Efficient and automatic shopping mall data quantitative analysis method

Abstract

The invention discloses an efficient and automatic shopping mall data quantitative analysis method, which expands and applies quantitative analysis methods used in the fields of finance, securities, spot goods, virtual currency and the like to shopping mall data analysis. The characteristics of multi-dimensional data of logs, commodities, contents, activities, search, transactions and the like of a shopping mall and the requirements of a quantitative analysis method for the data are comprehensively considered, collection is carried out in a reasonable mode, structured and standardized processing is carried out, a unique index, factor and strategy model system is established, and a factor value space is designed; by applying a periodic, classified and layered factor and strategy model training method, the influence of various main data of the shopping mall on the operating indexes can be comprehensively, accurately and efficiently analyzed.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for efficient and automatic quantitative analysis of shopping mall data. Background Art

[0002] With the rapid development of e-commerce, the amount of mall data has exploded. How to effectively and efficiently process and analyze this data and fully tap its value has become the key to mall operations. Traditional data analysis methods usually use tools and software to analyze the operating results after taking marketing activities, determine the effectiveness of marketing activities, and then formulate subsequent marketing activity plans. However, there is no conclusion on which type of marketing activities are most effective in which environment, and how the marketing activities of this product should be coordinated with the marketing activities of the upper-level product category, the upper-upper-level product category, and the upper-upper-upper-level product category. For the product attribute data and merchant attribute data seen by users, such as unit specification price, discount rate, historical sales volume, number of reviews, praise rate, logistics timeliness, etc., the operation staff can confirm their positive effect on improving the mall's operating indicators, but it is still difficult to determine how to determine the value of each attribute, how to coordinate between the attributes, and how to specifically and finely reflect the prototype and page design of the mall platform product. The commodity attribute data and merchant attribute data that are less concerned by operations and product personnel, such as inventory, are not clear in terms of their role in improving business indicators. There is currently a lack of a systematic and accurate analysis method to solve the above problems. Therefore, it is very urgent to develop an efficient and automatic quantitative analysis method for mall data. This method extends the quantitative analysis methods used in finance, securities, futures and spot, virtual currency and other fields to mall data analysis, comprehensively considers the characteristics of multi-dimensional data such as mall logs, commodities, content, activities, searches, transactions and the requirements of quantitative analysis methods for data, collects and structures and standardizes data in a reasonable way, establishes a unique system of indicators, factors and strategy models, designs factor value space, and uses periodic, classified and layered factor and strategy model training methods to comprehensively, accurately and efficiently analyze the impact of various major mall data on business indicators, which has great reference significance. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an efficient and automatic method for quantitative analysis of shopping mall data, so as to solve the problems in the current shopping mall data analysis, such as the inability to accurately analyze data, one-sided analysis of part of the data but not comprehensive and integrated analysis, and the unclear role of certain data, which leads to inaccurate and incomplete analysis conclusions.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: An efficient and automatic method for quantitative analysis of mall data includes the following steps: S1, collect data from the mall operation database, and perform structured and standardized processing on the collected data to build a data collection database. The data collection database includes a user database, a merchant database, a product database, a search database, a transaction order database, a payment database, a logistics database, and a marketing database. The data collection method is: Database synchronization realizes data collection: The stock data of users, merchants, products, searches, transaction orders, payments, logistics and marketing are collected by synchronizing the mall operation database once a day. This type of data is stored in a structured manner in database tables and updated daily. API interface collection: The daily incremental data of users, merchants, products, searches, transaction orders, payments, logistics, and marketing are collected by reading from the mall operation database every m minutes through the API interface; this type of data is structured and stored in database tables and updated in real time; File ETL synchronization mode collection: For article, image, video data, and log file file data, they are collected from the file system of the mall operation database in the file ETL synchronization mode; the collection cycle is once a day or every n minutes; Articles, pictures and videos include three types of articles, pictures and videos: merchants, products and marketing; Streaming data reading method collection: For intraday log data, streaming data reading is used to collect data; S2, set quantitative indicators for mall operations and manage them; S3, factor extraction and multi-level classification processing from the data collection database; S4, set the factor value space and step size; S5, establish an initial strategy model, which is an analysis model between factors and quantitative indicators of mall operations; S6, conduct multi-cycle single-factor training; S7, conduct multi-period factor combination training; S8, executes the generation and management of seed strategy models in multiple cycles, S9, optimize the seed strategy model.

[0005] Furthermore, the core data of the mall includes users, merchants, products, searches, transaction orders, payments, logistics, marketing, logs, etc. Basic data such as users, merchants, products, searches, transaction orders, payments, logistics, marketing, etc. are stored in the database, and merchant pictures and videos, product pictures and videos, and system logs are stored in files. The intraday changes in data such as merchants and products are relatively slow, while data such as users, searches, product inventory, transaction orders, payments, logistics, and marketing may change significantly during the day. The collection of log data is usually formed into files through streaming processing, and other data is stored in the database through business interfaces. Different collection and preprocessing methods are adopted according to the different storage methods of data, the daily change patterns, and the formation methods.

[0006] Preferably, in step S1, the method for performing structured processing on the collected data is: Extract the activity theme, corresponding merchant, corresponding product and generation time information of marketing articles, pictures and video data, and insert them into the marketing library; Extract the corresponding merchants, corresponding products and generation time information of merchant type articles, pictures and video data, and insert them into the merchant database; Extract the corresponding merchants, corresponding products and generation time information of the articles, pictures and video data of the product type, and insert them into the product library; Extract the log business classification and generation time information in the log file, and insert them into the user library, merchant library, product library, retrieval library, transaction order library, payment library, logistics library and marketing library according to the log business classification.

[0007] Preferably, in step S1, the method for standardizing the collected data is: Construct product attribute tables, merchant attribute tables, and marketing attribute tables; construct search tables, transaction order tables, payment tables, logistics tables, and user tables; Based on the product ID, extract the product ID, product attributes, and product attribute change record timestamp from the product database and store them in the product attribute table; Based on the product ID, extract the product ID, merchant attributes, and merchant attribute change record timestamp from the merchant database and store them in the merchant attribute table; Based on the product ID, the product ID, marketing activity attributes, and marketing activity attribute change record timestamp are extracted from the marketing database and stored in the marketing attribute table; The product attribute table, the merchant attribute table and the marketing attribute table are combined into an independent variable library, wherein the marketing attribute table and the merchant attribute table are associated with the product attribute table through the product ID; Based on the product ID, extract the product ID, search attributes, and the timestamp of the search attribute change record from the search library and store them in the search table; Based on the product ID, extract the product ID, transaction order attributes, and transaction order attribute change record timestamp from the transaction order database and store them in the transaction order table; Based on the product ID, extract the product ID, payment attributes, and payment attribute change record timestamp from the payment database and store them in the payment table; Based on the product ID, extract the product ID, logistics attributes, and logistics attribute change record timestamp from the logistics database and store them in the logistics table; Extract user attributes and user record change timestamps from the user database and store them in the user table; The user table, search table, logistics table, transaction table and payment table form a dependent variable database.

[0008] Furthermore, during the data standardization phase, duplicate rows are not deleted, missing values ​​are not filled, and abnormal data is not processed.

[0009] Preferably, in step S2, the method for setting the quantitative indicators of the mall operation and managing them is: Quantitative indicators include analyzing the number of times the strategy model is triggered within multiple time periods T (SigTimes), the probability of improving the commodity's operating indicators by 20% (IR20Ratio), the probability of improving by 3% (IR03Ratio), the probability of improvement (SigWinRatio), the probability of maximum reduction of no more than 3% (MaxLoss3), the probability of maximum reduction of no more than 7% (MaxLoss7), the maximum possible reduction percentage (MaxLoss), the maximum possible reduction days (MaxLossDays), the ratio of improvement probability or reduction probability greater than 3 (Eratio3), the Sharpe ratio (PR) and the convergence degree (ConValue); Quantitative indicator management is the management of all quantitative indicators, including adding, deleting, modifying, checking and changing priorities.

[0010] Furthermore, the mall's operating indicators include sales volume, sales revenue, profit, unit capital profit or other core goals that management personnel and product personnel are concerned about; when considering the impact of a certain factor on the commodity operating indicators, it is necessary to consider the inconsistencies in backgrounds such as different time periods, different categories, and different regions.

[0011] Preferably, in step S3, when factor extraction and multi-level classification processing are performed from the data collection database, the factors are extracted from the commodity attribute table, the marketing attribute table and the merchant attribute table, and the method is: Original factor extraction: Original factors include unit price factor and discount rate factor in commodity attributes; Statistical factor extraction: Statistical factors include historical sales factor, inventory factor, number of reviews factor and favorable review rate factor in product attributes, activity factor in marketing attributes and logistics timeliness factor in merchant attributes; The factor multi-level classification process includes classifying factors by: Divide product categories into three levels, including level 1 product type, level 2 product type, and level 3 product type; The factors corresponding to the original factors and statistical factors are extracted at each layer, namely, the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the first-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the second-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the third-level commodity type.

[0012] Preferably, in step S4, the method for setting the factor value space and step size is: Corresponding histograms are established for the unit price factor, discount rate factor, historical sales factor, and inventory factor. The same-category and cross-category products are analyzed, and the intervals and step lengths of the initial value spaces of different factors are set; Corresponding histograms are established for the evaluation number factor, the favorable comment rate factor, and the logistics timeliness factor, and cross-category products are analyzed to set the interval and step length constants of the initial value space of different factors; For activity factors, a histogram of activity frequency is established, and analysis is performed on applicable merchandise, activity start time, and activity end time. Constants for the intervals and step lengths of the initial value spaces of different factors are set.

[0013] Preferably, in step S5, the method for establishing the initial strategy model is: S501, quantitative indicators: The operation indicators of the mall include sales volume, sales revenue, profit, unit capital profit or other core goals that business personnel and product personnel pay attention to. When considering the impact of a certain factor on the commodity operation indicators, it is necessary to consider the inconsistency of the macro environment in different time periods; based on different time periods t, the impact of the factor on the commodity operation indicators is quantified to form a configurable time period matrix Tnx2 and a quantitative indicator matrix Gnx11; set the priority of each quantitative indicator and form a priority matrix Pn×3; ; The Tnx2 time period matrix represents the time period used to verify the strategy model. It has n rows and 2 columns in each row. Each row represents an optional time period. The first element represents the start time and the second element represents the end time. When the second element is "now", it means that the end time is the latest time. Quantitative indicators can be maintained in the configuration file. The quantitative indicator details of all strategy models in the same strategy model family are as follows: ; The Gnx11 quantitative indicator matrix has n rows, each with 11 columns. Each row represents a quantitative indicator set of a strategy, and each element in each row represents a specific indicator value; The following are examples of the priority details of each quantitative indicator: ; In the formula, P n×3 The matrix represents the priority matrix of each quantitative indicator; Z1 and Z2 are the serial numbers of the quantitative indicators, and IR20Ratio and SigWinRatio are the names of the quantitative indicators; S502, construct factor matrix: The factors are divided into 4 layers, including the original layer, the first-level commodity type layer, the second-level commodity type layer and the third-level commodity type layer; each layer has 8 factors, including unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor; the 8 factors are divided into 4 categories, including price category, quantity category, evaluation category and marketing category; Set each factor to have positive and negative impacts on product operation indicators; Set factor attributes including factorID, name, direction, type, level, start value, end value and step size; construct factor matrix , , ; Among them, F0 is the original layer factor matrix, with a total of 16 rows and 8 columns; each row represents the specific configuration of a factor, the first column of each row is the sequence number, the second column is the alias, and the third column is the use direction. When it is up, the specific value of the factor in the quantitative analysis model must be greater than a certain value to take effect; when it is down, the specific value of the factor in the quantitative analysis model must be greater than a certain value to take effect; the fourth column is the type, the fifth column is the level, the original layer factor level value is 0, the sixth column is the starting value, the seventh column is the ending value, and the eighth column is the step size; S503, constructing an initial strategy model: Establish the correlation between the initial strategy model and the factors to form a step-length validation correlation matrix W; the correlation coefficient of the correlation relationship ranges from 0.00 uncorrelated to 1.00 fully correlated, and the initial value is 1.00; The step verification relationship matrix can be maintained in the configuration file. The details are as follows: ; In the formula, the correlation matrix includes n rows and 64 columns, and each row represents the correlation between a quantitative analysis model and a factor. As shown above, F0, F1, F2, and F3 have a total of 64 factors. The first column of each row of the correlation matrix represents the degree of correlation between a quantitative analysis model and the first factor of F0, 0 for no correlation and 1 for complete correlation. The second column of each row of the correlation matrix represents the degree of correlation between a quantitative analysis model and the second factor of F0, 0 for no correlation and 1 for complete correlation, and so on. The formula of the quantitative analysis model is: G×T=S(W×(F0+F1+F2+F3)); Where G represents the quantitative indicator matrix, T represents the time period matrix, W represents the step verification correlation matrix, S represents the quantitative analysis model algorithm; F0, F1, F2 and F3 are factor matrices.

[0014] Preferably, in step S6, the method for performing multi-cycle single factor training is: In each time period t, the correlation matrix is ​​verified by step size within the starting and ending values ​​of the factors in the factor matrix The validity of the middle factor is verified by fixing the values ​​of other factors as the mean μ and obtaining the temporary conclusion matrix and the temporary correlation coefficient matrix ; Taking the w008 factor as an example, we get the temporary conclusion matrix and temporary correlation coefficient matrix: According to the indicator priority matrix P n×3 , sort and verify the comparison temporary conclusion matrix For each conclusion in the temporary correlation coefficient matrix, remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary correlation coefficient matrix to obtain a usable temporary conclusion matrix. and the available correlation coefficient matrix ; Verify all 64 factors in the factor matrix in turn and obtain a usable temporary conclusion matrix and the available correlation coefficient matrix ; Comprehensive all from arrive The available correlation coefficient matrix is ​​obtained by .

[0015] Preferably, in step S7, the method for performing multi-period internal factor combination training is: Among different factors, factors of the same type and at the same level have the strongest correlation, followed by factors of the same type and at different levels, and then factors of different types and at the same level. Factors of different types and at different levels have weaker correlation. Therefore, combination training is first performed on factors of the same type and at the same level, such as price, volume, evaluation, and promo factors of the original layer, and then combination training is performed on factors of the same type and across levels. Combination training of factors of the same type and level: In each time period t, Valid factor combinations are verified by combination within the same layer; During verification, the values ​​of other factors are fixed to The mean obtained in the above equation is used to obtain the temporary same-layer and same-type combination conclusion matrix and temporary same-layer and same-type combination correlation coefficient matrix ; Taking the promo combination factor as an example, the temporary same-level and same-type combination correlation coefficient matrix is ​​as follows: ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the temporary same-level and similar combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary same-level and similar combination correlation coefficient matrix, and obtain the available temporary same-level and similar combination conclusion matrix And the available same-level and same-type combination correlation coefficient matrix ; Comprehensive all from arrive The available temporary same-level and same-type combination correlation coefficient matrix is ​​obtained to obtain the comprehensive available same-level and same-type combination correlation coefficient matrix ; Cross-layer similar combination training: In each time period t, Verify effective factor combinations by combination across layers; During verification, the values ​​of other factors are fixed to the mean values ​​obtained in WCS(0−>n)n×64, and the temporary cross-layer similar combination conclusion matrix is ​​obtained. and temporary cross-layer homogeneous combination correlation coefficient matrix ; According to the indicator priority matrix P n×3, sort and verify the conclusions in the temporary cross-layer similar combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary cross-layer similar combination correlation coefficient matrix to obtain the available temporary cross-layer similar combination conclusion matrix and temporary cross-layer homogeneous combination correlation coefficient matrix ; Verify the cross-layer price, volume, evaluation, and promo combinations of all layers in the factor matrix in turn to obtain the final available temporary cross-layer similar combination conclusion matrix And the final available temporary cross-layer homogeneous combination correlation coefficient matrix ; Comprehensive all from arrive The available temporary cross-layer similar combination correlation coefficient matrix is ​​obtained to obtain the comprehensive available cross-layer similar combination correlation coefficient matrix : .

[0016] Preferably, in step S7, the method for executing the generation and management of the seed strategy model in multiple cycles is: In each time period t, Validate all factor combinations across portfolios, including two methods: S701, directly verify all factor value combinations: verify All factor value combinations in the combination are combined to obtain the final temporary correlation coefficient matrix containing all factor value combinations and the final interim conclusion matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary conclusion matrix in turn, remove the conclusions that rank last in all indicators, and obtain the final available correlation coefficient matrix And the final available conclusion matrix ; S702, indirectly verify all factor value combinations: Verify first Combine the first half of the factor values ​​in the combination to get the temporary intermediate combination 1, and then verify The second half of the factor value combinations in the combination are combined to obtain the temporary intermediate combination 2. Finally, the temporary intermediate combination 1 and the temporary intermediate combination 2 are verified to obtain all available factor value combinations. verify When the first half of the factor values ​​in the combination are price and volume, the factor values ​​in the fixed evalu and promo combinations are The mean obtained in The second half of the factor value combination in the combination is to verify the evaluation and promotion combination, and the factor value in the fixed price and volume combination is The final temporary intermediate correlation coefficient matrix is ​​obtained , and the final temporary intermediate conclusion matrix and ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary intermediate conclusion matrix in turn, remove the conclusions that rank last in all indicators, for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the final temporary intermediate correlation coefficient matrix to obtain the available final temporary intermediate correlation coefficient matrix , and the final temporary intermediate conclusion matrix and ; verify All factor value combinations in the combination are combined to obtain the final temporary correlation coefficient matrix containing all factor value combinations and the final interim conclusion matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary conclusion matrix in turn, remove the conclusions that rank last in all indicators. For example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the final temporary correlation coefficient matrix to obtain the final available correlation coefficient matrix And the final available conclusion matrix ; The above two methods obtained All factor value combinations in the strategy model set are used as the factor correlation coefficient matrix.

[0017] Furthermore, in step S8, the method for optimizing the seed strategy model is: From the strategy model set obtained in the previous step, cross-validate various quantitative indicators and select the most suitable seed strategy model as the initial strategy model. Then perform multi-cycle single factor training, multi-cycle factor combination training, and multi-cycle seed strategy model generation in sequence to obtain a new round of seed strategy models until the convergence ConValue of the strategy model indicator reaches the target value, which can be completed after 3 to 5 rounds.

[0018] The beneficial effects of the method for efficient and automatic quantitative analysis of mall data provided by the present invention are as follows: 1. This method extends the quantitative analysis methods used in finance, securities, futures and spot, virtual currency and other fields to the mall data analysis. It comprehensively considers the characteristics of the mall's logs, products, content, activities, searches, transactions and other multi-dimensional data and the requirements of quantitative analysis methods, collects and structures and standardizes them in a reasonable way, establishes a unique indicator, factor, model system, designs the factor value space, and uses periodic, classified, and hierarchical distribution factors and model strategy training methods. It can comprehensively, accurately and efficiently analyze the impact of the mall's main data on operating indicators, which has great reference significance.

[0019] 2. The mall data quantitative analysis system built based on the quantitative analysis method mentioned in the present invention can automatically carry out each stage from factor validity verification and training, same-layer and similar factor training, cross-layer and similar factor training, seed strategy model generation and optimization until the strategy model family meets the convergence conditions and automatically exits, which can greatly reduce the workload of strategy model analysis. And when there is an update of analysis data, the factor and strategy model can be automatically trained and updated in time. 3. This method uses factors divided into periods, layers, classifications, reasonable combinations, and step-by-step screening, which can greatly reduce the number of verifications, improve data analysis efficiency, and reduce labor and time costs without losing the accuracy of the analysis results. Taking the strategy model mentioned in the article as an example, considering 10 time periods, the strategy model contains 64 factors, and the number of times each factor takes values ​​in the value space ranges from 100 to 1000. The conventional quantitative analysis method completes an iteration cycle, and the number of times the available strategy model list needs to be verified is the number of time periods × the number of times factor 1 takes values ​​in the value space × ... the number of times factor 64 takes values ​​in the value space = 2.5 × 1037 times; while this method only needs the number of time periods × (single factor training verification times + factor combination training times + strategy model generation training times) × strategy model optimization iteration rounds ≈ 1.26 × 1012 times.

[0020] 4. The time period, quantitative indicators of the strategy model, quantitative indicator priorities, factors, correlation relationships, etc. in this method are set in the configuration file and can be quickly adjusted without modifying the strategy model. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 It is the full flow chart of the quantitative analysis method of the present invention; Figure 2 The present invention is a flow chart of data collection, structuring and standardization processing; Figure 3 It is a flow chart of single factor training in multiple cycles of the present invention; Figure 4 It is a flow chart of the multi-period factor combination training of the present invention. DETAILED DESCRIPTION

[0022] Embodiment 1: like Figure 1~Figure 4 As shown, an efficient and automatic method for quantitative analysis of mall data includes the following steps: S1, collect data from the mall operation database, and perform structured and standardized processing on the collected data to build a data collection database. The data collection database includes a user database, a merchant database, a product database, a search database, a transaction order database, a payment database, a logistics database, and a marketing database. The data collection method is: Database synchronization realizes data collection: The stock data of users, merchants, products, searches, transaction orders, payments, logistics and marketing are collected by synchronizing the mall operation database once a day. This type of data is stored in a structured manner in database tables and updated daily. API interface collection: The daily incremental data of users, merchants, products, searches, transaction orders, payments, logistics, and marketing are collected by reading from the mall operation database every m minutes through the API interface; this type of data is structured and stored in database tables and updated in real time; File ETL synchronization mode collection: For article, image, video data, and log file file data, they are collected from the file system of the mall operation database in the file ETL synchronization mode; the collection cycle is once a day or every n minutes; Articles, pictures and videos include three types of articles, pictures and videos: merchants, products and marketing; Streaming data reading method collection: For intraday log data, streaming data reading is used to collect data; S2, set quantitative indicators for mall operations and manage them; S3, factor extraction and multi-level classification processing from the data collection database; S4, set the factor value space and step size; S5, establish an initial strategy model, which is an analysis model between factors and quantitative indicators of mall operations; S6, conduct multi-cycle single-factor training; S7, conduct multi-period factor combination training; S8, executes the generation and management of seed strategy models in multiple cycles, S9, optimize the seed strategy model.

[0023] Furthermore, the core data of the mall includes users, merchants, products, searches, transaction orders, payments, logistics, marketing, logs, etc. Basic data such as users, merchants, products, searches, transaction orders, payments, logistics, marketing, etc. are stored in the database, and merchant pictures and videos, product pictures and videos, and system logs are stored in files. The intraday changes in data such as merchants and products are relatively slow, while data such as users, searches, product inventory, transaction orders, payments, logistics, and marketing may change significantly during the day. The collection of log data is usually formed into files through streaming processing, and other data is stored in the database through business interfaces. Different collection and preprocessing methods are adopted according to the different storage methods of data, the daily change patterns, and the formation methods.

[0024] Preferably, in step S1, the method for performing structured processing on the collected data is: Extract the activity theme, corresponding merchant, corresponding product and generation time information of marketing articles, pictures and video data, and insert them into the marketing library; Extract the corresponding merchants, corresponding products and generation time information of merchant type articles, pictures and video data, and insert them into the merchant database; Extract the corresponding merchants, corresponding products and generation time information of the articles, pictures and video data of the product type, and insert them into the product library; Extract the log business classification and generation time information in the log file, and insert them into the user library, merchant library, product library, retrieval library, transaction order library, payment library, logistics library and marketing library according to the log business classification.

[0025] Preferably, in step S1, the method for standardizing the collected data is: Construct product attribute tables, merchant attribute tables, and marketing attribute tables; construct search tables, transaction order tables, payment tables, logistics tables, and user tables; Based on the product ID, extract the product ID, product attributes, and product attribute change record timestamp from the product database and store them in the product attribute table; Based on the product ID, extract the product ID, merchant attributes, and merchant attribute change record timestamp from the merchant database and store them in the merchant attribute table; Based on the product ID, the product ID, marketing activity attributes, and marketing activity attribute change record timestamp are extracted from the marketing database and stored in the marketing attribute table; The product attribute table, the merchant attribute table and the marketing attribute table are combined into an independent variable library, wherein the marketing attribute table and the merchant attribute table are associated with the product attribute table through the product ID; Based on the product ID, extract the product ID, search attributes, and the timestamp of the search attribute change record from the search library and store them in the search table; Based on the product ID, extract the product ID, transaction order attributes, and transaction order attribute change record timestamp from the transaction order database and store them in the transaction order table; Based on the product ID, extract the product ID, payment attributes, and payment attribute change record timestamp from the payment database and store them in the payment table; Based on the product ID, extract the product ID, logistics attributes, and logistics attribute change record timestamp from the logistics database and store them in the logistics table; Extract user attributes and user record change timestamps from the user database and store them in the user table; The user table, search table, logistics table, transaction table and payment table form a dependent variable database.

[0026] Furthermore, during the data standardization phase, duplicate rows are not deleted, missing values ​​are not filled, and abnormal data is not processed.

[0027] Preferably, in step S2, the method for setting the quantitative indicators of the mall operation and managing them is: Quantitative indicators include analyzing the number of times the strategy model is triggered within multiple time periods T (SigTimes), the probability of improving the commodity's operating indicators by 20% (IR20Ratio), the probability of improving by 3% (IR03Ratio), the probability of improvement (SigWinRatio), the probability of maximum reduction of no more than 3% (MaxLoss3), the probability of maximum reduction of no more than 7% (MaxLoss7), the maximum possible reduction percentage (MaxLoss), the maximum possible reduction days (MaxLossDays), the ratio of improvement probability or reduction probability greater than 3 (Eratio3), the Sharpe ratio (PR) and the convergence degree (ConValue); Quantitative indicator management is the management of all quantitative indicators, including adding, deleting, modifying, checking and changing priorities.

[0028] Furthermore, the mall's operating indicators include sales volume, sales revenue, profit, unit capital profit or other core goals that management personnel and product personnel are concerned about; when considering the impact of a certain factor on the commodity operating indicators, it is necessary to consider the inconsistencies in backgrounds such as different time periods, different categories, and different regions.

[0029] Preferably, in step S3, when factor extraction and multi-level classification processing are performed from the data collection database, the factors are extracted from the commodity attribute table, the marketing attribute table and the merchant attribute table, and the method is: Original factor extraction: Original factors include unit price factor and discount rate factor in commodity attributes; Statistical factor extraction: Statistical factors include historical sales factor, inventory factor, number of reviews factor and favorable review rate factor in product attributes, activity factor in marketing attributes and logistics timeliness factor in merchant attributes; The factor multi-level classification process includes classifying factors by: Divide product categories into three levels, including level 1 product type, level 2 product type, and level 3 product type; The factors corresponding to the original factors and statistical factors are extracted at each layer, namely, the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the first-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the second-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the third-level commodity type.

[0030] Preferably, in step S4, the method for setting the factor value space and step size is: Corresponding histograms are established for the unit price factor, discount rate factor, historical sales factor, and inventory factor. The same-category and cross-category products are analyzed, and the intervals and step lengths of the initial value spaces of different factors are set; Corresponding histograms are established for the evaluation number factor, the favorable comment rate factor, and the logistics timeliness factor, and cross-category products are analyzed to set the interval and step length constants of the initial value space of different factors; For activity factors, a histogram of activity frequency is established, and analysis is performed on applicable merchandise, activity start time, and activity end time. Constants for the intervals and step lengths of the initial value spaces of different factors are set.

[0031] Preferably, in step S5, the method for establishing the initial strategy model is: S501, quantitative indicators: The operation indicators of the mall include sales volume, sales revenue, profit, unit capital profit or other core goals that business personnel and product personnel pay attention to. When considering the impact of a certain factor on the commodity operation indicators, it is necessary to consider the inconsistency of the macro environment in different time periods; based on different time periods t, the impact of the factor on the commodity operation indicators is quantified to form a configurable time period matrix Tnx2 and a quantitative indicator matrix Gnx11; set the priority of each quantitative indicator and form a priority matrix Pn×3; ; The Tnx2 time period matrix represents the time period used to verify the strategy model. It has n rows and 2 columns in each row. Each row represents an optional time period. The first element represents the start time and the second element represents the end time. When the second element is "now", it means that the end time is the latest time. Quantitative indicators can be maintained in the configuration file. The quantitative indicator details of all strategy models in the same strategy model family are as follows: ; The Gnx11 quantitative indicator matrix has n rows, each with 11 columns. Each row represents a quantitative indicator set of a strategy, and each element in each row represents a specific indicator value; The following are examples of the priority details of each quantitative indicator: ; In the formula, P n×3 The matrix represents the priority matrix of each quantitative indicator; Z1 and Z2 are the serial numbers of the quantitative indicators, and IR20Ratio and SigWinRatio are the names of the quantitative indicators; S502, construct factor matrix: The factors are divided into 4 layers, including the original layer, the first-level commodity type layer, the second-level commodity type layer and the third-level commodity type layer; each layer has 8 factors, including unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor; the 8 factors are divided into 4 categories, including price category, quantity category, evaluation category and marketing category; Set each factor to have positive and negative impacts on product operation indicators; Set factor attributes including factorID, name, direction, type, level, start value, end value and step size; construct factor matrix , , ; Among them, F0 is the original layer factor matrix, with a total of 16 rows and 8 columns; each row represents the specific configuration of a factor, the first column of each row is the sequence number, the second column is the alias, and the third column is the use direction. When it is up, the specific value of the factor in the quantitative analysis model must be greater than a certain value to take effect; when it is down, the specific value of the factor in the quantitative analysis model must be greater than a certain value to take effect; the fourth column is the type, the fifth column is the level, the original layer factor level value is 0, the sixth column is the starting value, the seventh column is the ending value, and the eighth column is the step size; S503, constructing an initial strategy model: Establish the correlation between the initial strategy model and the factors to form a step-length validation correlation matrix W; the correlation coefficient of the correlation relationship ranges from 0.00 uncorrelated to 1.00 fully correlated, and the initial value is 1.00; The step verification relationship matrix can be maintained in the configuration file. The details are as follows: ; In the formula, the correlation matrix includes n rows and 64 columns, and each row represents the correlation between a quantitative analysis model and a factor. As shown above, F0, F1, F2, and F3 have a total of 64 factors. The first column of each row of the correlation matrix represents the degree of correlation between a quantitative analysis model and the first factor of F0, 0 for no correlation and 1 for complete correlation. The second column of each row of the correlation matrix represents the degree of correlation between a quantitative analysis model and the second factor of F0, 0 for no correlation and 1 for complete correlation, and so on. The formula of the quantitative analysis model is: G×T=S(W×(F0+F1+F2+F3)); Where G represents the quantitative indicator matrix, T represents the time period matrix, W represents the step verification correlation matrix, S represents the quantitative analysis model algorithm; F0, F1, F2 and F3 are factor matrices.

[0032] Preferably, in step S6, the method for performing multi-cycle single factor training is: In each time period t, the correlation matrix is ​​verified by step size within the starting and ending values ​​of the factors in the factor matrix The validity of the middle factor is verified by fixing the values ​​of other factors as the mean μ and obtaining the temporary conclusion matrix and the temporary correlation coefficient matrix ; Taking the w008 factor as an example, we get the temporary conclusion matrix and temporary correlation coefficient matrix: According to the indicator priority matrix P n×3 , sort and verify the comparison temporary conclusion matrix For each conclusion in the temporary correlation coefficient matrix, remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary correlation coefficient matrix to obtain a usable temporary conclusion matrix. and the available correlation coefficient matrix ; Verify all 64 factors in the factor matrix in turn and obtain a usable temporary conclusion matrix and the available correlation coefficient matrix ; Comprehensive all from arrive The available correlation coefficient matrix is ​​obtained by .

[0033] Preferably, in step S7, the method for performing multi-period internal factor combination training is: Among different factors, factors of the same type and at the same level have the strongest correlation, followed by factors of the same type and at different levels, and then factors of different types and at the same level. Factors of different types and at different levels have weaker correlation. Therefore, combination training is first performed on factors of the same type and at the same level, such as price, volume, evaluation, and promo factors of the original layer, and then combination training is performed on factors of the same type and across levels. Combination training of factors of the same type and level: In each time period t, Valid factor combinations are verified by combination within the same layer; During verification, the values ​​of other factors are fixed to The mean obtained in the above equation is used to obtain the temporary same-layer and same-type combination conclusion matrix and temporary same-layer and same-type combination correlation coefficient matrix ; Taking the promo combination factor as an example, the temporary same-level and same-type combination correlation coefficient matrix is ​​as follows: ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the temporary same-level and similar combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary same-level and similar combination correlation coefficient matrix, and obtain the available temporary same-level and similar combination conclusion matrix And the available same-level and same-type combination correlation coefficient matrix ; Comprehensive all from arrive The available temporary same-level and same-type combination correlation coefficient matrix is ​​obtained to obtain the comprehensive available same-level and same-type combination correlation coefficient matrix ; Cross-layer similar combination training: In each time period t, Verify effective factor combinations by combination across layers; During verification, the values ​​of other factors are fixed to the mean values ​​obtained in WCS(0−>n)n×64, and the temporary cross-layer similar combination conclusion matrix is ​​obtained. and temporary cross-layer homogeneous combination correlation coefficient matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the temporary cross-layer similar combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the temporary cross-layer similar combination correlation coefficient matrix to obtain the available temporary cross-layer similar combination conclusion matrix and temporary cross-layer homogeneous combination correlation coefficient matrix ; Verify the cross-layer price, volume, evaluation, and promo combinations of all layers in the factor matrix in turn to obtain the final available temporary cross-layer similar combination conclusion matrix And the final available temporary cross-layer homogeneous combination correlation coefficient matrix ; Comprehensive all from arrive The available temporary cross-layer similar combination correlation coefficient matrix is ​​obtained to obtain the comprehensive available cross-layer similar combination correlation coefficient matrix : .

[0034] Preferably, in step S7, the method for executing the generation and management of the seed strategy model in multiple cycles is: In each time period t, Validate all factor combinations across portfolios, including two methods: S701, directly verify all factor value combinations: verify All factor value combinations in the combination are combined to obtain the final temporary correlation coefficient matrix containing all factor value combinations and the final interim conclusion matrix ; According to the indicator priority matrix P n×3, sort and verify the conclusions in the final temporary conclusion matrix in turn, remove the conclusions that rank last in all indicators, and obtain the final available correlation coefficient matrix And the final available conclusion matrix ; S702, indirectly verify all factor value combinations: Verify first Combine the first half of the factor values ​​in the combination to get the temporary intermediate combination 1, and then verify The second half of the factor value combinations in the combination are combined to obtain the temporary intermediate combination 2. Finally, the temporary intermediate combination 1 and the temporary intermediate combination 2 are verified to obtain all available factor value combinations. verify When the first half of the factor values ​​in the combination are price and volume, the factor values ​​in the fixed evalu and promo combinations are The mean obtained in The second half of the factor value combination in the combination is to verify the evaluation and promotion combination, and the factor value in the fixed price and volume combination is The final temporary intermediate correlation coefficient matrix is ​​obtained , and the final temporary intermediate conclusion matrix and ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary intermediate conclusion matrix in turn, remove the conclusions that rank last in all indicators, for example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the final temporary intermediate correlation coefficient matrix to obtain the available final temporary intermediate correlation coefficient matrix , and the final temporary intermediate conclusion matrix and ; verify All factor value combinations in the combination are combined to obtain the final temporary correlation coefficient matrix containing all factor value combinations and the final interim conclusion matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary conclusion matrix in turn, remove the conclusions that rank last in all indicators. For example, when comparing two conclusions, remove the conclusion rows where both IR20Ratio and SigWinRatio lag behind, and simultaneously remove the corresponding correlation coefficient rows in the final temporary correlation coefficient matrix to obtain the final available correlation coefficient matrix And the final available conclusion matrix ; The above two methods obtained All factor value combinations in the strategy model set are used as the factor correlation coefficient matrix.

[0035] Furthermore, in step S8, the method for optimizing the seed strategy model is: From the strategy model set obtained in the previous step, cross-validate various quantitative indicators and select the most suitable seed strategy model as the initial strategy model. Then perform multi-cycle single factor training, multi-cycle factor combination training, and multi-cycle seed strategy model generation in sequence to obtain a new round of seed strategy models until the convergence ConValue of the strategy model indicator reaches the target value, which can be completed after 3 to 5 rounds.

[0036] Embodiment 2: In step S4, the method for setting the factor value space and step size is: Corresponding histograms are established for the unit price factor, discount rate factor, historical sales factor, and inventory factor. The same-category and cross-category products are analyzed, and the intervals and step lengths of the initial value spaces of different factors are set; Corresponding histograms are established for the evaluation number factor, the favorable comment rate factor, and the logistics timeliness factor, and cross-category products are analyzed to set the interval and step length constants of the initial value space of different factors; For activity factors, a histogram of activity frequency is established, and analysis is performed on applicable merchandise, activity start time, and activity end time. Constants for the intervals and step lengths of the initial value spaces of different factors are set.

[0037] Unit specification price factor: A unit specification price range histogram is established. Through the analysis of the same category and cross-category products, all unit specification prices show an obvious log-normal distribution with μ=0 and extremely small δ; the initial value space of the unit specification price distribution factor can be [0,1000], with a step size of 1.

[0038] Discount rate factor: A discount rate histogram is established. Through the analysis of the same category and cross-category products, all discount rates show an obvious left-skewed distribution with minimal delta. The minimum discount rate is 0.00 and the maximum is 1.00; the initial value space of the discount rate factor can be [0.00, 1.00], with a step size of 0.01.

[0039] Historical sales factor: A historical sales histogram is established. Through the analysis of the same category and cross-category products, all sales show an obvious log-normal distribution with minimal delta. The initial value space of the historical sales factor can be [0,1000], with a step size of 1.

[0040] Inventory Factor: The inventory histogram is established. Through the analysis of the same category and cross-category products, all inventory levels show obvious randomness; there is no obvious pattern in the inventory levels of each product, and it is unclear whether there is a correlation between inventory levels and indicators. In summary, the initial value space of the inventory factor can be [0,10000], with a step size of 1.

[0041] Evaluation factor: The evaluation number histogram is established. Through the analysis of cross-category products, the evaluation numbers of all products show an obvious log-normal distribution with a minimum delta. The minimum value is 0, the maximum value is 502, and 99.99% of the product evaluation numbers are within 500; the initial value space of the evaluation number factor can be [0,500], with a step size of 1.

[0042] Favorable Rate Factor: A histogram of the favorable review rate is established. Through the analysis of cross-category products, all favorable review rates show obvious randomness, and most values ​​are close to 100%. The minimum value is 0% and the maximum value is 100%. The initial value space of the favorable review rate factor can be [0.00, 1.00], with a step size of 0.01.

[0043] Activity Factor: A histogram of the number of activities is established. By analyzing the applicable products, the start time of the activity, and the end time of the activity, all the numbers of activities show obvious randomness; the initial value space of the activity number factor can be [0,200], with a step size of 1.

[0044] Logistics Timeliness Factor A histogram of logistics timeliness is established. Through the analysis of cross-category products, all logistics timeliness shows obvious randomness, and most values ​​are close to 5 points. The minimum value is 0 points and the maximum value is 5 points; the initial value space of the logistics timeliness factor can be [0.00,5.00], with a step size of 0.01.

Claims

1. An efficient and automatic method for quantitative analysis of mall data, characterized in that: The following steps are involved: S1, collect data from the mall operation database, and perform structured and standardized processing on the collected data to build a data collection database. The data collection database includes a user database, a merchant database, a product database, a search database, a transaction order database, a payment database, a logistics database, and a marketing database. The data collection method is: Database synchronization realizes data collection: The stock data of users, merchants, products, searches, transaction orders, payments, logistics and marketing are collected by synchronizing the mall operation database once a day. The stock data of users, merchants, products, searches, transaction orders, payments, logistics and marketing are stored in a structured manner in database tables and updated daily. API interface collection: The daily incremental data of users, merchants, products, searches, transaction orders, payments, logistics, and marketing are collected by reading from the mall operation database every m minutes through the API interface; the daily incremental data of users, merchants, products, searches, transaction orders, payments, logistics, and marketing are stored in a structured manner through database tables and updated in real time; File ETL synchronization mode collection: For article, image, video data, and log file file data, they are collected from the file system of the mall operation database in the file ETL synchronization mode; the collection cycle is once a day or every n minutes; Articles, pictures and videos include three types of articles, pictures and videos: merchants, products and marketing; Streaming data reading method collection: For intraday log data, streaming data reading is used to collect data; S2, set quantitative indicators for mall operations and manage them; S3, factor extraction and multi-level classification processing from the data collection database; S4, set the factor value space and step size; S5, establish an initial strategy model, which is an analysis model between factors and quantitative indicators of mall operations; S6, conduct multi-cycle single-factor training; S7, conduct multi-period factor combination training; S8, executes the generation and management of seed strategy models in multiple cycles, S9, optimize the seed strategy model.

2. According to claim 1, a method for efficient and automatic quantitative analysis of shopping mall data is characterized in that: In step S1, the method for performing structured processing on the collected data is: Extract the activity theme, corresponding merchant, corresponding product and generation time information of marketing articles, pictures and video data, and insert them into the marketing library; Extract the corresponding merchants, corresponding products and generation time information of merchant type articles, pictures and video data, and insert them into the merchant database; Extract the corresponding merchants, corresponding products and generation time information of the articles, pictures and video data of the product type, and insert them into the product library; Extract the log business classification and generation time information in the log file, and insert them into the user library, merchant library, product library, retrieval library, transaction order library, payment library, logistics library and marketing library according to the log business classification.

3. The method for efficient and automatic quantitative analysis of mall data according to claim 2, characterized in that: In step S1, the method for standardizing the collected data is: Construct product attribute tables, merchant attribute tables, and marketing attribute tables; construct search tables, transaction order tables, payment tables, logistics tables, and user tables; Based on the product ID, extract the product ID, product attributes, and product attribute change record timestamp from the product database and store them in the product attribute table; Based on the product ID, extract the product ID, merchant attributes, and merchant attribute change record timestamp from the merchant database and store them in the merchant attribute table; Based on the product ID, the product ID, marketing activity attributes, and marketing activity attribute change record timestamp are extracted from the marketing database and stored in the marketing attribute table; The product attribute table, the merchant attribute table and the marketing attribute table are combined into an independent variable library, wherein the marketing attribute table and the merchant attribute table are associated with the product attribute table through the product ID; Based on the product ID, extract the product ID, search attributes, and the timestamp of the search attribute change record from the search library and store them in the search table; Based on the product ID, extract the product ID, transaction order attributes, and transaction order attribute change record timestamp from the transaction order database and store them in the transaction order table; Based on the product ID, extract the product ID, payment attributes, and payment attribute change record timestamp from the payment database and store them in the payment table; Based on the product ID, extract the product ID, logistics attributes, and logistics attribute change record timestamp from the logistics database and store them in the logistics table; Extract user attributes and user record change timestamps from the user database and store them in the user table; The user table, search table, logistics table, transaction table and payment table form a dependent variable database.

4. The method for efficient and automatic quantitative analysis of mall data according to claim 3 is characterized in that: In step S2, the method for setting the quantitative indicators of the mall operation and managing them is: Quantitative indicators include analyzing the number of times the strategy model is triggered within multiple time periods T (SigTimes), the probability of improving the commodity's operating indicators by 20% (IR20Ratio), the probability of improving by 3% (IR03Ratio), the probability of improvement (SigWinRatio), the probability of maximum reduction of no more than 3% (MaxLoss3), the probability of maximum reduction of no more than 7% (MaxLoss7), the maximum possible reduction percentage (MaxLoss), the maximum possible reduction days (MaxLossDays), the ratio of improvement probability or reduction probability greater than 3 (Eratio3), the Sharpe ratio (PR) and the convergence degree (ConValue); Quantitative indicator management is the management of all quantitative indicators, including adding, deleting, modifying, checking and changing priorities.

5. The method for efficient and automatic quantitative analysis of mall data according to claim 4 is characterized in that: In step S3, when factor extraction and multi-level classification processing are performed from the data collection database, factors are extracted from the commodity attribute table, the marketing attribute table and the merchant attribute table, and the method is: Original factor extraction: Original factors include unit price factor and discount rate factor in commodity attributes; Statistical factor extraction: Statistical factors include historical sales factor, inventory factor, number of reviews factor and favorable review rate factor in product attributes, activity factor in marketing attributes and logistics timeliness factor in merchant attributes; The factor multi-level classification process includes classifying factors by: Divide product categories into three levels, including level 1 product type, level 2 product type, and level 3 product type; The factors corresponding to the original factors and statistical factors are extracted at each layer, namely, the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the first-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the second-level commodity type; the unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor under the third-level commodity type.

6. The method for efficient and automatic quantitative analysis of mall data according to claim 5, characterized in that: In step S4, the method for setting the factor value space and step size is: Corresponding histograms are established for the unit price factor, discount rate factor, historical sales factor, and inventory factor. The same-category and cross-category products are analyzed, and the intervals and step lengths of the initial value spaces of different factors are set; Corresponding histograms are established for the evaluation number factor, the favorable comment rate factor, and the logistics timeliness factor, and cross-category products are analyzed to set the interval and step length constants of the initial value space of different factors; For activity factors, a histogram of activity frequency is established, and analysis is performed on applicable merchandise, activity start time, and activity end time. Constants for the intervals and step lengths of the initial value spaces of different factors are set.

7. The method for efficient and automatic quantitative analysis of mall data according to claim 6, characterized in that: In step S5, the method for establishing the initial strategy model is: S501, quantitative indicators: Based on different time periods t, the influence of factors on commodity operation indicators is quantified to form a configurable time period matrix Tnx2 and a quantitative indicator matrix Gnx11; each quantitative indicator is given a priority and a priority matrix Pn×3 is formed; S502, construct factor matrix: The factors are divided into 4 layers, including the original layer, the first-level commodity type layer, the second-level commodity type layer and the third-level commodity type layer; each layer has 8 factors, including unit specification price factor, discount rate factor, historical sales factor, inventory factor, evaluation number factor, favorable rate factor, activity factor and logistics timeliness factor; the 8 factors are divided into 4 categories, including price category, quantity category, evaluation category and marketing category; Set each factor to have positive and negative impacts on product operation indicators; Set factor attributes including factorID, name, direction, type, level, start value, end value and step size; construct factor matrix , , ; S503, constructing an initial strategy model: Establish the correlation between the initial strategy model and the factors to form a step-length validation correlation matrix W; the correlation coefficient of the correlation ranges from 0.00 uncorrelated to 1.00 fully correlated, and the initial value is 1.00; The formula of the quantitative analysis model is: G×T=S(W×(F0+F1+F2+F3)); Where G represents the quantitative indicator matrix, T represents the time period matrix, W represents the step verification correlation matrix, S represents the quantitative analysis model algorithm, and F0, F1, F2 and F3 are factor matrices.

8. The method for efficient and automatic quantitative analysis of mall data according to claim 7, characterized in that: In step S6, the method for performing multi-cycle single factor training is: In each time period t, the correlation matrix is ​​verified by step size within the starting and ending values ​​of the factors in the factor matrix The validity of the middle factor is verified by fixing the values ​​of other factors as the mean μ and obtaining the temporary conclusion matrix and the temporary correlation coefficient matrix ; According to the indicator priority matrix P n×3 , sort and verify the comparison temporary conclusion matrix Among the conclusions, remove the conclusions that rank last in all indicators; verify all 64 factors in the factor matrix in turn, and obtain all available temporary conclusion matrices and available correlation coefficient matrices; Combining all available temporary conclusion matrices and available correlation coefficient matrices, we can obtain a comprehensive available correlation coefficient matrix .

9. The method for efficient and automatic quantitative analysis of mall data according to claim 8, characterized in that: In step S7, the method for performing multi-period internal factor combination training is: First, conduct combined training on factors of the same type and level, and then conduct combined training on factors of the same type and level; Combination training of factors of the same type and level: In each time period t, Valid factor combinations are verified by combination within the same layer; During verification, the values ​​of other factors are fixed to The mean obtained in the above equation is used to obtain the temporary same-layer and same-type combination conclusion matrix and temporary same-layer and same-type combination correlation coefficient matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the temporary same-layer and same-type combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; All available temporary same-layer and same-type combination correlation coefficient matrices are integrated to obtain a comprehensive available same-layer and same-type combination correlation coefficient matrix; Cross-layer similar combination training: In each time period t, Verify effective factor combinations by combination across layers; During verification, the values ​​of other factors are fixed to The mean obtained in the above equation is used to obtain the temporary cross-layer similar combination conclusion matrix and temporary cross-layer homogeneous combination correlation coefficient matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the temporary cross-layer similar combination conclusion matrix in turn, and remove the conclusions that rank last in all indicators; Comprehensive all The available temporary cross-layer similar combination correlation coefficient matrix is ​​obtained to obtain the comprehensive available cross-layer similar combination correlation coefficient matrix .

10. The method for efficient and automatic quantitative analysis of mall data according to claim 9, characterized in that: In step S7, the method for executing the generation and management of the seed strategy model in multiple cycles is: In each time period t, Validate all factor combinations across portfolios, including two methods: S701, directly verify all factor value combinations: verify All factor value combinations in the combination are combined to obtain the final temporary correlation coefficient matrix containing all factor value combinations and the final interim conclusion matrix ; According to the indicator priority matrix P n×3 , sort and verify the conclusions in the final temporary conclusion matrix in turn, remove the conclusions that rank last in all indicators, and obtain the final available correlation coefficient matrix And the final available conclusion matrix ; S702, indirectly verify all factor value combinations: Verify first Combine the first half of the factor values ​​in the combination to get the temporary intermediate combination 1, and then verify The second half of the factor value combinations in the combination are combined to obtain the temporary intermediate combination 2. Finally, the temporary intermediate combination 1 and the temporary intermediate combination 2 are verified to obtain all available factor value combinations. The two methods obtained All factor value combinations in the strategy model set are used as the factor correlation coefficient matrix.

Images