A furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II

Abstract

The present invention discloses a furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II, which belongs to the field of management service mechanism and matching technology, including, firstly, dividing product modules and service modules based on product and service characteristics; secondly, extracting customer satisfaction by using topic model to better fit the application scenario of the enterprise; then, on the basis of extracting customer satisfaction, establishing a multi-objective optimization model with customer satisfaction and enterprise revenue as the goal, and solving it by NSGA-II to obtain a set of non-dominated solution sets; finally, using TOPSIS algorithm to select the best solution from multiple evaluation solutions. The present invention adopts the above-mentioned furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II, which is conducive to enterprises meeting the diversified needs of customers, ensuring enterprise revenue, and providing different furniture product service matching solutions for different customers.

Description

Technical Field

[0001] The present invention relates to the field of management service mechanism and matching technology, and in particular to a furniture product and service matching optimization method based on a WES-BTM topic model and NSGA-II. Background Art

[0002] With the development of economy, consumers' requirements for furniture products are increasing day by day, and traditional inventory mode and production mode are gradually impacted. Service-oriented transformation is conducive to enterprises to improve their competitiveness and sustainable development level, and has become the main transformation direction of manufacturing enterprises. However, service-oriented transformation is easy to cause waste of product functions or service resources, personalized products and services will lead to an increase in the rate of decline of corporate profits, and the price of service modules will affect the overall price, which may cause customers to be unwilling to pay. To this end, the present invention proposes a furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II. First, the product and service modules are divided based on the characteristics of products and services. Secondly, customer satisfaction is extracted through the topic model to better fit the application scenario of the enterprise. Then, on the basis of extracting customer satisfaction, a multi-objective optimization model is established with customer satisfaction and corporate profits as the goals, and then the model is solved by NSGA-Ⅱ to obtain a set of non-dominated solution sets. Finally, the TOPSIS algorithm is used to evaluate multiple product service combination schemes in the non-dominated solution set to obtain the optimal decision solution.

[0003] In the existing technology, when matching optimization goals, scholars will consider the needs of customers and enterprises. However, when selecting customer satisfaction indicators, most existing studies select goals based on past experience, which may lead to inaccurate indicators. In terms of solution method selection, when solving multi-objective problems, multi-objective optimization algorithms often produce a set of non-dominated solutions, making it impossible to determine the final solution. Summary of the invention

[0004] The purpose of the present invention is to provide a furniture product and service matching optimization method based on the WES-BTM topic model and NSGA-II to solve the problems existing in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides a furniture product and service matching optimization method based on the WES-BTM topic model and NSGA-II, comprising the following steps:

[0006] S1. Product and service module division: according to the characteristics of products and services, divide the product modules and service modules that match the products and services as matching objects;

[0007] S2. Use the WES-BTM model to analyze and mine customer needs. Enterprises collect customer needs and opinions through customer feedback and platform customer information. Based on the collected customer information, the WES-BTM model is used to mine customers' inherent needs and potential problems. The WES-BTM model is built on the basis of the BTM model. It converts word pairs into word vectors by loading the Word2Vec pre-training model, and completes word pair filtering by calculating the similarity of word pairs.

[0008] S3, building a multi-objective optimization model with customer satisfaction and corporate profits as the goals;

[0009] S4. Design the chromosome encoding method, population initialization method, genetic operator and environment selection of NSGA-Ⅱ, and output a set of non-dominated solutions through NSGA-Ⅱ;

[0010] S5. Use weighted TOPSIS algorithm to make decisions.

[0011] Preferably, step S1 specifically includes:

[0012] S11. Product service module division refers to the enterprise dividing its products into several independently operated components, each with specific functions, for users to select and combine as needed;

[0013] S12. Service modules abstract services and break them down into independent modules, allowing users to freely combine them according to their own needs. Service components of the same service module are interrelated and interdependent, while different service modules are independent of each other. Each service module has attributes that describe itself, including service level, service response, and service time. These attributes are used to identify and describe service modules. Different values ​​are set for these attributes in specific scenarios to obtain different module instances.

[0014] Preferably, step S2 specifically includes:

[0015] S21, word pair screening, first, load a pre-trained Word2Vec model, which is trained on a large-scale Chinese text corpus and provides high-quality word vector embedding; second, calculate the cosine similarity between word pairs; two vectors representing word pairs need to be input, the closer the cosine value is to 1 and the closer the angle is to 0, the higher the similarity between the two vectors; finally, filter out word pairs with small similarity;

[0016] S22. Confirmation of the optimal number of topics. Since the WES-BTM model belongs to unsupervised learning, it is necessary to determine the number of topics before modeling. The consistency score is used to determine the number of topics. The higher the topic consistency score, the higher the semantic similarity within a single topic, and the better the topic division effect.

[0017] S23, topic generation;

[0018] S24. Topic result analysis: After obtaining the topic results through topic model training, topics with high similarity are manually divided, so that the mined topic information is used as a customer satisfaction indicator.

[0019] Preferably, the process of generating the theme in step S23 includes:

[0020] S231. For each topic , calculate the topic distribution of the topic ;

[0021] S232. Calculate the topic distribution of the entire corpus ;

[0022] S233, calculate the similarity of the word pairs, filter the word pairs, and filter the word pairs. Each word pair in Do the following:

[0023] Assume that the filtered word pairs ① From Randomly select topics from the entire set , ②From the theme Randomly select a set of word pairs from , When modeling with WES-BTM, Gibbs sampling is used to adjust the global parameters and Perform approximate inference; first randomly select the initial state of the Markov chain, and then calculate each The conditional distribution of ,in Indicates except For each word pair except , by applying the chain rule to the joint probability of the entire data, we get the conditional probability; finally, we get the topic-word distribution by the number of word-to-topic assignments and word co-occurrences and global topic distribution .

[0024] Preferably, the customer satisfaction in step S3 is the customer satisfaction index using the subject result analysis in step S24; the enterprise profit is the total profit of the product module and the service module as the optimization target of the enterprise profit; and the set Represents the set of all product modules in the product-service matching process, where: Indicates the number of product categories, Represents product instances under each module; use collections Represents a set of all service modules, where: Indicates the category number of the service module. represents the service module instance under each module, and makes the following assumptions: at least one product and service of each type is selected, and at most one product and service of each type is selected. The decision variables of this model are set to and is a 0-1 variable. For example, when the value is 1, it means that the Module No. When the value is 0, it means that the Module No. instances.

[0025] Preferably, step S4 specifically includes:

[0026] S41, use integer coding to encode the chromosome; each gene in the chromosome is represented by an integer, the length of the chromosome represents the number of decision variables, and the number in the code represents the selected module;

[0027] S42, using integer random sampling strategy to generate the initial population, generating the initial solution set by randomly selecting integer values, each individual is a product and service selection plan;

[0028] S43, select the parent generation through the tournament selection method, and select from the population By comparing the ranking and crowding values ​​of individuals, a single individual with a higher ranking and crowding value than other individuals is selected as the winner; by adjusting value, adjust the size of individuals selected for each tournament;

[0029] Use uniform crossover to generate new offspring individuals. Select two parents from the population for crossover and generate masks of the same length for the two parents. The mask consists of a series of randomly generated binary numbers. Each binary number in the sequence is mapped to the corresponding allele on the parent chromosome. In inheritance, the specific value of the mask sequence determines the genetic orientation of the offspring at each allele position.

[0030] The mutation operation is implemented using the integer polynomial mutation method. The genes of several individuals are randomly selected for mutation, and the new gene values ​​are calculated using the polynomial mutation formula. The selected gene values ​​are replaced with the calculated new values ​​to complete the mutation.

[0031] S44. Divide the evolving population into several layers through non-dominated sorting, where the individuals in the first layer are the set of individuals that are not dominated by other individuals, the second layer is the set of individuals that are not dominated by other individuals in the evolving population after removing the first layer, and so on, to form several layers of individual sets. When making a selection, first consider the non-dominated set of the first layer, then the non-dominated set of the second layer, and so on, until the individual quantity requirement is met. Secondly, when two individuals belong to the same non-dominated level, they are sorted by comparing the crowding distance. The crowding distance is an indicator that measures the density of neighboring individuals of an individual in its target space. An individual with a larger crowding distance means that there are not too many other individuals around it, which helps to maintain population diversity. Individuals with a large crowding distance are selected first;

[0032] S45. Repeat the iteration until the termination condition is met, so as to find a set of non-dominated solutions to the multi-objective problem, that is, the Pareto optimal solution set.

[0033] Preferably, the TOPSIS algorithm in step S5 includes: first, constructing a standard decision matrix based on several Pareto optimal solutions obtained by NSGA-Ⅱ, standardizing each indicator in the decision matrix, eliminating the dimensional influence between the indicators, and assigning a weight to each indicator to achieve weighted processing; secondly, the algorithm determines the positive idealized target and the negative idealized target; then, for each scheme, calculating its Euclidean distance from the positive and negative ideal solutions, and using these distances to calculate the comprehensive evaluation index of each scheme; finally, sorting all the schemes according to the comprehensive evaluation index, and the scheme with a high comprehensive evaluation index value is determined as the optimal solution, that is, the scheme with the closest distance to the positive ideal solution and the farthest distance from the negative ideal solution.

[0034] Therefore, the present invention adopts the above-mentioned furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II, considers the challenges of furniture enterprises' service-oriented transformation, establishes and solves a product-service matching model with maximizing enterprise profits and maximizing customer satisfaction as objective functions, obtains the optimal product-service combination solution, and provides new ideas for the service-oriented transformation of manufacturing enterprises.

[0035] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the WES-BTM probability graph model of the present invention;

[0037] Figure 2 This is the flow chart of NSGA-Ⅱ of the present invention. DETAILED DESCRIPTION

[0038] Example

[0039] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0040] A furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II includes the following steps:

[0041] S1. Product and service module division: Enterprises divide product and service matching product modules and service modules according to product and service characteristics as matching objects.

[0042] S11. The essence of modular products is that enterprises divide products into multiple independently operable components, each with specific functions, for users to select and combine as needed. Taking cabinet furniture products as an example, in practice, cabinet products are generally divided into seven modules: cabinet body, door panels, countertops, decorative components, functional accessories, hardware accessories and drawers. Therefore, product module instances are characterized as product modules with different attributes, and one product module can be divided into multiple product instances.

[0043] S12. Service modules involve abstracting and decomposing services into independent modules, so that users can freely combine them according to their own needs. The service components of the same service module are interrelated and interdependent, and different service modules are independent of each other. Among them, each service module has attributes that describe itself, including service level, service response, and service time. These service attributes can be used to identify and describe service modules. By setting different values ​​for these attributes in specific scenarios, different module instances can be obtained. Taking furniture product services as an example, the services of the entire life cycle of furniture products include pre-sales services, in-sales services, and after-sales services. Pre-sales services mainly refer to personalized design services; in-sales services mainly refer to logistics services; after-sales services mainly include on-site repair services, remote fault diagnosis, and parts supply services. The instances of each module are different, and the values ​​of each parameter are also different. For example, personalized design services can be divided into customized design services and advanced customized design services; logistics services can be divided into daily logistics and same-day delivery; on-site repair services can be divided into basic repair services and advanced repair services; remote fault diagnosis can be divided into intelligent diagnosis and manual diagnosis; parts supply services can be divided into non-original parts and original parts supply services, etc.

[0044] S2. Use the WES-BTM model to analyze and mine customer needs. Enterprises collect customer needs and opinions through customer feedback and customer information obtained from the platform. Based on the collected customer information, the WES-BTM model is used to mine customers' inherent needs and potential problems. The WES-BTM model is built on the basis of the BTM model. It converts word pairs into word vectors by loading the Word2Vec pre-trained model, and completes word pair filtering by calculating the similarity of word pairs. Figure 1 shown.

[0045] S21. Word pair screening. First, load a pre-trained Word2Vec model. This pre-trained model is trained on a large-scale Chinese text corpus and provides high-quality word vector embedding. Secondly, calculate the cosine similarity between word pairs. This method requires the input of two vectors representing word pairs. The closer the cosine value is to 1 and the closer the angle is to 0, the higher the similarity of the two vectors. Finally, filter out the word pairs with less similarity.

[0046] S22, confirm the optimal number of topics. Since the WES-BTM model belongs to unsupervised learning, its number of topics needs to be determined before modeling. The present invention uses consistency scoring to determine the number of topics. The higher the topic consistency score, the higher the semantic similarity within a single topic, and the better the topic division effect.

[0047] S23, topic generation. The process of topic generation is mainly divided into the following three steps:

[0048] S231. For each specific topic , calculate the topic distribution for a specific topic ;

[0049] S232. Calculate the topic distribution of the entire corpus ;

[0050] S233, calculate the similarity of the word pairs and filter the word pairs. Each word pair in Do the following.

[0051] Assume that the filtered word pairs ① From Randomly select topics from the entire set ,Right now ②From the theme Randomly select a set of word pairs from ,Right now When modeling with WES-BTM, Gibbs sampling is used to quantify the global parameters. and Perform approximate inference. First, randomly select the initial state of the Markov chain. Then calculate each The conditional distribution of ,in Indicates except For each , by applying the chain rule to the joint probability of the entire data, we can get the conditional probability. Finally, we estimate the topic-word distribution by the number of word-to-topic assignments and word co-occurrences. and global topic distribution .

[0052] S24. Topic result analysis: After obtaining the topic results through the training of the topic model, the topics with high similarity are further divided manually, so that the mined topic information is used as a customer satisfaction indicator.

[0053] S3. Model construction is carried out with customer satisfaction and enterprise revenue as the goals. In terms of customer satisfaction, the results of topic analysis are used as customer satisfaction indicators. In terms of enterprise revenue, the total revenue of product modules and service modules is used as the optimization goal of enterprise revenue. To represent the set of all product modules in the product service matching process, Indicates the number of product categories, Represents the product instance under each module. Use collection To represent the set of all service modules, Indicates the category number of the service module. represents the service module instance under each module. The following assumptions are made: at least one product and service of each type is selected, and at most one product and service of each type is selected. The decision variables of this model are set to and is a 0-1 variable. For example, when the value is 1, it means that the Module No. When the value is 0, it means that the Module No. instances.

[0054] S4. Design the chromosome encoding method, population initialization method, genetic operator and environment selection of NSGA-Ⅱ. Use NSGA-Ⅱ to find a set of non-dominated solutions to the multi-objective problem, that is, the Pareto optimal solution set, such as Figure 2 shown.

[0055] S41. Encode using integer encoding. Each gene in the chromosome is represented by an integer. The length of the chromosome represents the number of decision variables, and the number in the encoding represents which module is selected.

[0056] S42. Use integer random sampling strategy to generate the initial population, and generate the initial solution set by randomly selecting integer values. Each individual is a possible product and service selection plan.

[0057] S43. Parent selection is performed through tournament selection. By comparing the ranking and crowding values ​​of individuals, a single individual with a higher ranking and crowding value than other individuals is selected and retained as the winner. The value can adjust the pressure level of the competing individuals. The smaller the value, the less pressure on the individual, and the greater the probability that individuals with low fitness will be selected.

[0058] New offspring individuals are generated using uniform crossover. Two parents are selected from the population for crossover, and masks of the same length are generated for the two parents. The mask consists of a series of randomly generated binary numbers, each binary number of which is mapped to the corresponding allele on the parent chromosome. In inheritance, the specific value (0 or 1) of the mask sequence determines the genetic orientation of the offspring at each allele position.

[0059] The mutation operation is implemented using the integer polynomial mutation method. The mutation is completed by randomly selecting one or more individual genes for mutation, calculating the new gene value using the polynomial mutation formula, and replacing the selected gene value with the calculated new value.

[0060] S44. Divide the evolving population into several layers by non-dominated sorting. Among them, the individuals in the first layer are the set of individuals that are not dominated by other individuals, the second layer is the set of individuals that are not dominated by other individuals in the evolving population after removing the first layer, and so on, to form several layers of individual sets. When making a selection, first consider the non-dominated set of the first layer, then consider the non-dominated set of the second layer, and so on, until the individual quantity requirement is met. Secondly, when two individuals belong to the same non-dominated level, they are sorted by comparing the crowding distance. The crowding distance is an indicator to measure the density of neighboring individuals of an individual in its target space. An individual with a larger crowding distance means that there are not too many other individuals around it, which helps to maintain population diversity. Individuals with a large crowding distance are selected first.

[0061] S45. Repeat the iteration until the termination condition is met, so as to find a set of balanced solutions to the multi-objective problem, namely, the Pareto optimal solution set.

[0062] S5. Use the weighted TOPSIS method to make decisions. The TOPSIS algorithm first standardizes the indicators in the decision matrix, eliminates the dimensional influence between the indicators, and assigns each indicator a certain weight to achieve weighted processing. Secondly, the algorithm determines the idealized target and the negative idealized target. Then, for each plan, the Euclidean distance between it and the positive and negative ideal solutions is calculated, and these distances are used to calculate the comprehensive evaluation index of each plan. Finally, all plans are sorted according to the comprehensive evaluation index, and the plan with the highest comprehensive evaluation index value is determined as the optimal solution, that is, the plan that is closest to the positive ideal solution and farthest from the negative ideal solution.

[0063] Therefore, the present invention adopts the above-mentioned furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II, takes product modules and service modules as objects, matches and combines product modules with service modules according to customer needs and corporate profits, thereby providing customers with product service integration solutions and realizing value co-creation between enterprises and customers.

[0064] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II, characterized in that: The following steps are involved: S1. Product and service module division: according to the characteristics of products and services, divide the product modules and service modules that match the products and services as matching objects; S2. Use the WES-BTM model to analyze and mine customer needs. Enterprises collect customer needs and opinions through customer feedback and platform customer information channels. Based on the collected customer information, the WES-BTM model is used to mine customers' inherent needs and potential problems. The WES-BTM model is built on the basis of the BTM model. It converts word pairs into word vectors by loading the Word2Vec pre-training model, and completes word pair filtering by calculating the similarity of word pairs. S3, building a multi-objective optimization model with customer satisfaction and corporate profits as the goals; S4. Design the chromosome encoding method, population initialization method, genetic operator and environment selection of NSGA-Ⅱ, and output a set of non-dominated solutions through NSGA-Ⅱ; S5. Use weighted TOPSIS algorithm to make decisions; Step S2 specifically includes: S21, word pair screening, first, load a pre-trained Word2Vec model, which is trained on a large-scale Chinese text corpus and provides word vector embedding; second, calculate the cosine similarity between word pairs; finally, filter out word pairs with small similarity; S22, optimal number of topics confirmed, using consistency scoring to determine the number of topics; S23, topic generation; S24, topic result analysis, after obtaining the topic results through topic model training, the topics with high similarity are manually divided, so that the mined topic information is used as a customer satisfaction indicator; The customer satisfaction in step S3 is the customer satisfaction index based on the subject result analysis in step S24; the enterprise profit is the total profit of the product module and service module as the optimization target of the enterprise profit; the set Represents the set of all product modules in the product-service matching process, where: Indicates the number of product categories, Represents product instances under each module; use collections Represents a set of all service modules, where: Indicates the category number of the service module. Represents the service module instance under each module; Step S4 specifically includes: S41, use integer coding to encode the chromosome; each gene in the chromosome is represented by an integer, the length of the chromosome represents the number of decision variables, and the number in the code represents the selected module; S42, using integer random sampling strategy to generate the initial population, generating the initial solution set by randomly selecting integer values, each individual is a product and service selection plan; S43, select the parent generation through the tournament selection method, and select from the population By comparing the ranking and crowding values ​​of individuals, a single individual with a higher ranking and crowding value than other individuals is selected as the winner; by adjusting value, adjust the size of individuals selected for each tournament; Use uniform crossover to generate new offspring individuals. Select two parents from the population for crossover and generate masks of the same length for the two parents. The mask consists of a series of randomly generated binary numbers. Each binary number in the sequence is mapped to the corresponding allele on the parent chromosome. In inheritance, the specific value of the mask sequence determines the genetic orientation of the offspring at each allele position. The mutation operation is implemented using the integer polynomial mutation method. The genes of several individuals are randomly selected for mutation, and the new gene values ​​are calculated using the polynomial mutation formula. The selected gene values ​​are replaced with the calculated new values ​​to complete the mutation. S44. The evolving population is divided into several layers by non-dominated sorting. When two individuals belong to the same non-dominated layer, they are compared by crowding distance, which is an indicator to measure the density of neighboring individuals of an individual in its target space. S45, iterate repeatedly until the termination condition is met, so as to find a set of non-dominated solutions to the multi-objective problem, which is also the Pareto optimal solution set; The TOPSIS algorithm in step S5 includes: first, according to several Pareto optimal solutions obtained by NSGA-Ⅱ, a standardized decision matrix is ​​constructed, and the dimensional influence between the indicators is eliminated by standardizing the indicators in the decision matrix, and each indicator is given a weight to achieve weighted processing; secondly, the algorithm determines the positive idealized target and the negative idealized target; then, for each plan, the Euclidean distance between it and the positive and negative ideal solutions is calculated, and the comprehensive evaluation index of each plan is calculated using these distances; finally, all plans are sorted according to the comprehensive evaluation index, and the plan with a high comprehensive evaluation index value is determined as the optimal solution.

2. The furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II according to claim 1 is characterized in that: Step S1 specifically includes: S11. Product service module division refers to the enterprise dividing its products into several independently operated components, each with specific functions, for users to select and combine as needed; S12. Service modules abstract services and break them down into independent modules, allowing users to freely combine them according to their own needs. Service components of the same service module are interrelated, while different service modules are independent of each other. Each service module has attributes that describe itself, including service level, service response, and service time. These attributes are used to identify and describe the service module.

3. The furniture product and service matching optimization method based on WES-BTM topic model and NSGA-II according to claim 1 is characterized in that: The process of theme generation in step S23 includes: S231. For each topic , calculate the topic distribution of the topic ; S232. Calculate the topic distribution of the entire corpus ; S233, calculate the similarity of the word pairs, filter the word pairs, and filter the word pairs. Each word pair in Do the following: Assume that the filtered word pairs ① From Randomly select topics from the entire set , ②From the theme Randomly select a set of word pairs from , When modeling with WES-BTM, Gibbs sampling is used to adjust the global parameters and To perform inference, we first randomly select the initial state of the Markov chain and then calculate each The conditional distribution of ,in Indicates except For each word pair except , by applying the chain rule to the joint probability of the entire data, we get the conditional probability; finally, we get the topic-word distribution by the number of word-to-topic assignments and word co-occurrences and global topic distribution .

Images