Intelligent filling system and method based on large model

Abstract

The invention discloses an intelligent filling system and method based on a large model, and the method comprises the steps: collecting the current call content, converting the call content into a text through a voice recognition large model, and obtaining a call text; carrying out preprocessing operation on the call text, and carrying out data analysis on the preprocessed call text by utilizing a text analysis large model to obtain an analysis result; making a plurality of work order templates, forming a work order database, and matching the analysis result with the work order database to obtain a current work order template; automatically filling an analysis result into a current work order template; and obtaining the emergency degree of the current service according to the analysis result, updating the emergency degree to the current work order template, and obtaining the service work order of the current call. The invention relates to the technical field of customer service, and solves the technical problems of single application scene and low intelligent degree in the existing intelligent form filling method.

Description

Technical Field

[0001] The present invention belongs to the field of customer service and relates to machine learning technology, specifically to an intelligent form filling system and method based on a large model. Background Art

[0002] Modern customer service systems are undergoing rapid development and change, facing the growing diversity and personalization of the public, as well as the ever-increasing work efficiency requirements. In this context, customer service systems not only need to be able to respond to customer questions and requests quickly and accurately, but also need to be able to provide efficient and convenient service experience. However, traditional customer service work methods, such as manual form filling, often have problems such as low work efficiency, low information accuracy, and long customer waiting time. These problems have seriously restricted the development of customer service systems and the improvement of service quality.

[0003] The patent with application number 202410300765.0 discloses "an intelligent automatic form filling method", which includes: building a user history form library; obtaining the form information and activity type filled in by the user; matching the corresponding relevant time, relevant content, and related activities from the historical form according to the current time or set time, the parsed form information, and the activity type, and filling them into the current form; based on the matched filled content and the user's historical filling habits, recommending the content items that need to be filled in.

[0004] The patent with application number 202010684711.0 discloses "a real-time call automatic form filling system and form filling method", in which the system includes: a real-time information point recognition module: used to convert real-time voice into recognized text; a real-time information point extraction module: used to extract business information points from the recognized text; an information point verification module: used to verify the business information points and push the verification results to the agent end; an information point modification module: used for the agent end to modify the information points according to the verification results; an information point saving module: used to save the modified information points; an information point push module: used to push the saved list to the customer system.

[0005] However, the existing intelligent form filling technology has the problem of single application scenario and low intelligence. For example, the above-mentioned intelligent automatic form filling method, by matching the current form filling information and the historical form information, fills the historical information into the current form, and then recommends the content to be filled in according to the user's filling habits. This method has an obvious technical problem of single application scenario, that is, the intelligent form filling method is only suitable for collecting information, requiring customers to fill in the form manually, and cannot automatically extract and fill in information in the real-time call scenario; and in the above-mentioned real-time call automatic form filling system and form filling method, although it can recognize and extract business information points from real-time voice, and perform operations such as verification, modification, saving and pushing, the degree of intelligence still needs to be improved. For example, in the process of voice recognition, it may be affected by factors such as environmental noise and accent, resulting in inaccurate recognition of text, which in turn affects the accuracy of extraction of business information points. At the same time, the system has the link of manual verification and modification, which to a certain extent reduces the degree of automation and work efficiency, and manual operation may introduce inconsistency and errors. Summary of the invention

[0006] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent form filling system and method based on a large model, which is used to solve the technical problems existing in the existing intelligent form filling methods, such as the single application scenario and the low degree of intelligence.

[0007] To achieve the above objectives, the first aspect of the present invention provides an intelligent form filling system based on a large model, comprising:

[0008] Speech transcription module: used to collect the current call content and convert the current call content into text using a speech recognition model to obtain the call text; wherein the speech recognition model is built based on a machine learning algorithm;

[0009] Data analysis module: used to pre-process the call text, and use the text analysis model to perform data analysis on the pre-processed call text to obtain analysis results; wherein the text analysis model is built based on a machine learning algorithm, and the analysis results include the current customer's personal information, product type, service type, processing flow, and main issues;

[0010] Work order matching module: used to create several work order templates and form a work order database, match the analysis result with the work order database, and obtain the current work order template;

[0011] Automatic form filling module: used to automatically fill the analysis results into the current work order template; and,

[0012] The urgency of the current service is obtained according to the analysis result, and the urgency is updated to the current work order template to obtain the service work order for the current call.

[0013] Based on the above-mentioned technical modules, an efficient and accurate intelligent form-filling process can be realized; through the voice transcription module, the call content can be quickly and accurately converted into text, saving manual recording time and improving the efficiency of information acquisition; through the data analysis module, in-depth mining of call texts and extraction of key information can be realized, providing accurate data support for subsequent work order processing; through the work order matching module, accurate work order template selection can be achieved, ensuring that the service process is highly consistent with customer needs; through the automatic form-filling module, automatic filling of analysis results into work order templates and accurate assessment and marking of service urgency can be achieved, optimizing the overall efficiency of service resource allocation and work order processing, improving service quality and response speed, and providing enterprises with more scientific decision-making basis and more efficient customer service solutions.

[0014] Furthermore, the speech recognition model is constructed based on a machine learning algorithm, including:

[0015] A1, extracting the regional information of all customers from the customer database to obtain several target regions;

[0016] A2, collecting a number of Mandarin speech data and a number of dialect speech data of the audience area to obtain a speech data set;

[0017] A3, preprocessing the speech data set, and annotating the preprocessed data to obtain a text annotation set, and dividing the speech data set and the text data set into a training set and a verification set according to a preset ratio;

[0018] A4, use machine learning algorithms to build a neural network model, use training sets and validation sets to train the neural network model, and iteratively optimize the neural network model based on the validation results;

[0019] A5, save the neural network model with the best verification result as output, and obtain a large speech recognition model with speech data as input and text data as output.

[0020] By extracting the audience areas of the customer groups and collecting the dialect speech data sets of the audience areas, combined with the Mandarin speech data sets, we can comprehensively cover the language expression patterns of customers, broaden the scope of application of large speech recognition models, effectively reduce the misreading or omission of information caused by language diversity, and improve the accuracy and completeness of speech transcription.

[0021] Furthermore, the text analysis model is constructed based on a machine learning algorithm, including:

[0022] B1, collecting a number of call texts, and performing a preprocessing operation of text cleaning and text normalization on the number of call texts to obtain a number of preprocessed call texts;

[0023] B2, manually annotating the call texts to obtain an annotated data set corresponding to the call texts, wherein the annotated data set includes personal information, product type, service type, processing flow, and main issues;

[0024] B3, using a text representation method to convert a number of call texts into numerical feature vectors to obtain a number of text features, and combining the number of text features and the corresponding annotation data sets into a model data set;

[0025] B4, using machine learning algorithms to build a text sequence model, inputting the model data set into the text sequence model for training and verification, and iteratively optimizing the text sequence model based on the verification results to obtain a text sequence model with the best verification results;

[0026] B5, mark the text sequence model with the best verification result as the text analysis big model, and obtain the text analysis big model whose input is the call text and output is the data analysis result.

[0027] By using the text analysis big model to obtain the customer's personal information, the type of product being consulted, the type of service being requested, the processing procedures that need to be carried out, and the main issues of this call, we can accurately grasp and fully understand the customer's needs, provide an accurate data basis for the subsequent automatic filling of forms, improve the accuracy of information and filling efficiency, and help staff understand the key points of customer demands at a glance.

[0028] Furthermore, matching the analysis result with the work order database includes:

[0029] C1, extract key information of several work order templates in the work order database to obtain several key information; the key information includes: work order template ID, product type, service type, and processing flow;

[0030] C2, construct a product index tree based on the product type according to some key information, and match the product type in the analysis result from the product index tree to obtain a preliminary matching set;

[0031] C3, using the cosine similarity calculation formula to obtain the similarity between the service type in the preliminary matching set and the service type in the analysis result, and obtain several similarities;

[0032] C4, select the work order template ID with the highest similarity score from several similarities in the preliminary matching set to obtain the current work order template.

[0033] By calculating the similarity between several work order templates and the current analysis results, the most matching work order template can be obtained, which can achieve accurate work order selection and adaptation, avoiding service delays or errors caused by mismatched work orders. Synchronization also helps companies collect and organize more standardized and consistent data, helping to gain in-depth insights into customer demand patterns, product problem distribution, and the effectiveness of service processes, and then adjust product strategies and optimize service processes in a targeted manner.

[0034] Furthermore, obtaining the urgency of the current service according to the analysis result includes:

[0035] D1, connect the current customer's historical database according to the current customer's personal information in the analysis results, and count the service records and consumption records in the historical database; the service records include the number of services, service time, problem level and service score, and the problem level is represented by numbers, 1 represents a simple problem, 2 represents a medium problem, and 3 represents a complex problem. The consumption records include the number of consumptions and the amount of consumption;

[0036] D2, according to the formula Calculate the service frequency a 1 , average consumption amounta 2 And the average problem complexity a 3 , where S n Indicates the number of services, S t2 Indicates the last service time, S t1 Indicates the first service time, C m represents the consumption amount of the mth consumption, P j represents the problem level of the jth service, S j Indicates the service rating for each time;

[0037] D3, according to the formula Calculate the urgency value E of the current service; where ω j (i=1,2,3) represents the weight value of each item, which is obtained from the weight division table, S represents the service type in the analysis result, P represents the product type in the analysis result, and L represents the user level score;

[0038] D4, obtaining the urgency of the current service according to the urgency value E and the urgency level classification table.

[0039] The customer's user level is obtained based on the customer's historical database, and combined with the analysis results of the current call, the urgency of the current service can be obtained. The service priority within Jingzhu can be obtained, so that service resources can be allocated more reasonably, ensuring that those truly urgent and important customer needs can receive the most timely response and processing, thereby enhancing customer trust and loyalty to the company.

[0040] A second aspect of the present invention provides an intelligent form filling method based on a large model, comprising:

[0041] S1, collect the current call content, convert the call content into text using the speech recognition model, and obtain the call text;

[0042] S2, preprocessing the call text, and using the text analysis model to perform data analysis on the preprocessed call text to obtain analysis results;

[0043] S3, preparing several work order templates and forming a work order database, matching the analysis result with the work order database, and obtaining the current work order template;

[0044] S4, automatically filling the analysis result into the current work order template; and,

[0045] The urgency of the current service is obtained based on the analysis results, and the urgency is updated to the current work order template to obtain the service work order for the current call.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] By collecting Mandarin and dialect speech data from the target area to train the speech recognition model, the language expression patterns of the target customers are fully covered, the application scenarios of the speech recognition model are broadened, the information misinterpretation or omission caused by language diversity is avoided, and the efficiency of information acquisition is improved; the text analysis model is used to accurately extract multi-dimensional information such as customer personal information, product type, service type, processing flow and main issues, and work order matching and urgency assessment are performed based on this information, making the system and method applicable to various business scenarios related to products and services. Whether it is product consultation, after-sales service or technical support, efficient work order processing can be performed through the system; by constructing a product index tree and calculating similarity, the most matching work order template in the current service scenario can be found from the work order database, avoiding service delays or errors caused by mismatched work orders; the urgency of the service is calculated based on the customer's historical database and the current call analysis results, realizing the reasonable allocation of service resources and ensuring that urgent and important customer needs are responded to in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] 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 work.

[0049] Figure 1A technical flow chart of an intelligent form filling system based on a large model provided by the present invention;

[0050] Figure 2 A framework diagram of an intelligent form filling system based on a large model provided by the present invention;

[0051] Figure 3 This is a technical flow chart of the work order matching module provided by the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. 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.

[0053] See also Figure 1 - Figure 3 The first embodiment of the present invention provides an intelligent form filling system based on a large model, including:

[0054] Speech transcription module: used to collect the current call content and convert the call content into text using a large speech recognition model to obtain the call text; the large speech recognition model is built based on a machine learning algorithm;

[0055] Data analysis module: used to pre-process the call text and perform data analysis on the pre-processed call text to obtain analysis results; the text analysis model is built based on the machine learning algorithm, and the data analysis results include the current customer's personal information, product type, service type, processing flow, and main issues;

[0056] Work order matching module: used to create several work order templates and form a work order database, match the analysis results with the work order database, and obtain the current work order template;

[0057] Automatic form filling module: used to automatically fill in the analysis results into the corresponding fields in the current work order template, and obtain the urgency of the current service based on the analysis results and update it into the current work order template to obtain the service work order for the current call;

[0058] It should be noted that the speech transcription module, data analysis module, work order matching module and automatic form filling module in this system are communicatively connected.

[0059] In this embodiment, before constructing the speech recognition model, the influence of dialects on call-to-text conversion was taken into consideration. Therefore, the regional information of all customers was first extracted from the customer database to clarify the target audience of the company's products and ensure that the model can adapt to the language characteristics of customers in different regions. Then, a number of Mandarin speech data and dialect speech data of the corresponding target audience areas were widely collected and integrated to obtain a rich speech data set, which improved the applicability and accuracy of the model in practical applications.

[0060] Then, the collected speech data set is preprocessed, including denoising, normalization and other operations to improve data quality, and the preprocessed data is rigorously annotated to form a text annotation set; then the speech data set and text data set are divided into training set and validation set according to the preset ratio, where the training set is used for model learning and parameter adjustment, and the validation set is used to evaluate the performance and generalization ability of the model, providing a basis for subsequent optimization;

[0061] Afterwards, a neural network model is constructed using a machine learning algorithm to learn the complex mapping relationship between speech features and text from the speech data; the neural network model is trained using the training set to continuously adjust its parameters to adapt to the data pattern, and during the training process, the neural network model is iteratively optimized based on the results of the validation set, and the accuracy and stability of the model are improved by continuously adjusting the model's structure and parameters, so that it can better fit the actual data; finally, the neural network model with the best validation result is saved as the output, thereby obtaining a large speech recognition model with speech data as input and text data as output, which can accurately convert the customer's speech into text in actual applications, provide reliable speech recognition support for the intelligent form filling system, effectively solve the impact of factors such as dialects on call-to-text conversion, and improve the intelligence level and service efficiency of the entire system.

[0062] After obtaining the speech recognition model, the current call content is collected and converted into text using the speech recognition model to obtain the call text;

[0063] It should be noted that the primary purpose of collecting call data in the present invention is to improve the quality and efficiency of customer service, and only collect call data directly related to achieving the primary purpose, strictly abide by the "Personal Information Protection Law of the People's Republic of China" and other relevant laws and regulations, ensure that all operations have a clear legal basis, and do not engage in any illegal or irregular data processing activities.

[0064] After preprocessing the call text, the text analysis model is used to perform data analysis on the preprocessed call text to obtain analysis results including the current customer's personal information, product type, service type, processing flow, and main issues. The construction process of the text analysis model is as follows:

[0065] First, collect a large amount of call text data from sources such as customer service systems and call record databases. These data should cover different customer groups, business scenarios, and time ranges to ensure that the model has wide adaptability; organize professionals to annotate the collected call texts, including customer personal information (such as name, contact information, age, etc.), product type (clearly specify the name or category of the product being consulted), service type (such as product consultation, sales support, technical support, after-sales service, account management, other services), and main issues (summarize the customer's core demands in concise language);

[0066] Then, the call text data is cleaned and preprocessed to remove noise data in the call text, such as irrelevant symbols, garbled characters, repeated characters, etc., and the abbreviations, slang and colloquial expressions in the text are processed; then a vocabulary is constructed to count the frequency of occurrence of each word in the call text, which can be easily implemented using Python's Counter library. For example, for a call text "I want to consult about a mobile phone issue", it is counted that "I" appears 1 time, "want" appears 1 time, "consult" appears 1 time, etc.; then the TF-IDF (term frequency-inverse document frequency) value of the word is calculated to measure the importance of the word in the text; then the n-grams (n-tuple) features in the text are extracted, such as bi-grams (bigrams), tri-grams (triplets), etc., to capture the order and combination information between words, and obtain the vocabulary features in several call texts;

[0067] Use natural language processing tools such as NLTK or spaCy to identify the part of speech of each word in the text, such as noun, verb, adjective, etc., and analyze the sentence structure to extract the length, type (declarative sentence, interrogative sentence, exclamatory sentence, etc.) and subject, predicate, and object components of the sentence to obtain the syntactic features of several call texts;

[0068] Use pre-trained word vector models, such as Word2Vec and GloVe, to convert words in the text into vector representations to capture the semantic information of the words. For sentence-level semantic representation, consider using methods based on recurrent neural networks (RNNs) or convolutional neural networks (CNNs) to obtain sentence vector representations and obtain semantic features of several call texts.

[0069] Next, the lexical features, syntactic features, and semantic features are used as text features, and the text features and the corresponding annotation data sets are combined into a model data set;

[0070] Finally, based on the machine learning algorithm, a suitable model structure is selected for construction and training, such as recurrent neural networks (RNN) and its variants, convolutional neural networks (CNN), models based on the Transformer architecture, etc., to obtain a text sequence model; then the model dataset is input into the text sequence model for training and verification, and the text sequence model is iteratively optimized according to the verification results to obtain the text sequence model with the best verification results, and marked as a text analysis model, thus obtaining a text analysis model with call text as input and data analysis results as output.

[0071] Through the large text analysis model, key information can be quickly and accurately extracted from seemingly disorganized call content, which not only greatly improves the efficiency of information processing and reduces the time and cost of manual analysis, but also avoids information omissions or misjudgments due to human factors, helping service personnel to fully understand customer needs at the first time and providing an accurate data basis for subsequent automatic form filling.

[0072] In the actual customer service system, different products and service types may correspond to different work order templates. In the process of manual form filling, the staff may not be able to identify the work order properly, resulting in incorrect work order selection, which will affect the service efficiency and quality, and may even delay the resolution of customer problems. Therefore, in this system, a work order matching module is set up to match the most suitable work order template based on the analysis results of the call content, improve the accuracy and efficiency of work order selection, and ensure that each work order can be correctly assigned to the corresponding processing flow.

[0073] Specifically, first, several work order templates in the work order database are processed to extract key information from them; the key information includes the work order template ID, product type, service type and processing flow. For example, a work order template has a unique ID for identification, such as "GT-001", product types include but are not limited to "smartphones", "smart speakers", "electric cars", etc., service types include but are not limited to "product consulting", "sales support", "technical support", "after-sales service", "account management", etc., and the processing flow includes but is not limited to "customer repair report-equipment inspection-determine the maintenance plan-maintenance operation-customer acceptance"; by performing such key information extraction on each work order template, the key information in several work order templates is obtained;

[0074] Next, the product type in the extracted key information is used as the basis to construct a product index tree. The construction of the index tree is similar to a tree structure, and each node represents a product type and its related work order template key information. For example, with electronic products as the root node, there may be sub-nodes such as smartphones and tablets under it, and each sub-node is associated with the corresponding work order template key information; then, based on the product type in the analysis results, search and match in the constructed product index tree. For example, if the analysis results show that the product type is a smartphone, then find the node corresponding to the smartphone in the product index tree, and use all the work order templates associated with this node as the preliminary matching set;

[0075] For the work order templates in the preliminary matching set, further calculate the similarity between their service type and processing flow and the service type and processing flow in the analysis results. Here, the cosine similarity calculation formula is used to calculate the service type similarity and processing flow similarity respectively:

[0076] Assume that there are work order template A and work order template B in the preliminary matching set, the service type in work order template A is "after-sales service", the service type in work order template B is "technical support", and the service type in the analysis result is "technical assistance". Then, the vector of the service type in the analysis result is obtained through the pre-trained Word2Vec model as V s , the vector of service types in work order template A is V sA , the vector of service types in work order template B is V sB , calculate the similarity S between the analysis result and the work order template A according to the cosine similarity formula A for: Similarly, the similarity between the analysis result and the work order template B is S B ;

[0077] Finally, the similarity scores of each work order template are compared, and the work order template ID with the highest similarity score is selected. For example, after comparison, it is found that the comprehensive similarity score of work order template A is the highest, so work order template A is determined as the current work order template.

[0078] After determining the current work order template, the analysis results are mapped one by one to the corresponding fields in the work order template according to "personal information, product type, service type, processing flow, and main issues", and connected to the current customer's historical database based on the personal information in the analysis results. The service records and consumption records in the historical database are counted, and then the urgency of the current service is obtained based on the service records, consumption records, and the product type and service type in the analysis results. The urgency is then updated in the current work order template to obtain the service work order for the current call.

[0079] Specifically, first, according to the personal information of the current customer in the analysis results, the historical database of the current customer is connected to count the service records and consumption records in the historical database; the service records include the number of services, service time, problem level and service score, and the problem level is represented by numbers, 1 represents a simple problem, 2 represents a medium problem, and 3 represents a complex problem. The consumption records include the number of consumptions and the amount of consumption;

[0080] Then, according to the formula Calculate the service frequency a 1 , average consumption amounta 2 And the average problem complexity a 3 , where S n Indicates the number of services, S t2 Indicates the last service time, S t1 Indicates the first service time, C m represents the consumption amount of the mth consumption, P j represents the problem level of the jth service, S j Indicates the service rating for each time;

[0081] Then, according to the formula Calculate the urgency value E of the current service; where ω i (i=1,2,3) represents the weight value of each item, which is obtained from the weight division table, such as Table 1, S represents the service type in the analysis result, P represents the product type in the analysis result, L represents the user level score, and the user level of the current user is obtained by comparing the user level score and the preset level value;

[0082]

[0083] Next, the urgency of the current service is obtained according to the urgency value E and the urgency level classification table; for example, a value greater than a is set as high urgency, a value greater than b but less than a is set as medium urgency, and so on. By comparing the range of the calculated urgency E, the corresponding urgency can be obtained.

[0084] The second aspect of the present invention provides an intelligent form filling method based on a large model, comprising:

[0085] S1, collect the current call content, convert the call content into text using the speech recognition model, and obtain the call text;

[0086] S2, preprocessing the call text, and using the text analysis model to perform data analysis on the preprocessed call text to obtain analysis results;

[0087] S3, making several work order templates and forming a work order database, matching the analysis results with the work order database, and obtaining the current work order template;

[0088] S4, automatically filling the analysis results into the corresponding fields in the current work order template; and, obtaining the urgency of the current service according to the analysis results, updating the urgency into the current work order template, and obtaining the service work order for the current call.

[0089] Part of the data in the above formula is calculated by removing the dimension and taking its numerical value. The formula is a formula closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0090] Working principle of the present invention:

[0091] Through the speech transcription module, the regional information extracted from the customer database is used to determine the audience area and collect Mandarin and dialect speech data to build and train a large speech recognition model to obtain the text of the call content, that is, the call text;

[0092] Through the data analysis module, by cleaning, labeling, and feature conversion of call texts, as well as building and training a large text analysis model, we can obtain analysis results including customer personal information, product type, service type, processing flow, and major issues;

[0093] Through the work order matching module, the product index tree constructed by extracting the key information of the work order template and the calculated service type similarity are used to obtain the current work order template that best matches the analysis result;

[0094] Through the automatic form filling module, using the analysis results and the urgency information calculated by connecting to the customer history database, a service ticket for the current call filled in with the analysis results and marked with the urgency is obtained.

[0095] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An intelligent form filling system based on a large model, characterized in that: include: Speech transcription module: used to collect the current call content and convert the current call content into text using a speech recognition model to obtain the call text; wherein the speech recognition model is built based on a machine learning algorithm; Data analysis module: used to pre-process the call text, and use the text analysis model to perform data analysis on the pre-processed call text to obtain analysis results; wherein the text analysis model is built based on a machine learning algorithm, and the analysis results include the current customer's personal information, product type, service type, processing flow, and main issues; Work order matching module: used to create several work order templates and form a work order database, match the analysis result with the work order database, and obtain the current work order template; Automatic form filling module: used to automatically fill the analysis results into the current work order template; and, The urgency of the current service is obtained according to the analysis result, and the urgency is updated to the current work order template to obtain the service work order for the current call.

2. According to claim 1, the intelligent form filling system based on a large model is characterized in that: The speech recognition model is built based on a machine learning algorithm, including: A1, extracting the regional information of all customers from the customer database to obtain several target regions; A2, collecting a number of Mandarin speech data and a number of dialect speech data of the audience area to obtain a speech data set; A3, preprocessing the speech data set, and annotating the preprocessed data to obtain a text annotation set, and dividing the speech data set and the text data set into a training set and a verification set according to a preset ratio; A4, use machine learning algorithms to build a neural network model, use training sets and validation sets to train the neural network model, and iteratively optimize the neural network model based on the validation results; A5, save the neural network model with the best verification result as output, and obtain a large speech recognition model with speech data as input and text data as output.

3. According to claim 1, the intelligent form filling system based on a large model is characterized in that: The text analysis model is built based on a machine learning algorithm and includes: B1, collecting a number of call texts, and performing a preprocessing operation of text cleaning and text normalization on the number of call texts to obtain a number of preprocessed call texts; B2, manually annotating the call texts to obtain an annotated data set corresponding to the call texts, wherein the annotated data set includes personal information, product type, service type, processing flow, and main issues; B3, using a text representation method to convert a number of call texts into numerical feature vectors to obtain a number of text features, and combining the number of text features and the corresponding annotation data sets into a model data set; B4, using machine learning algorithms to build a text sequence model, inputting the model data set into the text sequence model for training and verification, and iteratively optimizing the text sequence model based on the verification results to obtain a text sequence model with the best verification results; B5, mark the text sequence model with the best verification result as the text analysis big model, and obtain the text analysis big model whose input is the call text and output is the data analysis result.

4. According to claim 1, the intelligent form filling system based on a large model is characterized in that: Matching the analysis result with the work order database includes: C1, extract key information of several work order templates in the work order database to obtain several key information; the key information includes: work order template ID, product type, service type, and processing flow; C2, construct a product index tree based on the product type according to some key information, and match the product type in the analysis result from the product index tree to obtain a preliminary matching set; C3, using the cosine similarity calculation formula to obtain the similarity between the service type in the preliminary matching set and the service type in the analysis result, and obtain several similarities; C4, select the work order template ID with the highest similarity score from several similarities in the preliminary matching set to obtain the current work order template.

5. According to the large model-based intelligent form filling system of claim 1, it is characterized in that: The urgency of the current service is obtained according to the analysis result, including: D1, connect the current customer's historical database according to the current customer's personal information in the analysis results, and count the service records and consumption records in the historical database; the service records include the number of services, service time, problem level and service score, and the problem level is represented by numbers, 1 represents a simple problem, 2 represents a medium problem, and 3 represents a complex problem. The consumption records include the number of consumptions and the amount of consumption; D2, according to the formula The service frequency a1, average consumption amount a2 and average problem complexity a3 are calculated, where S n Indicates the number of services, S t2 Indicates the last service time, S t1 Indicates the first service time, C m represents the consumption amount of the mth consumption, P j represents the problem level of the jth service, S j Indicates the service rating for each time; D3, according to the formula Calculate the urgency value E of the current service; where ω i (i=1,2,3) represents the weight value of each item, which is obtained from the weight division table, S represents the service type in the analysis result, P represents the product type in the analysis result, and L represents the user level score; D4, obtaining the urgency of the current service according to the urgency value E and the urgency level classification table.

6. An intelligent form filling method based on a large model, applied to an intelligent form filling system based on a large model as claimed in any one of claims 1 to 5, characterized in that: include: S1, collect the current call content, convert the call content into text using the speech recognition model, and obtain the call text; S2, preprocessing the call text, and using the text analysis model to perform data analysis on the preprocessed call text to obtain analysis results; S3, preparing several work order templates and forming a work order database, matching the analysis result with the work order database, and obtaining the current work order template; S4, automatically filling the analysis result into the current work order template; as well as, The urgency of the current service is obtained based on the analysis results, and the urgency is updated to the current work order template to obtain the service work order for the current call.

Images