Machine learning-based store grouping and grading method and system

Abstract

The invention discloses a store grouping and grading method and system based on machine learning. The method comprises the following steps of: 1, acquiring required basic related data in an information system of an enterprise; 2, calculating a store city feature vector according to the commodity information in the store, the store commodity sales volume and the store order information; 3, based on the feature vectors of the cities where the stores are located, urban agglomeration division is carried out through a clustering algorithm; 4, calculating store characteristics according to the store information and the sales volume information; step 5, based on the store feature vectors, performing store grade division through a clustering algorithm; and step 6, outputting urban groups and store grade division corresponding to the stores. By collecting and analyzing the sales data, the product information data and other related information of the stores, the performance and potential of the stores can be accurately evaluated, so that more accurate city grouping and store grading are obtained. Meanwhile, dynamic grouping and grading can be achieved, market changes are adapted, and the operation efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and specifically relates to a store grouping and grading method and system based on machine learning, and in particular to a grouping and grading method for offline stores of a home furnishing enterprise. Background Art

[0002] With the booming market economy and increasingly fierce competition, offline store grading is of great significance to home furnishing companies in terms of optimizing resource allocation, improving market competitiveness, meeting consumer needs, promoting store development, and supporting decision-making. This is directly related to the company's operational efficiency, market competitiveness, and long-term development. With the development of information technology and software and hardware, relying on big data to perform more scientific grouping and grading of stores has become a future development trend.

[0003] At present, offline stores are mainly graded according to store sales performance, geographical location, store size and other dimensions, and the subjective experience of managers plays an important role. However, due to problems in early data and technology accumulation, less consideration is given to the use of big data analysis. Therefore, the present invention proposes a more general data governance and store grouping and grading method based on machine learning algorithms for the problem of offline store grouping and grading based on big data, which provides a basis for enterprises to formulate more scientific and reasonable business strategies and strategies. Summary of the invention

[0004] The purpose of the present invention is to propose and develop a data governance and store grouping and grading method based on machine learning algorithms to address the shortcomings of existing offline store grading methods for home furnishing companies.

[0005] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0006] The scenario that the present invention can support is: offline store classification of a home furnishing company that has multiple stores in multiple regions, and the company has collected basic information, products, and sales data of the stores or has the ability to collect these data.

[0007] Based on the supported scenarios, the implementation of the present invention is as follows:

[0008] Step 1: Obtain the required basic relevant data in the enterprise's information system.

[0009] Step 2: Calculate the store city feature vector based on the store product information, store product sales, and store order information.

[0010] Step 3: Based on the city feature vectors where the stores are located, city clusters are divided using a clustering algorithm.

[0011] Step 4: Calculate store characteristics based on store information and sales information.

[0012] Step 5: Based on the store feature vectors, store levels are divided using a clustering algorithm.

[0013] Step 6: Output the city clusters and store level classifications corresponding to the stores.

[0014] Furthermore, the acquisition of the basic relevant data described in step 1 includes: store size, in-store product information, store product sales, and store order information.

[0015] Furthermore, step 2 is specifically implemented as follows:

[0016] A method for calculating city feature vectors based on product specifications and sales information is proposed. First, the product specifications of the top 80% sofas and beds in all stores in the city are counted, and the city feature vectors are formed after multihot encoding. Finally, the feature vectors of all cities are obtained.

[0017] Furthermore, the commodity specification information is mainly used to describe the size of the commodity and is divided into M categories.

[0018] Furthermore, step 3 divides the city clusters through a clustering algorithm based on the city feature vectors where the stores are located. The specific implementation is as follows: first, initialize the number of clusters and cluster centroids, and then classify the cities into corresponding classes by calculating the cosine similarity between the feature vectors of each city and the cluster centroids; for each class, calculate the average value of all members of the class as the new cluster centroid; repeat the steps until the position of the cluster centroid no longer changes; finally, obtain the divided city clusters through the clustering algorithm.

[0019] Furthermore, the calculation of store features based on store information and sales information as described in step 4 is specifically implemented as follows: the store size, store sales, types of store sales, and the number of sales of goods in each price range of the store are multihot encoded to form a store feature vector, and finally all store feature vectors are obtained.

[0020] Furthermore, the sales volume of goods in different price ranges of stores is specifically divided as follows: the sales proportion of goods with different price positioning in the corresponding stores is counted; according to the 80 / 20 principle, the number of all specifications of goods in the country is first counted, denoted as N, which is regarded as the dimension size of the multihot vector; secondly, the sales volume generated by goods of different specifications in each city is counted, and the products of corresponding specifications are taken from the product specification with the highest sales volume in turn, and the component value of the corresponding position of the vector is set to 1, until the top 80% of the sales volume of the products is obtained, and finally the values ​​of the remaining components of the vector are set to 0 to obtain the corresponding store feature vector.

[0021] Furthermore, the present invention also provides a data governance and store grouping and grading system based on machine learning algorithms, which includes four modules: a data reading module, a city grouping module, a store grading module, and a log module.

[0022] The data reading module will obtain all the data required by the clustering and grading algorithm from the enterprise's information system by executing SQL statements;

[0023] The city clustering module calculates the characteristic vector of a city by counting the sales volume information and product specification information of all stores in the city, and divides all cities into K different city clusters through a clustering algorithm.

[0024] The store grading module calculates the store feature vector by counting the store size, product information and product sales information in the urban agglomeration, and divides all stores into N different store grades through a clustering algorithm.

[0025] The log module writes all calculation results into the log file of the corresponding date. Users can view it by themselves, and can display the execution information such as the executed SQL, the city grouping of the store and the store classification results.

[0026] Beneficial effects of the present invention:

[0027] The present invention proposes a store clustering and grading method based on machine learning through big data analysis and machine learning algorithms. Aiming at the city clustering and store grading problems of offline stores of home furnishing enterprises, the present invention realizes automatic city clustering and store grading through statistics and calculation of information such as cities, stores, products, and sales volume, which is beneficial to the enterprise's optimal resource allocation, accurate product delivery, and improved customer satisfaction.

[0028] Advantages of the present invention: By collecting and analyzing the sales data, product information data and other relevant information of the store, the present invention can accurately evaluate the performance and potential of the store, thereby obtaining more accurate city clustering and store classification. At the same time, the present invention can accurately capture the characteristic changes of stores and cities, realize dynamic clustering and classification, adapt to market changes, and improve operational efficiency. From the implementation level, this method is implemented through Python development, and the computational complexity of the model is not high. On the premise that the enterprise provides relevant data, it does not require excessive computing resources to perform calculations, and has strong feasibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a distribution information diagram of commodity specifications in different urban agglomerations of the present invention.

[0030] Figure 2 It is a specific flow chart of the present invention for realizing automatic grouping and grading of stores.

[0031] Figure 3 This is an example of the present invention realizing automatic grouping and grading of stores. DETAILED DESCRIPTION

[0032] The present invention will be described in detail below with reference to the accompanying drawings, and the purpose and results of the present invention will become more apparent. Figure 1 shown.

[0033] like Figure 2 and 3 As shown, a data governance and store clustering and grading method based on machine learning algorithm.

[0034] The technical solution adopted by the present invention to solve the technical problem is as follows:

[0035] The scenario that the present invention can support is: offline store classification of a home furnishing company that has multiple stores in multiple regions, and the company has collected basic information, products, and sales data of the stores or has the ability to collect these data.

[0036] Based on the supported scenarios, the implementation of the present invention is as follows:

[0037] Step 1: Basic data acquisition:

[0038] The basic data required by the present invention includes store size, in-store merchandise information, store merchandise sales, store order information, etc.;

[0039] Step 2: Calculate the store city feature vector based on the store product information, store product sales and store order information;

[0040] Different from the general method of dividing urban agglomerations by geographical location, the present invention obtains the city feature vector of the store through the product information in the store and the sales volume of the store's products.

[0041] For home furnishing companies, dividing urban clusters based on housing type information is a more reasonable way to classify cities. This can alleviate the problem of product unsalability caused by the release of products that do not conform to the main housing types in a certain area. However, since such data is difficult and costly to obtain, we proposed a method to calculate the city feature vector through product specifications and sales information.

[0042] Specifically, the present invention collects the commodity specification information of the top 80% sofas and beds in all stores in the city, uses multihot encoding to form a city feature vector, and finally obtains the feature vectors of all cities. The commodity specification information is mainly used to describe the size of the commodity, which is divided into N categories, and N is preferably 4, such as Figure 1 , showing the distribution information of commodity specifications in different urban agglomerations after urban agglomeration division (taking sofas as an example).

[0043] Step 3: Based on the city feature vectors where the stores are located, city clusters are divided using a clustering algorithm:

[0044] The k-means algorithm is used to divide urban agglomerations through urban feature vectors. The specific implementation is as follows:

[0045] First, the number of clusters and cluster centroids are initialized, and then the city is classified into the corresponding class by calculating the cosine similarity between the feature vector of each city and the cluster centroid; for each class, the average value of all members of the class is calculated as the new cluster centroid; the above steps are repeated until the position of the cluster centroid no longer changes. Finally, the divided urban agglomeration is obtained through the clustering algorithm.

[0046] Step 4: Calculate store characteristics based on store information and sales information:

[0047] The store characteristics are calculated by the store size, store product information and store product sales in the city cluster. Specifically, the store characteristics are calculated by the store size, store sales, types of store sales and the number of store sales in each price range. The above information is encoded using multihot to form a store feature vector, and finally all store feature vectors are obtained;

[0048] The sales volume of goods in each price range of the store is specifically divided into the following: the sales proportion of goods with different price positioning (high-end, mid-range and cost-effective) in the corresponding store is counted, and then according to the 80 / 20 principle, the number of all specifications of goods in the country is first counted, recorded as N, which is regarded as the dimension size of the multihot vector. Secondly, the sales volume generated by goods of different specifications is counted in each city, and the products of corresponding specifications are taken from the product specification with the highest sales volume in turn, and the component value of the corresponding position of the vector is set to 1, until the top 80% of the sales volume of the products is obtained, and finally the values ​​of the remaining components of the vector are set to 0 to obtain the corresponding vector.

[0049] Step 5: Based on the store feature vector, store levels are divided using a clustering algorithm:

[0050] Similar to the method in step 3, the store feature vector and k-means clustering method are also used to obtain a more reasonable store level division;

[0051] Step 6: Output the city clusters and store level classifications corresponding to the stores.

[0052] Output the city grouping and store grading results of all stores and store them in the database.

[0053] The technical architecture of the method for grouping and grading offline stores of home furnishing enterprises includes four modules: a data reading module, a city grouping module, a store grading module, and a log module.

[0054] The data reading module will obtain all the data required by the clustering and grading algorithm from the enterprise's information system by executing SQL statements;

[0055] The city clustering module calculates the characteristic vector of a city by counting the sales volume information and product specification information of all stores in the city, and divides all cities into K different city clusters through a clustering algorithm.

[0056] The store grading module calculates the store feature vector by counting the store size, product information and product sales information in the urban agglomeration, and divides all stores into N different store grades through a clustering algorithm.

[0057] The log module writes all calculation results into the log file of the corresponding date. Users can view it by themselves, and can display the execution information such as the executed SQL, the city grouping of the store and the store classification results.

[0058] The present invention is not limited to the above specific implementations. A person skilled in the art can implement the present invention by using other various specific implementations based on the content disclosed in the present invention. Therefore, any design that adopts the design structure and ideas of the present invention and makes some simple changes or modifications falls within the protection scope of the present invention.

Claims

1. A store grouping and grading method based on machine learning, characterized in that: The steps include: Step 1: Obtain the required basic relevant data in the enterprise's information system; Step 2: Calculate the store city feature vector based on the store product information, store product sales and store order information; Step 3: Based on the city feature vectors where the stores are located, city clusters are divided using a clustering algorithm; Step 4: Calculate store characteristics based on store information and sales information; Step 5: Based on the store feature vector, store levels are divided through clustering algorithm; Step 6: Output the city clusters and store level classifications corresponding to the stores.

2. The store grouping and grading method based on machine learning according to claim 1 is characterized in that: The scenario supported by this invention is: offline store classification of a home furnishing company with multiple stores in multiple regions, where the company has collected basic information, products, and sales data of the stores or has the ability to collect these data.

3. The method for store grouping and grading based on machine learning according to claim 2, characterized in that: The acquisition of the basic relevant data described in step 1 includes: store size, in-store product information, store product sales, and store order information.

4. The method for store grouping and grading based on machine learning according to claim 3 is characterized in that: Step 2 is implemented as follows: A method for calculating city feature vectors based on product specifications and sales information is proposed. First, the product specifications of the top 80% sofas and beds in all stores in the city are counted, and the city feature vectors are formed after multihot encoding. Finally, the feature vectors of all cities are obtained.

5. The method for store grouping and grading based on machine learning according to claim 4, characterized in that: The commodity specification information is mainly used to describe the size of the commodity and is divided into M categories.

6. The method for store grouping and grading based on machine learning according to claim 4, characterized in that: Step 3 divides the city clusters based on the city feature vectors where the stores are located through a clustering algorithm. The specific implementation is as follows: First, initialize the number of clusters and cluster centroids, and then classify the cities into corresponding classes by calculating the cosine similarity between the feature vector of each city and the cluster centroid; for each class, calculate the average value of all members of the class as the new cluster centroid; repeat the steps until the position of the cluster centroid no longer changes; finally, obtain the divided urban agglomeration through the clustering algorithm.

7. A store grouping and grading method based on machine learning according to claim 4 or 6, characterized in that: The calculation of store characteristics based on store information and sales information described in step 4 is specifically implemented as follows: The store size, store sales, types of store sales, and sales quantity of goods in each price range of the store are encoded using multihot to form a store feature vector, and finally the feature vectors of all stores are obtained.

8. The method for store grouping and grading based on machine learning according to claim 7, characterized in that: The sales volume of goods in different price ranges of stores is specifically divided as follows: the sales proportion of goods with different price positioning in the corresponding stores is counted; according to the 80 / 20 principle, the number of all specifications of goods in the country is first counted, recorded as N, which is regarded as the dimension size of the multihot vector; secondly, the sales volume generated by goods of different specifications in each city is counted, and the products of corresponding specifications are taken from the product specification with the highest sales volume in turn, and the component value of the corresponding position of the vector is set to 1, until the top 80% of the sales volume of the products is obtained, and finally the values ​​of the remaining components of the vector are set to 0 to obtain the corresponding store feature vector.

9. The store grouping and grading system based on machine learning according to claim 7, characterized in that: The system consists of 4 modules: data reading module, city grouping module, store classification module, and log module; The data reading module will obtain all the data required by the clustering and grading algorithm from the enterprise's information system by executing SQL statements; The city clustering module calculates the characteristic vector of a city by counting the sales volume information and product specification information of all stores in a city, and divides all cities into K different city clusters through a clustering algorithm; The store grading module calculates the store feature vector by counting the store size, product information and product sales information in the urban cluster, and divides all stores into N different store grades through a clustering algorithm; The log module writes all calculation results into the log file of the corresponding date; Users can view it by themselves, and it can display execution information such as the executed SQL, store city grouping and store classification results.

Images