and analysis module, connected to the question preprocessing and classification unit signal, is used to retrieve relevant information from the local knowledge base and external data sources based on the classification results, and analyze the retrieved information using natural language analysis technology to extract key content and logical relationships; A hybrid processing module, connected to the question retrieval and analysis module signal, used to process the analyzed question data, the hybrid processing module includes a natural language processing subunit and an AI large model access subunit, and the natural language processing subunit and the AI ​​large model access subunit are both provided with a complexity determination threshold; When the complexity of the parsed data is lower than the complexity determination threshold of the natural language processing subunit, it is processed by the natural language processing subunit; When the complexity of the parsed data exceeds the complexity judgment threshold of the natural language processing sub-unit, the AI ​​big model access sub-unit calls the external AI big model for processing; A result integration and output unit module is connected to the hybrid processing unit module and is used to integrate the processing results. The result integration and output unit module includes a natural language processing result output module and an AI large model result output module, which are used to output the processing results of the natural language processing sub-unit and the AI ​​large model access sub-unit respectively; An interactive module, used to provide an interactive interface between the user and the system, the interactive module includes a visual operation screen and a multimodal interactive support component, and the interactive module includes an answer display module; The answer display module is connected to the result integration and output unit signal, and includes an answer retrieval submodule, an answer acquisition submodule, an answer summary submodule, an answer integration submodule, and an answer generation submodule.

2. According to claim 1, a hybrid interactive system based on an AI large model is characterized by: In the hybrid processing module, as the complexity of the problem gradually approaches the threshold, the performance of the natural language processing sub-unit will change. Within the threshold range, its processing time is relatively stable and short, and the accuracy is high. However, when the complexity of the problem approaches the upper limit of the threshold, the processing time may begin to increase, requiring more computing resources and time to analyze complex grammatical structures and semantic relationships. Since large AI models usually require a lot of computing resources and a long processing time, when the complexity of the problem reaches above the threshold, the access of the AI ​​large model to sub-unit processing can provide more valuable solutions for complex problems. This resource allocation method is based on the reasonable setting of the threshold to balance the processing efficiency and processing quality of the system.

3. The AI ​​large model hybrid interactive system according to claim 1 is characterized by: In the multimodal question acquisition module, the voice collection unit interacts with data from other parts of the system, and the input text questions are promptly transmitted to the question preprocessing and classification unit for subsequent processing. During the data transmission process, an efficient data transmission protocol is adopted to ensure the integrity and accuracy of the text data. At the same time, the text input unit is connected to the system's history module, which can automatically record historical questions entered by users, making it convenient for users to view and reuse them.

4. The AI ​​large model hybrid interactive system according to claim 1 is characterized by: Improve the recognition accuracy of multiple dialects. Through transfer learning methods, based on the general language model, fine-tune the training using a small amount of sample data from specific dialects, so that the model can quickly adapt to the characteristics of new dialects and achieve accurate recognition of multiple dialects.

5. The AI ​​large model-based hybrid interactive system according to claim 4 is characterized by: While the transfer learning method is fine-tuning the sample data, the real-time feedback and adaptive adjustment function will automatically analyze the reasons when encountering low recognition accuracy, such as poor voice signal quality and language model mismatch. If it is a voice signal quality problem, the system will prompt the user to adjust the speaking method or environment; if it is a language model problem, the system will automatically trigger the online learning mechanism and use the new voice data to update and optimize the model to ensure recognition accuracy.

6. The AI ​​large model-based hybrid interactive system according to claim 5 is characterized by: A classification algorithm based on machine learning is used to classify questions. The classification model is trained and optimized using training data with labeled classification labels. The sample data is fine-tuned through transfer learning methods. During the training process, the model parameters are optimized so that the model can accurately predict the classification category based on the input question features.

7. The AI ​​large model-based hybrid interactive system according to claim 5, characterized in that: Model training and optimization include training data preparation, model training process, and model evaluation and optimization. Training data preparation collects a large amount of labeled problem data as a training set to ensure that each category has a sufficient number of samples and that the samples are diverse, covering different language expressions and problem scenarios. The model training process uses the training set data to train the selected machine learning algorithm. During the training process, some hyperparameters are set, and the optimal hyperparameter combination can be found through grid search and random search methods to improve model performance. Model evaluation and optimization evaluates the performance of the trained model using the validation set data. Common evaluation indicators include accuracy and precision. Accuracy indicates the proportion of correctly classified problems to the total number of problems. Precision indicates the proportion of problems predicted to be positive and actually positive to the number of problems predicted to be positive.

8. The AI ​​large model hybrid interactive system according to claim 1 is characterized by: The answer retrieval optimization submodule is used to further retrieve optimization information in the knowledge base based on user questions and processing results. The answer integration generation submodule is used to integrate processing results and optimization information to generate a final answer. The multi-format display submodule is used to display the final answer in multiple formats such as text, voice, and charts, and the display format can be automatically adapted according to the user terminal device type and user preferences.

Images