Method and system for vehicle model recommendation based on large model, and medium
Through the large-model model recommendation method, the model search engine of the bert and sbert model combined with the Neo4j database has been solved, and the problem of inefficient recommendation of traditional models has been achieved, personalized and accurate model recommendations have been achieved, and user experience and sales conversion rates have been improved. .
Patent Information
- Application Number
- CN202411378123.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2025-01-07
AI Technical Summary
Traditional car model recommendation methods are inefficient and difficult to meet consumers' personalized needs. The existing recommendation algorithm cannot fully cover the complex factors of car purchase decisions, resulting in limited recommendation results.
Using a large model-based method to conduct in-depth analysis of user demand text, combined with professional knowledge in the automotive field, feature extraction and matching through bert-base-chinese pre-trained model and sbert semantic search model, and using the model search engine of Neo4j graph database Perform personalized recommendations and continuously optimize through conversational recommendation interfaces and microservice interfaces.
It achieves more efficient and accurate model recommendations, improves user experience and sales conversion rates, reduces repeated order submissions and manual review waste, and quickly meets customer needs.
Smart Images
Figure CN119271891A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of artificial intelligence, and in particular, to a method, system and medium for recommending vehicle models based on a large model. Background Art
[0002] With the development of Internet technology and artificial intelligence, the demand for personalized services is growing, especially in the automobile sales industry, where consumers have a strong demand to quickly and accurately find models that meet their needs. The traditional model recommendation method mainly relies on the experience of sales staff or simple questionnaires, which is often inefficient and difficult to meet the personalized needs of consumers.
[0003] In recent years, the advancement of natural language processing technology, especially the emergence of large pre-trained models, has provided new possibilities for solving the above problems. Large pre-trained models such as BERT and GPT have powerful language understanding and generation capabilities through large-scale corpus training. These models can extract deep semantic information from text, including emotions, intentions, and needs, thereby achieving more accurate personalized recommendations.
[0004] However, there are still challenges in applying large models to car model recommendations, such as how to effectively transform user needs into structured data, how to match car model features with user preferences, and how to ensure the real-time and accuracy of recommendation results. In addition, considering the complexity of car purchase decisions, including multiple dimensions such as price, performance, safety, and brand preference, existing recommendation algorithms may not be able to fully cover all important factors, resulting in limited recommendation results.
[0005] Therefore, an innovative method is needed to use large models to conduct in-depth analysis of user demand texts and combine them with expertise in the automotive field to provide users with car model recommendations in a more efficient and accurate manner, thereby improving user experience and sales conversion rates. Summary of the invention
[0006] In view of the defects in the prior art, the purpose of the present invention is to provide a method, system and medium for vehicle model recommendation based on a big model, which uses the big model to conduct in-depth analysis of user demand text, and combines professional knowledge in the automotive field to provide users with vehicle model recommendations in a more efficient and accurate manner, thereby improving user experience and sales conversion rate.
[0007] In order to achieve the above technical effects, the present invention adopts the following technical solutions:
[0008] According to a first aspect of the present invention, a method for recommending a vehicle model based on a large model is provided, which specifically comprises the following steps:
[0009] Step S1. First, obtain user demand text data, including historical orders and data generated using a large model, and use the data to build a detailed database containing vehicle model characteristics;
[0010] Step S2. Large model training:
[0011] Feature extraction model selection: Use the bert-base-chinese pre-trained model and combine the Multi-Head mechanism to process token pairs and generate a score matrix;
[0012] Select the semantic search model for clustering and semantic search;
[0013] Fine-tune the model: Fine-tune the feature extraction model and semantic search model based on the annotated data to improve their understanding of the context related to the vehicle model;
[0014] Step S3. After the user inputs the requirement text, the fine-tuned feature extraction model is used to extract key information, and the semantic search model is used to search in the standard feature vector library;
[0015] Step S4. Model matching algorithm: Combine the standard features defined by the user with the features recommended by AI and adjust the weight of each feature according to the importance of the demand, and use the model search engine to make personalized model recommendations;
[0016] Step S5: By providing a conversational recommendation interface and microservice interface, users are allowed to provide feedback on the recommendation results, and the recommendation algorithm is continuously optimized;
[0017] Step S6: Regularly update the model with new data and feedback, monitor system performance and user satisfaction, and adjust the strategy in time to optimize the recommendation effect.
[0018] Preferably, step S1 also includes a data cleaning and labeling step, cleaning the text data and labeling key information related to the vehicle model and features.
[0019] Preferably, the vehicle type search engine in step S4 is a vehicle type search engine developed in Java based on the Neo4j graph database.
[0020] Preferably, selecting a semantic search model for clustering and semantic search in step S2 specifically includes selecting sbert's paraphrase-mult ilingual-MiniLM-L12-v2 model, which is suitable for clustering and semantic search and mapped to a 384-dimensional dense vector space.
[0021] According to a second aspect of the present invention, a system for recommending vehicle types based on a large model is provided, for implementing the above-mentioned method for recommending vehicle types based on a large model.
[0022] Preferably, the following functions are also included:
[0023] (1) Provide a conversational interactive assistant to facilitate rapid effect verification after model training;
[0024] (2) The initial training data can be imported offline, and the data version can be controlled for model training;
[0025] (3) Access AI model matching results and make annotation corrections based on them; the corrected data will be automatically added to the training data set and support model training;
[0026] (4) The matching score rate of each vehicle model feature decomposed by the model can be checked, and low matching rate features can be monitored and trained in a timely manner.
[0027] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement the above-mentioned method for recommending vehicle models based on a large model.
[0028] Compared with the prior art, the present invention has the following beneficial effects:
[0029] The method for recommending vehicle models based on a big model provided by the present invention can utilize the big model to conduct in-depth analysis of user demand texts, and combined with the professional knowledge of automobile product configuration within the factory, realize automatic recommendation of vehicle models / orders when special modified vehicles are submitted, thereby quickly meeting customer needs while reducing repeated order submissions, avoiding repeated review labor waste and promoting the concentration of sales models. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:
[0031] Figure 1 This is a flowchart of the steps of the method for recommending vehicle models based on a large model described in the first embodiment. DETAILED DESCRIPTION
[0032] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.
[0033] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.
[0034] First Embodiment
[0035] like Figure 1 As shown, this embodiment provides a method for recommending vehicle models based on a large model, which uses the large model to conduct in-depth analysis of user demand texts and combines the professional knowledge of automobile product configuration in the factory to achieve automatic recommendation of vehicle models / orders when special modified vehicles are submitted. Specifically, the following steps are included:
[0036] Step S1. Data collection and preprocessing
[0037] S1.1 Text data collection: Collect a large amount of user demand text data, such as user descriptions in historical orders, and also combine other large models such as GPT to assist in generating training data.
[0038] S1.2 Vehicle type database: Establish a detailed vehicle type database, including information such as the characteristic values of each vehicle type.
[0039] S1.3 Data cleaning and annotation: Clean the text data to remove irrelevant information, and annotate the text to mark the key information related to the vehicle model and vehicle model characteristics.
[0040] Step S2. Large model training
[0041] S2.1 Feature extraction model selection: The bert-base-chinese pre-trained model is used as the base model for feature extraction. "bert-base-chinese" is a version of the BERT model specifically designed for processing Chinese text. BERT (Bidirectional Encoder Representations from Transformers) is a transformer-based pre-trained model developed by Google that has achieved excellent performance in natural language processing (NLP) tasks.
[0042] In the lower network, a Multi-Head mechanism based on token pairs is used. Specifically, Multi-Head uses the first token and the last token of the fragment to represent a span (hence it is called token pairs). On this basis, the interaction layer and the classification layer are added to finally generate a Multi-Head matrix. The numerical expression is: f(h_i,h_j,t)=score_{ij,t}, where h represents the output of BERT, and score_{ij,t} represents the score of the text fragment with index_{start}=i and index_{end}=j on the tth entity class. The Multi-Head mechanism can naturally solve the problems of inter-group nesting and intra-group nesting.
[0043] S2.2 Semantic search model selection: Considering the accuracy and efficiency of reasoning, the paraphrase-multilingual-MiniLM-L12-v2 model of SBERT is selected. This model has 117 million parameters and can map sentences and paragraphs to a 384-dimensional dense vector space. This model is suitable for tasks such as clustering and semantic search.
[0044] S2.3 Fine-tuning the model: After collecting the labeled data, fine-tune the two pre-trained models mentioned above. Fine-tuning enables the model to better understand the specific vocabulary and context related to the vehicle model, improving the relevance and accuracy of the task.
[0045] Step S3. Feature extraction and demand understanding: When the user enters the demand text, the fine-tuned feature extraction model is used to extract key information, such as vehicle model features, etc. Then, the semantic search model is used to perform semantic search in the pre-standard feature vector library based on the extracted feature information to obtain the standard product definition features.
[0046] Step S4. Model matching algorithm: Use a model search engine developed in Java based on the Neo4j graph database. Users define standard features, relationships between models, and other relevant model information in advance. The search engine combines the features recommended by the AI model to make personalized model recommendations. The weight of each feature is adjusted according to the importance of the demand to make the recommendation more accurate.
[0047] Step S5. User interface and feedback mechanism: recommend car models through conversations with users and provide various microservice interfaces for external calls. At the same time, users are allowed to give feedback on the recommendation results, continuously optimize the recommendation algorithm, and improve the user experience.
[0048] Step S6. Continuous optimization and updating: Regularly update the model with new data and feedback to adapt to market changes and the evolution of user preferences. At the same time, monitor the performance of the system and user satisfaction, and adjust the strategy in time to optimize the recommendation effect.
[0049] When a customer has special modification needs, he or she will usually find a car model that is closest to his or her requirements as the base car, and add a paragraph of text to describe the modification requirements he or she needs, and then generate a C order and submit it to the factory.
[0050] When the factory receives a C order, it needs to pass product review, price review, and delivery period review before it can support formal order acceptance. The whole process takes about 3-7 days. When a customer submits a C order, after selecting a reference model and filling in the modification requirements, the "AI check similar model orders" function can be performed through the method for recommending models based on a large model provided in this embodiment. After the AI is called, it will return the model and historical reviewed orders that are most similar to the customer's current needs. If the customer finds a favorite model in the recommended models / orders, he can click it directly and place an order. The method for recommending models based on a large model provided in this embodiment can quickly meet customer needs while reducing duplicate order submissions, avoiding manual waste from repeated reviews, and promoting the concentration of sales models.
[0051] Second Embodiment
[0052] This embodiment provides a system for recommending vehicle models based on a large model, which is used to implement the method for recommending vehicle models based on a large model described in the first embodiment. The system includes an ISVO management platform independently built based on an open source environment, which integrates the feedback data of the front-end business system for monitoring-labeling-training-verification, forms a management closed loop, and optimizes the operation and management process of the enterprise. In addition to being used to implement the method for recommending vehicle models based on a large model described in the first embodiment, the system also includes the following main functions:
[0053] 1. Provide a conversational interactive assistant to facilitate rapid effect verification after model training.
[0054] 2. The initial training data can be imported offline, and the data version can be controlled for model training.
[0055] 3. Each order will store and retrieve AI model matching results, and annotation corrections can be made based on this. The corrected data will be automatically added to the training data set to support model training.
[0056] 4. The matching score rate of each vehicle model decomposed by the model can be checked, and the low matching rate features can be monitored and trained in time.
[0057] Embodiment 3
[0058] This embodiment provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement the method for recommending vehicle models based on a large model described in Embodiment 1.
[0059] The above describes the specific embodiments of the present invention. Based on the above description, relevant personnel can make various changes and modifications without departing from the technical concept of the present invention.
Claims
1. A method for recommending vehicle models based on a large model, characterized in that: The specific steps include: Step S1. First, obtain user demand text data, including historical orders and data generated using a large model, and use the data to build a detailed database containing vehicle model characteristics; Step S2. Large model training: Feature extraction model selection: Use the bert-base-chinese pre-trained model and combine the Multi-Head mechanism to process token pairs and generate a score matrix; Select the semantic search model for clustering and semantic search; Fine-tune the model: Fine-tune the feature extraction model and semantic search model based on the annotated data to improve their understanding of the context related to the vehicle model; Step S3. After the user inputs the requirement text, the fine-tuned feature extraction model is used to extract key information, and the semantic search model is used to search in the standard feature vector library; Step S4. Model matching algorithm: Combine the standard features defined by the user with the features recommended by AI and adjust the weight of each feature according to the importance of the demand, and use the model search engine to make personalized model recommendations; Step S5: By providing a conversational recommendation interface and microservice interface, users are allowed to provide feedback on the recommendation results, and the recommendation algorithm is continuously optimized; Step S6: Regularly update the model with new data and feedback, monitor system performance and user satisfaction, and adjust the strategy in time to optimize the recommendation effect.
2. The method for vehicle model recommendation based on a large model according to claim 1, characterized in that: Step S1 also includes data cleaning and labeling steps, cleaning text data and labeling key information related to vehicle models and features.
3. The method for vehicle model recommendation based on a large model according to claim 1, characterized in that: The vehicle type search engine in step S4 is a vehicle type search engine developed in Java based on the Neo4j graph database.
4. The method for vehicle model recommendation based on a large model according to claim 1, characterized in that: Selecting a semantic search model for clustering and semantic search in step S2 specifically includes selecting sbert's paraphrase-multilingual-MiniLM-L12-v2 model, which is suitable for clustering and semantic search and mapped to a 384-dimensional dense vector space.
5. A system for recommending vehicle models based on a large model, characterized in that: Used to implement the method for recommending vehicle models based on a large model as described in any one of claims 1-3.
6. The system for recommending vehicle models based on a large model according to claim 5, characterized in that: Also includes the following features: (1) Provide a conversational interactive assistant to facilitate rapid effect verification after model training; (2) The initial training data can be imported offline, and the data version can be controlled for model training; (3) Access AI model matching results and make annotation corrections based on them; the corrected data will be automatically added to the training data set and support model training; (4) The matching score rate of each vehicle model feature decomposed by the model can be checked, and low matching rate features can be monitored and trained in a timely manner.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method for recommending vehicle models based on a large model as described in any one of claims 1 to 4.