Automatic work order processing system and method based on large language model

The automated work order processing system based on a large language model enables the automatic generation and dispatch of work orders, solving the problems of low processing quality and efficiency in traditional work order systems and improving the quality and efficiency of unstructured information processing.

CN118569601BActive Publication Date: 2024-11-22QIMO TECH (GUANGZHOU) CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410885350.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-22
Estimated Expiration
2044-07-03

AI Technical Summary

Technical Problem

Traditional work order systems rely on manual processing of requests, resulting in low processing quality and efficiency, especially when dealing with unstructured information, making it difficult to effectively obtain the complete semantics of the text.

Method used

An automated work order processing system based on a large language model is adopted. Service requests are entered through user terminals, the server analyzes and generates work orders, and communication and information collection are carried out in combination with the large language model. Work orders are automatically allocated and processed, and the large language model is used for the automatic generation and dispatch of work orders.

Benefits of technology

It improved the efficiency and accuracy of work order processing, reduced manual intervention, and enhanced the quality of unstructured information processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118569601B_ABST
    Figure CN118569601B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field of work order processing, and discloses an automatic work order processing system and method based on a large language model, wherein the system comprises: a user terminal, a server and a processing terminal, wherein the user terminal and the processing terminal are both connected to the server in communication; the user terminal is used to input the user's service requirements and upload the user's service requirements to the server; the server is used to create and allocate service work orders, and to call a work order processing interface, generate a work order processing task based on the service work order, and issue the work order processing task to the processing terminal; the processing terminal is used to record the execution record of the work order processing task in the process of executing the work order processing task, and upload the execution record to the server. The present invention uses a large language model to call the work order interface, realize the automatic generation and distribution of work orders, and thus improve the efficiency and accuracy of work order processing.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of work order processing, and in particular relates to an automatic work order processing system and method based on a large language model. Background Art

[0002] Work orders are an important tool in the service system of an enterprise or organization. They are mainly used to record employees' demands, needs or problems, as well as the responses and treatment measures of the enterprise or organization. Work orders can be regarded as the evidence of the interaction between employees and the enterprise or organization, and are also the basis for the enterprise or organization to track, solve problems and evaluate service effects.

[0003] In the era of digital transformation of enterprises, with the cloudification of the third-generation support system, system changes are becoming more and more frequent. At the same time, in order to quickly respond to market changes and customer needs, the business scope of enterprises is constantly expanding, and the complexity of the system is growing rapidly. These current situations have brought new challenges to the processing of business demands. In the enterprise work order system, the processing of unstructured information in business demands is of utmost importance. Unstructured data includes text, pictures, audio and video, etc. At the same time, external input information has a series of problems such as ambiguity and uncertainty.

[0004] The traditional work order system relies solely on manual processing of requests, which has problems with low processing quality and efficiency. At the same time, the request information is mainly short text, which has the characteristics of semantic sparsity, fuzziness and uncertainty. Traditional feature extraction methods cannot obtain the complete semantics of the text. Summary of the invention

[0005] The purpose of the present invention is to provide an automatic work order processing system and method based on a large language model to solve the problem of low processing quality and efficiency in traditional work order systems that rely solely on manual processing of requests.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides an automatic work order processing system based on a large language model, the system comprising: a user terminal, a server and a processing terminal, the user terminal and the processing terminal are both communicatively connected to the server;

[0008] The user terminal is used to input the user's service requirements and upload the user's service requirements to the server;

[0009] The server comprises:

[0010] Work Order createThe module is used to obtain the user's service request, analyze the user's service request based on the large language model, obtain the user's service demand, and determine whether the user's service demand is an existing demand. If so, match the existing work order; if not, generate a work order filling task to communicate with the user, collect the necessary information required for the work order, and create a new work order based on the necessary information; save the new work order or the existing work order as a service work order in the database, and mark the service work order as a pending assignment state;

[0011] The work order allocation module is used to obtain the service work order in the pending allocation state, call the automatic allocation interface, determine the processing department to which the service work order belongs; count the work order quantity of each processing personnel in the processing department, and allocate the service work order in the pending allocation state to the processing personnel corresponding to the processing department according to the work order quantity of each processing personnel in the processing department, and mark the service work order as pending;

[0012] The work order processing module is used to call the work order processing interface, generate work order processing tasks based on the service work order, and send the work order processing tasks to the processing terminal;

[0013] The processing terminal is used to record the execution record of the work order processing task and upload the execution record to the server during the execution of the work order processing task. The server is also used to mark the service work order as processed after the work order processing task is completed; and synchronize the service work order in the processed state to the user terminal.

[0014] In a second aspect, the present invention provides an automatic work order processing method based on a large language model, the method comprising:

[0015] Analyze the service requests generated by users based on the large language model to automatically create service tickets for the users;

[0016] Determine the processing department to which the user's service ticket belongs, count the existing number of work orders of each processing personnel in the processing department, and assign the service ticket to the corresponding processing personnel in the processing department according to the existing number of work orders of each processing personnel;

[0017] After the service work order is assigned, a work order processing task is generated based on the service work order, and during the execution of the work order processing task, an execution record of the work order processing task is recorded.

[0018] Preferably, the service request generated by the user is analyzed based on the large language model to automatically create a service ticket for the user, including:

[0019] Obtaining a user's service request, inputting the service request into a large language model, and extracting a summary of the service request;

[0020] Determine whether there are keywords similar or identical to the summary of the service request in the pre-built work order library. If so, match the existing work orders based on the keywords. If not, expand the summary based on the large language model to obtain the communication question and answer text.

[0021] Based on the Q&A text, obtain the user's question intent;

[0022] Based on the large language model, necessary information in the user's question intent is extracted, and a new work order is created according to the necessary information, and the newly created work order or the matched existing work order is used as the service work order.

[0023] Preferably, after creating the service ticket of the user, the method further includes:

[0024] Get the historical work orders created by the user;

[0025] Determine whether the newly created service ticket of the user is the same as the user's historical ticket;

[0026] If yes, delete the newly created service ticket and generate a prompt text to inform the user;

[0027] Count the number of repetitions between the newly created service tickets of the user and the historical tickets of the user;

[0028] Build the processing priority of the user's historical work orders, and generate a priority increase prompt based on the number of repetitions to inform the user whether to increase the processing priority of the historical service work orders.

[0029] Preferably, determining whether the newly created service ticket of the user is a duplicate of the user's historical ticket includes:

[0030] The historical work orders of the user are split into content to obtain a first character set, and the newly created service work orders of the user are split into content to obtain a second character set;

[0031] Encode the first character set to obtain a first word vector, a first position vector, and a first embedding vector for each character in the first character set; superimpose the first word vectors, the first position vectors, and the first embedding vectors of all characters in the first character set and input them into a pre-trained text extraction model to obtain a first text vector;

[0032] Encode the second character set to obtain a second word vector, a second position vector, and a second embedding vector for each character in the second character set; superimpose the second word vectors, the second position vectors, and the second embedding vectors of all characters in the second character set and input them into a pre-trained text extraction model to obtain a second text vector;

[0033] Performing average pooling processing on the first text vector to obtain a processed first text vector; performing average pooling processing on the second text vector to obtain a processed second text vector;

[0034] Calculating the similarity between the processed first text vector and the processed second text vector;

[0035] Determine whether the similarity between the processed first text vector and the processed second text vector is greater than a preset value;

[0036] If yes, the newly created service ticket is a non-duplicate ticket; if no, the newly created service ticket is a duplicate ticket.

[0037] Preferably, the calculation expression of the similarity between the processed first text vector and the processed second text vector is:

[0038] ;

[0039] In the formula, F is the first text vector after processing With the processed second text vector The similarity of is the cosine similarity function.

[0040] Preferably, when the service work order is an existing work order, the work order processing task includes at least one solution, and the solution is used to solve the problem that needs to be solved in the service work order;

[0041] Generate a work order processing task based on the service work order, including:

[0042] Extract the issues that need to be resolved in the service ticket;

[0043] Recommend at least one solution from the solution library to solve the problem that needs to be solved in the service ticket.

[0044] Preferably, at least one solution for solving the problem to be solved in the service ticket is recommended from the solution library, including:

[0045] Based on the preset retrieval model, the problem to be solved in the service ticket is segmented to obtain several words;

[0046] Calculate the relevance of each word to each solution document in the solution library;

[0047] According to the relevance of each word to each solution document in the solution library, determine the text relevance score between the problem to be solved in the service ticket and each solution document;

[0048] Screening out solution documents whose text relevance scores are greater than a preset score to obtain a candidate solution set, wherein the candidate solution set includes a plurality of candidate solutions;

[0049] Calculate the similarity between each candidate solution and the problem to be solved in the service ticket, and obtain several similarity values;

[0050] Sort several similarity values ​​from large to small to obtain a similarity sequence;

[0051] The candidate solutions corresponding to the first K similarity values ​​in the similarity sequence are taken as the solutions to the problems that need to be solved in the service ticket.

[0052] Preferably, the functional expression of the relevance of each word to each solution document in the solution library is:

[0053] ;

[0054] In the formula, For the i Words With solution documentation d The correlation of For the i Words In the program documentation d The frequency of occurrence in k 1 is the first adjustment factor, b is the second adjustment factor, For solution documentation d The ratio of the length to the average length.

[0055] Preferably, the function expression of the text relevance score between the problem to be solved in the service ticket and each solution document is:

[0056] ;

[0057] In the formula, S Score the textual relevance of the problem to be solved in the service ticket to each solution document, For the i Words With solution documentation d The correlation of To contain words The number of documents, N is the total number of solution documents, k 0 is the adjustment coefficient, n The total number of words.

[0058] Beneficial effects:

[0059] The present invention utilizes a large language model to analyze user service requests and communicates and exchanges through natural language to complete the necessary information required for filling out work orders; at the same time, the large language model is used to call the work order interface to realize the automatic generation and distribution of work orders, thereby improving the efficiency and accuracy of work order processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0061] Figure 1 is a block diagram of an automatic work order processing system based on a large language model provided by an embodiment of the present invention;

[0062] Figure 2 It is a flow chart of an automatic work order processing method based on a large language model provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0063] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0064] Embodiment 1

[0065] Figure 1 is a block diagram of an automatic work order processing system based on a large language model provided by an embodiment of the present invention, such as Figure 1 As shown, this embodiment provides an automatic work order processing system based on a large language model, and the system includes: a user terminal, a server and a processing terminal, and the user terminal and the processing terminal are both communicatively connected to the server; wherein the user terminal can be but not limited to a personal computer (personal computer, PC), a tablet computer or a smart phone; the processing terminal can be but not limited to a personal computer (personal computer, PC), a tablet computer or a smart phone; the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud computing, network services, and middleware services.

[0066] The user terminal is used to enter the user's service requirements and upload the user's service requirements to the server. The user can submit the service requirements through applications such as applications, web applications, etc. (abbreviated as APP in English) on the user terminal. The service requirements can be data in the format of text, pictures, voice, etc., and the service requirements include a description of the problem.

[0067] The server includes: a work order creation module, a work order allocation module and a work order processing module;

[0068] The work order creation module is used to obtain the user's service request, analyze the user's service request based on the large language model, obtain the user's service demand, and determine whether the user's service demand is an existing demand. If so, match the existing work order; if not, generate a work order filling task to communicate with the user, collect the necessary information required for the work order, and create a new work order based on the necessary information; save the new work order or the existing work order as a service work order in the database, and mark the service work order as a pending assignment state;

[0069] The work order allocation module is used to obtain the service work order in the pending allocation state, call the automatic allocation interface, determine the processing department to which the service work order belongs; count the work order quantity of each processing personnel in the processing department, and allocate the service work order in the pending allocation state to the processing personnel corresponding to the processing department according to the work order quantity of each processing personnel in the processing department, and mark the service work order as pending status;

[0070] The work order processing module is used to call the work order processing interface, generate work order processing tasks based on the service work order, and send the work order processing tasks to the processing terminal;

[0071] The processing terminal is used to record the execution record of the work order processing task and upload the execution record to the server during the execution of the work order processing task. The server is also used to mark the service work order as processed after the work order processing task is completed; and synchronize the service work order in the processed state to the user terminal.

[0072] The present invention utilizes a large language model to analyze user service requests and communicates and exchanges through natural language to complete the necessary information required for filling out work orders; at the same time, the large language model is used to call the work order interface to realize the automatic generation and distribution of work orders, thereby improving the efficiency and accuracy of work order processing.

[0073] Embodiment 2

[0074] Figure 2 is a flowchart of an automatic work order processing method based on a large language model provided by an embodiment of the present invention, such as Figure 2As shown, this embodiment provides an automatic work order processing method based on a large language model. The method is implemented based on the automatic work order processing system based on a large language model in Embodiment 1. The method is mainly applied on the server side of the system. The method includes:

[0075] Step S10: Analyze the service request generated by the user based on the large language model to automatically create a service ticket for the user.

[0076] In this embodiment, a service demand is submitted in an application program (APP) on a user terminal such as a personal computer (PC), a tablet computer or a smart phone, or a web (World Wide Web) application program. The service demand may be data in the format of text, image, voice or the like. The service demand includes a description of the problem. The user terminal then uploads the service demand to the server. The server analyzes the service request generated by the user based on a large language model to automatically create a service ticket for the user.

[0077] Specifically, the service request generated by the user is analyzed based on the large language model to automatically create a service ticket for the user, including:

[0078] Step S101: obtaining a user's service request, inputting the service request into a large language model, and extracting a summary of the service request;

[0079] Step S102: determine whether there are keywords similar to or identical to the summary of the service request in the pre-built work order library. If so, match the existing work order according to the keywords; if not, expand the summary based on the large language model to obtain the communication question and answer text;

[0080] Step S103: based on the exchange question and answer text, obtaining the user's question intention;

[0081] Step S104: extracting necessary information from the user's question intention based on the large language model, creating a new work order according to the necessary information, and using the created new work order or the matched existing work order as the service work order.

[0082] In this embodiment, a large language model (LLM) is obtained by pre-training and fine-tuning a large amount of data, and can generate natural language text or understand the meaning of language text. Large language models are characterized by their large scale and billions of parameters, which help them learn complex patterns in language data. These models are usually based on deep learning architectures, such as Transformer structures, which help them achieve impressive performance on various natural language processing tasks. After fine-tuning, large language models can be widely used in downstream tasks. With authorization, large language models can use the GPT (Generative Pre-Trainin) series, the PaLM (Path-Aggregated learning for multi-task) series, the LLaMa (Large language Model MetaAi) series, etc.

[0083] As a further optimization of this embodiment, since some users may submit multiple service requests in a short period of time, and the server may create duplicate service tickets according to the service requests, in order to prevent the server from repeatedly creating service tickets; after creating the service ticket of the user, the method further includes:

[0084] Step a01: Get the historical work orders created by the user;

[0085] Step a02: Determine whether the newly created service ticket of the user is the same as the user's historical ticket;

[0086] Step a03: If yes, delete the newly created service ticket and generate a prompt text to inform the user;

[0087] Step a04: Count the number of repetitions between the newly created service ticket of the user and the historical ticket of the user;

[0088] Step a05: construct the processing priority of the user's historical work orders, and generate a priority improvement prompt according to the number of repetitions to inform the user whether to increase the processing priority of the historical service work orders.

[0089] In this embodiment, when the user submits a service request again, if the server determines that the service ticket created by the service request is a duplicate of the user's historical ticket, the newly created service ticket of the user needs to be deleted to avoid duplicate creation. At this time, the server generates a request duplicate; when the user makes service requests multiple times, it means that the problem of the service request may be aggravated. Since each service ticket has a processing priority, when there are two or more repetitions, the user is informed whether to increase the processing priority of the service ticket. If the user increases the processing priority of the service ticket, the service ticket will be allocated and processed first.

[0090] For step a02, specifically, determining whether the newly created service work order of the user is the same as the historical work order of the user includes:

[0091] Step a021: split the content of the user's historical work orders to obtain a first character set, and split the content of the newly created service work order of the user to obtain a second character set;

[0092] Step a022: Encode the first character set to obtain a first word vector, a first position vector, and a first embedding vector for each character in the first character set; superimpose the first word vectors, the first position vectors, and the first embedding vectors of all characters in the first character set and input them into a pre-trained text extraction model to obtain a first text vector;

[0093] Step a023: Encode the second character set to obtain a second word vector, a second position vector and a second embedding vector for each character in the second character set; superimpose the second word vectors, the second position vectors and the second embedding vectors of all characters in the second character set and input them into a pre-trained text extraction model to obtain a second text vector;

[0094] Step a024: performing average pooling processing on the first text vector to obtain a processed first text vector; performing average pooling processing on the second text vector to obtain a processed second text vector;

[0095] Step a025: Calculate the similarity between the processed first text vector and the processed second text vector;

[0096] Step a026: determining whether the similarity between the processed first text vector and the processed second text vector is greater than a preset value;

[0097] Step a027: If yes, the newly created service work order is a non-duplicate work order; if no, the newly created service work order is a duplicate work order.

[0098] In this embodiment, the text extraction model is constructed by training with the BERT algorithm. During the training process, the historical sample data of the enterprise is used for training. The BERT (Bidirectional Encoder Representations from Transformers) algorithm is a pre-trained language representation model that can provide powerful language understanding capabilities for downstream tasks.

[0099] In this embodiment, the calculation expression of the similarity between the processed first text vector and the processed second text vector is:

[0100] ;

[0101] In the formula, F is the first text vector after processing With the processed second text vector The similarity of is the cosine similarity function.

[0102] In this embodiment, the present invention inputs the newly created service work order and the historical work order into the text extraction model respectively, obtains the semantic vector through the pooling layer, and then calculates the cosine similarity, and deletes the newly created service work order with a similarity higher than a preset value to avoid duplicate creation of work orders, reduce the resource utilization of the server, and improve the efficiency of the server in processing work orders.

[0103] Step S20: Determine the processing department to which the user's service work order belongs, count the existing work orders of each processing personnel in the processing department, and allocate the service work order to the corresponding processing personnel in the processing department according to the existing work orders of each processing personnel.

[0104] In this embodiment, each service ticket includes the ticket number, creation time, problem content, user information and expedited status. The department to which the service ticket belongs is matched according to the problem content of each service ticket, and the service ticket is handled by relevant staff of the department.

[0105] During the allocation process, work orders cannot be allocated to processing personnel whose existing work order volume exceeds the work order volume threshold, to prevent service work orders from waiting for a long time, thereby reducing the efficiency and timeliness of service work order processing.

[0106] Step S30: after completing the allocation of the service work order, generating a work order processing task based on the service work order, and recording the execution record of the work order processing task during the execution of the work order processing task.

[0107] The present invention utilizes a large language model to analyze user service requests and communicates and exchanges through natural language to complete the necessary information required for filling out work orders; at the same time, the large language model is used to call the work order interface to realize the automatic generation and distribution of work orders, thereby improving the efficiency and accuracy of work order processing.

[0108] As a further optimization of this embodiment, when the service ticket is an existing ticket, the problem in the service ticket is a historical problem, and there is a solution to the problem; therefore, the ticket processing task includes at least one solution, and the solution is used to solve the problem that needs to be solved in the service ticket.

[0109] Specifically, a work order processing task is generated based on the service work order, including:

[0110] Step b01: extract the problems that need to be solved in the service ticket;

[0111] Step b02: recommend at least one solution from the solution library for solving the problem that needs to be solved in the service ticket.

[0112] In this embodiment, a solution that can solve the problem in the service ticket is found in the solution library, and then the solution is sent to the user terminal to prompt the user with a solution to the current problem. The user can refer to the solution to try to solve the current problem, which can improve the efficiency of solving the user's problem and avoid the user waiting for a long time for the processing personnel to solve the problem.

[0113] For step b02, specifically, at least one solution for solving the problem to be solved in the service ticket is recommended from the solution library, including:

[0114] Step b021: Segment the problems to be solved in the service ticket based on a preset retrieval model to obtain a number of words.

[0115] In this embodiment, the preset retrieval model can be a Boolean model, a vector space model, a BM25 model, a prime Bayesian classifier, and a deep learning-based retrieval model.

[0116] Step b022: Calculate the relevance of each word to each solution document in the solution library.

[0117] In this embodiment, the function of the relevance of each word to each solution document in the solution library is expressed as:

[0118] ;

[0119] In the formula, For the i Words With solution documentation dThe correlation of For the i Words In the program documentation d The frequency of occurrence in k 1 is the first adjustment factor, b is the second adjustment factor, For solution documentation d The ratio of the length to the average length.

[0120] Step b023: According to the relevance of each word to each solution document in the solution library, determine the text relevance score between the problem to be solved in the service ticket and each solution document.

[0121] In this embodiment, the function expression of the text relevance score between the problem to be solved in the service ticket and each solution document is:

[0122] ;

[0123] In the formula, S Score the textual relevance of the problem to be solved in the service ticket to each solution document, For the i Words With solution documentation d The correlation of To contain words The number of documents, N is the total number of solution documents, k 0 is the adjustment coefficient, n The total number of words.

[0124] Step b024: Filter out solution documents whose text relevance scores are greater than a preset score to obtain a candidate solution set, which includes a number of candidate solutions.

[0125] In this embodiment, based on the text relevance scores of the problems to be solved in the service ticket and each solution document, a certain number of solutions to be solved can be screened out. These solutions to be solved have a high degree of similarity to the solutions required to solve the problems in the user's service needs and can be used to solve the problems in the user's service needs.

[0126] Step b025: Calculate the similarity between each candidate solution and the problem to be solved in the service ticket to obtain a number of similarity values.

[0127] In this embodiment, the function for calculating the similarity between each candidate solution and the problem to be solved in the service ticket can be calculated using a cosine similarity function to obtain a number of similarity values.

[0128] Step b026: Sort a number of similarity values ​​from large to small to obtain a similarity sequence; select the candidate solutions corresponding to the first K similarity values ​​in the similarity sequence as solutions to the problem to be solved in the service ticket.

[0129] In this embodiment, after using the cosine similarity function to calculate the similarity value between each solution to be solved and the problem to be solved in the service ticket, the K solutions corresponding to the similarity values ​​with the highest similarity values ​​are selected as the solutions to solve the problem to be solved in the service ticket. In this embodiment, the solutions corresponding to the first 5 similarity values ​​can be taken; by calculating the similarity value between each solution to be solved and the problem to be solved in the service ticket, the solution screening range can be further narrowed, and the accuracy and effect of solution screening can be improved.

[0130] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0131] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. An automatic work order processing method based on a large language model, characterized in that: The method comprises: Analyze the service requests generated by users based on large language models to automatically create service tickets for the users; Determine the processing department to which the user's service ticket belongs, count the existing number of work orders of each processing personnel in the processing department, and assign the service ticket to the corresponding processing personnel in the processing department according to the existing number of work orders of each processing personnel; After the service work order is assigned, a work order processing task is generated based on the service work order, and during the execution of the work order processing task, an execution record of the work order processing task is recorded; Wherein, when the service work order is an existing work order, the work order processing task includes at least one solution, and the solution is used to solve the problem that needs to be solved in the service work order; Generate a work order processing task based on the service work order, including: Extract the issues that need to be resolved in the service ticket; Recommend at least one solution from the solution library to solve the problem in the service ticket; Recommend at least one solution from the solution library to solve the problem in the service ticket, including: Based on the preset retrieval model, the problem to be solved in the service ticket is segmented to obtain several words; Calculate the relevance of each word to each solution document in the solution library; According to the relevance of each word to each solution document in the solution library, determine the text relevance score between the problem to be solved in the service ticket and each solution document; Screening out solution documents whose text relevance scores are greater than a preset score to obtain a candidate solution set, wherein the candidate solution set includes a plurality of candidate solutions; Calculate the similarity between each candidate solution and the problem to be solved in the service ticket, and obtain several similarity values; Sort several similarity values ​​from large to small to obtain a similarity sequence; Take the candidate solutions corresponding to the first K similarity values ​​in the similarity sequence as the solutions to the problems that need to be solved in the service ticket; The function expression of the relevance of each word to each solution document in the solution library is: ; In the formula, For the i Words With solution documentation d The correlation of For the i Words In the program documentation d The frequency of occurrence in k 1 is the first adjustment factor, b is the second adjustment factor, For solution documentation d The ratio of the length to the average length.

2. The automatic work order processing method based on a large language model according to claim 1 is characterized in that: Analyze the service requests generated by users based on a large language model to automatically create service tickets for the users, including: Obtaining a user's service request, inputting the service request into a large language model, and extracting a summary of the service request; Determine whether there are keywords similar or identical to the summary of the service request in the pre-built work order library. If so, match the existing work order based on the keywords. If not, expand the summary based on the large language model to obtain the communication question and answer text. Based on the Q&A text, obtain the user's question intent; Based on the large language model, necessary information in the user's question intent is extracted, and a new work order is created according to the necessary information, and the newly created work order or the matched existing work order is used as the service work order.

3. The automatic work order processing method based on a large language model according to claim 1 is characterized in that: After creating the service ticket for the user, the method further includes: Get the historical work orders created by the user; Determine whether the newly created service ticket of the user is the same as the user's historical ticket; If so, delete the newly created service ticket and generate a prompt text to inform the user.

4. The automatic work order processing method based on a large language model according to claim 3 is characterized in that: Determine whether the newly created service ticket of the user is the same as the user's historical ticket, including: The historical work orders of the user are split into content to obtain a first character set, and the newly created service work orders of the user are split into content to obtain a second character set; Encode the first character set to obtain a first word vector, a first position vector, and a first embedding vector for each character in the first character set; superimpose the first word vectors, the first position vectors, and the first embedding vectors of all characters in the first character set and input them into a pre-trained text extraction model to obtain a first text vector; Encode the second character set to obtain a second word vector, a second position vector, and a second embedding vector for each character in the second character set; superimpose the second word vectors, the second position vectors, and the second embedding vectors of all characters in the second character set and input them into a pre-trained text extraction model to obtain a second text vector; Performing average pooling processing on the first text vector to obtain a processed first text vector; performing average pooling processing on the second text vector to obtain a processed second text vector; Calculating the similarity between the processed first text vector and the processed second text vector; Determine whether the similarity between the processed first text vector and the processed second text vector is greater than a preset value; If yes, the newly created service ticket is a non-duplicate ticket; if no, the newly created service ticket is a duplicate ticket.

5. The automatic work order processing method based on a large language model according to claim 4 is characterized in that: The calculation expression of the similarity between the processed first text vector and the processed second text vector is: ; In the formula, F is the first text vector after processing With the processed second text vector The similarity of is the cosine similarity function.

6. The automatic work order processing method based on a large language model according to claim 1 is characterized in that: The function expression of the text relevance score between the problem to be solved in the service ticket and each solution document is: ; In the formula, S Score the textual relevance of the problem to be solved in the service ticket to each solution document, For the i Words With solution documentation d The correlation of To contain words The number of documents, N is the total number of solution documents, k 0 is the adjustment coefficient, n The total number of words.

7. An automatic work order processing system based on a large language model, the system is used to implement the automatic work order processing method based on a large language model according to any one of claims 1 to 6, characterized in that: The system comprises: a user terminal, a server and a processing terminal, wherein the user terminal and the processing terminal are both communicatively connected to the server; The user terminal is used to input the user's service requirements and upload the user's service requirements to the server; The server comprises: The work order creation module is used to obtain the user's service request, analyze the user's service request based on a large language model, obtain the user's service demand, and determine whether the user's service demand is an existing demand. If so, match the existing work order; if not, generate a work order filling task to communicate with the user, collect the necessary information required for the work order, and create a new work order based on the necessary information; save the new work order or the existing work order as a service work order in the database, and mark the service work order as a pending assignment state; The work order allocation module is used to obtain the service work order in the pending allocation state, call the automatic allocation interface, determine the processing department to which the service work order belongs; count the work order quantity of each processing personnel in the processing department, and allocate the service work order in the pending allocation state to the processing personnel corresponding to the processing department according to the work order quantity of each processing personnel in the processing department, and mark the service work order as pending; The work order processing module is used to call the work order processing interface, generate work order processing tasks based on the service work order, and send the work order processing tasks to the processing terminal; The processing terminal is used to record the execution record of the work order processing task and upload the execution record to the server during the execution of the work order processing task. The server is also used to mark the service work order as processed after the work order processing task is completed; and synchronize the service work order in the processed state to the user terminal.

Citation Information

Cited By

  • Work order processing system and method

    CN119668566A