Intelligent digital employee voice dialogue type tax declaration method based on large language model

Abstract

The invention discloses an intelligent digital employee voice dialogue type tax declaration method based on a large language model, and the method comprises the following steps: converting a tax declaration intention target voice into characters, obtaining a first text and a feature vector thereof, inputting the first text and the feature vector into a large language model, outputting the probability distribution information of the first text, and inputting the probability distribution information into an intention classifier, obtaining an intention prediction result of the first text, if the intention prediction result hits a tax declaration situation, waking up intelligent digital employee process automation, calling a dialogue voice engine, guiding a user to output required filling information through voice, obtaining a second text set through character conversion, automatically reminding the user to check information through the tax declaration intelligent digital employee process, and after the user confirms an instruction, completing the tax declaration. And the user operation is automatically replaced. According to the invention, by combining a dialogue voice engine, a large language model and intelligent digital employees, the tax declaration work of memorizing a path or contrasting an operation guide and manually operating an electronic tax bureau system is changed into voice type dialogue tax declaration, so that the workload of tutoring staff of the tax bureau is reduced.

Description

Technical Field

[0001] The present invention relates to the field of tax declaration business, and in particular to an intelligent digital employee voice conversational tax declaration method based on a large language model. Background Art

[0002] With economic development, taxpayers have more and more frequent demands for tax returns, but the existing tax return business system has certain drawbacks, and there is limited room for improvement in convenient tax returns. Taxpayers need to remember cumbersome operating steps, and the gradual improvement of the tax system has made the tax return workflow complicated, requiring a large number of grassroots tax personnel to guide taxpayers in operation. Existing tax hall solutions, such as teaching videos, comparative dual-screen step-by-step operation instructions or step-by-step instructions, require taxpayers to read and understand the content before operating, which is time-consuming and laborious, and the effect is not ideal. In order to improve taxpayers' self-service tax return experience, a new tax service model needs to be provided. However, the business system of the electronic tax bureau lacks simple service operations, and the third-party business system does not have permission optimization. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, provide a voice conversational tax reporting method for intelligent digital employees based on a large language model, provide conversational tax reporting services by connecting external intelligent digital employees and using an artificial intelligence conversation engine, and further enhance the taxpayer experience.

[0004] Another object of the present invention is to provide an intelligent digital employee voice conversational tax reporting system based on a large language model.

[0005] The purpose of the present invention is achieved through the following technical solutions: The intelligent digital employee voice conversation tax declaration method based on a large language model includes the following steps: S1, receiving the target voice of tax filing intention input by the user; S2. The conversational speech engine converts the target speech of the tax filing intention into text to obtain a first text, and then performs preprocessing and feature extraction on the first text to obtain a feature vector of the first text; S3, the feature vector of the first text is input into the large language model, and the large language model is given the probability distribution of the next word in the context of the first text, and finally the probability distribution information of the first text is obtained; S4, inputting the probability distribution information of the first text into the intent classifier to obtain the intent prediction result of the first text; S5. If the intention prediction result of the first text matches the tax declaration situation, the intelligent digital employee process automation is awakened and the process goes to step S6; if not, a reminder is given that the service is not in scope; S6. The intelligent digital employee process automation interface calls the conversational voice engine to guide the user to voice-output the required reporting information, and obtains the second text set through text conversion. The tax filing intelligent digital employee process automation reminds the user to check the information. After obtaining the user's confirmation instruction, it automatically replaces the user to operate the natural person electronic tax bureau to submit the reporting information.

[0006] In step S2, the preprocessing and feature extraction are completed through a preprocessing model. The preprocessing model converts the words of the first text into vector representations through word embedding, captures the semantic relationship and contextual information between words, and realizes the conversion of the first text data into a vector representation with higher semantic information.

[0007] In step S2, the feature vector of the first text is obtained in the following manner: a) Corpus initialization: Collect text data sets generated by tax declaration dialogue data and tax proper name data for training word embedding models; b) The preprocessing model performs a word segmentation operation on the first text to split it into single sequences, removes stop word information according to a public stop word database, and obtains processed data; c) The preprocessing model counts the word frequencies in the word segmentation results of the processed data, and builds a vocabulary based on the frequencies. Then, based on the vocabulary and corpus, it inputs the word embedding model Word2Vec to obtain the final word embedding representation, that is, the final word embedding vector; d) Feature extraction: The user inputs data into the trained word embedding model, converts each word into a corresponding vector representation, and calculates the feature vector f(x) using the following formula: f(x) = [w_1, w_2, ..., w_n]; Among them, w_i represents the word embedding vector of the i-th word.

[0008] In step S6, the second text set is a preset information group that needs to be collected for process automation, including declaration month, declaration cost, and total income, and the corresponding value content is the user input content.

[0009] In step S6, the operation process of the intelligent digital employee process automation is as follows: i. Call the conversational voice engine to obtain the form filling parameter information required for the process; ii. Automatically obtain the QR code for logging into the Electronic Taxation Bureau for natural persons and guide the user to scan the code to log into the system; iii. When logged in, the tax declaration intelligent digital employee process will automatically click on the corresponding page HTML element according to the script preset path, open the electronic tax bureau tax declaration process menu page, and enter the tax declaration parameter page; iv. Fill the second text set parameter content obtained above into the corresponding input box according to the preset position. After completing the information filling, check whether the required elements on the page are completed according to the required * mark. If all the information is filled in, click the submit button element to complete the process instead of manually submitting the tax return process; v. Collect the submission success mark based on the submitted page, call the aforementioned conversational voice engine on the front end through the interface, and announce the successful tax return result.

[0010] Another object of the present invention is achieved by the following technical solutions: The intelligent digital employee voice conversational tax declaration system based on a large language model includes a conversational voice engine, a preprocessing model, a large language model, and an intelligent digital employee process automation module; The conversational speech engine converts the target speech of the tax filing intention into text to obtain a first text; The preprocessing model then preprocesses and extracts features from the first text to obtain a feature vector of the first text; The feature vector of the first text is input into the large language model, and the large language model is given the probability distribution of the next word in the context of the first text, and finally the probability distribution information of the first text is obtained; If the intention prediction result of the first text hits the tax filing situation, the intelligent digital employee process automation will be awakened. The intelligent digital employee process automation interface will call the conversational voice engine to guide the user to voice output the required reporting information, and obtain the second text set through text conversion. The tax filing intelligent digital employee process automation will remind the user to check the information. After obtaining the user's confirmation instruction, it will automatically replace the user to operate the natural person electronic tax bureau to submit the reporting information.

[0011] The large language model is Tongyi Qianwen 14B.

[0012] At the same time, the present invention provides: A server includes a processor and a memory, wherein the memory stores at least one program, and the program is loaded and executed by the processor to implement the above-mentioned intelligent digital employee voice conversational tax reporting method based on a large language model.

[0013] A computer-readable storage medium stores at least one program, which is loaded and executed by a processor to implement the above-mentioned intelligent digital employee voice conversational tax reporting method based on a large language model.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects: The present invention combines a conversational speech engine, a large language model and intelligent digital employees to realize the transformation of the tax declaration work in the electronic tax bureau system from the original work of remembering the path or referring to the operation guide to manual operation to voice-based conversational tax declaration, thereby reducing the workload required for guidance by tax bureau staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of the intelligent digital employee voice conversational tax reporting method based on a large language model described in the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0017] like Figure 1 The intelligent digital employee voice conversation tax declaration method based on a large language model includes the following steps: 1. The program receives the target voice of tax filing intention input by the user.

[0018] 2. The conversational speech engine converts the target speech into text to obtain the first text, and then preprocesses the first text to extract features and obtain feature vector information. The preprocessing model uses word embedding technology to convert words into vector representations, capture the semantic relationship and context information between words, and realize the conversion of the first text data into a vector representation with higher semantic information.

[0019] a) Corpus initialization: Collect a large amount of tax declaration dialogue data and tax proper name data to generate text data sets for training word embedding models.

[0020] b) The preprocessing model performs a word segmentation operation on the first text to split it into single sequences, removes stop word information according to a public stop word database, and obtains processed data.

[0021] c) The preprocessing model will count the word frequencies in the word segmentation results of the processed data, and build a vocabulary based on the frequencies. Then, based on the aforementioned vocabulary and corpus, it is input into the word embedding model Word2Vec to obtain the final word embedding representation.

[0022] In natural language processing (NLP), "word embedding representation" and "word embedding vector" usually refer to the same concept. They are both used to describe the representation of words in vector space, which is learned by algorithms and can capture the semantic and syntactic information of words.

[0023] "Word embedding representation" and "word embedding vector" are the same concept, both referring to the representation of words in vector space. This representation enables computers to understand and process language through mathematical operations between vectors, such as calculating the similarity between words or the overall meaning of a sentence.

[0024] The preprocessing model processes the data and counts the word frequency, and then builds a vocabulary. The vocabulary is a collection of all words that will be used to train word embedding models such as Word2Vec. Through the Word2Vec model, each word will be mapped to a point in a high-dimensional space, and this point is a vector, which is also a word embedding vector.

[0025] d) Extract features: Apply the trained word embedding model to the user input data and convert each word into a corresponding vector representation. The feature vector can be calculated using the following formula: f(x) = [w_1, w_2, ..., w_n]; Among them, w_i represents the word embedding vector of the i-th word.

[0026] "Extracting features" means converting each word in the text data input by the user into a corresponding vector representation through a trained word embedding model. The "word embedding vector" here is the vector representation of each word obtained by Word2Vec in step c).

[0027] 3. Input the feature vector obtained above into the large language model. The large language model gives the probability distribution of the next word in the context, and finally obtains the probability distribution information. Input the probability distribution into the intent classifier to obtain the final intent prediction result. If the prediction result hits the preset tax declaration intelligent digital employee process automation, the process script will be awakened.

[0028] Flow example: Assume that the trained large language model is M and the intent classifier is C, the first text is X, and the user's intent needs to be identified. First, the first text entered by the user needs to be preprocessed and feature extracted to obtain a feature vector f(x). Then, the feature vector is input into the large language model to obtain a probability distribution P(y|x, M), where y represents the user's intent category. Finally, the probability distribution is input into the intent classifier to obtain the final intent prediction result C(P(y|x, M)).

[0029] 4. The intelligent digital employee process automation interface for tax reporting calls the conversational voice engine to guide the user to voice-output the required reporting information, and obtains the second text set through text conversion. The intelligent digital employee process automation for tax reporting reminds the user to check the information, and after obtaining the user's confirmation instruction, it automatically replaces the user to operate the natural person electronic tax bureau to submit the reporting information.

[0030] a) Second text set: presets the information groups that need to be collected for process automation, such as declaration month, declaration cost, total revenue and other parameter information, and the corresponding value content is the user input content.

[0031] b) Steps for automating the tax declaration process using the intelligent digital employee: i. Call the conversational speech engine to obtain the form filling parameter information required for the process.

[0032] ii. Automatically obtain the natural person electronic tax bureau login QR code and guide the user to scan the code to log in to the system.

[0033] iii. When logged in, the intelligent digital employee tax declaration process will be automatically automated according to the script preset path, automatically click on the corresponding page HTML element, open the electronic tax bureau tax declaration process menu page, and enter the tax declaration parameter page.

[0034] iv. Fill the second text set parameter content obtained above into the corresponding input box according to the preset position. After completing the information filling, check whether the required elements on the page are completed according to the required * mark. If all the information is filled in, click the Submit button element to complete the process instead of manually submitting the tax return process.

[0035] v. Collect the submission success mark based on the submitted page, call the aforementioned conversational voice engine on the front end through the interface, and announce the successful tax return result.

[0036] The intelligent digital employee voice conversational tax declaration system based on a large language model includes a conversational voice engine, a preprocessing model, a large language model, and an intelligent digital employee process automation module; The conversational speech engine converts the target speech of the tax filing intention into text to obtain a first text; The preprocessing model then preprocesses and extracts features from the first text to obtain a feature vector of the first text; The feature vector of the first text is input into the large language model, and the large language model is given the probability distribution of the next word in the context of the first text, and finally the probability distribution information of the first text is obtained; If the intention prediction result of the first text hits the tax filing situation, the intelligent digital employee process automation will be awakened. The intelligent digital employee process automation interface will call the conversational voice engine to guide the user to voice output the required reporting information, and obtain the second text set through text conversion. The tax filing intelligent digital employee process automation will remind the user to check the information. After obtaining the user's confirmation instruction, it will automatically replace the user to operate the natural person electronic tax bureau to submit the reporting information.

[0037] The large language model is Tongyi Qianwen 14B.

[0038] At the same time, the present invention provides: A server includes a processor and a memory, wherein the memory stores at least one program, and the program is loaded and executed by the processor to implement the above-mentioned intelligent digital employee voice conversational tax reporting method based on a large language model.

[0039] A computer-readable storage medium stores at least one program, which is loaded and executed by a processor to implement the above-mentioned intelligent digital employee voice conversational tax reporting method based on a large language model.

[0040] The intelligent digital employee in this embodiment is a virtual workforce based on artificial intelligence and robotic process automation (RPA) technology, also known as a digital virtual workforce. It can automatically perform repetitive, regular and predictable tasks. It is suitable for business scenarios with large data processing volume, high work repetition, clear process rules and cross-business system operations. By automating existing business processes and quickly responding to future new businesses, it can effectively improve the operational efficiency of business departments and reduce operating costs.

[0041] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. An intelligent digital employee voice conversational tax reporting method based on a large language model, characterized in that: The following steps are involved: S1, receiving the target voice of tax filing intention input by the user; S2. The conversational speech engine converts the target speech of the tax filing intention into text to obtain a first text, and then performs preprocessing and feature extraction on the first text to obtain a feature vector of the first text; S3, the feature vector of the first text is input into the large language model, and the large language model is given the probability distribution of the next word in the context of the first text, and finally the probability distribution information of the first text is obtained; S4, inputting the probability distribution information of the first text into the intent classifier to obtain the intent prediction result of the first text; S5. If the intention prediction result of the first text matches the tax declaration situation, the intelligent digital employee process automation is awakened and the process goes to step S6; if not, a reminder is given that the service is not in scope; S6. The intelligent digital employee process automation interface calls the conversational voice engine to guide the user to voice-output the required reporting information, and obtains the second text set through text conversion. The tax filing intelligent digital employee process automation reminds the user to check the information. After obtaining the user's confirmation instruction, it automatically replaces the user to operate the natural person electronic tax bureau to submit the reporting information.

2. According to claim 1, the intelligent digital employee voice conversational tax reporting method based on a large language model is characterized in that: In step S2, the preprocessing and feature extraction are completed through a preprocessing model. The preprocessing model converts the words of the first text into vector representations through word embedding, captures the semantic relationship and contextual information between words, and realizes the conversion of the first text data into a vector representation with higher semantic information.

3. According to claim 1, the intelligent digital employee voice conversational tax reporting method based on a large language model is characterized in that: In step S2, the feature vector of the first text is obtained in the following manner: a) Corpus initialization: Collect text data sets generated by tax declaration dialogue data and tax proper name data for training word embedding models; b) The preprocessing model performs a word segmentation operation on the first text to split it into single sequences, removes stop word information according to a public stop word database, and obtains processed data; c) The preprocessing model counts the word frequencies in the word segmentation results of the processed data, and builds a vocabulary based on the frequencies. Then, based on the vocabulary and corpus, it inputs the word embedding model Word2Vec to obtain the final word embedding representation, that is, the final word embedding vector; d) Feature extraction: The user inputs data into the trained word embedding model, which converts each word into a corresponding vector representation; The eigenvector f(x) is calculated using the following formula: f(x) = [w_1, w_2, ..., w_n]; Among them, w_i represents the word embedding vector of the i-th word.

4. According to claim 1, the intelligent digital employee voice conversational tax reporting method based on a large language model is characterized in that: In step S6, the second text set is a preset information group that needs to be collected for process automation, including declaration month, declaration cost, and total income, and the corresponding value content is the user input content.

5. According to claim 1, the intelligent digital employee voice conversation tax reporting method based on a large language model is characterized in that: In step S6, the operation process of the intelligent digital employee process automation is as follows: i. Call the conversational voice engine to obtain the form filling parameter information required for the process; ii. Automatically obtain the QR code for logging into the Electronic Taxation Bureau for natural persons and guide the user to scan the code to log into the system; iii. When logged in, the tax declaration intelligent digital employee process will automatically click on the corresponding page HTML element according to the script preset path, open the electronic tax bureau tax declaration process menu page, and enter the tax declaration parameter page; iv. Fill the second text set parameter content obtained above into the corresponding input box according to the preset position. After completing the information filling, check whether the required elements on the page are completed according to the required * mark. If all the information is filled in, click the submit button element to complete the process instead of manually submitting the tax return process; v. Collect the submission success mark based on the submitted page, call the aforementioned conversational voice engine on the front end through the interface, and announce the successful tax return result.

6. Intelligent digital employee voice conversational tax declaration system based on large language model, characterized by: It includes a conversational speech engine, a preprocessing model, a large language model, and an intelligent digital employee process automation module; The conversational speech engine converts the target speech of the tax filing intention into text to obtain a first text; The preprocessing model then preprocesses and extracts features from the first text to obtain a feature vector of the first text; The feature vector of the first text is input into the large language model, and the large language model is given the probability distribution of the next word in the context of the first text, and finally the probability distribution information of the first text is obtained; If the intention prediction result of the first text hits the tax filing situation, the intelligent digital employee process automation will be awakened. The intelligent digital employee process automation interface will call the conversational voice engine to guide the user to voice output the required reporting information, and obtain the second text set through text conversion. The tax filing intelligent digital employee process automation will remind the user to check the information. After obtaining the user's confirmation instruction, it will automatically replace the user to operate the natural person electronic tax bureau to submit the reporting information.

7. The intelligent digital employee voice conversational tax declaration system based on a large language model according to claim 6 is characterized in that: The large language model is Tongyi Qianwen 14B.

8. A server, comprising a processor and a memory, characterized in that: At least one program is stored in the memory, and the program is loaded and executed by the processor to implement the intelligent digital employee voice conversational tax reporting method based on a large language model as described in any one of claims 1 to 5.

9. A computer-readable storage medium, wherein at least one program is stored in the storage medium, characterized in that: The program is loaded and executed by the processor to implement the intelligent digital employee voice conversational tax reporting method based on a large language model as described in any one of claims 1 to 5.

Images