Welfare program product selection device based on large model and supply chain

Abstract

This invention relates to the field of employee welfare selection technology, and specifically discloses an employee welfare program selection device based on a large model and supply chain, comprising: a demand acquisition module that collects employee welfare demand data and service provider supply chain data; a model selection module that generates multiple employee welfare program selection plans based on a pre-trained large model; a suitability assessment module that analyzes the welfare demand suitability of each employee welfare program selection plan based on a multi-layered sentiment keyword set, the multi-dimensional industry attributes of the employee, employee age distribution, employee geographic distribution, and each selection object in each employee welfare program selection plan; an optimality assessment module that assesses the optimality of each employee welfare program selection plan based on the welfare demand suitability and total gross profit; and a program selection module that selects the employee welfare program selection plan with the highest optimality as the final employee welfare program selection plan; thereby helping enterprises efficiently formulate employee welfare programs that meet their own needs, improve employee satisfaction, and are economically beneficial.

Description

Technical Field

[0001] This invention relates to the field of welfare product selection technology, and in particular to a welfare program selection device based on large-scale models and supply chains. Background Technology

[0002] In today's corporate management field, employee benefits are increasingly important as a means to attract and retain talent and improve employee satisfaction and loyalty. For companies, the selection process is crucial in developing a benefits program that meets their own needs and satisfies employees. However, traditional methods of selecting benefits programs face numerous challenges. With the increasing variety of goods on the market and the diversification of corporate needs, relying solely on human experience and simple market research for benefits selection is insufficient to comprehensively and accurately meet the needs of both the company and its employees. Furthermore, there is a lack of scientific and reasonable evaluation methods for cost-effectiveness. Benefit program selection devices based on large-scale models and supply chains are of great significance in this context. With the continuous development of artificial intelligence technology and companies' pursuit of refined management, benefit program selection devices based on large-scale models and supply chains have broad application prospects in the field of corporate benefits management and are expected to become a key tool for improving the level of corporate benefits management.

[0003] However, existing welfare program selection technologies have many shortcomings. For welfare program selection based on large-scale models, there is a lack of a scientific evaluation system. It cannot accurately analyze the suitability of each welfare program to employee needs based on multi-dimensional factors, nor can it comprehensively evaluate the suitability to employee needs and total gross profit. This results in selected programs either failing to meet employee needs or performing poorly in terms of cost-effectiveness. Ultimately, due to the lack of a scientific evaluation and screening mechanism, it is difficult to select the optimal welfare program from numerous options, thus failing to effectively help improve the quality and effectiveness of corporate welfare management.

[0004] Therefore, this invention proposes a welfare program selection device based on a large model and supply chain. Summary of the Invention

[0005] This invention provides a welfare program selection device based on a large-scale model and supply chain analysis. The large-scale model, with its powerful data analysis and processing capabilities, can deeply mine and analyze massive amounts of welfare demand and supply chain data. Combined with supply chain data, it ensures the feasibility and cost-effectiveness of program selection. Furthermore, by considering the multidimensional industry attributes of the served enterprise, employee age distribution, and employee geographic distribution, it achieves optimal screening of welfare program selection. This device can provide enterprises with more targeted welfare program selection solutions that better meet employee expectations and are cost-effective.

[0006] This invention provides a welfare program selection device based on a large-scale model and supply chain, comprising:

[0007] The demand collection module is used to collect welfare demand data from the service recipients and supply chain data from the service providers.

[0008] The model selection module is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions.

[0009] The fit assessment module is used to analyze the fit of each welfare selection scheme with welfare needs based on the multi-layered sentiment keyword set of welfare purpose tags in the welfare needs data of the service recipient company, the multi-dimensional industry attributes of the service recipient company, the age distribution of employees, the geographical distribution of employees, and each selection object in each welfare selection scheme.

[0010] The optimization evaluation module is used to evaluate the optimization of each welfare product selection scheme based on the welfare demand fit and total gross profit of each welfare product selection scheme.

[0011] The scheme selection module is used to select the welfare scheme with the highest priority among all welfare schemes as the final welfare scheme.

[0012] Preferably, welfare demand data includes: number of people receiving welfare, budget amount, and welfare purpose tag.

[0013] Preferably, the service provider's supply chain data includes: service provider's product inventory information and gross profit threshold.

[0014] The preferred model selection module includes:

[0015] The model building sub-module is used to train a large product selection model by using a large number of historical product selection schemes, the corresponding welfare needs data of the service recipient companies and the corresponding service provider supply chain data as training samples.

[0016] The model selection submodule is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions.

[0017] Preferably, the fit evaluation module includes:

[0018] The responsiveness analysis submodule is used to analyze the responsiveness of each selected item in each welfare selection plan to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise, the multi-dimensional industry attributes of the service enterprise, and the corresponding employee age distribution and employee geographic distribution.

[0019] The fit analysis submodule is used to analyze the fit of each welfare selection scheme with the welfare needs of the client company based on the multi-layered sentiment keyword set of welfare purpose tags in the welfare needs data of each selected object in each welfare selection scheme, the multi-dimensional industry attributes of the client company, and the corresponding employee age distribution and employee geographic distribution.

[0020] Preferably, the responsiveness analysis submodule includes:

[0021] The graph and vector generation unit is used to generate a sentiment graph of the welfare needs of the serviced enterprises based on the correlation strength between word vectors of two keywords in a multi-layer sentiment keyword set of welfare purpose tags in the welfare needs data of the serviced enterprises. It encodes the descriptive text of each selection object in each welfare selection plan into a vector to obtain the descriptive vector of each selection object.

[0022] The first responsiveness analysis unit is used to calculate the responsiveness of each selected object in each welfare selection scheme to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service enterprise, based on the description vector of each selected object and the welfare demand sentiment map of the service enterprise.

[0023] The industry attribute matrix building unit is used to generate a multi-dimensional industry attribute-sentiment association matrix based on the association strength of each keyword in the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise. It also generates a multi-dimensional industry attribute-selection object adaptation matrix based on the fit between the multi-dimensional industry attributes of the service enterprise and each text description keyword of each selection object in each welfare selection plan.

[0024] The second responsiveness analysis unit is used to analyze the responsiveness of each product selection object in each welfare product selection scheme to the multidimensional industry attributes of the served enterprise based on the multidimensional industry attribute-sentiment association matrix and the multidimensional industry attribute-product selection object adaptation matrix of each product selection object.

[0025] The third responsiveness analysis unit is used to evaluate the responsiveness of each selected item in each welfare selection plan to the corresponding employee age distribution and the corresponding employee geographic distribution.

[0026] Preferably, the first responsiveness analysis unit includes:

[0027] The first adaptation determination subunit is used to combine at least two keywords in the welfare demand sentiment graph of the served enterprise to obtain a recombined keyword set, mine all the relevant statements in all recombined keyword sets, calculate the similarity between the description vector of each selected object and the statement vector of each relevant statement in each recombined keyword set, and calculate the semantic description adaptation between each selected object and each recombined keyword set based on the similarity between the description vector of each selected object and the statement vector of all relevant statements in each recombined keyword set.

[0028] The second adaptation determination subunit is used to determine the weight of each reorganized keyword set based on the welfare needs sentiment graph of the served enterprise, and to calculate the relative adaptation of each product selection object and each reorganized keyword set based on the weight of each reorganized keyword set, each product selection object and the semantic description adaptation of each reorganized keyword set.

[0029] The first relevance analysis subunit is used to analyze the relevance of each product selection object to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service recipient enterprise based on the relative fit of all recombined keyword sets of each product selection object and welfare demand sentiment map.

[0030] Preferably, the first responsiveness analysis subunit includes:

[0031] The decreasing topology path fitting end is used to sort and connect all recombined keyword sets in the welfare demand sentiment graph according to the principle of descending relative fit between each product selection object and the welfare demand sentiment graph of the served enterprise, so as to obtain a decreasing fit topology path.

[0032] The responsiveness analysis uses the ratio of the length of the decreasing fit topology path to the length of the maximum topology path in the welfare demand sentiment graph as the fit deviation of each selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise, and the difference between 1 and the fit deviation is used as the responsiveness of the corresponding selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise.

[0033] Preferably, the second responsiveness analysis unit includes:

[0034] The indirect inference subunit is used to reassign the multi-dimensional industry attribute-emotional association matrix based on the relevance of each product selection object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service recipient enterprise, and obtain the multi-dimensional industry attribute-product selection object adaptation inference matrix for each product selection object.

[0035] The second responsiveness analysis subunit is used to calculate the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object based on the multidimensional industry attribute-selection object adaptation estimation matrix and the multidimensional industry attribute-selection object adaptation matrix for each selection object. Based on the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object, the responsiveness of each selection object in each welfare product selection plan to the multidimensional industry attributes of the serviced enterprise is analyzed.

[0036] Preferably, the third responsiveness analysis unit includes:

[0037] The third responsiveness analysis subunit is used to obtain the evaluation scores of a large number of consumers of different ages and regions for each selected item in each welfare product selection plan. Based on the age distribution of the corresponding employees, the evaluation scores of consumers of different ages for each selected item in each welfare product selection plan are calculated to obtain the responsiveness of each selected item to the corresponding age distribution of employees.

[0038] The fourth responsiveness analysis subunit is used to calculate the evaluation scores of consumers in different regions for each product selection option in each welfare product selection scheme based on the geographical distribution of the corresponding employees, and to obtain the responsiveness of each product selection option to the geographical distribution of the corresponding employees.

[0039] The beneficial effects of this invention compared to existing technologies are as follows: It collects welfare demand data from the service recipient company and supply chain data from the service provider, comprehensively integrating key information to provide rich and accurate data support for subsequent product selection. Utilizing a pre-trained product selection model, multiple welfare product selection plans are generated based on the collected data, fully leveraging the powerful analytical capabilities of the model to improve the efficiency and scientific rigor of product selection. Taking into account multi-dimensional factors such as multi-layered sentiment keyword sets for welfare objectives, multi-dimensional industry attributes of the company, and employee age and geographical distribution, the invention deeply analyzes the welfare demand fit of each product selection plan, ensuring that the plan aligns with the actual needs of the company and its employees. Combining welfare demand fit with total gross profit to evaluate the plan's optimality, the invention balances the achievement of corporate welfare goals with cost-effectiveness. The plan with the highest optimality is selected as the final welfare product selection plan, providing companies with the optimal welfare product selection decision-making, helping them efficiently formulate welfare plans that meet their own needs, improve employee satisfaction, and are economically beneficial.

[0040] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0041] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0043] Figure 1 This is a schematic diagram of a welfare program selection device based on a large model and supply chain in an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the model selection module in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the fit evaluation module in an embodiment of the present invention. Detailed Implementation

[0046] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0047] like Figure 1 As shown, the present invention provides an implementation method for a welfare program selection device based on a large model and supply chain, comprising:

[0048] The demand collection module is used to collect welfare demand data from the service recipients and supply chain data from the service providers.

[0049] The model selection module is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions.

[0050] The fit assessment module is used to analyze the fit of each welfare selection scheme with welfare needs based on the multi-layered sentiment keyword set of welfare purpose tags in the welfare needs data of the service recipient company, the multi-dimensional industry attributes of the service recipient company, the age distribution of employees, the geographical distribution of employees, and each selection object in each welfare selection scheme.

[0051] The optimization evaluation module is used to evaluate the optimization of each welfare product selection scheme based on the welfare demand fit and total gross profit of each welfare product selection scheme.

[0052] The scheme selection module is used to select the welfare scheme with the highest priority among all welfare schemes as the final welfare scheme.

[0053] In this embodiment, the service recipient company refers to an enterprise that needs to develop employee benefit plans and is the main party requesting the benefit selection service. For example, a manufacturing company that wants to improve employee satisfaction and loyalty through reasonable benefit selection is the service recipient company, and the device will carry out the benefit plan selection work based on its situation.

[0054] In this embodiment, the welfare needs data includes information such as the number of employees receiving benefits, the budget amount, and the benefit purpose label. For example, a company plans to provide welfare benefits to 300 employees with a budget of 50,000 yuan, and the benefit purpose label is "expressing care." This data provides key information for welfare selection and program evaluation.

[0055] In this embodiment, the service provider's supply chain data includes information on the service provider's product library and a gross profit threshold. The service provider's product library contains details of various products that can be offered as benefits, and the gross profit threshold specifies the minimum profit margin that must be achieved when selecting products. For example, if the service provider's product library has a wide variety of gifts, food, and other products, the gross profit threshold is set at 15%, and product selection must ensure that this profit requirement is met.

[0056] In this embodiment, the pre-trained product selection model is trained using deep learning with a large number of historical product selection schemes (as input samples during training) and corresponding employee welfare demand data and service provider supply chain data (as output samples during training). It can generate multiple welfare product selection schemes based on new input data. For example, by training on numerous welfare product selection cases from different industries, the model can generate suitable product selection schemes for the needs of new companies.

[0057] In this embodiment, the employee welfare product selection plan is generated by a large-scale product selection model based on welfare demand data and supply chain data, and includes a series of product selection combinations. For example, the employee welfare product selection plan generated for a certain enterprise might include branded sports backpacks, health check-up packages, and online training courses, thereby meeting the enterprise's welfare needs while taking into account the supply chain situation.

[0058] In this embodiment, the welfare purpose label indicates the core intention of the enterprise providing welfare benefits, such as "incentivizing innovation" or "promoting teamwork." It provides guidance for evaluating whether the welfare selection plan meets the needs of the enterprise. If the label is "incentivizing innovation," the selection should focus on items or services that stimulate innovation.

[0059] In this embodiment, the multi-layered sentiment keyword set of the welfare purpose label is a set of keywords refined from the welfare purpose label, reflecting the emotions and intentions implied by the welfare purpose. Taking the welfare purpose label "incentivize innovation" as an example, its multi-layered sentiment keyword set may include words such as "breakthrough," "creativity," and "pioneering," used to deeply evaluate the matching degree between the selected products and the welfare purpose.

[0060] In this embodiment, the multidimensional industry attributes, employee age distribution, and employee geographic distribution of the served enterprise are important factors influencing the selection of employee benefits. Multidimensional industry attributes reflect the characteristics of the industry to which the enterprise belongs; for example, the cultural and creative industries emphasize art and creativity. Employee age distribution reflects the preferences of employees of different age groups; for example, younger employees may favor electronic products, while older employees may need more health-related products. Employee geographic distribution reveals the differences in needs among employees in different regions; for example, employees in the south may have a greater need for heatstroke prevention products, while employees in the north may need more warm clothing. These factors help to make employee benefit selection more closely aligned with their actual needs.

[0061] In this embodiment, the selected items are specific goods or services in the welfare selection plan, such as skin care products of a specific brand, tickets to a concert, professional skills training services, etc. These constitute the actual content of welfare and need to be selected by taking into account multiple factors in order to meet the company's welfare objectives and the needs of employees.

[0062] In this embodiment, the welfare needs fit of the welfare product selection plan refers to the degree to which the welfare product selection plan matches the welfare needs of the enterprise being served. A plan with a high fit is better able to meet the needs of both the enterprise and its employees. For example, a plan that targets the enterprise's innovation needs and the preferences of young employees has a high fit.

[0063] In this embodiment, total gross profit refers to the total profit that can be generated by all selected items in the welfare product selection scheme while meeting the service provider's supply chain gross profit threshold requirements. It is an important indicator for evaluating the economic benefits of the welfare product selection scheme. For example, if a scheme's selected items meet the 20% gross profit threshold, the total gross profit is calculated by comprehensively evaluating all selected items, which is used to measure the economic feasibility of the scheme.

[0064] In this embodiment, the optimality of the welfare product selection scheme is an indicator derived by evaluating the scheme based on the overall welfare demand fit and total gross profit. It takes into account both the achievement of corporate welfare goals and cost-effectiveness. A scheme with a high optimality not only meets the needs of the company and its employees but also has good economic benefits. The device will use this as the basis to select the final welfare product selection scheme.

[0065] To clarify the composition of welfare demand data and provide clear demand dimensions for subsequent product selection and evaluation, it is proposed that welfare demand data include: number of people receiving welfare, budget amount, and welfare purpose tags.

[0066] In this embodiment, the number of employees to whom benefits are to be distributed refers to the number of employees that the service recipient company plans to provide benefits to. This data plays a crucial quantitative role in the selection of benefits. For example, if the number of employees to whom benefits are to be distributed is 100, then when selecting benefits, it is necessary to ensure that the quantity of the selected benefits or services can meet the needs of these 100 people.

[0067] In this embodiment, the budget amount is the upper limit set by the service provider for this welfare distribution. It directly limits the scope and level of welfare items selected. For example, if the budget amount is 50,000 yuan, the selection of items needs to be considered within this budget. This means that it is necessary to both meet the goals that the company hopes to achieve through welfare and select suitable items within a limited budget.

[0068] In this embodiment, the welfare purpose label clearly defines the core objective that the recipient company wants to achieve by distributing benefits. It provides guidance for the entire benefits selection process. For example, if the welfare purpose label is "enhancing employee cohesion," then the selection should revolve around items or activities that promote employee communication and cooperation, such as team-building activities and group sports equipment. Different welfare purpose labels will lead to drastically different selection directions. If it's "rewarding outstanding employees," the selection might lean towards items with higher value and symbolic meaning, such as customized trophies and high-end business gifts.

[0069] To clearly define the service provider's supply chain data and provide the necessary supply chain information foundation for the product selection model and subsequent evaluation, it is proposed that the service provider's supply chain data includes: service provider's product inventory information and gross profit threshold.

[0070] In this embodiment, the service provider's product database information includes detailed information on various products that the service provider can offer for employee welfare selection. This covers product categories, such as food, electronic products, daily necessities, and cultural and entertainment products; product brands, as different brands differ in quality, brand awareness, and style; product specifications, such as performance indicators for electronic products, and ingredients and packaging specifications for food; and information on the product's sourcing channels, i.e., the current quantity of each product available.

[0071] In this embodiment, the gross profit threshold is a profit standard set by the service provider to limit the minimum profit margin expected when selecting products from the product library as welfare items. It has a significant impact on the cost-effectiveness of the welfare item selection plan. For example, a gross profit threshold of 20% means that the service provider hopes that the gross profit from the sale of the selected product combination will reach 20% of the product cost. This affects the product selection decision. On the one hand, companies need to consider the relationship between product price and cost when selecting products to ensure that the gross profit threshold requirement is met; on the other hand, the service provider will also screen and recommend products in the product library based on this, prioritizing those that meet the gross profit requirement to ensure its own economic benefits. This also encourages companies to plan their welfare budget more rationally while meeting welfare needs.

[0072] like Figure 2As shown, in order to train and build a large-scale product selection model using historical data, and to generate welfare product selection solutions based on this model, a product selection module is proposed, including:

[0073] The model building sub-module is used to train a large product selection model by using a large number of historical product selection schemes, the corresponding welfare needs data of the service recipient companies and the corresponding service provider supply chain data as training samples.

[0074] The model selection submodule is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions.

[0075] In this embodiment, historical product selection plans, corresponding employee welfare needs data, and corresponding service provider supply chain data form a crucial data foundation for training the large-scale product selection model. Historical product selection plans include various welfare product combinations developed for different companies in the past, detailing the specific goods or services selected. The corresponding employee welfare needs data covers information such as the number of employees receiving benefits, budget amount, and welfare purpose tags at the time, reflecting the company's expectations and requirements for benefits. The corresponding service provider supply chain data records information on available products in the service provider's product library at the time, as well as the set gross profit threshold. For example, a welfare product selection plan was developed for an internet company, selecting benefits such as smart bracelets and online learning courses. The company at the time had 200 employees receiving benefits, a budget of 150,000 yuan, and the welfare purpose tag was "improving employee health and skills." The service provider's product library offered various brands of smart bracelets and course resources, with a gross profit threshold of 18%. These data collectively constitute a set of historical data. Numerous such historical data sets provide rich examples for model training.

[0076] In this embodiment, a large-scale product selection model is built by using a large amount of historical product selection plans, corresponding employee welfare needs data, and corresponding service provider supply chain data as training samples. This involves inputting this historical data into a specific machine learning algorithm or model framework. By continuously learning the patterns, relationships, and rules within this data, the model can understand how different welfare needs data (such as welfare objectives, budgets, number of employees, etc.) and service provider supply chain data (product inventory information, gross profit thresholds) influence product selection decisions, thereby generating reasonable welfare product selection plans. For example, during training, the model may discover that for technology companies with a high budget and a welfare objective of "incentivizing innovation," selecting high-end creative design tools and cutting-edge technology training courses as welfare product selection plans is often more popular, while also meeting the service provider's gross profit threshold requirements. After training with a large amount of such data, the model gradually masters these rules, enabling it to generate reasonable welfare product selection plans that meet actual needs for newly input employee welfare needs data and service provider supply chain data.

[0077] like Figure 3 As shown, in order to accurately assess the suitability of welfare product selection schemes for welfare needs by analyzing the correlation between the target audience and multi-dimensional factors of welfare needs, a suitability assessment module is proposed, including:

[0078] The responsiveness analysis submodule is used to analyze the responsiveness of each selected item in each welfare selection plan to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise, the multi-dimensional industry attributes of the service enterprise, and the corresponding employee age distribution and employee geographic distribution.

[0079] The fit analysis submodule is used to analyze the fit of each welfare selection scheme with the welfare needs of the client company based on the multi-layered sentiment keyword set of welfare purpose tags in the welfare needs data of each selected object in each welfare selection scheme, the multi-dimensional industry attributes of the client company, and the corresponding employee age distribution and employee geographic distribution.

[0080] In this embodiment, the degree of correspondence between each selected item in each welfare selection scheme and the multi-layered sentiment keyword set of welfare purpose tags in the welfare demand data of the served enterprise, the multi-dimensional industry attributes of the served enterprise, and the corresponding employee age distribution and employee geographical distribution is an indicator for measuring the degree of fit between the selected item and the enterprise's various welfare needs.

[0081] The relevance of the multi-layered affective keyword set to the welfare objective label is important. For example, if the welfare objective label is "improving employee well-being," its multi-layered affective keyword set might include "comfort," "pleasure," and "care." If the selected product is a high-quality massage chair, its description and functions will be highly relevant to keywords such as "comfort" and "care," indicating a high degree of relevance to this keyword set. This means that the selected product can effectively reflect the welfare objective of "improving employee well-being."

[0082] The relevance of the product selection to the multidimensional industry attributes of the served companies is crucial. Taking the technology industry as an example, its industry attributes may emphasize innovation and efficiency. If the selected product is a brand-new tablet computer, its ability to improve work efficiency and its innovative features resonate with the attributes of the technology industry, thus resulting in a high degree of relevance to those attributes.

[0083] Regarding the relevance of the employee age distribution, assuming that the company has a large number of young employees, and the target product is fashionable sports equipment, since young employees generally love sports, this sports equipment resonates with the age distribution of young employees and has a high relevance.

[0084] Regarding the relevance of the product selection based on the geographical distribution of employees, if most of the company's employees are located in the north where winters are cold, selecting high-quality warm clothing products would resonate with the geographical distribution of employees and have a high degree of relevance.

[0085] In this embodiment, the welfare demand fit of each welfare selection scheme is analyzed based on the relevance of each selected item in each welfare selection scheme to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise, the multi-dimensional industry attributes of the served enterprise, and the corresponding employee age distribution and employee geographic distribution. This analysis comprehensively considers the relevance of the selected items in the above aspects to determine the degree of matching between the entire welfare selection scheme and the enterprise's welfare needs. This can be obtained by weighting the relevance of each selected item in each welfare selection scheme to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise, the multi-dimensional industry attributes of the served enterprise, and the corresponding employee age distribution and employee geographic distribution based on preset weights. For example:

[0086] Suppose there is a welfare product selection plan, which includes product A (a smart health bracelet), and the service recipient is an internet technology company.

[0087] The relevance of target product A to the multi-layered sentiment keyword set of the welfare purpose tag: The welfare purpose tag is "improving employee health and work efficiency", and the multi-layered sentiment keyword set includes terms such as "health", "convenience", and "efficiency". Through a series of analyses (such as generating sentiment maps and calculating semantic description fit as described above), the relevance of target product A to these keyword sets is found to be 0.8.

[0088] The degree to which product selection target A corresponds to multi-dimensional industry attributes:

[0089] The multidimensional industry attributes of internet technology companies include innovation and a technological feel. Analysis (including steps such as generating a multidimensional industry attribute-sentiment association matrix and a multidimensional industry attribute-product selection object adaptation matrix) shows that product selection object A has a relevance of 0.7 to the company's multidimensional industry attributes.

[0090] The relevance of product selection target A to the age distribution of employees:

[0091] The company's employees are distributed as follows: 60% are 25-35 years old, 30% are 36-45 years old, and 10% are over 45 years old. Evaluation scores of the smart health bracelet were collected from consumers of different ages. The average score for the 25-35 year olds was 8 points, the average score for the 36-45 year olds was 7 points, and the average score for the over 45 year olds was 6 points.

[0092] The weighted calculation based on the employee age distribution is: (8×60%+7×30%+6×10%)÷10=0.75, that is, the responsiveness of product selection A to the employee age distribution is 0.75.

[0093] The relevance of product selection target A to the geographical distribution of employees:

[0094] The geographical distribution of the company's employees is as follows: 40% are from the northern region and 60% are from the southern region. Evaluation scores of the smart health bracelets were collected from consumers in different regions; the average score for consumers in the northern region was 7.5 points, and the average score for consumers in the southern region was 8 points.

[0095] The weighted calculation based on the geographical distribution of employees is: (7.5×40%+8×60%)÷10=0.78, that is, the responsiveness of product selection A to the geographical distribution of employees is 0.78.

[0096] Assuming the preset weights are: 0.4 for the relevance of the multi-layered emotional keyword set of the welfare purpose tag, 0.3 for the relevance of the multi-dimensional industry attributes, 0.2 for the relevance of the employee age distribution, and 0.1 for the relevance of the employee geographical distribution.

[0097] The suitability of the welfare needs of the selected product A is: 0.8×0.4+0.7×0.3+0.75×0.2+0.78×0.1=0.32+0.21+0.15+0.078=0.758.

[0098] Through such calculations, the suitability of product selection A to the company's welfare needs can be quantified. For the entire welfare product selection plan, after each product selection has undergone similar calculations, and the suitability of all product selections is comprehensively considered, the degree of matching between the welfare product selection plan and the company's welfare needs can be fully analyzed.

[0099] To comprehensively analyze the relevance between target audiences and welfare purpose tags, industry attributes, age, and geographic distribution by generating sentiment maps, descriptive vectors, and correlation matrices, a relevance analysis submodule is proposed, including:

[0100] The graph and vector generation unit is used to generate a sentiment graph of the welfare needs of the serviced enterprises based on the correlation strength between word vectors of two keywords in a multi-layer sentiment keyword set of welfare purpose tags in the welfare needs data of the serviced enterprises. It encodes the descriptive text of each selection object in each welfare selection plan into a vector to obtain the descriptive vector of each selection object.

[0101] The first responsiveness analysis unit is used to calculate the responsiveness of each selected object in each welfare selection scheme to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service enterprise, based on the description vector of each selected object and the welfare demand sentiment map of the service enterprise.

[0102] The industry attribute matrix building unit is used to generate a multi-dimensional industry attribute-sentiment association matrix based on the association strength of each keyword in the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise. It also generates a multi-dimensional industry attribute-selection object adaptation matrix based on the fit between the multi-dimensional industry attributes of the service enterprise and each text description keyword of each selection object in each welfare selection plan.

[0103] The second responsiveness analysis unit is used to analyze the responsiveness of each product selection object in each welfare product selection scheme to the multidimensional industry attributes of the served enterprise based on the multidimensional industry attribute-sentiment association matrix and the multidimensional industry attribute-product selection object adaptation matrix of each product selection object.

[0104] The third responsiveness analysis unit is used to evaluate the responsiveness of each selected item in each welfare selection plan to the corresponding employee age distribution and the corresponding employee geographic distribution.

[0105] In this embodiment, keyword word vectors are a representation of keywords in text converted into numerical vectors. They can be generated based on existing bag-of-words (BOW) models or the Continuous Bag-of-Words (CBOW) model. For example, the word vectors of the keywords "happiness" and "pleasure" are relatively close in the vector space, indicating semantic similarity. In this scenario, each keyword in the multi-layered sentiment keyword set for welfare purpose labels is represented as a word vector for subsequent analysis of the relationships between them.

[0106] In this embodiment, the correlation strength between word vectors of pairs of keywords in the multi-layered sentiment keyword set of the welfare purpose tag in the welfare needs data of the served enterprises is used to measure the semantic tightness of these keywords. The correlation strength is determined by methods such as calculating the similarity between word vectors (e.g., cosine similarity). For example, for the two keywords "team" and "cooperation" in the multi-layered sentiment keyword set of the welfare purpose tag "enhancing employee sense of belonging", the cosine similarity of their word vectors is calculated. If the similarity is high, it indicates that the two keywords have a strong correlation, meaning that they are semantically closely related and both are closely linked to the purpose of enhancing employee sense of belonging.

[0107] In this embodiment, a welfare needs sentiment graph is generated based on the correlation strength between word vectors of pairs of keywords in a multi-layered sentiment keyword set of welfare purpose tags from the welfare needs data of the served enterprises. This graph graphically illustrates the relationships between these keywords, with nodes representing keywords, edges representing the associations between keywords, and the weight of the edges determined by the correlation strength. For example, if the keywords "team" and "cooperation" have a high correlation strength, then the edge weight between them in the graph will be larger, and the connecting line will be thicker, intuitively showing the close connection between these two keywords in welfare purposes. Through the welfare needs sentiment graph, the emotional and semantic structure contained in welfare purposes can be more clearly understood, providing a foundation for subsequent analysis of the fit between selected products and welfare purposes.

[0108] In this embodiment, the descriptive text of each selected item in each welfare item selection scheme is encoded into a vector to obtain a descriptive vector for each selected item. This is the process of converting the textual description of the selected item into a numerical vector, similar to the word vector representation of keywords. This can be generated based on existing bag-of-words (BOW) models or the Continuous Bag-of-Words (CBOW) model. Through this encoding method, the descriptive information of the selected item can be expressed in vector form in space, facilitating comparison and analysis with keyword vectors in the welfare demand sentiment graph. For example, if the selected item is a "customized team culture T-shirt," its descriptive text, after encoding, yields a descriptive vector. This vector contains semantic information related to team culture and can be used for subsequent relevance calculations to the multi-layered sentiment keyword set of welfare purpose labels.

[0109] In this embodiment, a multi-dimensional industry attribute-sentiment association matrix is ​​generated based on the association strength of each keyword in the multi-layered sentiment keyword set of the welfare purpose tag in the welfare needs data of the served enterprise, according to the multi-dimensional industry attributes of the served enterprise. The rows of the matrix may represent different industry attributes (such as innovation and forward-thinking in the technology industry), and the columns represent the keywords in the multi-layered sentiment keyword set of the welfare purpose tag. The value of each matrix element indicates the association strength between the corresponding industry attribute and the keyword. For example, for the technology industry, the association strength with the keyword "innovation" may be high, resulting in a larger element value in the corresponding position in the matrix, indicating a close relationship between the technology industry attribute and the keyword "innovation," reflecting the technology industry's need to enhance employees' innovative awareness through benefits. Assuming the served enterprise is a fintech company, its multi-dimensional industry attributes may include professionalism, stability, innovation, and risk control. The welfare purpose tag is "enhancing employees' professional competence and innovative ability," and the multi-layered sentiment keyword set may include "professional knowledge," "innovative thinking," "risk awareness," and "stable development."

[0110] A review of reports on the fintech industry reveals that they often emphasize the importance of innovation for fintech companies to break through traditional financial models and expand into new businesses in the current market environment. For example, a report might point out that innovation enables companies to gain a greater advantage in a highly competitive market and is closely related to enhancing employees' "innovative thinking." Therefore, the correlation strength between "innovation" and "innovative thinking" could be set at 0.8 (out of 1). Simultaneously, the report might also mention that professional competence is the foundation for the stable operation of fintech companies, so the correlation strength between "professionalism" and "professional knowledge" could be set at 0.9.

[0111] Experts in the fintech field were invited to score the evaluation based on their experience and in-depth understanding of the industry. The experts believe that in the fintech industry, risk control is always integrated into business processes and is closely related to cultivating employees' "risk awareness." Therefore, the correlation between "risk control" and "risk awareness" can be rated as 0.7. While "stability" is important for business development, its direct correlation with "innovative thinking" is relatively weak, and can be rated as 0.4.

[0112] We collected internal documents and meeting minutes from the fintech company and analyzed their statements regarding welfare objectives and industry attributes. If we found that mentions of "stable development" and "stability" were frequent and closely related, data analysis determined the correlation strength between "stability" and "stable development" to be 0.75.

[0113] In this embodiment, a multi-dimensional industry attribute-product selection matching matrix is ​​generated for each product selection object based on the compatibility between the multi-dimensional industry attributes of the served enterprise and the keywords in the text description of each product selection object in each welfare product selection scheme. This matrix also uses rows to represent different industry attributes and columns to correspond to keywords in the product selection object's description text. The values ​​of the matrix elements reflect the degree of compatibility between the industry attributes and the keywords in the product selection object's description. For example, for the technology industry and the product selection object "smart bracelet," if its description text contains keywords such as "smart technology" and "convenient innovation," it has a high degree of compatibility with the innovativeness and convenience attributes of the technology industry, and the element value at the corresponding position in the matrix will be larger, indicating that the smart bracelet performs well in terms of industry attribute compatibility.

[0114] Taking the aforementioned fintech company as an example, suppose a welfare product selection plan includes product A: professional financial training courses. Its text description keywords may include "in-depth explanation of financial knowledge", "analysis of cutting-edge financial technology trends", "case practice exercises", etc.

[0115] Semantic analysis: Utilizing natural language processing technology, semantic matching is performed on the descriptive text and industry attributes of the selected products. For example, for the industry attribute of "professionalism," "In-depth Explanation of Financial Knowledge" clearly reflects professional content, and through semantic similarity calculation (such as cosine similarity), its fit with "professionalism" reaches 0.8. "Analysis of Cutting-Edge Financial Technology Trends" involves both innovative financial technology content and is related to industry professionalism, achieving a fit of 0.7 with "professionalism" and 0.8 with "innovation."

[0116] Collect feedback from fintech industry professionals regarding similar product selections. If most practitioners believe that "case study exercises" are very helpful in improving risk control capabilities, based on this feedback, the fit between "case study exercises" and "risk control" can be set at 0.7.

[0117] Based on the generally accepted training course standards and requirements in the fintech industry, the content of the selected course was evaluated to ensure it met the criteria. If the standards emphasize the systematic transmission of financial knowledge, and the course excels in this aspect, the fit between "in-depth explanation of financial knowledge" and "professionalism" could be further improved to 0.9. Simultaneously, comparing this to the industry's pursuit of innovation, the course's in-depth analysis of cutting-edge fintech trends also improved its fit with "innovation" to 0.85.

[0118] To accurately analyze the responsiveness of target products to a multi-layered sentiment keyword set for welfare-oriented tags by combining keywords, calculating semantic description fit, and relative fit, a first responsiveness analysis unit is proposed, including:

[0119] The first adaptation determination subunit is used to combine at least two keywords in the welfare demand sentiment graph of the served enterprise to obtain a recombined keyword set, mine all the relevant statements in all recombined keyword sets, calculate the similarity between the description vector of each selected object and the statement vector of each relevant statement in each recombined keyword set, and calculate the semantic description adaptation between each selected object and each recombined keyword set based on the similarity between the description vector of each selected object and the statement vector of all relevant statements in each recombined keyword set.

[0120] The second adaptation determination subunit is used to determine the weight of each reorganized keyword set based on the welfare needs sentiment graph of the served enterprise, and to calculate the relative adaptation of each product selection object and each reorganized keyword set based on the weight of each reorganized keyword set, each product selection object and the semantic description adaptation of each reorganized keyword set.

[0121] The first relevance analysis subunit is used to analyze the relevance of each product selection object to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service recipient enterprise based on the relative fit of all recombined keyword sets of each product selection object and welfare demand sentiment map.

[0122] In this embodiment, all statements involving all recombined keyword sets are extracted, that is, all statements containing these recombined keyword sets are found from relevant text data (such as corporate benefit-related documents, industry reports, employee feedback, etc.). These statements can more clearly demonstrate the meaning and application of the recombined keyword sets in the actual context. For example, in a company's discussion records about improving the office environment, there might be a statement like "By providing ergonomic office chairs, we can improve employee comfort and thus increase work efficiency," which contains the recombined keyword set of "comfort" and "work efficiency" mentioned above.

[0123] In this embodiment, the statement vectors are generated by converting the mined statements containing the recombined keyword set into vector form. Similar to word vectors for keywords, statement vectors can numerically represent the semantic features of a statement, facilitating subsequent operations such as similarity calculations. Natural language processing techniques, such as the bag-of-words model and deep learning models, can encode statements into vectors. For example, processing the statement "By providing ergonomic office chairs..." using a specific neural network model yields a vector representing the semantic information of that statement.

[0124] In this embodiment, the similarity between the description vector of each selected product and the statement vector of each related statement in each recombined keyword set is calculated to measure the semantic similarity between the selected product and the statements containing the recombined keyword set. Commonly used similarity calculation methods include cosine similarity and Euclidean distance.

[0125] In this embodiment, the semantic description fit between each product selection object and each recombined keyword set is calculated based on the similarity between the description vector of each product selection object and the statement vectors of all relevant statements in each recombined keyword set. This step is a comprehensive quantification of the semantic matching degree between the product selection object and the recombined keyword set. For example, the semantic description fit can be represented by averaging the similarity between the description vector of the product selection object and the statement vectors of all relevant statements in the recombined keyword set.

[0126] In this embodiment, the weight of each restructured keyword set is determined based on the welfare needs sentiment graph of the served enterprise. The welfare needs sentiment graph displays the strength of association between keywords. Based on these associations, as well as factors such as the keyword's position in the graph and its closeness to the welfare purpose label, each restructured keyword set can be assigned a corresponding weight. Assume the served enterprise is a creative design company, and its welfare purpose label is "stimulating employee creativity and improving team collaboration." A welfare needs sentiment graph is constructed based on relevant text data, where multi-layered sentiment keyword sets include keywords such as "creative inspiration," "team communication," "design innovation," "collaborative atmosphere," and "brainstorming."

[0127] First, these keywords are combined to obtain a recombined keyword set, for example:

[0128] Reorganized keyword set 1: "Creative Inspiration" + "Design Innovation";

[0129] Reorganization keyword set 2: "team communication" + "cooperative atmosphere";

[0130] Reorganized keyword set 3: "Brainstorming" + "Creative Inspiration";

[0131] The next step is to analyze how to determine the weight of each recombined keyword set based on the welfare demand sentiment graph:

[0132] Association Strength: In the welfare needs sentiment map, if the line connecting "creative inspiration" and "design innovation" is thick (indicating high association strength), it means that these two keywords are closely related semantically and in terms of achieving welfare goals. Let's assume, through some quantitative method, their association strength score is 0.8 (out of 1). In comparison, the association strength score for "team communication" and "collaborative atmosphere" is 0.7, and the association strength score for "brainstorming" and "creative inspiration" is 0.75. Higher association strength is more advantageous when determining weights.

[0133] Keyword Position in the Graph: If "Creative Inspiration" is located at the core of the graph, surrounded by multiple other keywords, it indicates that it plays a crucial role in achieving the welfare objective. While "Team Communication" is important, its position is relatively peripheral. The recombined keyword sets composed of keywords in core positions should have relatively higher weights. For example, due to the core position of "Creative Inspiration," recombined keyword sets 1 and 3, which contain it, will have a greater advantage in weight considerations.

[0134] The relevance of the benefit objective label to the overall goal: The benefit objective label "Stimulating employee creativity and enhancing teamwork" emphasizes both creativity and teamwork. Keywords such as "creative inspiration," "design innovation," and "brainstorming" directly revolve around "stimulating employee creativity," showing a high degree of relevance to the benefit objective label. "Team communication" and "collaborative atmosphere" also revolve around "enhancing teamwork," and are closely related. However, considering the overall focus of the benefit objective, creativity is more crucial for creative design companies. Therefore, reorganized keyword sets involving creativity-related keywords may be given higher weight.

[0135] Taking all the above factors into account, a weight is determined for each recombined keyword set:

[0136] Reorganized Keyword Set 1: Since “creative inspiration” and “design innovation” have a high correlation (0.8), and “creative inspiration” is in the core position and closely related to the welfare purpose label, it is assigned a weight of 0.35.

[0137] Reorganized keyword set 2: The correlation strength of "team communication" and "cooperative atmosphere" is 0.7, which is relatively marginal and slightly less closely related to the welfare purpose label. It is assigned a weight of 0.25.

[0138] Reorganized keyword set 3: The correlation strength between "brainstorming" and "creative inspiration" is 0.75. The core position of "creative inspiration" and its close connection with the welfare purpose label are assigned a weight of 0.4.

[0139] In this way, based on the welfare demand sentiment graph, and taking into account factors such as the strength of the association between keywords, their position in the graph, and their closeness to welfare purpose tags, each recombined keyword set is assigned a corresponding weight. These weights will be used for subsequent analysis of the suitability of the selected products to the welfare purpose.

[0140] In this embodiment, the relative fit between each product selection object and each recombined keyword set is calculated based on the weight of each recombined keyword set, the semantic description fit of each recombined keyword set, and the semantic description fit of each recombined keyword set. This can typically be calculated by multiplying the weight by the semantic description fit. For example, if a recombined keyword set has a weight of 0.8 and a product selection object has a semantic description fit of 0.7 with that recombined keyword set, then their relative fit might be 0.8 × 0.7 = 0.56. This relative fit considers both the semantic matching degree between the product selection object and the recombined keyword set, as well as the importance of the recombined keyword set in the welfare objective, thus more accurately reflecting the actual fit between the product selection object and the recombined keyword set.

[0141] In this embodiment, the relative fit analysis of all recombined keyword sets based on each selected product and welfare demand sentiment map assesses the responsiveness of each selected product to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise. By comprehensively considering the relative fit between the selected product and all recombined keyword sets, the degree of fit between the selected product and the multi-layered sentiment keyword set of the welfare purpose tag can be fully evaluated. For example, statistical operations such as summing and averaging the relative fit between the selected product and all recombined keyword sets can be performed to obtain a numerical value to represent the responsiveness. If the value is high, it indicates that the responsiveness between the selected product and the multi-layered sentiment keyword set of the welfare purpose tag is high, that is, the selected product can better reflect the emotions and intentions contained in the enterprise's welfare purpose; conversely, it indicates that the responsiveness is low, and the fit between the selected product and the welfare purpose needs to be improved.

[0142] To determine the relevance of selected products to a multi-layered sentiment keyword set for welfare objectives in an intuitive and quantitative way by fitting a suitable decreasing topology path, a first relevance analysis subunit is proposed, including:

[0143] The decreasing topology path fitting end is used to sort and connect all recombined keyword sets in the welfare demand sentiment graph according to the principle of descending relative fit between each product selection object and the welfare demand sentiment graph of the served enterprise, so as to obtain a decreasing fit topology path.

[0144] The responsiveness analysis uses the ratio of the length of the decreasing fit topology path to the length of the maximum topology path in the welfare demand sentiment graph as the fit deviation of each selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise, and the difference between 1 and the fit deviation is used as the responsiveness of the corresponding selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise.

[0145] In this embodiment, based on the principle of descending relative fit between each selected product and the welfare needs sentiment graph of the served enterprise, all recombined keyword sets in the welfare needs sentiment graph are sorted and connected to obtain a decreasing fit topology path. This means comparing the relative fit calculated between the selected product and each recombined keyword set, placing the recombined keyword sets with higher relative fit first, and the lower ones last, and then connecting these recombined keyword sets sequentially to form a specific path. For example, for a certain selected product, the relative fit with recombined keyword set A is 0.8, with recombined keyword set B is 0.6, and with recombined keyword set C is 0.4. Then, connecting these three recombined keyword sets in descending order yields the decreasing fit topology path ABC. During the connection process, all keywords contained in the recombined keyword sets are merged and replaced into a new node before being connected. This path can intuitively show the changing trend of the fit between the selected product and different recombined keyword sets, providing a basis for further analysis of the fit between the selected product and welfare needs.

[0146] In this embodiment, the ratio of the length of the decreasing fit topology path to the length of the longest topology path in the welfare demand sentiment graph is a quantitative indicator used to measure the overall fit between the selected product and the welfare demand. By calculating the ratio of their lengths, we can understand the extent to which the selected product can encompass the welfare intentions contained in the welfare demand sentiment graph. For example, if the length of the decreasing fit topology path is 8 (assuming the path contains 8 keyword sets), and the length of the longest topology path in the welfare demand sentiment graph is 10 (containing 100 keywords), then their length ratio is 0.08. The closer this ratio is to 1, the higher the deviation between the selected product and the welfare intentions reflected in the welfare demand sentiment graph, that is, the less fully the selected product can respond to the meaning expressed by the multi-layered sentiment keyword set of the welfare purpose label; conversely, the smaller the ratio, the higher the fit.

[0147] To calculate and comprehensively analyze the relevance of selected products to multi-dimensional industry attributes based on the relevance of the target audience to the welfare purpose label, a second relevance analysis unit is proposed, including:

[0148] The indirect inference subunit is used to reassign the multi-dimensional industry attribute-emotional association matrix based on the relevance of each product selection object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service recipient enterprise, and obtain the multi-dimensional industry attribute-product selection object adaptation inference matrix for each product selection object.

[0149] The second responsiveness analysis subunit is used to calculate the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object based on the multidimensional industry attribute-selection object adaptation estimation matrix and the multidimensional industry attribute-selection object adaptation matrix for each selection object. Based on the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object, the responsiveness of each selection object in each welfare product selection plan to the multidimensional industry attributes of the serviced enterprise is analyzed.

[0150] In this embodiment, based on the responsiveness of each selected product to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare needs data of the service recipient enterprise, the multi-dimensional industry attribute-sentiment association matrix is ​​re-assigned to obtain the multi-dimensional industry attribute-selected product fit estimation matrix for each selected product. The core of this step is to use the responsiveness of the selected product to the multi-layered sentiment keyword set of the welfare purpose tag to adjust the correlation strength between the multi-dimensional industry attributes and sentiment keywords, thereby obtaining a new matrix related to the selected product. For example, suppose that in the multi-dimensional industry attribute-sentiment association matrix, the correlation strength between the technology industry and the keyword "innovation" is originally 0.7. If a selected product (such as an innovative smart office device) has a higher responsiveness to the multi-layered sentiment keyword set of the welfare purpose tag, for example, 0.8. When reassigning values, the matrix elements might be adjusted based on this responsiveness, with the new value potentially becoming 0.7 × 0.8 = 0.56 (this is just an illustrative calculation). This process is repeated for all elements in the matrix to obtain a multi-dimensional industry attribute-product selection fit calculation matrix specific to each product selection target. The significance of this is that it integrates the alignment between the product selection target and the welfare objective into the consideration of industry attributes and emotional connections, making subsequent analysis more closely aligned with the actual relationship between product selection and industry needs.

[0151] In this embodiment, a comprehensive multidimensional industry attribute-product selection object adaptation matrix is ​​calculated for each product selection object based on the multidimensional industry attribute-product selection object adaptation calculation matrix and the multidimensional industry attribute-product selection object adaptation matrix. These two matrices reflect the relationship between product selection objects and industry attributes from different perspectives. By adding them together, a more comprehensive matrix reflecting the adaptation of product selection objects to multidimensional industry attributes can be obtained.

[0152] In this embodiment, the responsiveness of each selected item in each welfare selection scheme to the multi-dimensional industry attributes of the served enterprise is obtained based on the multi-dimensional industry attribute-selected item adaptation comprehensive matrix analysis. The element values ​​in the adaptation comprehensive matrix reflect the degree of correlation between the selected item and different industry attributes under comprehensive consideration. By analyzing these element values, the responsiveness of the selected item to the multi-dimensional industry attributes of the served enterprise can be obtained. For example, the element values ​​representing different industry attributes in the matrix can be summarized and statistically analyzed, such as summing or averaging. If the average value of the element values ​​related to the technology industry in the adaptation comprehensive matrix of a certain selected item is high, it indicates that the selected item has a high responsiveness to the technology industry attributes, which means that the selected item has a good fit in meeting the welfare needs of technology industry enterprises; conversely, if the average value is low, the responsiveness is low, and it may not meet the welfare selection needs of the industry. In this way, the matching degree between each selected item and the multi-dimensional industry attributes of the served enterprise can be accurately evaluated, providing an important basis for optimizing the welfare selection scheme.

[0153] To accurately analyze the relevance of product selection to the age and geographic distribution of employees by acquiring and calculating consumer evaluation scores from different ages and regions, a third relevance analysis unit is proposed, including:

[0154] The third responsiveness analysis subunit is used to obtain the evaluation scores of a large number of consumers of different ages and regions for each selected item in each welfare product selection plan. Based on the age distribution of the corresponding employees, the evaluation scores of consumers of different ages for each selected item in each welfare product selection plan are calculated to obtain the responsiveness of each selected item to the corresponding age distribution of employees.

[0155] The fourth responsiveness analysis subunit is used to calculate the evaluation scores of consumers in different regions for each product selection option in each welfare product selection scheme based on the geographical distribution of the corresponding employees, and to obtain the responsiveness of each product selection option to the geographical distribution of the corresponding employees.

[0156] In this embodiment, the evaluation scores of consumers of different ages for each product in each welfare product selection scheme are calculated based on the corresponding employee age distribution to obtain the responsiveness of each product to the corresponding employee age distribution. Specifically, a large number of evaluation scores from consumers of different ages for each product are first collected. These evaluation scores reflect the degree of liking or satisfaction of different age groups with the product selection. For example, evaluation scores from consumers aged 18-25, 26-35, and 36-45 for a certain brand of sports watches (product selection) are collected. Then, the calculation is performed in conjunction with the employee age distribution of the served company. Assuming that the 26-35 age group has the largest proportion of employees in the company, if consumers in this age group generally give higher evaluation scores for sports watches, the evaluation scores of this age group will have a relatively large weight in the calculation. Through a specific calculation method, such as weighted average (multiplying the evaluation scores of different age groups by the proportion of that age group in the company's employees and then summing them), the final result is the responsiveness of the product selection to the corresponding employee age distribution. The higher the responsiveness, the more the selected product matches the preferences of the company's current age distribution, and the more suitable it is as a welfare product for the company's employees.

[0157] In this embodiment, the evaluation scores of consumers from different regions for each product in each welfare product selection scheme are calculated based on the geographical distribution of the corresponding employees to obtain the responsiveness of each product to the corresponding geographical distribution of employees. First, evaluation scores of consumers from different regions for each product are collected extensively. These scores reflect the acceptance or preference of different regional groups for the product. For example, evaluation scores of a warm blanket (product) are collected from consumers from different regions such as the North, South, East, and West. Next, the calculation is performed based on the geographical distribution of the employees of the served company. If the company has a large number of employees from the South, and consumers from the South give higher evaluation scores to the warm blanket, the evaluation scores from the South will have a relatively higher weight in the calculation process. Similarly, a weighted average calculation method is used (multiplying the evaluation scores from different regions by the proportion of that region among the company's employees, and then summing them up). The resulting value is the responsiveness of the product to the corresponding geographical distribution of employees. The higher the responsiveness, the better the selected product can meet the needs of employees in the current geographical distribution of the company, and the more likely it is to be welcomed by employees when selected as employee benefits.

[0158] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A welfare program product selection device based on a large model and a supply chain, characterized by, include: The demand collection module is used to collect welfare demand data from the service recipients and supply chain data from the service providers. The model selection module is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions. The fit assessment module is used to analyze the fit of each welfare selection scheme with welfare needs based on the multi-layered sentiment keyword set of welfare purpose tags in the welfare needs data of the service recipient company, the multi-dimensional industry attributes of the service recipient company, the age distribution of employees, the geographical distribution of employees, and each selection object in each welfare selection scheme. The optimization evaluation module is used to evaluate the optimization of each welfare product selection scheme based on the welfare demand fit and total gross profit of each welfare product selection scheme. The solution selection module is used to select the welfare product selection solution with the highest priority among all welfare product selection solutions as the final welfare product selection solution; The compatibility assessment module includes: The responsiveness analysis submodule is used to analyze the responsiveness of each selected item in each welfare selection plan to the multi-layered sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise, the multi-dimensional industry attributes of the service enterprise, and the corresponding employee age distribution and employee geographic distribution. The fit analysis submodule is used to analyze the fit of each welfare selection scheme with the welfare needs based on the multi-layer sentiment keyword set of the welfare purpose tag in the welfare needs data of the service enterprise, the multi-dimensional industry attributes of the service enterprise, and the corresponding employee age distribution and employee geographic distribution of each selection object in each welfare selection scheme. The responsiveness analysis submodule includes: The graph and vector generation unit is used to generate a sentiment graph of the welfare needs of the serviced enterprise based on the correlation strength between word vectors of two keywords in a multi-layer sentiment keyword set of welfare purpose tags in the welfare needs data of the serviced enterprise. It encodes the descriptive text of each selection object in each welfare selection plan into a vector to obtain the descriptive vector of each selection object. The first responsiveness analysis unit is used to calculate the responsiveness of each selected object in each welfare selection scheme to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service enterprise, based on the description vector of each selected object and the welfare demand sentiment map of the service enterprise. The industry attribute matrix building unit is used to generate a multi-dimensional industry attribute-sentiment association matrix based on the association strength of each keyword in the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service enterprise. It also generates a multi-dimensional industry attribute-selection object adaptation matrix based on the fit between the multi-dimensional industry attributes of the service enterprise and each text description keyword of each selection object in each welfare selection plan. The second responsiveness analysis unit is used to analyze the responsiveness of each product selection object in each welfare product selection scheme to the multidimensional industry attributes of the served enterprise based on the multidimensional industry attribute-sentiment association matrix and the multidimensional industry attribute-product selection object adaptation matrix of each product selection object. The third responsiveness analysis unit is used to evaluate the responsiveness of each selected item in each welfare selection plan to the corresponding employee age distribution and the corresponding employee geographic distribution.

2. The welfare program selection device based on a large model and supply chain as described in claim 1, characterized in that, The welfare demand data includes: the number of people receiving welfare, the budget amount, and the welfare purpose tag.

3. The welfare program selection device based on a large model and supply chain as described in claim 1, characterized in that, The service provider's supply chain data includes: information on the service provider's product inventory and gross profit threshold.

4. The welfare program selection device based on a large model and supply chain as described in claim 1, characterized in that, The model selection module includes: The model building sub-module is used to train a large product selection model by using a large number of historical product selection schemes, the corresponding welfare needs data of the service recipient companies and the corresponding service provider supply chain data as training samples. The model selection submodule is used to input the welfare needs data of the service recipient company and the supply chain data of the service provider into a pre-trained product selection model to obtain multiple welfare product selection solutions.

5. The welfare program selection device based on a large model and supply chain according to claim 1, characterized in that, The first responsiveness analysis unit includes: The first adaptation determination subunit is used to combine at least two keywords in the welfare demand sentiment graph of the served enterprise to obtain a recombined keyword set, mine all the relevant statements in all recombined keyword sets, calculate the similarity between the description vector of each selected object and the statement vector of each relevant statement in each recombined keyword set, and calculate the semantic description adaptation between each selected object and each recombined keyword set based on the similarity between the description vector of each selected object and the statement vector of all relevant statements in each recombined keyword set. The second adaptation determination subunit is used to determine the weight of each reorganized keyword set based on the welfare needs sentiment graph of the served enterprise, and to calculate the relative adaptation of each product selection object and each reorganized keyword set based on the weight of each reorganized keyword set, each product selection object and the semantic description adaptation of each reorganized keyword set. The first relevance analysis subunit is used to analyze the relevance of each product selection object to the multi-layer sentiment keyword set of welfare purpose tags in the welfare demand data of the service recipient enterprise based on the relative fit of all recombined keyword sets of each product selection object and welfare demand sentiment map.

6. The welfare program selection device based on a large model and supply chain according to claim 5, characterized in that, The first responsiveness analysis subunit includes: The decreasing topology path fitting end is used to sort and connect all recombined keyword sets in the welfare demand sentiment graph according to the principle of descending relative fit between each product selection object and the welfare demand sentiment graph of the served enterprise, so as to obtain a decreasing fit topology path. The responsiveness analysis uses the ratio of the length of the decreasing fit topology path to the length of the maximum topology path in the welfare demand sentiment graph as the fit deviation of each selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise, and the difference between 1 and the fit deviation is used as the responsiveness of the corresponding selected object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the served enterprise.

7. The welfare program selection device based on a large model and supply chain according to claim 1, characterized in that, The second responsiveness analysis unit includes: The indirect inference subunit is used to reassign the multi-dimensional industry attribute-emotional association matrix based on the relevance of each product selection object to the multi-layer sentiment keyword set of the welfare purpose tag in the welfare demand data of the service recipient enterprise, and obtain the multi-dimensional industry attribute-product selection object adaptation inference matrix for each product selection object. The second responsiveness analysis subunit is used to calculate the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object based on the multidimensional industry attribute-selection object adaptation estimation matrix and the multidimensional industry attribute-selection object adaptation matrix for each selection object. Based on the multidimensional industry attribute-selection object adaptation comprehensive matrix for each selection object, the responsiveness of each selection object in each welfare product selection plan to the multidimensional industry attributes of the serviced enterprise is analyzed.

8. The welfare program selection device based on a large model and supply chain according to claim 1, characterized in that, The third responsiveness analysis unit includes: The third responsiveness analysis subunit is used to obtain the evaluation scores of a large number of consumers of different ages and regions for each selected item in each welfare product selection plan. Based on the age distribution of the corresponding employees, the evaluation scores of consumers of different ages for each selected item in each welfare product selection plan are calculated to obtain the responsiveness of each selected item to the corresponding age distribution of employees. The fourth responsiveness analysis subunit is used to calculate the evaluation scores of consumers in different regions for each product selection option in each welfare product selection scheme based on the geographical distribution of the corresponding employees, and to obtain the responsiveness of each product selection option to the geographical distribution of the corresponding employees.

Images