CN110633401A - A prediction model of store data and its establishment method - Google Patents

Abstract

The embodiment of the invention discloses a prediction model of store data and an establishment method thereof, wherein the prediction method of the store data comprises the following steps: acquiring first store information around stores through a web crawler technology, acquiring second store information through an enterprise internal management system, cleaning and performing structured data processing to obtain different types of sales data, dividing the sales data into continuous variable data and discrete variable data, performing normalization correction on the continuous variable data, and performing assignment continuous optimization on the discrete variable data; and performing training set learning by using a machine learning method to obtain a prediction model of the sales scale, and performing replication optimization on the prediction model by using actual sales scale data of the store to be predicted. The prediction model of store data and the establishment method thereof disclosed by the embodiment of the invention can reasonably and accurately predict the sales data of stores, further reasonably arrange the operation of the stores and improve the terrace effect of the stores.

Description

A prediction model of store data and its establishment method

technical field

The invention relates to the field of big data, in particular to a prediction model for store data and a method for establishing the same.

Background technique

Traditionally, companies have often used manual forecasting based on business experts to predict the sales volume of stores, which is based on the experience of sales personnel combined with manual analysis to predict the sales of stores. This forecasting method is called the expert method, which can quickly and easily obtain However, this kind of prediction did not use the sales data to make a reasonable and convincing judgment, nor did it have theoretical support, resulting in uneven prediction results, which in turn led to the establishment, personnel, stocking, operation and management of the store. influences.

Due to the wide application of machine learning methods, at present, enterprises will directly import the data of different stores into the computer, and simply predict the sales situation of the stores through machine learning. This prediction is only a simple application of machine learning, but the company's store sales It is not only affected by the management data of the store itself, but also affected by multiple factors of different geographical locations, urban development and other enterprise stores. This prediction method does not analyze and process various factors. The application of store management data will have greater uncertainty, which will also affect the final forecasting effect.

SUMMARY OF THE INVENTION

In order to solve the problems of the prior art, the embodiment of the present invention provides a method and system for establishing a store data prediction model, which can reasonably and accurately predict the sales data of the store, and then reasonably arrange the operation of the store and improve the floor efficiency of the store.

In order to solve the above-mentioned technical problems, the technical scheme adopted in the present invention is:

In a first aspect, an embodiment of the present invention provides a method for establishing a store data prediction model, comprising the following steps:

Obtain the information of the first store around the store through the web crawler technology, and obtain the information of the second store through the internal management system of the enterprise, and store the information of the first store and the information of the second store in the database;

Perform cleaning and structured data processing on the first store information and the second store information to obtain different types of sales data, divide the sales data into continuous variable data and discrete variable data, and analyze the continuous variable data. Carry out normalization correction, and carry out continuous optimization of assignment to the discrete variable data;

Use the method of machine learning to score the feature data after the normalization correction and the continuous optimization, and automatically select the feature data with high scores for training set learning to obtain a prediction model of sales scale, and use the The actual sales scale data of the store to be predicted is reviewed and optimized for the prediction model.

Further, according to the specific location and retail scene of the store to be predicted, the prediction model is used to output the predicted value of the sales scale of the store to be predicted.

Further, the crawling methods of the crawler technology include at least crawling map data, crawling online shopping data, and crawling official statistical data.

Further, the first store information includes city data features, business district data features, and geographic data features, and the second store information includes store membership data features and store item data features.

Further, the normalization correction includes: first performing logarithmic bias correction on the continuous variable data that is close to a normal distribution, and then performing mean normalization processing on the continuous variable data after bias correction.

Further, the continuous optimization of the assignment includes: performing a univariate analysis of variance on the discrete variable data, screening the discrete variable data that has a greater impact on the sales performance of the store, and then converting the screened discrete variable data into storefronts. For the corresponding mean value of the plateau effect, the different mean values of the plateau effect are successively sorted according to different quantitative values.

Further, before using the prediction model to output the prediction result, it also includes specifying the business type of the retail scene, and according to the business type, automatically associate and match to the corresponding specified prediction model.

On the other hand, an embodiment of the present invention also provides a prediction model for store data, including:

The data collection module includes a web crawler unit and an enterprise data collection unit, the web crawler unit is used to obtain the information of the first store around the store through the web crawler technology, and the enterprise data collection unit is used to obtain the second store information through the internal management system of the enterprise. Store information, the data collection module then stores the first store information and the second store information into a database;

The data processing module is used for cleaning and structured data processing on the first store information and the second store information, obtaining different types of sales data, and dividing the sales data into continuous variable data and discrete variable data, Carry out normalization and correction to the continuous variable data, and carry out assignment continuous optimization to the discrete variable data;

The machine learning module is used to score the feature data after the normalization correction and the continuous optimization, and automatically select the feature data with high scores for training set learning to obtain a prediction model of the sales scale, and use the The actual sales scale data of the store to be predicted is used to review and optimize the prediction model.

Further, the data processing module includes a continuous variable correction unit, which is used to first perform logarithmic correction on the continuous variable data that is close to the normal distribution, and then perform mean normalization processing on the continuous variable data after the correction. .

Further, the data processing module includes a discrete variable optimization unit for performing univariate analysis of variance on the discrete variable data, screening the discrete variable data that has a greater impact on the store's floor efficiency, and then performing a univariate analysis on the filtered data. The discrete variable data is converted into the average per floor efficiency corresponding to the store, and the different average per floor efficiency values are sequentially sorted according to different quantitative values.

The beneficial effects brought by the technical solutions provided in the embodiments of the present invention are:

The embodiment of the present invention provides a method and system for establishing a store data prediction model, wherein the information around the store is collected through the network crawler technology of the computer, and the network crawler can obtain the information of the first store in a large amount and accurately, which significantly improves the information acquisition. At the same time, the second store information is obtained in combination with the internal management system of the enterprise, and the first store information and the second store information are analyzed by computer to clean and optimize the reasonable and accurate target characteristic data. The processing makes the various data collected before machine learning to eliminate multiple interference factors, which significantly improves the accuracy of the model obtained after machine learning; the actual sales data of the store is used to review and optimize the prediction model in the embodiment of the present invention. , the output accuracy of the prediction model is improved again, and the data predicted by the output of the prediction model is finally made reasonable and effective. Then through the effective prediction results, the operation of the store is reasonably arranged, and the floor efficiency of the store is improved.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

Fig. 1 is a kind of flow chart of the establishment method of the store data prediction model disclosed by the embodiment of the present invention;

Fig. 2 is a kind of flow chart for data processing analysis in the establishment method of store data prediction model disclosed in the embodiment of this schematic diagram;

FIG. 3 is a schematic structural diagram of a prediction model for store data disclosed in an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only Some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Example 1:

The establishment method of a kind of store data prediction model of the embodiment of the present invention disclosed in Figure 1, comprises the following steps:

S1: Obtain the information of the first store around the store through the web crawler technology, obtain the information of the second store through the internal management system of the enterprise, and store the information of the first store and the information of the second store in the database;

S2: Perform cleaning and structured data processing on the first store information and the second store information to obtain different types of sales data, and divide the sales data into continuous variable data and discrete variable data. The variable data is normalized and corrected, and the discrete variable data is assigned continuous optimization;

S3: Use the method of machine learning to score the feature data after the normalization correction and the continuous optimization, and automatically select the feature data with high scores for training set learning to obtain a prediction model of sales scale, and use The actual sales scale data of the store to be predicted is reviewed and optimized for the prediction model.

Preferably, according to the specific location and retail scene of the store to be predicted, the prediction model is used to output the predicted value of the sales scale of the store to be predicted, and then through the effective prediction result, the operation of the store is reasonably arranged, and the floor efficiency of the store is improved. .

Crawler technology is a program that "automatically browses the web", which automatically crawls the information needed by users on the World Wide Web according to certain rules. With the development of the Internet, the network has become the carrier of a large amount of information. Crawler technology has also become an important part of data collection and is the most basic step in big data analysis. Preferably, the crawling methods of the crawler technology include but are not limited to crawling map data, crawling online shopping data, and crawling official statistical data. Specifically, use web crawler tools to crawl website data such as AutoNavi maps, national data, Dianping, and Meituan to obtain information such as surrounding passenger flow, population, business distribution, point of interest distribution, and traffic conditions.

Preferably, the first store information includes city data features, business district data features, and geographic data features. Specifically, since the macroeconomic features of a city have a certain correlation with the quantity and quality of potential customers, it is necessary to first consider the city Data characteristics are also necessary for detailed analysis of population structure, so it is more reasonable to analyze within the radiation range of business districts. The collection of urban data characteristics includes: the number of permanent residents, the number of registered population, the total GDP, urbanization rate, The per capita disposable income of urban residents and the total retail sales of consumer goods; the collection of data characteristics of business districts includes: demographic data characteristics, passenger flow data characteristics, and portrait characteristics; according to the "retail cluster" theory, retail enterprises opening stores in the same area can produce "spillover effects" ”, thereby attracting greater passenger flow, so it is necessary to consider the number and distribution of competitors. Transportation facilities and POIs can obviously attract passenger flow. Usually, people choose places with more convenient transportation for consumption. At the same time, a large number of passenger gathering points can directly bring passenger flow. Therefore, the collection of geographic data features includes: competitor data, traffic Facilities, gathering point POI (point of interest).

The second store information includes store membership data features and store item data features. Specifically, since the members of the store are the main consumer groups, it is necessary to analyze their characteristics. The characteristics of the member data include: the number of members, the penetration rate of members, the annual turnover rate of members, the annual purchase amount of members, and the unit price of members; The characteristics of the project itself and sales are obviously directly related, and the subsequent characteristic analysis is mainly based on these data. The data characteristics of the store project include: city category, market level, city level division, business district, business district type , store nature, opening time, annual sales including tax, annual rent, floor distribution, interior area, building area.

As shown in the flowchart of data processing and analysis disclosed in FIG. 2 , the normalization correction includes: first performing logarithmic correction on the continuous variable data that is close to a normal distribution, and then performing mean normalization on the continuous variable data after correction. Unified processing. Specifically, after sorting out the above-mentioned various data, due to the diverse data types, the continuous variable data that has a greater impact on store sales, such as floor efficiency, sales, interior area and rent, etc., are subjected to linear analysis. It is found that it is relatively close to the normal distribution, but has obvious positive bias. If it is directly normalized, it will not achieve the desired effect. Therefore, it is necessary to take the logarithm log(1+x) first to alleviate this positive bias. Partial trend, where x represents the continuous variable data before taking the logarithm, and the result obtained after taking the logarithm is subjected to linear analysis again. It can be seen through computer analysis that it has reached a normal distribution. Do mean normalization: [x_scale=(x ₁ -mean)/std], where mean is the mean of a set of continuous variable data after logarithmic processing, and std is a set of continuous variable data standards after logarithmic processing Difference, x ₁ is an element of a group of continuous variable data after logarithmic processing, and x_scale represents the result after normalization and correction. Through mean normalization, the fluctuation of the continuous variable data can be reduced, thereby reaching the global minimum value, which facilitates the effect of the learner in the subsequent machine learning process to be more convincing and accurate, and reduces errors.

Preferably, the continuous optimization of the assignment includes: performing a one-way variance analysis on the discrete variable data, screening the discrete variable data that has a greater impact on the sales performance of stores, and then converting the filtered discrete variable data into storefronts For the corresponding mean value of the plateau effect, the different mean values of the plateau effect are successively sorted according to different quantitative values. Specifically, for the discrete variable data, if the value after calculating the unary variance is large, it proves that the impact on the store's floor efficiency is large. Characteristic analysis, it is found that multiple discrete variable data will have a greater impact on the store's floor efficiency, such as the nature of the store and city level. Set a mean of 1, 2, 3, and 4, and use the mean value to quantitatively describe their impact on the efficiency, which is equivalent to the assignment of discrete variable data to make it a continuous variable amount of data. This method uses a different method from One-Hot encoding to process discrete data. Compared with One-Hot encoding, it can retain more information, significantly improving the breadth of important sales data collection and the depth of processing, thus ensuring that subsequent machines Accuracy and representation of learning.

Preferably, before using the prediction model to output the prediction result, the method further includes clarifying the business type of the retail scene, and according to the business type, automatically associate and match to the corresponding specified prediction model. Specifically, since different retail scenarios will generate different prediction models, and different retail scenarios have different business types, after machine learning, different prediction models will be generated according to different retail scenarios. Before the prediction model, it is necessary to first define the retail scene of the store to be predicted. The computer will automatically associate and match the specified prediction model according to the input retail scene. This association and matching are compared with the traditional machine learning. It is more targeted and has more accurate forecast data. For the company's stores, the output sales scale is more accurate, so that the store can reasonably arrange the operation of the store and improve the floor efficiency of the store.

Machine learning is a multi-domain interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or realize human learning behaviors to acquire new knowledge or skills, and to reorganize existing knowledge structures to continuously improve their performance. Specifically, based on existing feature data and previous research, when a prediction model is automatically established, since machine learning includes two types of algorithms, bagging and boosting, this embodiment preferably uses random forest regression in bagging algorithms. Among them, random forest is a strong learner composed of multiple classification and regression trees as weak learners. Random forest can automatically select optimal features and score. Among them, when constructing the model for store forecasting, input 20 processed feature data that have a greater impact on the store, which are the following features: 'store level', 'region', 'city level', 'city category', 'Market level', 'Distance from core enterprises', 'Business network planning type', 'Store nature', 'Opening time', 'Whether it is a shopping mall store', 'Annual rent', 'Interior area_SUM' ,'Inner area_B2F','Inner area_B1F','Inner area_1F','Inner area_2F','Inner area_3F','Inner area_4F',' Inner area_5F', 'Inner area_6F', 'Inner area_7F', 'Inner area_8F', 'Inner area_interlayer', 'Tax-excluded sales floor effect in sleeve', 'Restore tax-included sales throughout the year'; then the computer automatically selects the 16 feature data with the highest scores, and selects 500 regression trees as weak learners; 80% of the sample set is selected as the training set, and the remaining 20% is used as the test set; The score value of the predictive model learned from the training set is 0.84.

Preferably, when using the prediction model in the embodiment of the present invention, the prediction model in the embodiment of the present invention can not only predict the store to be opened, but also predict the store that has been opened, wherein, for the store to be opened, The prediction model can clearly show the location where the store has the largest sales scale and the highest ping-pong efficiency. By comparison, while verifying the prediction results, the actual data is imported into the process of machine learning, so as to complete the review and optimization of the prediction model, so that the prediction model is in a process of dynamic optimization and correction, and further improve the availability of the prediction model. and accuracy, reducing the error margin of the prediction model.

Embodiment 2:

A schematic diagram of a structure of the prediction model of store data provided by the embodiment of the present invention disclosed in FIG. 3 includes:

The data collection module 1 includes a web crawler unit 11 and an enterprise data collection unit 12, the web crawler unit 11 is used to obtain the first store information around the store through the web crawler technology, and the enterprise data collection unit 12 is used to The management system obtains the second store information, and the data collection module 1 stores the first store information and the second store information into the database again;

The data processing module 2 is used for cleaning and structured data processing of the first store information and the second store information, obtaining different types of sales data, and dividing the sales data into continuous variable data and discrete variable data , carry out normalization and correction to the continuous variable data, and carry out assignment continuous optimization to the discrete variable data;

The machine learning module 3 is used to score the feature data after the normalization correction and the continuous optimization, and automatically select the feature data with high scores for training set learning to obtain a forecast model of sales scale, and use The actual sales scale data of the store to be predicted is reviewed and optimized for the prediction model.

Specifically, the data collected by the web crawler unit 11 includes: city data features, business district data features, and geographic data features, and the data collected by the enterprise data collection unit 12 includes membership data features and store item data features. Through the web crawler unit 11 and the enterprise data collection unit 12, almost all key features can be covered for the sales scale of the store to be predicted, and the details of each collected feature can be found in the introduction in the first embodiment, and will not be discussed here. More details.

Preferably, the data processing module 2 includes a continuous variable correction unit 21, which is used to first perform logarithmic correction on the continuous variable data that is close to a normal distribution, and then perform mean normalization on the continuous variable data after correction. Further, the data processing module 2 includes a discrete variable optimization unit 22, which is used to perform a one-way variance analysis on the discrete variable data, and screen the discrete variable data that has a greater impact on the store’s floor efficiency, and then The filtered discrete variable data is converted into the average per floor efficiency corresponding to the store, and the different average per floor efficiency values are sequentially sorted according to different quantitative values. Since the web crawler unit 11 and the enterprise data collection unit 12 are only for data collection and acquisition, the data input into the database needs to be further optimized and corrected, so that the data entered into the data prediction module has a common representativeness and accuracy. Specifically, the continuous variable correction unit 21 firstly takes the logarithm log(1+x) to alleviate this positive bias trend, wherein x represents the continuous variable data before taking the logarithm, and the result obtained after taking the logarithm Perform linear analysis again, and it can be seen through computer analysis that it has reached a normal distribution. On this basis, the mean is normalized: [x_scale=(x ₁ -mean)/std]; See the introduction in the above-mentioned first embodiment for the method of unification correction, which will not be described in detail again. The discrete variable optimization unit 22 first calculates the unary variance of the multiple sets of discrete variable data, and then through discrete characteristic analysis, it is found that the multiple discrete variable data will have a greater impact on the floor efficiency of the store, such as the nature of the store and the city level, etc. , so it is possible to delineate a mean value of 1, 2, 3, and 4 for each discrete variable data according to the mean value of the ping effect under the corresponding value of each discrete variable data, and quantitatively describe their effect on the ping effect through the mean value of the ping effect. The effect is equivalent to the assignment of discrete variable data to make it a continuous variable data volume.

Preferably, when using a prediction model to predict the sales scale of a store by a computer, a prediction model pairing unit of a computer needs to be used, and the prediction model pairing unit is used to specify the retail scene before using the prediction model to output the prediction result. According to the type of business format, the corresponding designated prediction model is automatically associated and matched; further, the machine learning module 3 also includes a model review and optimization unit 31, wherein, due to different retail scenarios, there will be Different forecast models, and different retail scenarios have different types of formats, directly input the retail scenario of the store, through the model pairing unit can match the corresponding designated forecast model, and then complete the sales scale of the store. Prediction, the prediction model will display the prediction result through the prediction result display unit on the computer. In order to further improve the accuracy of the prediction model, the model review and optimization unit 31 is used to compare the actual sales scale of the stores that have been opened with the predicted sales scale, and while verifying the prediction results, import the actual data into the machine learning In the process, the review and optimization of the prediction model is completed, so that the prediction model is in a process of dynamic optimization and correction, which further improves the usability and accuracy of the prediction model and reduces the error range of the prediction model.

All the above-mentioned optional technical solutions can be combined arbitrarily to form optional embodiments of the present invention, which will not be repeated here.

It should be noted that when the store data prediction model provided in the above embodiment predicts the sales scale of the store, only the division of the above functional modules is used as an example for illustration. The functional module of the store data is completed, that is, the internal structure of the store data prediction model is divided into different functional modules to complete all or part of the functions described above. In addition, the store data prediction model provided in the above embodiment and the store data prediction method embodiment belong to the same concept, and the specific implementation process thereof is detailed in the method embodiment, which will not be repeated here.

Those of ordinary skill in the art can understand that all or part of the steps of implementing the above embodiments can be completed by hardware, or can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The storage medium mentioned may be a read-only memory, a magnetic disk or an optical disk, etc.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection of the present invention. within the range.

Claims (10)

1. A building method of a store data prediction model is characterized by comprising the following steps:

acquiring first store information around a store through a web crawler technology, acquiring second store information through an enterprise internal management system, and storing the first store information and the second store information into a database;

cleaning and structuring data processing are carried out on the first store information and the second store information to obtain different types of sales data, the sales data are divided into continuous variable data and discrete variable data, normalization correction is carried out on the continuous variable data, and assignment continuous optimization is carried out on the discrete variable data;

and (3) scoring the normalized and corrected and continuously optimized feature data by using a machine learning method, automatically selecting the feature data with high score to perform training set learning to obtain a prediction model of sales scale, and then repeating and optimizing the prediction model by using the actual sales scale data of the store to be predicted.

2. The method for building the store data prediction model according to claim 1, wherein the prediction model is used to output the predicted value of the sales scale of the store to be predicted according to the specific location and retail scene of the store to be predicted.

3. The method of building a store data prediction model according to claim 1, wherein the crawled paths by crawler technology include at least crawling map data, crawling online shopping data, crawling official statistics.

4. The method for building the store data prediction model according to claim 1, wherein the first store information includes city data features, business circle data features and geographic data features, and the second store information includes store member data features and store project data features.

5. The store data prediction model building method according to claim 1, wherein the normalization correction includes: and carrying out logarithmic deviation correction on the continuous variable data which are approximately normally distributed, and then carrying out mean value normalization processing on the corrected continuous variable data.

6. The method for building the store data prediction model according to claim 1, wherein the assigning continuous optimization comprises: and carrying out unitary variance analysis on the discrete variable data, screening the discrete variable data which has a large influence on the store plateau effect, converting the screened discrete variable data into the store plateau effect mean value corresponding to the store, and continuously sequencing the different plateau effect mean values according to different quantitative values.

7. The method for building the store data prediction model according to claim 2, further comprising the step of specifying a business type of the retail scene before outputting the prediction result by using the prediction model, and automatically associating and matching the business type with the corresponding specified prediction model.

8. A predictive model of store data, comprising:

the system comprises a data acquisition module and a database, wherein the data acquisition module comprises a web crawler unit and an enterprise data acquisition unit, the web crawler unit is used for acquiring first store information around stores through a web crawler technology, the enterprise data acquisition unit is used for acquiring second store information through an enterprise internal management system, and the data acquisition module stores the first store information and the second store information into the database;

the data processing module is used for cleaning and carrying out structured data processing on the first store information and the second store information to obtain different types of sales data, dividing the sales data into continuous variable data and discrete variable data, carrying out normalization correction on the continuous variable data, and carrying out assignment continuous optimization on the discrete variable data;

and the machine learning module is used for scoring the characteristic data after the normalization correction and the continuous optimization, automatically selecting the characteristic data with high score to perform training set learning to obtain a prediction model of the sales scale, and performing repeated optimization on the prediction model by using the actual sales scale data of the store to be predicted.

9. The store data prediction model of claim 8, wherein the data processing module comprises a continuous variable correction unit, and is configured to perform logarithmic correction on the continuous variable data that is approximately normally distributed, and then perform mean normalization on the corrected continuous variable data.

10. The store data prediction model according to claim 8, wherein the data processing module comprises a discrete variable optimization unit, and is configured to perform unitary variance analysis on the discrete variable data, screen the discrete variable data that has a larger influence on store floor effects, convert the screened discrete variable data into store-corresponding floor effect means, and continuously sort the different floor effect means according to different quantitative values.

Images