Search

CN-122022886-A - Multi-mode price comparing and full-chain decision making method for cosmetic service integrating multi-source data

CN122022886ACN 122022886 ACN122022886 ACN 122022886ACN-122022886-A

Abstract

The invention provides a multi-mode price comparing and full-chain decision method for a cosmetic service fusing multi-source data, which realizes an end-to-end solution from multi-mode data acquisition to full-chain service optimization by constructing a three-layer architecture of a data layer fusion layer decision layer. The invention integrates price, attribute, market, image and text 5 types of data, improves decision dimension, improves accuracy compared with the traditional method, adopts a multi-mode model of an attention mechanism and LSTM, realizes accurate price prediction by a dynamic decision model, improves inventory turnover rate, breaks through purchasing inventory sales links, shortens decision response time from 72 hours to 2 hours, reduces annual operation cost and optimizes operation efficiency by full-chain cooperation.

Inventors

  • Feng Baodan
  • WANG HAIYONG

Assignees

  • 杭州吉智云信息技术有限公司

Dates

Publication Date
20260512
Application Date
20260121

Claims (10)

  1. 1. A cosmetic service multi-mode price comparison and full chain decision method integrating multi-source data is characterized by comprising the following steps: S1, multi-source heterogeneous data acquisition and preprocessing, namely acquiring multi-mode heterogeneous data including internal business data, external electronic commerce platform data, social media data, supply chain internet of things data, macro economic data and image video data, and performing cleaning, standardization and feature engineering processing; S2, constructing a multi-mode price comparison model and predicting the price, namely constructing a multi-mode fusion prediction model, processing the time sequence price data, the text semantic data and the visual characteristic data which are preprocessed by the S1, and outputting a price prediction sequence of a future appointed period; s3, identifying multi-modal similar products and generating a price comparison report, namely identifying a product set similar to a target product based on the multi-modal characteristics of the products extracted in the S2, calculating multi-dimensional price comparison indexes, and generating a visual price comparison report; S4, supplier evaluation and purchase strategy optimization, namely comprehensively evaluating suppliers based on a multidimensional evaluation system, and generating a dynamic purchase strategy through a multi-objective optimization algorithm based on an evaluation result, price prediction and price comparison report; S5, market trend-driven sales strategy formulation, namely extracting market trend characteristics based on social media and electronic commerce platform data, and dynamically formulating sales strategies comprising pricing, channel combination and resource allocation by combining price comparison reports; S6, full chain inventory optimization decision, namely integrating sales prediction and market trend data, calculating dynamic safety inventory, and performing inventory optimization management based on inventory classification and multi-level inventory cooperation strategies.
  2. 2. The method according to claim 1, wherein in the step S1, the collecting multi-source heterogeneous data specifically includes: The method comprises the steps of obtaining historical transaction and product data by interfacing an internal business system through an API interface; capturing bid price, promotion and evaluation data of an external electronic commerce platform in real time through a distributed crawler cluster; acquiring hot topics, KOL data and user generated content of social media through an open platform API; Collecting raw material prices, production progress and logistics data through edge computing equipment deployed at a provider end; Obtaining a macroscopic index by butting a macroscopic economic database; And collecting product packaging images and user color testing video data.
  3. 3. The method according to claim 1, wherein in the step S2, the multi-modal fusion prediction model constructed adopts a fusion architecture based on a transducer, and the method comprises: the time sequence price prediction sub-network adopts a cyclic neural network with an attention mechanism to process the historical price sequence and outputs a time sequence feature vector; the text semantic understanding sub-network adopts a pre-training semantic model to process product components and user evaluation texts and outputs text feature vectors; The visual characteristic extraction sub-network adopts a visual transducer model to process the product image and outputs an image characteristic vector; And the multi-mode fusion layer fuses the time sequence, the text and the visual feature vector by adopting a cross attention mechanism or a mode of a multi-layer perceptron after feature splicing, and outputs a price prediction result.
  4. 4. The method according to claim 1, wherein in step S3, identifying similar products specifically comprises: Performing preliminary clustering based on product attribute characteristics; In the cluster, calculating cosine similarity of comprehensive feature vectors among products, wherein the comprehensive feature vectors at least comprise price features formed by price prediction sequences, attribute features formed by product attribute codes, market features formed by market indexes, image features extracted by image models and text features extracted by text models; and selecting the products with the similarity higher than the dynamic threshold value or ranked at the top as a similar product set.
  5. 5. The method according to claim 1 or 4, wherein in the step S3, the calculated multidimensional scaling index comprises: price competitiveness index for measuring the price position of the target product in the similar product set; the feature value ratio is used for measuring the weighted feature value corresponding to the unit price of the target product; the market trend score is used for comprehensively reflecting the market trend of the product based on the search quantity, the social media popularity and the sales quantity increase rate; and the risk early warning index is used for evaluating potential risks by integrating price fluctuation, supply chain risks and policy risks.
  6. 6. The method of claim 1, wherein in the step S4, the multidimensional evaluation system for suppliers includes at least price competitiveness, quality stability, delivery capacity, financial health and sustainable development dimensions, and the suppliers are classified based on weighted scores of the dimensions, and the dynamic purchasing strategy is generated by constructing an optimization model targeting minimizing purchasing cost and risk penalty cost and meeting total demand and capacity of the suppliers as constraints, and solving purchasing quantity allocation schemes of the suppliers.
  7. 7. The method of claim 1, wherein in step S5, dynamically formulating the sales policy comprises: determining base pricing based on cost, target gross interest rate, brand premium, and bid price; dynamically adjusting basic pricing according to the real-time demand index and the bid price change rate; different sales promotion pricing schemes are designed aiming at different scenes of new product marketing, inventory cleaning and holiday marketing.
  8. 8. The method of claim 1, wherein in step S6, the formula for calculating the dynamic safety stock is: ss=z x sigma demand x LT, wherein: SS: secure inventory (units: pieces); z is a service level coefficient; Sigma demand: daily required standard deviation (reflecting the degree of fluctuation of demand, unit: parts/day); V LT: lead period (purchase cycle + transit time, units: day).
  9. 9. A computer readable storage medium having stored thereon a computer program, wherein the program when executed by a processor implements the cosmetic service multimodal price comparison and full chain decision method of any of claims 1 to 8 that fuses multisource data.
  10. 10. A cosmetic business multi-modal price ratio and full chain decision system for implementing the method of any one of claims 1 to 8, characterized by adopting a cloud side three-layer architecture, comprising: the cloud computing layer comprises a server cluster, a hybrid storage system and network equipment, and is used for deploying data preprocessing, model training and reasoning, business decision and visual core services; The edge computing layer is deployed on the core provider side, comprises an edge server and an Internet of things gateway, and is used for collecting and preprocessing supply chain production and logistics data in real time and uploading the data to the cloud end through the high-speed communication module; The terminal equipment layer comprises a data acquisition terminal deployed on the Internet, a user terminal for man-machine interaction and intelligent perception hardware deployed on a warehouse, and is used for acquiring merchant and social media data, providing a decision interaction interface and real-time inventory data.

Description

Multi-mode price comparing and full-chain decision making method for cosmetic service integrating multi-source data Technical Field The invention relates to the technical field of intelligent decision-making of cosmetic service, in particular to a multi-mode price-comparing and full-chain decision-making method of the cosmetic service integrating multi-source data, a multi-mode price-comparing and full-chain decision-making system of the cosmetic service integrating multi-source data, electronic equipment and a computer readable storage medium. Background The cosmetic industry as a core plate of the global quick-wear product market has broken through $5000 million in 2025, with the chinese market accounting for 35%. Along with the rising of the Z generation consumer groups (accounting for more than 45 percent), the electronic commercialization of social media (shaking sound/rapid hand makeup GMV annual acceleration of 120 percent) and the maturation of a cross-border supply chain, the industry presents new characteristics of 'demand individuation, channel diversification and competitive refinement'. Three major core pain points exist in the current cosmetic enterprise decision-making system: (1) The data islanding effect is remarkable, pricing decisions depend on ERP system historical data (coverage rate is less than 30%), purchasing decisions depend on static quotations of suppliers (update period is 72 hours), inventory management depends on experience values (safety inventory deviation rate reaches 28%), and full-chain data splitting is caused. (2) The multi-mode information fusion is missing, namely the existing system only processes the structured data (the occupied ratio is 65%), ignores multi-mode information such as product component maps (unstructured text), user UGC content (images/videos), supply chain Internet of things data (time sequence signals) and the like, and results in decision dimension one-sided. (3) The average 48-hour response delay exists in the link of pricing, purchasing and inventory sales, and the life cycle characteristics of the cosmetic product in 3090 days cannot be dealt with, so that the common problems of industries of 15% of the explosive and the backlog of the diapause product inventory are caused. According to De-service digital white paper in 2025 cosmetic industry, the average profit margin of enterprises adopting the traditional decision mode is 12.3 percent lower than that of industrial standard poles, wherein the decision error caused by low data processing efficiency accounts for 63 percent. Therefore, the construction of an intelligent decision system for fusing multi-source heterogeneous data and realizing full-chain dynamic optimization becomes a key of the breaking office of the cosmetic enterprise. Disclosure of Invention In order to solve the technical problems in the prior art, the invention provides the following technical scheme: On the one hand, a multi-mode price comparing and full-chain decision method for the cosmetic service integrating multi-source data is provided, which comprises the following steps: S1, multi-source heterogeneous data acquisition and preprocessing, namely acquiring multi-mode heterogeneous data including internal business data, external electronic commerce platform data, social media data, supply chain internet of things data, macro economic data and image video data, and performing cleaning, standardization and feature engineering processing; S2, constructing a multi-mode price comparison model and predicting the price, namely constructing a multi-mode fusion prediction model, processing the time sequence price data, the text semantic data and the visual characteristic data which are preprocessed by the S1, and outputting a price prediction sequence of a future appointed period; s3, identifying multi-modal similar products and generating a price comparison report, namely identifying a product set similar to a target product based on the multi-modal characteristics of the products extracted in the S2, calculating multi-dimensional price comparison indexes, and generating a visual price comparison report; S4, supplier evaluation and purchase strategy optimization, namely comprehensively evaluating suppliers based on a multidimensional evaluation system, and generating a dynamic purchase strategy through a multi-objective optimization algorithm based on an evaluation result, price prediction and price comparison report; S5, market trend-driven sales strategy formulation, namely extracting market trend characteristics based on social media and electronic commerce platform data, and dynamically formulating sales strategies comprising pricing, channel combination and resource allocation by combining price comparison reports; S6, full chain inventory optimization decision, namely integrating sales prediction and market trend data, calculating dynamic safety inventory, and performing inventory optimization management based on inventory