Data processing method based on big language model dialogue robot and related equipment

Abstract

The invention relates to the technical field of dialogue robots, and provides a big language model dialogue robot-based data processing method and related equipment, and the method comprises the steps: carrying out the text conversion and vectorization processing of voice input content, and obtaining text data and vector representation; performing similarity retrieval on the vector representation and vector data in a preset vector database to obtain associated knowledge content; performing keyword and key demand screening on the text data to generate cue words, and inputting the associated knowledge content and cue words into a preset large language model to generate a natural language answer text; and converting the natural language answer text into voice information and outputting the voice information. Through the scheme, the understanding of the voice input content of the user and the generation of the natural language answer are realized, and the problems of time and labor consumption in knowledge base construction and mechanization of the dialogue content due to the fact that an existing dialogue robot extracts predefined content or pre-recorded text from a knowledge base to answer the user are solved.

Description

Technical Field

[0001] The present application relates to the technical field of conversational robots, and in particular to a data processing method and related equipment for conversational robots based on a large language model. Background Art

[0002] In today's artificial intelligence field, conversational robots have become an important part of intelligent services. Conversational robots are widely used in various industries, including customer service, medical care, education, etc., greatly improving work efficiency and user experience. With the rapid development of big data technology and natural language processing technology, the intelligence level of conversational robots has been continuously improved, gradually developing from simple rule matching to complex semantic understanding and generation using large language models.

[0003] In the prior art, conversational robots usually rely on pre-built knowledge bases that store some pre-defined questions and answers, which can be obtained through traditional search engines or knowledge graphs. In addition, for specific scenarios, it is necessary to design the conversational robot's conversational flow or canvas. This process usually requires the participation of professional conversational design engineers, which is labor-intensive and time-consuming. In addition, the conversational robot can only extract pre-defined content or pre-recorded text from the knowledge base to answer users. There are problems such as time-consuming and labor-intensive knowledge base construction and mechanized conversation content. Summary of the invention

[0004] In order to improve the existing conversational robots that extract predefined content or pre-recorded text from the knowledge base to answer users, there are problems such as time-consuming and labor-intensive knowledge base construction and mechanized conversation content. This application provides a data processing method and related equipment for a conversational robot based on a large language model.

[0005] The present invention provides a data processing method based on a large language model dialogue robot, comprising the following steps: performing text conversion and vectorization processing on voice input content to obtain text data and vector representation; performing similarity retrieval on the vector representation and vector data in a preset vector database to obtain associated knowledge content; performing keyword and key demand screening on the text data to generate prompt words, inputting the associated knowledge content and the prompt words into a preset large language model to generate a natural language answer text; converting the natural language answer text into voice information and outputting it.

[0006] As a preferred solution, the step of converting the voice input content into text and vectorizing it to obtain text data and vector representation includes: using speech recognition technology to convert the voice input content into text to obtain text data, using a neural network algorithm to extract features from the text data to generate feature data; inputting the feature data into a preset embedding model for conversion processing to obtain a vector representation.

[0007] As a preferred solution, the step of performing similarity retrieval between the vector representation and the vector data in a preset vector database to obtain the associated knowledge content includes: calculating the cosine similarity between the vector representation and the vector data stored in the preset vector database; calculating the distance between the vector representation and each vector in the vector data based on the cosine similarity to obtain the most similar vector; The knowledge content corresponding to the most similar vector is used as the associated knowledge content.

[0008] As a preferred solution, the preset vector database includes multimodal knowledge content vector representations of text, image, and video.

[0009] As a preferred solution, the step of screening the text data for keywords and key requirements and generating prompt words includes: performing semantic analysis on the text data through a natural language processing algorithm to obtain semantic features, screening out keywords and key requirements based on the semantic features, and generating prompt words according to the keywords and key requirements.

[0010] As a preferred embodiment, the step of inputting the associated knowledge content and the prompt words into a preset large language model to generate a natural language answer text includes: semantically integrating the associated knowledge content and the prompt words through multimodal fusion technology to generate integrated semantic data; and performing natural language generation processing on the integrated semantic data using a preset large language model to obtain a natural language answer text.

[0011] As a preferred solution, after the step of inputting the associated knowledge content and the prompt words into a preset large language model to generate a natural language answer text, it includes: obtaining user feedback information on the natural language answer text, and dynamically adjusting the reply method of the large language model based on the feedback information.

[0012] The present application also provides an optimization system for a conversational robot based on a large language model, comprising: a processing unit, for performing text and vector processing on voice input content to obtain text data and vector representation; a retrieval unit, for performing similarity retrieval on the vector representation and vector data in a preset vector database to obtain associated knowledge content; a generation unit, for performing keyword and key requirement screening on the text data, generating prompt words, inputting the associated knowledge content and the prompt words into a preset large language model, and generating a natural language answer text; and an output unit, for converting the natural language answer text into voice information and outputting it.

[0013] The present application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, it implements any of the above-mentioned data processing methods based on a large language model dialogue robot.

[0014] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the data processing method based on a large language model dialogue robot as described in any one of the above.

[0015] Compared with the prior art, the present application has the following beneficial effects: accurate information and rich dialogue. By converting the voice input content into text and vectorization, the conversion of voice signals into text data and vector representation is realized. Through the vector similarity retrieval technology, the knowledge content most similar to the input vector can be quickly and effectively retrieved from the preset vector database; through the keyword and key demand screening technology, prompt words are generated, combined with the relevant knowledge content input into the large language model, so as to generate natural language answer text, and through the text-to-speech technology, the natural language answer text is converted into voice information and output, realizing the understanding of the user's voice input content and the generation of natural language answers, improving the existing dialogue robot to extract predefined content or pre-recorded text from the knowledge base to answer users, and there are problems such as time-consuming and labor-intensive knowledge base construction and mechanized dialogue content. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention 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 only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] The structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with this technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantive technical significance. Any structural modification, change in proportion or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical contents disclosed by the present invention.

[0018] Figure 1 It is a flowchart of a data processing method based on a large language model dialogue robot provided by an embodiment of the present invention; Figure 2is a schematic block diagram of the structure of a data processing system based on a large language model dialogue robot provided by an embodiment of the present invention; Figure 3 It is a schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0019] Description of reference numerals: 10. Data processing system of a conversational robot based on a large language model; 11. Processing unit; 12. Retrieval unit; 13. Generation unit; 14. Output unit; 20. Electronic device; 21. Memory; 22. Processor. DETAILED DESCRIPTION

[0020] 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 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.

[0021] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0022] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0023] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0024] The technical solution of the present invention is further described below with reference to the accompanying drawings and through specific implementation methods.

[0025] Embodiment 1: like Figure 1 As shown, the present application provides a data processing method based on a large language model dialogue robot, comprising steps S100 to S400: Step S100: convert the speech input content into text and perform vectorization processing to obtain text data and vector representation.

[0026] In this step, speech recognition technology is used to convert speech input content into text data; specifically, the input speech signal is analyzed and processed through a preset speech recognition model and converted into corresponding text data. At the same time, the text data is converted into a high-dimensional vector representation through a text vectorization method for subsequent similarity retrieval and semantic analysis.

[0027] For example, deep learning models are used to extract features and convert speech signals into text, and word embedding technology is used to convert text data into vector representations.

[0028] Step S200: perform similarity search between the vector representation and the vector data in a preset vector database to obtain related knowledge content.

[0029] In this step, vector similarity retrieval technology is used to perform similarity matching between the vector representation of the text data and the vector data in a preset vector database; specifically, the cosine similarity between the input vector and the vector in the database is calculated to retrieve the most similar knowledge content in order to generate the answer content.

[0030] For example, use algorithms such as KD-Tree or LSH in the vector database to quickly retrieve similar vectors and return the corresponding knowledge content.

[0031] Step S300: Screen the text data for keywords and key requirements, generate prompt words, input the associated knowledge content and prompt words into a preset large language model, and generate a natural language answer text.

[0032] In this step, the text data is screened for keywords and key requirements through text analysis methods; specifically, models such as TF-IDF or BERT are used to identify keywords in the text and generate prompt words based on user intent. The retrieved relevant knowledge content and generated prompt words are input into the preset large language model to generate natural language answer text that matches the user's question.

[0033] For example, the TF-IDF model is used to calculate the importance of words in a text, and the BERT model is used to further extract context-related key demand information to generate question prompt words that help the language model answer questions.

[0034] Step S400: convert the natural language answer text into voice information and output it.

[0035] In this step, the generated natural language answer text is converted into voice information through text-to-speech technology; specifically, the preset text-to-speech model is used to synthesize the text content into natural and fluent voice output to enhance the user's interactive experience.

[0036] For example, use the WaveNet or Tacotron2 text-to-speech models to convert natural language answer text into high-quality speech signals.

[0037] In this embodiment, first, the text data and vector representation are obtained by converting the speech input content into text and vectorizing it; then, the vector representation is searched for similarity with the vector data in the preset vector database to obtain the related knowledge content; then, the text data is screened for keywords and key requirements to generate prompt words, and the related knowledge content and prompt words are input into the preset large language model to generate natural language answer text; finally, the natural language answer text is converted into speech information and output. By converting the speech input content into text and vectorizing it, the conversion of speech signal to text data and vector representation is realized; through the vector similarity retrieval technology, the knowledge content most similar to the input vector can be quickly and effectively retrieved from the preset vector database; through the keyword and key requirement screening technology, prompt words are generated, and the large language model is input in combination with the relevant knowledge content to generate the natural language answer text; finally, through the text-to-speech technology, the natural language answer text is converted into speech information and output. This method can realize the accurate understanding of the user's speech input content and the generation of natural language answers, greatly improve the intelligence level and user experience of the dialogue robot, and effectively solve the problems of mechanization of dialogue content and poor semantic understanding effect in the prior art.

[0038] Embodiment 2: In step S100, the speech input content is converted into text using speech recognition technology to obtain text data, and the features of the text data are extracted using a neural network algorithm to generate feature data.

[0039] Through automatic speech recognition (ASR) technology, the speech signal is converted into text data; specifically, the text data is processed using a deep neural network algorithm to extract feature information and generate feature data for subsequent processing steps.

[0040] For example, by using ASR to parse the speech signal, the speech signal is converted into corresponding text data, and key features are further extracted.

[0041] The feature data is input into the preset embedding model for conversion processing to obtain a vector representation.

[0042] By using a pre-trained embedding model, the feature data is converted into a vector representation; specifically, word embedding technology such as Word2Vec, GloVe, or BERT is used to map text features into a vector space to generate a high-dimensional vector representation.

[0043] For example, the BERT model is used to convert text feature data into a vector representation with semantic information for subsequent similarity retrieval.

[0044] In step S200, the cosine similarity between the vector representation and the vector data stored in the preset vector database is calculated.

[0045] The cosine similarity between the vector representation and the vector data in the vector database is calculated to determine the degree of similarity between the two. Specifically, the cosine similarity formula is used to calculate the cosine value of the angle between the input vector and the vector in the database as a measure of similarity.

[0046] For example, by calculating the cosine similarity between the input text vector and each vector in the preset vector database, the most similar vectors are screened out.

[0047] Based on the cosine similarity, the distance between the vector representation and each vector in the vector data is calculated to obtain the most similar vector.

[0048] By sorting the cosine similarity calculation results, the vector representation is screened out, and the vector that is most similar to the preset vector database is selected; specifically, all calculated cosine similarity values ​​are arranged in descending order, and the vector with the highest similarity is selected as the matching result.

[0049] For example, by sorting the calculated cosine similarity values, the top N most similar vectors are selected for subsequent knowledge content retrieval.

[0050] The knowledge content corresponding to the most similar vector is taken as the associated knowledge content.

[0051] By searching a preset vector database, the knowledge content corresponding to the most similar vector is extracted as the associated knowledge content; specifically, according to the index of the similar vector, its associated text, image or video content is obtained.

[0052] For example, knowledge content such as FAQ answers, product information, or tutorial videos corresponding to the most similar vector is retrieved.

[0053] Among them, the preset vector database includes multimodal knowledge content vector representation of text, image, and video.

[0054] By constructing a multimodal knowledge content vector database, different types of data (such as text, images, and videos) are converted into a unified vector representation; specifically, using multimodal embedding technology, text, image, and video data are vectorized separately and stored in the same vector space.

[0055] For example, the multimodal CLIP model is used to jointly embed images and text to create a vector database containing multiple data types for cross-modal retrieval.

[0056] In step S300, the steps of screening the text data for keywords and key requirements and generating prompt words include: performing semantic analysis on the text data through a natural language processing algorithm to obtain semantic features, screening out keywords and key requirements based on the semantic features, and generating prompt words based on the keywords and key requirements.

[0057] By adopting natural language processing algorithms (such as BERT or GPT), deep semantic analysis of text data is performed to extract key semantic features. Specifically, these semantic features are used to filter out keywords and key requirements that best represent user intentions from the text, and prompt words are generated based on these keywords and key requirements.

[0058] For example, the BERT model is used for semantic analysis to identify the main keywords in user questions and generate corresponding prompt words, such as "repair steps", "payment method", etc.

[0059] In step S300, the associated knowledge content and prompt words are input into a preset large language model to generate a natural language answer text, including: semantically integrating the associated knowledge content and prompt words through multimodal fusion technology to generate integrated semantic data.

[0060] By using multimodal fusion technology, different types of knowledge content (such as text, images, videos) and prompt words are semantically integrated; specifically, text and non-text data are converted into a unified semantic representation, and combined with prompt words to generate integrated semantic data for input into the large language model.

[0061] For example, a multimodal fusion model is used to integrate image descriptions with text prompts to generate input data containing rich semantic information.

[0062] The preset large language model is used to perform natural language generation processing on the integrated semantic data to obtain a natural language answer text.

[0063] By calling the preset large language model, natural language generation is performed on the integrated semantic data; specifically, a natural language answer text that conforms to the context is generated based on the input semantic data to ensure the accuracy and fluency of the answer.

[0064] For example, using a large language model to generate answer text can generate accurate and natural responses based on the user's questions and knowledge content, such as a detailed description of a product failure.

[0065] After the steps of inputting the relevant knowledge content and prompt words into the preset large language model to generate the natural language answer text, it includes: obtaining the user's feedback information on the natural language answer text, and dynamically adjusting the reply mode of the large language model based on the feedback information.

[0066] By designing a user feedback collection mechanism, we can obtain users' satisfaction and opinions on the generated natural language answer texts; specifically, we establish a feedback loop to dynamically adjust the response method of the large language model based on user feedback information, and continuously optimize and improve the performance of the model.

[0067] For example, use user satisfaction questionnaires or conversation record analysis tools to collect user feedback, and adjust the parameters and generation strategies of the large language model based on the feedback results to improve the accuracy of answers and user satisfaction.

[0068] In this embodiment, the speech input content is converted into text by using speech recognition technology to obtain text data, and the text data is feature extracted by using a neural network algorithm to generate feature data, which is further converted by a preset embedding model to obtain a vector representation. Then, by calculating the cosine similarity between the vector representation and the vector data stored in the preset vector database, the most similar vector is screened out, and the associated knowledge content is obtained based on this. Among them, the preset vector database includes multimodal knowledge content vector representations of text, images, and videos to improve the accuracy and extensiveness of retrieval. In the text data processing, the text data is semantically analyzed by a natural language processing algorithm, keywords and key requirements are screened out, prompt words are generated, and the associated knowledge content and prompt words are semantically integrated by combining multimodal fusion technology to generate integrated semantic data, and finally the integrated semantic data is processed by a preset large language model for natural language generation to obtain a natural language answer text. At the same time, the user's feedback information on the natural language answer is obtained through the user feedback mechanism, and the reply method of the large language model is dynamically adjusted based on the feedback information to continuously optimize the answer effect.

[0069] Embodiment 3: The present application also provides an optimization system for a conversational robot based on a large language model, comprising a processing unit 11, a retrieval unit 12, a generation unit 13 and an output unit 14.

[0070] The processing unit 11 is mainly used to perform text and vector processing on the speech input content to obtain text data and vector representation.

[0071] The retrieval unit 12 is mainly used to perform similarity retrieval between the vector representation and the vector data in the preset vector database to obtain the associated knowledge content.

[0072] The generation unit 13 is mainly used to screen the text data for keywords and key requirements, generate prompt words, input the associated knowledge content and prompt words into a preset large language model, and generate a natural language answer text.

[0073] The output unit 14 is mainly used to convert the natural language answer text into voice information and output it.

[0074] In this embodiment, by using the processing unit 11 to perform text conversion and vectorization processing on the voice input content, high-quality text data and vector representation are obtained to ensure the accuracy of subsequent processing. The retrieval unit 12 uses similarity retrieval technology to match the input vector representation with the vector data in the preset vector database, thereby obtaining knowledge content that is highly relevant to the user input content. The generation unit 13 generates prompt words by screening keywords and key requirements in the text data, and inputs the large language model in combination with relevant knowledge content to generate a natural language answer text that meets the user's question. The output unit 14 converts the generated natural language answer text into voice information and outputs it to improve the user experience. Through the efficient collaboration of various units, the optimization system improves the intelligence level and response accuracy of the dialogue robot and significantly improves the user interaction experience.

[0075] It should be noted that technical personnel in the relevant technical field can clearly understand that, for the convenience and conciseness of description, the specific working process of the system and each unit described above can refer to the corresponding process in the aforementioned data processing method based on the large language model dialogue robot and related equipment embodiments, and will not be repeated here.

[0076] Embodiment 4: like Figure 2 As shown, the present application also provides an electronic device 20, including a memory 21 and a processor 22, the memory 21 stores a computer program that can be run on the processor 22, and the processor 22 implements the data processing method based on the large language model dialogue robot of embodiment 1 when executing the computer program In this embodiment, by integrating the memory 21 and the processor 22 in the electronic device 20, and running the computer program stored in the memory 21, the processor 22 can execute the data processing method based on the large language model dialogue robot described in Example 1. The method includes text conversion and vectorization processing of the voice input content, similarity retrieval, screening keywords and key requirements, generating prompt words, and generating natural language answer text through the large language model, and finally converting it into voice output. Through this method, the electronic device 20 can efficiently realize voice recognition, semantic understanding and natural language generation, significantly improving the intelligence level and user experience of the dialogue robot.

[0077] Embodiment 5: like Figure 3 As shown, the present application also provides a computer readable storage medium on which a computer program is stored, causing the processor to perform a data processing method based on a large language model dialogue robot as in Embodiment 1.

[0078] In this embodiment, by storing a computer program on a computer-readable storage medium, when the processor runs the program, the processor is enabled to execute the data processing method based on the large language model dialogue robot in Example 1. The method includes text conversion and vectorization processing of voice input content, similarity retrieval, screening keywords and key requirements, generating prompt words, and generating natural language answer text through a large language model, and finally converting it into voice output. Through the computer-readable storage medium, the data processing method of the dialogue robot based on the large language model can be conveniently applied to various computing devices to achieve efficient voice recognition and natural language processing, and enhance the user's interactive experience.

[0079] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data processing method based on a large language model dialogue robot, characterized in that: The following steps are involved: Perform text conversion and vectorization processing on the speech input content to obtain text data and vector representation; Performing similarity search between the vector representation and vector data in a preset vector database to obtain associated knowledge content; Screening the text data for keywords and key requirements, generating prompt words, inputting the associated knowledge content and the prompt words into a preset large language model, and generating a natural language answer text; The natural language answer text is converted into voice information and output.

2. The data processing method based on a large language model dialogue robot according to claim 1, characterized in that: The step of converting the speech input content into text and vectorizing it to obtain text data and vector representation includes: Using speech recognition technology to convert the speech input content into text to obtain text data, and using a neural network algorithm to extract features from the text data to generate feature data; The feature data is input into a preset embedding model for conversion processing to obtain a vector representation.

3. The data processing method based on a large language model dialogue robot according to claim 1, characterized in that: The step of performing similarity retrieval between the vector representation and vector data in a preset vector database to obtain associated knowledge content includes: By calculating the cosine similarity between the vector representation and the vector data stored in a preset vector database; Calculate the distance between the vector representation and each vector in the vector data based on the cosine similarity to obtain the most similar vector; The knowledge content corresponding to the most similar vector is used as the associated knowledge content.

4. The data processing method based on a large language model dialogue robot according to claim 3 is characterized in that: The preset vector database includes multimodal knowledge content vector representations of text, images, and videos.

5. The data processing method based on a large language model dialogue robot according to claim 1, characterized in that: The step of screening the text data for keywords and key requirements to generate prompt words includes: The text data is semantically analyzed by a natural language processing algorithm to obtain semantic features, keywords and key requirements are screened out based on the semantic features, and prompt words are generated according to the keywords and the key requirements.

6. The data processing method based on a large language model dialogue robot according to claim 1, characterized in that: The step of inputting the associated knowledge content and the prompt word into a preset large language model to generate a natural language answer text includes: Performing semantic integration on the associated knowledge content and the prompt words through multimodal fusion technology to generate integrated semantic data; The integrated semantic data is processed by natural language generation using a preset large language model to obtain a natural language answer text.

7. The data processing method based on a large language model dialogue robot according to claim 1, characterized in that: After the step of inputting the associated knowledge content and the prompt word into a preset large language model to generate a natural language answer text, the method includes: Acquire user feedback information on the natural language answer text, and dynamically adjust the answer mode of the large language model based on the feedback information.

8. An optimization system for a conversational robot based on a large language model, characterized in that: include: A processing unit, used for performing text and vector processing on the speech input content to obtain text data and vector representation; A retrieval unit, used to perform similarity retrieval between the vector representation and vector data in a preset vector database to obtain associated knowledge content; A generating unit, configured to screen the text data for keywords and key requirements, generate prompt words, input the associated knowledge content and the prompt words into a preset large language model, and generate a natural language answer text; The output unit is used to convert the natural language answer text into voice information and output it.

9. An electronic device, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and the processor implements the data processing method based on a large language model dialogue robot as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the processor executes the data processing method based on a large language model dialogue robot as described in any one of claims 1 to 7.

Images