Multi-level supply chain industrial and commercial collaborative distribution method and system

Abstract

The invention discloses a multi-level supply chain industrial and commercial collaborative distribution method, which comprises the steps of constructing an industrial and commercial collaborative cigarette delivery recommendation system framework based on consumer demands, carrying out multi-source comment data acquisition and preprocessing, carrying out consumer demand feature deep analysis based on preprocessed data, and carrying out cigarette delivery recommendation. Clustering and product similarity calculation are carried out on the demands obtained through analysis, and the standard putting proportion is adjusted according to the clustered demands and the calculated product similarity. According to the invention, through multi-source data acquisition and deep analysis, accurate capture of consumer demand characteristics, and combination of clustering analysis and product similarity calculation, dynamic intelligent adjustment of a product specification delivery proportion is realized, the precision and flexibility of industrial and commercial collaborative distribution are effectively improved, the problem of supply-demand imbalance is reduced, supply chain resource allocation is optimized, the market response capability is enhanced, and the market competitiveness is improved. And a scientific method support is provided for efficient collaboration of the multi-level supply chain.

Description

Technical Field

[0001] The present invention relates to the field of tobacco sales, and more particularly to a multi-level supply chain industry and commerce collaborative distribution method and system. Background Art

[0002] Currently, the main cigarette delivery models in the Chinese tobacco industry include delivery by grade, delivery by price range, delivery by grade + price range, point selection delivery, "quantitative label", "customer demand", etc. These traditional delivery methods are all based on the in-depth mining of tobacco industry data, mainly relying on the operating capabilities and historical purchase data of tobacco retailers. The collaborative decision-making between industrial enterprises and commercial enterprises also long relied on static delivery strategies driven by historical sales data. This model forms a regional delivery plan by statistically calculating indicators such as the monthly sales volume and inventory turnover rate of each product specification for each retailer. However, with the increasing diversification and personalization of the consumer market, this method has shown significant drawbacks: on the one hand, sales data only reflects the results of "completed transactions" and it is difficult to capture the dynamic changes in consumers' implicit demands such as price sensitivity and taste preferences; on the other hand, the micro-demand differences in different regional markets are masked by the extensive administrative level division, resulting in the unmet differentiated ordering demands of convenience stores, catering terminals, and entertainment venues within the same county-level unit. In response to the above demands, the academic and industrial circles have carried out a number of explorations. At the demand perception level, Patent CN113807092A proposes a method for analyzing online reviews of cigarette brands based on the LDA topic model. By constructing a topic dictionary containing 150 keywords such as "tar content" and "throat irritation", topic modeling is performed on user evaluations on e-commerce platforms, realizing the extraction of demand characteristics in dimensions such as price sensitivity and packaging satisfaction. In the field of market segmentation, Patent CN107103488B designs a hybrid judgment model that combines collaborative filtering and K-means clustering. Its innovation lies in introducing user portrait features, converting discrete variables such as age, occupation, and income level into coordinate points in a continuous vector space, and identifying similar consumer groups by calculating the Euclidean distance. In addition, the article "Research on the Precise Delivery Model of High-end and High-price Cigarettes Based on Business Districts - Taking Yulin City as an Example" attempts to introduce GIS (Geographic Information System) technology, incorporating 12 indicators such as the resident population density, per capita GDP, and the number of competing product stores within a 3-kilometer radius around the retailer into the business district evaluation system, delimiting 6 customer grades through on-site research, and formulating differentiated limited delivery strategies accordingly.

[0003] However, the above existing cigarette delivery technologies have significant limitations in terms of market demand response and precision decision-making. First of all, traditional static delivery strategies based on grades, price ranges, etc. overly rely on explicit indicators such as retail customers' historical sales data and inventory turnover rates. In essence, it is a passive regulation mechanism based on "post hoc attribution" and is difficult to effectively capture the dynamic evolution characteristics of consumers' implicit demands. For example, key consumption preferences such as price sensitivity and throat irritation can only be indirectly inferred through transaction data, resulting in the long-term masking of the differentiated demands of heterogeneous scenarios such as convenience stores and catering terminals within the same administrative region by administrative level divisions, forming a "supply-determined demand" extensive management model.

[0004] The demand perception technology has the defect of single-dimensionality: Although the review analysis based on the LDA topic model can extract explicit topics such as price and packaging, it lacks quantitative modeling of professional dimensions such as taste intensity and tar suitability, and does not consider the impact of the review time decay factor on demand trend prediction; while the collaborative filtering model integrating user portraits is limited by the precision bottleneck of the conversion of discrete variables into continuous variables and is prone to semantic drift phenomena in the representation learning of qualitative characteristics such as occupation and income.

[0005] Furthermore, the market segmentation method has not broken through the spatial granularity limit. Although the business district evaluation system driven by GIS technology introduces geographical information parameters such as population density and competitor distribution, it does not establish the mapping relationship between the consumer demand feature vector and the spatial coordinates, resulting in the difficulty of accurately matching the 6-category customer grade division with the dynamic demand spectrum of terminal types such as convenience stores and entertainment venues.

[0006] Finally, there are methodological defects in the product substitution relationship modeling: The current similarity calculation based on Euclidean distance or cosine similarity mainly focuses on single attributes such as price and tar content, and does not construct a product feature vector space including multi-modal features such as flavor characteristics and packaging design, resulting in a significant lag in the substitute recommendation strategy when dealing with sensory demand changes such as "spicy throat" and "long-lasting sweetness". These technical bottlenecks together have led to long-term structural contradictions in the industry such as supply-demand mismatch and low inventory turnover efficiency. Summary of the Invention

[0007] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a multi-level supply chain industry and commerce collaborative distribution method and system for solving key technical problems such as insufficient supply-demand matching accuracy, lag in demand dynamic response, deviation in product substitution relationship modeling, and extensive regional market segmentation in the cigarette business field.

[0008] To achieve the above purpose, the present invention provides the following technical solutions: A multi-level supply chain industry and commerce collaborative distribution method, characterized in that it includes the following steps: Step 1, construct a framework of an industry and commerce collaborative cigarette delivery recommendation system based on consumer demand; Step 2: Collect and preprocess multi-source review data; Step 3: Deeply analyze the consumer demand characteristics based on the data preprocessed in Step 2; Step 4: Cluster the demands obtained by the analysis in Step 3 and calculate the product similarity; Step 5: Adjust the product specification delivery ratio according to the demands completed by clustering and the calculated product similarity in Step 4.

[0009] As a further improvement of the present invention, the specific steps of collecting and preprocessing multi-source review data in Step 2 are as follows: Step 2-1: Obtain the cigarette product review data in the research area and construct the original review corpus , where the corpus includes cigarette product names, user IDs, product ratings, and review texts; Step 2-2: Compare the cigarette product names in the corpus with the product specifications that can be sold by retailers in this area, and remove the product review data that cannot be sold in this area; Step 2-3: Perform denoising processing on the review texts in the corpus, including filtering advertising texts, deleting duplicate reviews, removing non-target brand-related entries, and general praises, emotional outbursts or irrelevant reviews, to obtain an effective review set ; Step 2-4: Use a Chinese word segmentation toolkit to perform lexical analysis on , and combine it with the tobacco industry term dictionary to correct the word segmentation results to generate a standardized review summary set ; Step 2-5: Perform entity recognition on to construct a user-review association table , where the association table includes user ID, review time, review summary, and reviewed product.

[0010] As a further improvement of the present invention, the specific steps of deeply analyzing the consumer demand characteristics in Step 3 are as follows: Step 3-1: Load the fine-tuned BERT model dedicated to the tobacco field, and perform fine-grained sentiment analysis on each comment in , and output a four-dimensional feature vector: Among them, PS represents price sensitivity, TS represents taste intensity, DS represents packaging satisfaction, and JS represents tar suitability; Step 3-2: Apply the attention mechanism for weighted aggregation to to calculate the demand feature vector of a single user : : Among them, Indicates the time decay factor of the th comment, which is calculated based on two dimensions of semantic importance and time decay, through a learnable parameter matrix and bias to obtain semantic importance, and map the comment feature vector to the attention score space: where the tanh function is used for non-linear activation to make the score distribution more concentrated; Introduce a time decay coefficient to calculate the time decay factor, because recent comments are more valuable for reference than historical comments: Step Three: Statistically calculate the average demand feature vector of each cigarette specification : where represents the user number who comments on the cigarette specification .

[0011] As a further improvement of the present invention, the specific steps of clustering the demand and calculating the product similarity in the fourth step are as follows: Step Four One: Output the regional clustering result , and each cluster contains the regional plan and the demand feature distribution.

[0012] Step Four Two: Extract the list of retail customer sales specifications within the regional plan of this clustering. For each specification that the retail customer can sell, quantify the degree of meeting each demand according to its characteristics, and obtain , where the higher it is, the more the demand k is met.

[0013] Step Four Three: Calculate the comprehensive demand score of each product to measure the overall matching degree of the product to the consumer demand: where n represents the number of consumer demands, represents the weight of the th demand, which is obtained from the ratio of this demand to the total number of demands.

[0014] Step Four Four: Construct a product feature vector space: Use One-Hot encoding to process discrete variables; Step Four Five: Calculate the similarity between the specifications sold by each retail customer : Among them, is the price weight factor, indicating the cosine similarity between tar and fragrance type.

[0015] As a further improvement of the present invention, the specific steps for adjusting the product specification delivery ratio in step five are as follows: Step five one: Based on the similar products of each product specification, calculate the overall competitiveness between the similar products and this product specification; Balance its own needs and the influence of similar products: Among them , The larger it is, the more attention is paid to its own attributes; Step five two: Under ideal conditions, calculate the delivery ratio completely based on consumer demand matching: Step five three: Integrate historical data, control the adjustment intensity, and update the delivery ratio: Among them, is the current delivery ratio of each product specification; Step five four: Set the minimum delivery ratio to avoid out-of-stock of this product: Set the minimum delivery ratio And perform normalization processing: Step five five: Push the delivery adjustment suggestion to the enterprise, and the interface protocol field definition is as follows: { "Suggestion number": "RECOM_2024_Q1", "Adjustment suggestion": {"Nanjing Yuhuashi": -15%, "Nanjing Xuanhemen": +20%}, "Valid time": "2024-XX-XX" }

[0016] On the other hand, the present invention provides a system, including: Data acquisition module: used to capture multi-source heterogeneous review data and perform preprocessing; Demand feature mining module: using a deep learning model to extract multi-dimensional demand features of consumers; Demand clustering and similar product calculation module: Integrate demand characteristics and product similarity calculation to achieve micro-market segmentation; Product specification delivery ratio adjustment module: Generate substitute combination suggestions based on product attributes and demand matching.

[0017] Compared with the solutions in the prior art, the present invention has the following beneficial effects: 1. Dynamic demand perception: Compared with the limitation of only extracting explicit topics such as price and packaging through the LDA topic model in CN113807092A, the present invention innovatively uses a fine-tuned BERT model dedicated to the tobacco field, combines the attention mechanism and the time decay factor, and realizes the fine-grained quantitative analysis of the four-dimensional demand characteristics of price sensitivity, taste intensity, packaging satisfaction, and tar suitability.

[0018] 2. Multi-modal product similarity modeling: Breaking through the defect of the collaborative filtering model based only on discrete variables of user portraits in CN107103488B, the present invention constructs a hybrid similarity calculation model including price weights and cosine similarity, and forms a multi-dimensional feature vector space covering the core attributes of cigarettes through encoding discrete attributes, improving the accuracy of product substitution relationship recognition, especially showing stronger relevance when dealing with sensory demand changes such as "pungent throat" and "long-lasting aftertaste".

[0019] 3. Demand-driven matching mechanism: The present invention outputs clusters with differentiated demand characteristic distributions through the demand clustering algorithm, and generates a product-demand matching degree evaluation matrix in combination with the list of product specifications sold by retailers, realizing the refinement of spatial granularity from administrative level division to demand feature mapping.

[0020] 4. Dynamic delivery strategy: Aiming at the supply-demand mismatch problem in the traditional delivery mode according to grades, the present invention proposes a balance model that integrates its own demand score and the competitiveness of similar products, and designs a progressive adjustment algorithm including historical data fusion coefficient, minimum delivery constraint, and normalization processing, which not only avoids market fluctuations caused by radical adjustments but also ensures the supply-demand matching degree. Description of the Drawings

[0021] Figure 1 It is a schematic flowchart of the multi-level supply chain industrial and commercial collaborative distribution method of the present invention. Detailed Embodiment

[0022] The following will further elaborate on the present invention in combination with the embodiments given in the drawings.

[0023] Refer to Figure 1 As shown, a multi-level supply chain industrial and commercial collaborative distribution method of this embodiment mainly includes the following steps.

[0024] S1: Construct a framework for an industrial and commercial collaborative cigarette delivery recommendation system based on consumer demand The system includes the following functional modules: Data acquisition module: used to capture multi-source heterogeneous review data and perform preprocessing; Requirement feature mining module: adopts a deep learning model to extract multi-dimensional consumer requirement features; Requirement clustering and similar product calculation module: fuses requirement features and product similarity calculation to achieve micro-market segmentation; Product specification delivery ratio adjustment module: generates substitute combination suggestions based on the matching of product attributes and requirements.

[0025] S2. Multi-source review data acquisition and preprocessing S2.1. Obtain cigarette product review data in the research area and construct an original review corpus , the corpus includes cigarette product names, user IDs, product ratings, and review texts; S2.2. Compare the cigarette product names in the corpus with the product specifications that can be sold by retailers in this area, and remove the product review data that cannot be sold in this area; S2.3. Perform denoising processing on the review texts in the corpus, including filtering advertising texts, deleting duplicate reviews, removing non-target brand-related entries, and general praise, emotional outbursts or irrelevant reviews, to obtain an effective review set ; S2.4. Use a Chinese word segmentation toolkit to perform lexical analysis, and combine it with a tobacco industry term dictionary (including professional descriptions such as "pungent throat" and "long-lasting aftertaste") to correct the word segmentation results and generate a standardized review summary set ; S2.5. Perform entity recognition on to construct a user-review association table , the association table includes user ID, review time, review summary, and reviewed product.

[0026] S3. In-depth analysis of consumer requirement features S3.1. Load the fine-tuned BERT model dedicated to the tobacco field, and perform fine-grained sentiment analysis on each comment in , and output a four-dimensional feature vector: Among them, PS represents price sensitivity, TS represents taste intensity, DS represents packaging satisfaction, and JS represents tar suitability.

[0027] S3.2. Apply the attention mechanism for weighted aggregation to calculate the requirement feature vector of a single user : : Among them represents the time decay factor of the th comment, which is calculated based on two dimensions: semantic importance and time decay. Through a learnable parameter matrix and bias obtain semantic importance and map the comment feature vector to the attention score space: where the tanh function is used for non-linear activation to make the score distribution more concentrated.

[0028] Introduce the time decay coefficient to calculate the time decay factor, because recent comments are more valuable as references than historical comments: S3.3. Statistically calculate the average demand feature vector of each cigarette specification : Among them, represents the user number who comments on the cigarette specification .

[0029] S4. Demand clustering and product similarity calculation S4.1. Output the regional clustering results , and each cluster contains the regional plan and the demand feature distribution.

[0030] S4.2. Extract the list of retail customer sales specifications within the regional plan of this clustering. For each specification that the retail customer can sell , quantify the degree of meeting each demand according to its characteristics, and obtain , where the higher it is, the more the demand k is met.

[0031] S4.3. Calculate the comprehensive demand score of each product to measure the overall matching degree of the product to the consumer demand: Among them, n represents the number of consumer demands, represents the weight of the th demand, which is obtained from the ratio of this demand to the total number of demands.

[0032] S4.4. Construct the product feature vector space: Use One-Hot encoding to process discrete variables; S4.5. Calculate the similarity between the product specifications sold by each retailer : where is the price weight factor, represents the cosine similarity between tar and fragrance type.

[0033] S5. Product specification delivery ratio adjustment S5.1. Based on the similar products of each product specification, calculate the overall competitiveness between the similar products and this product specification.

[0034] Balance its own needs and the influence of similar products: where , The larger it is, the more it emphasizes its own attributes.

[0035] S5.2. Under ideal conditions, calculate the delivery ratio completely based on consumer demand matching: S5.3. Integrate historical data, control the adjustment strength, and update the delivery ratio: where, is the current delivery ratio of each product specification.

[0036] S5.4. Set the minimum delivery ratio to avoid out-of-stock of this product: Set the minimum delivery ratio And perform normalization processing: S5.5. Push the delivery adjustment suggestions to the enterprise, and the interface protocol fields are defined as follows: { "Suggestion number": "RECOM_2024_Q1", "Adjustment suggestion": {"Nanjing Yuhuashi": -15%, "Nanjing Xuanhemen": +20%}, "Valid time": "2024-04-01" }

[0037] In summary, the multi-level supply chain industrial and commercial collaborative distribution method of this embodiment includes: 1. Preprocessing method for comment data: By constructing a tobacco industry term dictionary to correct the Chinese word segmentation results, combining entity recognition technology to generate a standardized comment summary set, and establishing a user-comment association table, accurate cleaning and semantic enhancement of unstructured comment data are achieved.

[0038] 2. Demand feature extraction based on the fine-tuned BERT model: Use a BERT model dedicated to the tobacco field to perform fine-grained sentiment analysis on comments, output a four-dimensional feature vector of price sensitivity, taste intensity, packaging satisfaction, and tar suitability, and fuse the time decay factor and semantic importance through an attention mechanism to construct a dynamic demand feature vector generation method.

[0039] 3. Construction of a multi-modal product feature vector space and calculation of hybrid similarity: Comprehensively use encoding to process discrete attributes, combine price weights and cosine similarity metrics to construct a multi-dimensional feature vector space covering the core attributes of cigarettes, and solve the problem of insufficient fusion of multi-modal information such as flavor characteristics and packaging design in traditional similarity calculations.

[0040] 4. Demand-driven dynamic clustering and matching degree evaluation: Through a demand clustering algorithm, combine the list of product specifications sold by retailers to generate a product-demand matching degree evaluation matrix, breaking through the defect of rough granularity of the demand feature space caused by traditional administrative level division.

[0041] 5. Dynamic adjustment of delivery ratio: Based on a balance model of the comprehensive demand score of products and the overall competitiveness of similar products, design a progressive adjustment algorithm including historical data fusion, minimum delivery constraint, and normalization processing to achieve the coordinated optimization of supply-demand matching degree and market stability.

[0042] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A multi-level supply chain industrial and commercial collaborative goods distribution method, characterized in that: It includes the following steps: Step 1: Construct a collaborative industrial and commercial cigarette delivery recommendation system framework based on consumer needs; Step 2: Collect and preprocess multi-source review data; Step 3: Deeply analyze consumer demand characteristics based on the data preprocessed in Step 2; Step 4: Cluster the demands obtained by the analysis in Step 3 and calculate the product similarity; Step 5: Adjust the product specification delivery ratio according to the demands completed by clustering and the calculated product similarity in Step 4.

2. The multi-level supply chain industrial and commercial collaborative goods distribution method according to claim 1, wherein: The specific steps of collecting and preprocessing multi-source review data in Step 2 are as follows: Step 2-1: Obtain the cigarette product review data of the research area and construct the original review corpus , where the corpus includes cigarette product names, user IDs, product ratings, and review texts; Step 2.2: Compare the cigarette product names in the corpus with the product specifications that can be sold by local retailers, and remove the product review data that cannot be sold in this region; Step 2-3: Perform denoising processing on the review texts in the corpus, including filtering out advertising texts, deleting duplicate comments, removing entries unrelated to the target brand, as well as general praises, emotional outbursts or irrelevant comments, to obtain a set of valid comments ; Step 24: Use a Chinese word segmentation toolkit to perform lexical analysis, correct the word segmentation results by combining with the tobacco industry term dictionary, and generate a standardized comment abstract set ; Step 25. Perform entity recognition on to construct a user-comment association table . The association table includes user ID, comment time, comment summary, and commented product.

3. The multi-level supply chain industrial and commercial collaborative goods distribution method according to claim 1 or 2, characterized in that: The specific steps of deeply analyzing consumer demand characteristics in Step 3 are as follows: Step 3.1: Load the fine-tuned BERT model dedicated to the tobacco field, and perform fine-grained sentiment analysis on each comment in , and output a four-dimensional feature vector, which includes price sensitivity, taste intensity, packaging satisfaction, and tar suitability; Step 3.

2. Apply weighted aggregation using the attention mechanism to calculate the demand feature vector of a single user ; Then, obtain the semantic importance through the learnable parameter matrix and the bias , and map the comment feature vector to the attention score space; ​​ Subsequently, the time decay coefficient is calculated Time decay factor; Step 3: Statistically analyze the average demand eigenvector of each cigarette specification .

4. The multi-level supply chain industrial and commercial collaborative goods distribution method according to claim 1 or 2, characterized in that: The specific steps of clustering the demands and calculating the product similarity in Step 4 are as follows: Step 4.1: Output the regional clustering results, and each cluster includes regional planning and demand characteristic distribution; Step Four Two: Extract the list of retail customer sales product specifications within the regional plan of the cluster. For each product specification that the retail customer can sell , quantify the degree of satisfaction for each requirement according to its characteristics, and obtain , where The higher it is, the more the requirement k is satisfied; Step 43. Calculate the comprehensive demand score for each product to measure the overall matching degree of the product to consumer needs; Step 4.4: Construct a product feature vector space, and then use One-Hot encoding to process discrete variables; Step 4 and 5: Calculate the similarity between the product specifications on sale for each retailer .

5. The multi-level supply chain industrial and commercial collaborative goods distribution method according to claim 1 or 2, characterized in that: The specific steps of adjusting the product specification delivery ratio in Step 5 are as follows: Step 5.1: Calculate the overall competitiveness of similar products and this product specification based on the similar products of each product specification; After that, balance the own needs and the influence of similar products; Step 5.2: Calculate the delivery ratio that is fully matched based on consumer needs under ideal conditions; Step 5.3: Integrate historical data, control the adjustment intensity, and update the delivery ratio; Step 5.4: Set the minimum delivery ratio and perform normalization processing; Step 5.5: Push delivery adjustment suggestions to the enterprise.

6. A system applying the multi-level supply chain industrial and commercial collaborative goods allocation method according to any one of claims 1 to 5, characterized in that: It includes: Data collection module: used to capture multi-source heterogeneous review data and perform preprocessing; Demand feature mining module: uses a deep learning model to extract multi-dimensional consumer demand characteristics; Demand clustering and similar product calculation module: integrates demand characteristics and product similarity calculation to achieve micro-market segmentation; Product specification delivery ratio adjustment module: generates substitute combination suggestions based on the matching of product attributes and demands.

Images