Personalized product customization service method based on supply and demand fine-grained accurate cognition

Abstract

The invention discloses a personalized product customization service method based on supply and demand fine-grained accurate cognition, and the method combines CoT, KG and LLMs, can judge whether the daily dialogue of a user is related to air conditioner product selection and matching or not according to the daily dialogue of the user, and carries out the output through the universal LLMs if the daily dialogue is not related to the air conditioner product selection and matching; if the LLMs are related to air conditioner product matching, related air conditioner matching requirements are extracted through related intention understanding and extraction capacity of the LLMs, and factors such as colors, manufacturers, models and prices are gradually limited in a multi-round question and answer mode; and finally, an air conditioner product selecting and matching scheme matched with user requirements is given based on related limiting conditions given by the user. In addition, the highly personalized demand of the user is considered, and a part replacement module is added, so that the personalized customization demand of the user is further improved. According to the method, fine-grained accurate cognition is gradually performed on the user demands by utilizing a multi-round question and answer mechanism aiming at the user demands, and personalized product customization service is realized.

Description

Technical Field

[0001] The present invention belongs to the technical fields of computer services and industrial Internet, and relates to a personalized product customization service method, specifically to a personalized product customization service method based on fine-grained and accurate understanding of supply and demand. Background Art

[0002] With the rise and rapid development of information technology, the personalized customization production mode that fully caters to customer needs has gradually replaced the traditional manual production and large-scale mass production manufacturing modes. Mass customization is a production method that integrates enterprises, customers, suppliers, employees, etc., and uses the overall optimization concept to make full use of various existing resources of the enterprise. With the support of standard technology, modern design methods, information technology and advanced manufacturing technology, according to the personalized needs of customers, customized products and services are provided, which can significantly enhance the core competitiveness of enterprises. Currently, in the field of product selection and matching, existing research tends to describe the relevant process optimization in product selection and matching, and there is less research on user needs and service recommendations. Summary of the Invention

[0003] In order to solve the above problems existing in the prior art, the present invention provides a personalized product customization service method based on fine-grained and accurate understanding of supply and demand for the field of personalized selection and matching of air-conditioning products. This method aims at user needs, uses a multi-round question-and-answer mechanism to gradually and accurately understand user needs at a fine-grained level, and finally realizes personalized product customization services.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A personalized product customization service method based on fine-grained and accurate understanding of supply and demand includes the following steps:

[0006] Step S1: Based on the research and analysis of mainstream e-commerce platforms and the air-conditioning product market, integrate the technical parameters, functional characteristics and core component information of existing air-conditioning products, and construct an air-conditioning knowledge graph for product selection and matching;

[0007] Step S2: Based on the online review data of users for mainstream air-conditioning products, the demand feedback data from enterprise market research, and the analysis of industry white papers, construct a multi-dimensional user demand knowledge graph;

[0008] Step S3: Associate the air-conditioning knowledge graph constructed in step S1 with the user demand knowledge graph constructed in step S2 through the has_charactor relationship to form a fusion knowledge graph from user needs - related products and related components;

[0009] Step S4: According to the user requirement knowledge graph constructed in Step S2, select LLMs and combine relevant prompt words to generate an instruction set of user requirement points for the corresponding product selection and customization plan and perform LoRA fine-tuning work;

[0010] Step S5: Define multi-dimensional user requirement points through the analysis of user requirements, and use LLMs and relevant prompt words to judge the relevance of product personalization customization and identify product dimensions;

[0011] Step S6: Utilize the relevant guidelines of the CoT technology to gradually improve relevant personalized product customization tasks and finally generate a product-level assembly plan and a component-level replacement plan;

[0012] Step S7: If the user is not satisfied with the preliminary selection result, replace relevant components.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] 1. The present invention can construct knowledge graphs at the product level and component level of relevant air conditioners, decompose air conditioner products and enable component assembly. In the generation of the knowledge graph, a user requirement knowledge graph is constructed based on the multi-view idea.

[0015] 2. The present invention can realize the generation of a customized assembly plan for air conditioners, support extracting corresponding user requirement points by using LLMs based on the user's requirement description, matching them with feature points, and then realizing the generation of a customized assembly plan for air conditioners via the knowledge graph.

[0016] 3. The present invention can realize the generation of information in general daily conversations, and achieve high-standard innovation in personalized product customization services based on fine-grained and accurate cognition of supply and demand without affecting the communication of general large language models.

[0017] 4. The present invention analyzes the multi-dimensional characteristics of relevant products, and uses the CoT technology to realize the penetration and information confirmation for specific dimensions, ensuring the flexibility of the overall invention process. Description of the Drawings

[0018] Figure 1 It is the overall framework diagram of the personalized product customization service method based on fine-grained and accurate cognition of supply and demand.

[0019] Figure 2 It is the business process of the personalized product customization service method based on fine-grained and accurate cognition of supply and demand.

[0020] Figure 3 It is the implementation form of the product selection knowledge graph.

[0021] Figure 4 It is the implementation form of the user requirement knowledge graph.

[0022] Figure 5 For the implementation form of the user demand - air conditioner product fusion knowledge graph.

[0023] Figure 6 It is the business process idea for constructing relevant instruction sets based on user needs. Specific implementation manners

[0024] The technical solutions of the present invention will be further described below in conjunction with the accompanying drawings, but are not limited thereto. Any modification or equivalent replacement of the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention shall be covered by the protection scope of the present invention.

[0025] The present invention provides a personalized product customization service method based on fine-grained accurate cognition of supply and demand. The method combines the Chain of Thought (CoT), Knowledge Graph (KG), and Large Language Models (LLMs) technologies. It can determine whether the daily conversation of the user is related to the selection and matching of air conditioner products. If not, it uses the general LLMs for output; if it is related to the selection and matching of air conditioner products, it extracts relevant air conditioner selection and matching requirements through the relevant intention understanding and extraction capabilities of the LLMs, and gradually restricts factors such as color, manufacturer, model, and price through multiple rounds of questions and answers. Finally, based on the relevant restrictive conditions given by the user, an air conditioner product selection and matching plan that matches the user's needs is given. In addition, considering the highly personalized needs of users, a parts replacement module is added to further improve the personalized customization needs of users. After giving a preliminary recommendation plan according to the product-level needs of the user, if the user is not satisfied with the recommended product parts plan, parts with the same capabilities can be recommended. After the user actively selects parts, a product selection and customization plan that meets the user's more highly personalized needs can be recommended. As Figure 1 and Figure 2 shown, it specifically includes the following steps:

[0026] Step S1: Based on the research and analysis of mainstream e-commerce platforms and the air conditioner product market, integrate the technical parameters, functional characteristics, and core component information of existing air conditioner products, and construct an air conditioner knowledge graph for product selection and matching.

[0027] This step is mainly to obtain the data information of air conditioner products and components, and use the relevant information to construct an air conditioner product selection and matching knowledge graph. The specific steps are as follows:

[0028] Step S11: Obtain the data information of the air conditioner product and its components. The data information includes information such as the manufacturer, model, Chinese name, type, price, related assembled components of the product, and the characteristics of the product and its components.

[0029] Step S12: Use the relevant information to construct a product selection knowledge graph for the air conditioner product. Take the air conditioner system concept as the top-level node, and associate the specific information of the relevant air conditioner products and components in the form of triples. The information in the product selection knowledge graph includes the color, price, etc. of the product. The specific implementation form of the knowledge graph is as Figure 3 shown.

[0030] Step S2: Based on the online review data of mainstream air conditioner products by users, the demand feedback data from enterprise market research, and the analysis of industry white papers, construct a multi-dimensional user demand knowledge graph.

[0031] This step mainly performs fine-grained extraction on the review data of air conditioner products commonly seen in the market by users on the official website, gives a part of the demand nodes preferred by users, and then gives a part of the demand nodes preferred by users according to the relevant user demand data. The specific steps are as follows:

[0032] Step S21: Based on three data sources (user reviews on air conditioner brand official websites, enterprise market research data, industry white papers), adopt a fine-grained knowledge extraction method to systematically mine the core demand characteristics of Internet users for air conditioner products. The construction method of its knowledge graph is as Figure 4 shown.

[0033] Step S22: Combine the demand characteristics obtained in Step S21 and express them as entities in the knowledge graph to form the final required user demand knowledge graph. The user demand knowledge graph includes the following personalized demand characteristics: cooling and heating, air purification, stable operation, noise level, air supply level, intelligent temperature control, air quality, energy-saving performance, environmental protection certification, appearance style, size and installation, price, cost performance, children, teenagers, the elderly, allergic people, asthmatic people, chronic patients, long-term home office, long-term home study, frequent outdoor sports, hot and humid areas, cold areas, small spaces, large spaces, electrical safety, and its own protection function.

[0034] Step S3: Associate the air conditioner knowledge graph constructed in Step S1 with the user demand knowledge graph constructed in Step S2 through the has_charactor relationship to form a fusion knowledge graph from user needs - related products and related components.

[0035] This step mainly focuses on steps S1 and S2. In the product selection knowledge graph constructed in step S1, corresponding feature points exist for relevant products and components; in the user demand knowledge graph constructed in step S2, relevant descriptions also exist for relevant user demands. A user's demand may correspond to the features of multiple components, and the feature of one component usually corresponds to the characteristics of multiple demands. There is a "many-to-many" relationship between them. The present invention considers relevant factors in fusing the product selection knowledge graph and the demand graph, and uses the has_charactor relationship in the knowledge graph to fuse the relationships of the two view graphs, realizing a fusion knowledge graph from user demands - relevant products and relevant components. The specific implementation form of the knowledge graph is as Figure 5 shown.

[0036] Step S4: According to the user demand knowledge graph constructed in step S2, select LLMs and combine relevant prompt words to generate an instruction set of user demand points for the corresponding product selection customization plan and perform LoRA fine-tuning work.

[0037] This step mainly aims at step S2. Based on the relevant user demand knowledge graph constructed in step S2, LLMs are combined with relevant prompt words for instruction set generation. After constructing the relevant instruction set, use the LoRA fine-tuning method to fine-tune the relevant LLMs to form domain-specific LLMs that can finely identify users' daily functional requirements. The specific steps are as follows:

[0038] Step S41: Based on the user demand knowledge graph constructed in step S2, use LLMs and combine relevant prompt words to generate an instruction set.

[0039] In terms of prompt word design, the format of the instruction set is restricted in aspects including but not limited to daily colloquialism, processing fields, etc., such as Figure 6 shown. The relevant prompt word design steps are as follows: Use the API interface of ChatGPT3.5, combine the personalized demand features obtained in step S22, and design prompt words. For example, "You are an expert in instruction set processing. Your task is to ask questions about the selection of industrial product requirements according to different users in different scenarios. The selection of industrial products focuses on purchasing relevant products, and the form of the questions should also be close to daily life. Please construct relevant question-and-answer pairs with the demand feature of 'often doing outdoor sports'." Finally, use the personalized demand features obtained in step S22, the text understanding ability of the large language model, and the relevant prompt words designed by the present invention to obtain relevant question-and-answer pairs for fine-tuning the large language model. In the instruction set generation task, the present invention hopes that the input of the model is question-oriented, the output corresponds to the relevant demands of users, and the types of generated instruction sets cover all users' demands as much as possible.

[0040] Step S42: Due to the complexity of relevant requirements and scenarios in the product selection and matching field, and the fact that high-quality recommendations of LLMs often rely on the scale and quality of training data, the size of the instruction set is effectively expanded through operations such as synonym replacement, random insertion, random swapping, and random deletion.

[0041] Step S43: According to the instruction set constructed in Steps S41 to S42, use the LoRA fine-tuning method to fine-tune the relevant LLMs to form domain-specific LLMs that can finely identify users' daily functional requirements.

[0042] Step S5: Define multi-dimensional user requirement points through the analysis of user requirements, and use LLMs and relevant prompt words to perform product personalization customization relevance judgment and product dimension recognition.

[0043] This step mainly aims at Steps S2, S3, and S4. Through in-depth analysis of user comments and industry white papers, a multi-dimensional evaluation is carried out based on a fine-grained and accurate understanding of supply and demand in the air-conditioning field. Using the high-precision text understanding ability of LLMs and relevant dimension restrictions, a dimension recognition prompt word engineering is designed to perform product personalization customization relevance judgment and product dimension recognition. The specific steps are as follows:

[0044] Step S51: Through in-depth analysis of user comments and industry white papers in Step S2, conduct multi-dimensional considerations based on a fine-grained and accurate understanding of supply and demand in the air-conditioning field, including manufacturers, prices, colors, etc.

[0045] Step S52: On the basis of Step S51, utilize the high-precision text understanding ability of LLMs and combine the relevant dimension restrictions in Step S51 to design dimension recognition prompt words. In the selection of dimensions, the present invention obtains several dimensions that users are most concerned about in industrial product customization through research and relevant industry white papers, including daily requirements, prices, manufacturers, and used specifications. On this basis, use the large language model and relevant dimension recognition prompt words to judge user input. Use relevant agents to judge whether the user's relevant input is related to product personalization customization. If it is relevant, judge which dimension of the product the user input has a strong correlation with and extract the corresponding information; if it is not relevant, output the corresponding result through the general model.

[0046] Step S6: Use the relevant guidelines of the CoT technology to gradually improve the relevant personalized product customization tasks and finally generate a product-level assembly plan and a component-level replacement plan.

[0047] This step mainly aims at step S5. Based on the judgment of product personalization customization relevance and product dimension recognition in step S5, a random angle is selected to divide the product dimensions. In addition, the present invention also designs relevant CoT technology, enabling the present invention to guide users to gradually complete the information confirmation of all dimension points of the product, avoiding the situation where users input too much content but cannot complete the corresponding work. After multiple rounds of question-and-answer input and answers, the present invention finally outputs a product-level assembly plan and a component-level replacement plan related thereto based on multi-dimensional information confirmation. The specific steps are as follows:

[0048] Step S61: Based on the judgment of product personalization customization relevance and product dimension recognition in step S5, a random angle is selected to divide the product dimensions;

[0049] Step S62: Design CoT technology. Through the relevant elements that have not been perfected in the dimension list and the large language model, guide users to gradually fill in the dimensions for the blank elements in the dimension list, and gradually complete the information confirmation of all dimension points of the product;

[0050] Step S63: After multiple rounds of question-and-answer input and answers, finally output a product-level assembly plan and a component-level replacement plan related thereto based on multi-dimensional information confirmation.

[0051] Step S7: Users also have highly personalized requirements for the selection and matching of products. If users are not satisfied with the preliminary selection and matching results, they can replace relevant components.

[0052] This step mainly aims at step S6. Based on the product-level assembly plan and component-level replacement plan given in step S6, if users are not very satisfied with the current product-level assembly plan, the present invention can replace the product components through natural language description using the component-level replacement plan to achieve a higher degree of personalized customization.

[0053] The above steps S1 to S7 implement a personalized product customization service method based on fine-grained accurate cognition of supply and demand. Since the process of air conditioner product selection and customization is based on component-based customization rules, information resources of each component are required. For step S1, the present invention designs diversified information resources including an air conditioner product library, an air conditioner product feature library, an air conditioner compressor model library, an air conditioner motor model library, an air conditioner condenser model library, an air conditioner four-way valve model library, an air conditioner evaporator model library, etc.; for step S2, the present invention designs information resources such as a user demand library.

[0054] In addition, to facilitate knowledge reasoning for air-conditioning products oriented to personalized demand analysis, the design has the following requirements for the data acquisition in steps S1, S2, and S7: air-conditioning product library, user demand library, design management, supplier management, product design resource library, customer management, and statistical analysis, where:

[0055] Air-conditioning product library: All product-level air-conditioning information can be viewed, including color, price, type, etc. The component information of the air-conditioning, as well as the manufacturers and prices of each component, can also be viewed. Administrators can replace components of the air-conditioning products in the product library.

[0056] User demand library: It includes the relevant demands of most users on the market for air-conditioning products, such as personalized demands like rapid cooling and heating, suitable for the elderly, suitable for teenagers, etc.

[0057] Design management: It includes information updates at the product level, component level, and features of the air-conditioning to achieve real-time interactive design. The models selected for each module of the air-conditioning are chosen by the customer, and the design of each component of the air-conditioning involves human participation, which largely ensures the effectiveness of the air-conditioning assembly plan.

[0058] Supplier management: According to the supplier information of the components, corresponding management and regulation are carried out on whether the supplier can provide the corresponding component products.

[0059] Product design resource library: It includes air-conditioning product library, air-conditioning product feature library, air-conditioning compressor model library, air-conditioning motor model library, air-conditioning condenser model library, air-conditioning four-way valve model library, air-conditioning evaporator model library, etc.

[0060] Customer management: It includes operations such as adding, deleting, modifying, and querying customers. The basic information of customers includes name, account, country of origin, avatar, contact person, shipping address, etc. Through customers, the design and knowledge reasoning process of the air-conditioning can be bound to the customers.

[0061] Statistical analysis: It includes statistical analysis of air-conditioning products and component data, statistical analysis of design data, and statistical analysis of enterprise data. The statistical analysis of air-conditioning products and component data conducts a comprehensive analysis of air-conditioning products, including quantity analysis at the product level of air-conditioning, data analysis at the component level, and selection analysis at the feature level, etc.; the statistical analysis of design data includes design number analysis of customer series, data analysis of customer rankings by design number, etc.; the statistical analysis of enterprise data includes rankings of design quantities of each institution series, rankings of customers of each institution, and rankings of air-conditioning selection situations of each institution.

Claims

1. A personalized product customization service method based on fine-grained and accurate understanding of supply and demand, characterized in that The method comprises the following steps: Step S1: Based on the research and analysis of mainstream e-commerce platforms and air-conditioning product markets, the technical parameters, functional characteristics and core component information of existing air-conditioning products are integrated to construct an air-conditioning knowledge graph for product selection; Step S2: construct a multi-dimensional user demand knowledge graph based on online review data of users on mainstream air-conditioning products, demand feedback data from enterprise market research, and industry white paper analysis; Step S3: Associating the air-conditioning knowledge graph constructed in step S1 with the user demand knowledge graph constructed in step S2 through the has_character relationship to form a fusion knowledge graph from user demand to related products and related parts; Step S4: Based on the user demand knowledge graph constructed in step S2, select LLMs and related prompt words to generate a user demand point instruction set for the corresponding product selection and customization solution and perform LoRA fine-tuning; Step S5: define multi-dimensional user demand points by analyzing user needs, and use LLMs and related prompt words to determine the relevance of product customization and identify product dimensions; Step S6: Using the relevant guidance of CoT technology, the relevant personalized product customization tasks are gradually improved and finally a product-level assembly plan and a component-level replacement plan are generated; Step S7: If the user is not satisfied with the preliminary matching result, relevant parts are replaced.

2. The personalized product customization service method based on fine-grained accurate understanding of supply and demand according to claim 1, characterized in that The specific steps of step S1 are as follows: Step S11, acquiring data information of products and parts of air conditioner products; Step S12: Use relevant information to construct a product selection knowledge graph for air-conditioning products: take the air-conditioning system concept as the top-level node, and associate the specific information of related air-conditioning products and components in the form of triples.

3. The personalized product customization service method based on fine-grained accurate understanding of supply and demand according to claim 2, characterized in that The data information includes the product manufacturer, model, Chinese name, type, price, product-related assembly parts and characteristic information of the product and parts.

4. The personalized product customization service method based on fine-grained and accurate understanding of supply and demand according to claim 1, characterized in that The specific steps of step S2 are as follows: Step S21: Based on the triple data sources of user comments on the official website of the air conditioner brand, enterprise market research data, and industry white papers, a fine-grained knowledge extraction method is used to systematically mine the core demand characteristics of Internet users for air conditioner products; Step S22: Combine the demand features obtained in step S21 and express them in the knowledge graph to form the final user demand knowledge graph.

5. The personalized product customization service method based on fine-grained accurate perception of supply and demand according to claim 4, characterized in that The user demand knowledge graph includes the following personalized demand features: cooling and heating, air purification, stable operation, noise level, air supply level, intelligent temperature control, air quality, energy-saving performance, environmental certification, appearance style, size and installation, price, cost-effectiveness, children, teenagers, the elderly, people with allergies, people with asthma, people with chronic diseases, long-term home office, long-term home study, frequent outdoor sports, hot and humid areas, cold areas, small spaces, large spaces, electrical appliance safety, and self-protection functions.

6. The personalized product customization service method based on fine-grained accurate perception of supply and demand according to claim 1, characterized in that The specific steps of step S4 are as follows: Step S41, based on the user demand knowledge graph constructed in step S2, an instruction set is generated by combining LLMs with relevant prompt words; Step S42: Effectively expand the size of the instruction set through operations such as synonym replacement, random insertion, random swapping, and random deletion; Step S43: According to the instruction set constructed in Steps S41 - S42, fine-tune the relevant LLMs in the way of LoRA fine-tuning to form domain-specific LLMs that can finely identify users' daily functional requirements.

7. The personalized product customization service method based on fine-grained accurate perception of supply and demand according to claim 1, characterized in that In Step S41, in terms of prompt design, restrict the format of the instruction set to be in daily spoken language and in the processing domain. The steps of prompt design are as follows: Use the API interface of ChatGPT3.5 and combine the personalized demand characteristics obtained in Step S22 to design the prompt.

8. The personalized product customization service method based on fine-grained precise understanding of supply and demand according to claim 1, wherein The specific steps of Step S5 are as follows: Step S51: Through in-depth analysis of the user comments and industry white papers in Step S2, conduct multi-dimensional considerations based on a fine-grained and accurate understanding of the supply and demand in the air-conditioning field; Step S52: On the basis of Step S51, utilize the high-precision text understanding ability of LLMs, and combine the relevant dimensional restrictions in Step S51 to design dimensional recognition prompts. On this basis, use the large language model and the relevant dimensional recognition prompts to judge the user input, and judge whether the user's relevant input is related to product personalized customization. If it is relevant, judge which dimension of the product the user input has a strong correlation with and extract the corresponding information; if it is not relevant, output the corresponding result through the general model.

9. The personalized product customization service method based on fine-grained accurate perception of supply and demand according to claim 1, characterized in that The specific steps of Step S6 are as follows: Step S61: Based on the judgment of product personalized customization relevance and product dimension recognition in Step S5, randomly cut into an angle for product dimension division; Step S62: Design the CoT technology. Through the relevant elements that are not yet complete in the dimension list and the large language model, guide the user to gradually fill in the dimensions for the blank elements in the dimension list, and gradually complete the information confirmation of all dimension points of the product; Step S63: After multiple rounds of question-and-answer input and answers, finally output the product-level assembly plan and component-level replacement plan related to it to the user based on multi-dimensional information confirmation.

10. The personalized product customization service method based on fine-grained accurate perception of supply and demand according to claim 1, characterized in that The specific steps of Step S7 are as follows: Based on the product-level assembly plan and component-level replacement plan given in Step S6, if the user is not very satisfied with the current product-level assembly plan, replace the product components through natural language description using the component-level replacement plan to achieve a higher degree of personalized customization.

Images