A method and system for identifying insurance scenario intentions based on a large model
By building a categorical intent recognition large model and adopting batch model services and parameter fine-tuning methods, the problems of insufficient semantic understanding capabilities of traditional deep learning models and slow inference speed of large language models are solved, efficient intent recognition is achieved, and hardware resource consumption and deployment costs are reduced.
Patent Information
- Application Number
- CN202411654275.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-04-01
AI Technical Summary
Traditional deep learning models lack semantic understanding capabilities in intention recognition, and the intent recognition method based on large language models is slow in reasoning and high computing resources, resulting in high hardware deployment and maintenance costs, which limits its widespread popularity in practical applications.
Build a categorical intent recognition big model, transform the output layer of the generative large language model, and optimize the model training and inference process by transforming the output layer of the generative large language model, and adopting batch model services and parameter fine-tuning methods.
While ensuring strong semantic understanding capabilities, it greatly improves the speed of model inference, reduces hardware resource consumption and deployment costs, and significantly improves customer service efficiency and user experience.
Smart Images
Figure CN119740582A_ABST
Abstract
Description
Background Art
[0002] Current intent recognition methods are mainly divided into two categories: one is the intent recognition method based on traditional deep learning models, including but not limited to Bert (Bidirectional Encoder Representations from Transformers), LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Networks), and the other is the intent recognition method based on large language models (LLM, Large Language Model). Traditional deep learning models have achieved certain achievements in natural language processing (NLP) tasks, but their performance is still insufficient when facing complex semantic understanding tasks. In contrast, in recent years, large language models such as GPT (Generative Pre-trained Transformer) have gradually attracted attention and research due to their superior text generation and understanding capabilities.
[0003] Although traditional deep learning models have been widely used in intent recognition, they are obviously insufficient in semantic understanding. Traditional models often show limitations when dealing with complex and diverse user intent, resulting in unsatisfactory accuracy and usage results.
[0004] At the same time, the intent recognition method based on large language models has a slow reasoning speed due to its huge number of parameters and complex computing requirements, and has problems such as high computing resource consumption and high hardware deployment costs, which seriously hinders its widespread popularization and promotion in practical applications. The specific manifestations include: first, the slow reasoning speed of large models is difficult to meet the real-time response requirements, which has a negative impact on user experience; second, the high demand for hardware resources during model training and reasoning leads to high hardware deployment and maintenance costs, further limiting its large-scale application. Summary of the invention
[0005] The present application provides a large-model-based insurance scenario intent recognition method and system, which is used to greatly improve the model reasoning speed, reduce hardware resource consumption and deployment costs while ensuring powerful semantic understanding capabilities, thereby significantly improving customer service efficiency and user experience.
[0006] In a first aspect, a method for identifying insurance scenario intention based on a large model is provided, comprising the following steps:
[0007] Build a large model for classification-based intent recognition;
[0008] The classification-based intention recognition large model is used for reasoning to complete insurance scenario intention recognition.
[0009] In the above technical solution, a large classification-based intent recognition model is constructed; the classification-based intent recognition large model is used for reasoning to complete insurance scenario intent recognition; while ensuring strong semantic understanding capabilities, the model reasoning speed is greatly improved, hardware resource consumption and deployment costs are reduced, thereby significantly improving customer service efficiency and user experience.
[0010] In a specific implementation scheme, the classification-based intent recognition large model adopts a batch model service.
[0011] In a specific implementation scheme, the steps of constructing a large classification intent recognition model specifically include:
[0012] By modifying the output layer of the generative large language model, a classification large language model is obtained;
[0013] The classification-based large language model is trained to obtain the classification-based intent recognition large model.
[0014] In a specific implementation scheme, the steps of obtaining a classification-based large language model by transforming the output layer of the generative large language model are as follows:
[0015] The classification large language model is obtained by transforming the output layer (hidden layer size, number of vocabularies) of the generative large language model into the output layer (hidden layer size, number of intents).
[0016] In a specific implementation scheme, the step of training the classification-based large language model to obtain the classification-based intent recognition large model specifically includes:
[0017] Optimize the intent recognition training set to obtain a clearly labeled intent recognition dataset;
[0018] The categorical large language model is trained using the clearly labeled intent recognition data set to obtain the categorical intent recognition large model.
[0019] In a specific implementation scheme, the intention recognition training set is optimized using a parameter fine-tuning method.
[0020] In a specific implementation scheme, the classification-based intent recognition large model completes insurance scenario intent recognition through one-time reasoning.
[0021] In the second aspect, a large model-based insurance scenario intention recognition system is provided, including:
[0022] Large model module, used to build a large classification intent recognition model;
[0023] The intent recognition module is used to use the classification-based intent recognition large model for reasoning and complete insurance scenario intent recognition.
[0024] In the above technical solution, a large classification-based intent recognition model is constructed; the classification-based intent recognition large model is used for reasoning to complete insurance scenario intent recognition; while ensuring strong semantic understanding capabilities, the model reasoning speed is greatly improved, hardware resource consumption and deployment costs are reduced, thereby significantly improving customer service efficiency and user experience.
[0025] In a specific implementation scheme, the classification-based intent recognition large model adopts a batch model service.
[0026] In a specific implementation scheme, the categorized intent recognition big model includes a categorized intent recognition big model for single inference intent. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 A flowchart of a method for identifying insurance scenario intentions based on a large model provided in an embodiment of the present application;
[0028] Figure 2 A structural block diagram of the insurance scenario intention recognition system based on a large model provided in an embodiment of the present application;
[0029] Figure 3 A schematic diagram of constructing a large model for classification-based intent recognition provided in an embodiment of the present application;
[0030] Figure 4 A reasoning diagram of a large model of classified intent recognition for a single inference intention provided in an embodiment of the present application. DETAILED DESCRIPTION
[0031] The present application is further described in detail below through the accompanying drawings and embodiments. Through these descriptions, the characteristics and advantages of the present application will become clearer and more specific.
[0032] The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments. Although various aspects of the embodiments are shown in the drawings, the drawings are not necessarily drawn to scale unless otherwise noted.
[0033] In addition, the technical features involved in different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.
[0034] To facilitate understanding of the insurance scenario intention recognition method and system based on a large model provided in the embodiment of the present application, its application scenario is first explained. The insurance scenario intention recognition method and system based on a large model provided in the embodiment of the present application are used to greatly improve the model reasoning speed while ensuring strong semantic understanding capabilities, reduce hardware resource consumption and deployment costs, thereby significantly improving customer service efficiency and user experience. Current intention recognition methods are mainly divided into two categories: one is an intention recognition method based on traditional deep learning models, including but not limited to Bert (Bidirectional Encoder Representations from Transformers), LSTM (Long Short-Term Memory) and CNN (Convolutional Neural Networks), and the other is an intention recognition method based on a large language model (LLM, Large Language Model). Traditional deep learning models have achieved certain achievements in natural language processing (NLP) tasks, but when faced with complex semantic understanding tasks, their performance is still insufficient. In contrast, in recent years, large language models such as GPT (Generative Pre-trained Transformer) have gradually attracted attention and research due to their superior text generation and understanding capabilities. Although traditional deep learning models have been widely used in intent recognition, they have obvious deficiencies in semantic understanding capabilities. Traditional models often show limitations when dealing with complex and diverse user intentions, resulting in unsatisfactory accuracy and use effects. At the same time, the intent recognition method based on a large language model has a slow reasoning speed due to its huge number of parameters and complex computing requirements, and there are problems such as large consumption of computing resources and high cost of hardware deployment, which seriously hinders its widespread popularization and promotion in practical applications. The specific manifestations include: first, the large model has a slow reasoning speed and is difficult to meet the real-time response requirements, which has a negative impact on user experience; second, the high demand for hardware resources during model training and reasoning leads to high hardware deployment and maintenance costs, further limiting its large-scale application. For this reason, the embodiment of the present application provides an insurance scenario intent recognition method and system based on a large model, so as to greatly improve the model reasoning speed, reduce hardware resource consumption and deployment costs while ensuring strong semantic understanding capabilities, thereby significantly improving customer service efficiency and user experience. The following is a detailed description of the embodiment in conjunction with specific drawings.
[0035] refer to Figures 1 to 4 , Figure 1 A flowchart of a method for identifying insurance scenario intentions based on a large model provided in an embodiment of the present application; Figure 2 A structural block diagram of the insurance scenario intention recognition system based on a large model provided in an embodiment of the present application; Figure 3 A schematic diagram of constructing a large model for classification-based intent recognition provided in an embodiment of the present application; Figure 4 A reasoning diagram of a large model of classified intent recognition for a single inference intention provided in an embodiment of the present application.
[0036] exist Figure 1 In the embodiment of the present application, a method for identifying insurance scenario intention based on a large model is provided, comprising the following steps:
[0037] Build a large model for classification-based intent recognition;
[0038] The classification-based intention recognition large model is used for reasoning to complete insurance scenario intention recognition.
[0039] In the above technical solution, a large classification-based intent recognition model is constructed; the classification-based intent recognition large model is used for reasoning to complete insurance scenario intent recognition; while ensuring strong semantic understanding capabilities, the model reasoning speed is greatly improved, hardware resource consumption and deployment costs are reduced, thereby significantly improving customer service efficiency and user experience.
[0040] In a specific implementation scheme, the classification-based intent recognition large model adopts a batch model service.
[0041] In a specific implementation scheme, the steps of constructing a large classification intent recognition model specifically include:
[0042] By modifying the output layer of the generative large language model, a classification large language model is obtained;
[0043] The classification-based large language model is trained to obtain the classification-based intent recognition large model.
[0044] In a specific implementation scheme, the steps of obtaining a classification-based large language model by transforming the output layer of the generative large language model are as follows:
[0045] The classification large language model is obtained by transforming the output layer (hidden layer size, number of vocabularies) of the generative large language model into the output layer (hidden layer size, number of intents).
[0046] In a specific implementation scheme, the step of training the classification-based large language model to obtain the classification-based intent recognition large model specifically includes:
[0047] Optimize the intent recognition training set to obtain a clearly labeled intent recognition dataset;
[0048] The categorical large language model is trained using the clearly labeled intent recognition data set to obtain the categorical intent recognition large model.
[0049] In a specific implementation scheme, the intention recognition training set is optimized using a parameter fine-tuning method.
[0050] In a specific implementation scheme, the classification-based intent recognition large model completes insurance scenario intent recognition through one-time reasoning.
[0051] Specifically, the insurance scenario intention recognition method based on the big model includes:
[0052] a) Structural transformation of large language models
[0053] First, the output layer (Encoder output) structure of the existing generative large language model (CausalModel) is modified. The output layer structure of the traditional generative large language model is usually (hidden layer size, number of vocabularies), where the number of vocabularies represents the number of target words that the model can generate. This application optimizes this output layer structure to (hidden layer size, number of intents) to form a classification large language model (Sequence Classification Model), such as Figure 3 As shown in the figure. Through this adjustment, the model output is transformed from generating a single word to generating a specific intent category, thus achieving the purpose of intent classification. This not only simplifies the output space of the model, but also ensures that the model can learn and recognize intent-level information more effectively.
[0054] b) Training of large models for classification-based intent recognition
[0055] During the model training phase, this application uses efficient parameter fine-tuning methods, such as LoRA (Low-Rank Adaptation), to further optimize the classified large language model on the intent recognition training set. LoRA is a parameter fine-tuning technology designed specifically for large-scale language models, which can significantly reduce the consumption of training resources while maintaining training efficiency. By training on a large number of clearly labeled intent recognition data sets, a large classified intent recognition model with high accuracy and strong transfer learning capabilities is obtained. This model can efficiently identify the intent of input data and has wide applicability in a variety of application scenarios.
[0056] c) Classification-based intent recognition large model reasoning
[0057] When traditional generative intent recognition models output specific intents (such as "claim consultation"), they usually need to perform multiple iterative reasoning to achieve the expected results, which increases the reasoning time and computing resource consumption. However, the classification intent recognition model proposed in this application can produce clear intent results through one-time reasoning, such as "claim consultation". Figure 4As shown in the figure. Since the output layer of the improved model is smaller, the required computational complexity is greatly reduced, which significantly improves the reasoning speed and efficiency. Especially in real-time application environments, this improvement can ensure faster response speed and lower computing resource utilization.
[0058] d) Model service transformation
[0059] In order to further improve the overall service response efficiency of the model in high-concurrency scenarios, this application has transformed the traditional single-processing model service and adopted a continuous batch processing method. New requests are placed in the queue asynchronously, and the model processes this batch of requests uniformly within a predetermined time interval. This batch processing method can not only effectively share the resource consumption of a single inference task, but also achieve more efficient concurrent processing capabilities while ensuring accuracy. This transformation is particularly suitable for practical application scenarios that need to process a large number of requests, such as intelligent customer service. Through the above-mentioned improvement measures, the processing efficiency and user experience of the entire chain can be significantly improved.
[0060] In the above technical solution, the generative intent recognition large model that reasons about intent word by word in a loop is transformed into a classification intent recognition large model that reasons about intent once, and the single processing model service is changed to a batch processing model service, which effectively improves the model response speed and the throughput of the model service.
[0061] exist Figure 2 In the embodiment of the present application, a large model-based insurance scenario intention recognition system is provided, including:
[0062] Large model module, used to build a large classification intent recognition model;
[0063] The intent recognition module is used to use the classification-based intent recognition large model for reasoning and complete insurance scenario intent recognition.
[0064] In the above technical solution, a large classification-based intent recognition model is constructed; the classification-based intent recognition large model is used for reasoning to complete insurance scenario intent recognition; while ensuring strong semantic understanding capabilities, the model reasoning speed is greatly improved, hardware resource consumption and deployment costs are reduced, thereby significantly improving customer service efficiency and user experience.
[0065] In a specific implementation scheme, the classification-based intent recognition large model adopts a batch model service.
[0066] In a specific implementation scheme, the categorized intent recognition big model includes a categorized intent recognition big model for single inference intent.
[0067] Those skilled in the art will appreciate that the present application may be implemented as a system, method or computer program product.
[0068] Therefore, the present disclosure may be specifically implemented in the following forms, namely: it may be completely hardware, it may be completely software (including firmware, resident software, microcode, etc.), or it may be a combination of hardware and software, generally referred to herein as a "circuit", "module" or "system". In addition, in some embodiments, the present application may also be implemented in the form of a computer program product in one or more computer-readable media, and the computer-readable medium may contain computer-readable program code.
[0069] Any combination of one or more computer-readable media can be used. Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or devices, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: electrical connections with one or more wires, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this document, computer-readable storage media can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.
[0070] Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art can change, modify, replace and modify the above embodiments within the scope of the present application. On this basis, a variety of replacements and improvements can be made to the present application, all of which fall within the scope of protection of the present application.
Claims
1. A method for identifying insurance scenario intentions based on a large model, characterized in that: The following steps are involved: Build a large model for classification-based intent recognition; The classification-based intention recognition large model is used for reasoning to complete insurance scenario intention recognition.
2. The insurance scenario intention recognition method based on a large model according to claim 1 is characterized in that: The classification-based intent recognition large model adopts batch processing model service.
3. The insurance scenario intention recognition method based on a large model according to claim 2 is characterized in that: The steps to build a large classification intent recognition model include: By modifying the output layer of the generative large language model, a classification large language model is obtained; The classification-based large language model is trained to obtain the classification-based intent recognition large model.
4. The insurance scenario intention recognition method based on a large model according to claim 3 is characterized in that: By transforming the output layer of the generative large language model, the steps to obtain the classification large language model are as follows: The classification large language model is obtained by transforming the output layer (hidden layer size, number of vocabularies) of the generative large language model into the output layer (hidden layer size, number of intents).
5. The insurance scenario intention recognition method based on a large model according to claim 4 is characterized in that: The step of training the classification-based large language model to obtain the classification-based intent recognition large model specifically includes: Optimize the intent recognition training set to obtain a clearly labeled intent recognition dataset; The categorical large language model is trained using the clearly labeled intent recognition data set to obtain the categorical intent recognition large model.
6. The insurance scenario intention recognition method based on a large model according to claim 5 is characterized in that: The intention recognition training set is optimized by using a parameter fine-tuning method.
7. The insurance scenario intention recognition method based on a large model according to claim 6 is characterized in that: The classification-based intent recognition large model completes insurance scenario intent recognition through one-time reasoning.
8. An insurance scenario intention recognition system based on a large model, characterized in that: include: Large model module, used to build a large classification intent recognition model; The intent recognition module is used to use the classification-based intent recognition large model for reasoning and complete insurance scenario intent recognition.
9. The insurance scenario intention recognition system based on a large model according to claim 8 is characterized in that: The classification-based intent recognition large model adopts batch processing model service.
10. The insurance scenario intention recognition system based on a large model according to claim 9 is characterized in that: The classification-based intent recognition big model includes a classification-based intent recognition big model for single-time inference intent.