Commodity uploading screening system and method based on data analysis

Abstract

The invention relates to the technical field of commodity management, in particular to a commodity upload screening system and method based on data analysis, and the system comprises a contribution distribution analysis module which collects user purchase data of commodities, ranks the user purchase data of the commodities according to proportions, converts the user purchase data into a commodity user purchase cumulative contribution curve, and transmits the cumulative contribution curve to a database; and generating user contribution cumulative distribution data. According to the invention, through grouping arrangement and amount distribution statistics of user purchase data, in combination with formatting processing and accumulated contribution curve generation, contribution distribution of users in commodity purchase can be accurately quantified, and commodities with balanced user distribution or high market potential can be effectively screened out. And the limitation that potential commodities may be missed in simple sales volume sorting is avoided. On the basis of purchase behavior analysis, commodity purchase concentration degree attributes are further calculated, commodities meeting concentration degree requirements are accurately screened through integral values and Gini coefficients, and quantitative standards are provided for recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity management, and in particular to a commodity broadcast screening system and method based on data analysis. Background Art

[0002] The field of commodity management technology refers to the management and technical means involved in each link of the commodity life cycle, including commodity selection, pricing, inventory management, promotion strategy, sales analysis, etc. This field improves the efficiency and accuracy of commodity management through data-driven and information-based means, helping enterprises optimize commodity structure, control costs, increase sales and enhance market competitiveness.

[0003] Among them, the product live broadcast screening system is a tool that combines algorithms and data analysis, which is specifically used to help e-commerce platforms or merchants select products suitable for live broadcast promotion from a large number of products. Its main purpose is to screen products with high conversion potential or high attention through comprehensive analysis of product sales data, audience preferences, market trends and other information, so as to improve the transaction rate of live broadcasts.

[0004] It is difficult for existing technologies to effectively process the contribution distribution characteristics in the purchase data of commodity users, which makes it difficult to accurately distinguish commodities with uneven user contributions. For example, if the ranking is based only on sales volume, it is easy to ignore commodities with balanced user distribution but potential, thus failing to tap into more market value. In commodity screening, existing technologies lack quantitative analysis of purchase concentration, making it difficult to screen out commodities that meet the concentration requirements and have promotion value through indicators. For example, commodities with even user distribution may be ignored because of low total sales. In the identification of commodity collaborative relationships, it is difficult to effectively combine purchase behavior with evaluation data for cross-analysis, resulting in the recommendation results being unable to accurately reflect the actual correlation between commodities. For example, commodities with high correlation may not form a combination recommendation because the evaluation data is not fully utilized. For unpopular commodities, it is difficult for existing technologies to establish a systematic analysis model through user interaction data, and the potential value of low-attention commodities is difficult to quantify, thereby limiting the mining and recommendation of unpopular commodities. In addition, during the recommendation process, abnormal sales data such as brushing will appear, reducing the accuracy of the recommendation system and user experience satisfaction. Summary of the invention

[0005] In order to solve the technical problems existing in the prior art, the embodiments of the present invention provide a product broadcast screening system and method based on data analysis. The technical solution is as follows:

[0006] A commodity broadcast screening system based on data analysis, the system comprising:

[0007] The contribution distribution analysis module collects the user purchase data of the product, sorts the user purchase data of the product by proportion, converts it into the cumulative contribution curve of the user purchase of the product, and generates the user contribution cumulative distribution data;

[0008] The commodity concentration optimization module calculates the Gini coefficient of the commodity based on the integral value of the commodity user purchase cumulative contribution curve in the user contribution cumulative distribution data, screens commodities with concentration within a threshold range, and generates a commodity concentration screening result;

[0009] The product collaborative mining module extracts the user purchase records and user evaluation data of the screened products based on the product concentration screening results, identifies the cluster center products and related products in the product similarity matrix through the co-occurrence purchase times of the products and the overlap of user evaluations, and generates product collaborative combination data;

[0010] The unpopular product identification module builds a user interaction network of products based on product collaborative combination data, adds unpopular products as nodes to the user interaction network, connects the interaction paths between users and products according to the behavioral data weights of unpopular products, and generates a potential unpopular product set;

[0011] The product recommendation pool optimization module obtains the current product sales time series data and constructs a product sales distribution curve based on the potential unpopular product set, identifies abnormal peaks and deviation intervals of the product sales distribution curve, and generates an optimized live broadcast recommendation product pool.

[0012] The improvements of the present invention are that the user contribution cumulative distribution data includes user purchase ratio ranking, commodity cumulative purchase ratio curve and commodity amount distribution mapping; the commodity concentration screening results include a commodity list that meets the concentration threshold, priority ranking scores and inventory status data; the commodity synergy combination data includes a commodity synergy strength matrix, commodity clustering center identification data and conversion rate correlation results; the potential unpopular commodity set includes unpopular commodity node network centrality data, interaction path weight distribution data and time series consumption preference characteristics; the live broadcast recommended commodity pool includes sales distribution optimized commodity data, commodity marking results after anomaly elimination and a recommended preferred commodity set.

[0013] The present invention is improved in that the contribution distribution analysis module comprises:

[0014] The user purchase data collection submodule is based on the user purchase data of commodities, including user purchase records and corresponding amount information, and groups the user purchase records by commodity, counts the distribution of the amount in each group, and performs data formatting to generate commodity user purchase records and amount distribution data;

[0015] The user purchase data accumulation submodule accumulates the purchase amount of each commodity user based on the commodity user purchase records and amount distribution data, calculates the cumulative proportion of user purchases, sorts the data by proportion, and generates the cumulative proportion data of commodity user purchases;

[0016] The contribution curve generation submodule converts the cumulative proportion data of user purchases of the commodities into a cumulative contribution curve of user purchases of the commodities, and generates cumulative distribution data of user contributions in combination with the user purchase situation of each commodity.

[0017] The present invention is improved in that the commodity concentration optimization module comprises:

[0018] The commodity concentration attribute analysis submodule calculates the integral value of the commodity user purchase cumulative contribution curve based on the user contribution cumulative distribution data, sums up the integral value to represent the concentration attribute of each commodity, and generates the commodity Gini coefficient;

[0019] The commodity concentration screening submodule compares the Gini coefficient with the set concentration threshold based on the commodity Gini coefficient, screens commodities whose Gini coefficients are within the threshold range, and generates concentration screening commodity data;

[0020] The priority sorting submodule filters the commodity data based on the concentration, performs sales statistics analysis on the filtered commodities, obtains the sales data, inventory data and user evaluation data of the commodities, sorts the commodities according to the set weights, and generates commodity concentration screening results.

[0021] The present invention is improved in that the integral value of the cumulative contribution curve of commodity user purchases is calculated by using the formula:

[0022]

[0023] Get the integral value I, which represents the commodity concentration attribute;

[0024] Where x represents the cumulative proportion of the purchase amount, F(x) represents the function of the user contribution cumulative curve, ∫ represents the integral symbol, which is used to calculate the area under the curve, [0, 1] represents the range of the integral, and dx represents the change value of the cumulative proportion x;

[0025] According to the integral value I, the formula is adopted:

[0026] G = 2·I-1;

[0027] Calculate the Gini coefficient G;

[0028] Among them, the value range of G is 0≤G≤1. When G is closer to 0, it means that the purchase distribution is more uniform. When it is closer to 1, it means that the purchase concentration is higher.

[0029] The present invention is improved in that the commodity collaborative mining module comprises:

[0030] The user purchase record extraction submodule extracts the user purchase record data of the screened products based on the product concentration screening result, groups and organizes the user purchase information by product, and generates user purchase and evaluation data of the screened products;

[0031] The similarity matrix construction submodule is based on the user purchase and evaluation data of the selected products, calculates the synergy strength value of each pair of products by counting the number of co-occurrence purchases and the overlap of user evaluations between the products, and quantifies the correlation between each pair of products based on the synergy strength value to construct a similarity matrix of the products;

[0032] The synergistic combination generation submodule identifies the cluster center products and related products in the similarity matrix of the products, analyzes the cross data of product purchase conversion rate and user behavior, determines the synergistic combination relationship between products, and obtains product synergistic combination data.

[0033] The present invention is improved in that the synergy strength value of each pair of commodities is calculated by using the formula:

[0034]

[0035] Get S ij Represents the synergy strength value S of product i and product j ij ;

[0036] Among them, C ij represents the number of co-occurrence purchases of product i and product j, R ij represents the number of overlapping user reviews for product i and product j, C i and C j Represents the total number of purchases of product i and product j, R i and R j Respectively represent the total number of evaluations of product i and product j, W 1 and W 2 They are the co-occurrence purchase weight and user review overlap weight respectively.

[0037] The present invention is improved in that the unpopular product identification module comprises:

[0038] The user interaction network construction submodule establishes an interaction network between products and users based on the product collaborative combination data, extracts click behavior data, collection behavior data and purchase behavior data of unpopular products, connects the interaction paths between nodes according to the weight of the behavior data, and generates product user interaction network data;

[0039] The unpopular product analysis submodule counts the interaction frequency between unpopular products and other products or user nodes based on the product user interaction network data, analyzes the position relationship and neighbor relationship of unpopular product nodes in the interaction network, and generates unpopular product network centrality data;

[0040] The unpopular product marking submodule compares the centrality value of the unpopular product node with a preset centrality threshold based on the unpopular product network centrality data, marks the unpopular product nodes whose centrality values ​​meet the threshold range, and generates a potential unpopular product set.

[0041] The present invention is improved in that the product recommendation pool optimization module includes:

[0042] The commodity sales data collection submodule extracts the sales time series data of each commodity based on the potential unpopular commodity set, organizes the original sales data by segment according to the time axis, draws the commodity sales distribution curve by connecting the sales data of the time periods, and generates commodity sales distribution curve data;

[0043] The sales anomaly elimination submodule identifies the abnormal peak points and deviation intervals in the sales distribution based on the sales distribution curve data of the commodities, extracts the sales records and user behavior characteristic data of the corresponding abnormal commodities, cross-validates and analyzes the sales records of the abnormal commodities with the user behavior characteristics, eliminates the sales records of the commodities determined to be abnormal, and generates optimized commodity sales data;

[0044] The recommendation pool marking submodule integrates the optimized product sales records with the potential unpopular product set based on the optimized product sales data, marks the optimized product set as being in a recommendable state, and generates an optimized live broadcast recommendation product pool.

[0045] The method for selecting products to be broadcasted based on data analysis is performed based on the above-mentioned system for selecting products to be broadcasted based on data analysis, and includes the following steps:

[0046] S1: Based on the product purchase data, generate the product user purchase cumulative contribution curve after sorting, and obtain the user contribution cumulative distribution data;

[0047] S2: Based on the cumulative distribution data of user contributions, the Gini coefficient is calculated using the cumulative contribution curve, and the products whose concentrations meet the threshold range are screened to generate a product concentration screening result;

[0048] S3: Based on the product concentration screening results, analyze the co-occurrence purchases and evaluation overlaps, construct a similarity matrix, extract cluster center products and their related products, and generate product synergy combination data;

[0049] S4: Build a user interaction network based on the synergistic combination data of the products, identify potential unpopular products, optimize the sales curve to screen products, and generate an optimized live broadcast recommendation product pool.

[0050] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0051] By grouping and sorting user purchase data and sum distribution statistics, combined with formatting processing and the generation of cumulative contribution curves, it is possible to accurately quantify the contribution distribution of users in product purchases, effectively screen out products with balanced user distribution or high market potential, and avoid the limitation that simple sales ranking may miss potential products. Based on the purchase behavior analysis, the product purchase concentration attribute is further calculated, and the integral value and Gini coefficient are used to accurately screen out products that meet the concentration requirements, providing a quantitative standard for recommendation. The identification of the synergistic relationship between products is through the cross-analysis of the co-occurrence purchase frequency and the overlapping data of user evaluations, and the similarity matrix is ​​constructed in combination with the synergy strength value, which can clearly reveal the correlation and combination potential between products and avoid the limitations that may be caused by the recommendation of a single product. For unpopular products, by constructing a user interaction network and analyzing the interaction frequency and node centrality, potential unpopular products can be identified. Combined with time series analysis and abnormal sales data elimination, the optimized product sales data is integrated with the potential product set to form a product pool with more recommendation value, further improving the accuracy and reliability of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0053] Figure 1 It is a system module diagram of the present invention;

[0054] Figure 2 It is a system framework diagram of the present invention;

[0055] Figure 3 A schematic diagram of a contribution distribution analysis module of the present invention;

[0056] Figure 4 It is a schematic diagram of a commodity concentration optimization module of the present invention;

[0057] Figure 5 It is a schematic diagram of the commodity collaborative mining module of the present invention;

[0058] Figure 6 This is a schematic diagram of a module for identifying unpopular products according to the present invention;

[0059] Figure 7 Schematic diagram of the commodity recommendation pool optimization module of the present invention

[0060] Figure 8 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0062] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0063] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding, relevant" and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0064] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be written in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0065] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0066] See also Figure 1 The embodiment of the present invention provides a commodity broadcast screening system based on data analysis, the system comprising:

[0067] The contribution distribution analysis module collects the user purchase data of the product, sorts the user purchase data of the product by proportion, converts it into the cumulative contribution curve of the user purchase of the product, and generates the user contribution cumulative distribution data;

[0068] The commodity concentration optimization module calculates the Gini coefficient of the commodity based on the integral value of the commodity user purchase cumulative contribution curve in the user contribution cumulative distribution data, screens the commodities whose concentration is within the threshold range, and generates the commodity concentration screening results;

[0069] The product collaborative mining module extracts the user purchase records and user evaluation data of the screened products based on the product concentration screening results. It identifies the cluster center products and related products in the product similarity matrix through the co-occurrence purchase times and user evaluation overlaps, and generates product collaborative combination data.

[0070] The unpopular product identification module builds a user interaction network of products based on product collaborative combination data, adds unpopular products as nodes to the user interaction network, connects the interaction paths between users and products according to the behavioral data weights of unpopular products, and generates a potential unpopular product set;

[0071] The product recommendation pool optimization module obtains the current product sales time series data and constructs the product sales distribution curve based on the set of potential unpopular products, identifies the abnormal peaks and deviation intervals of the product sales distribution curve, and generates an optimized live broadcast recommendation product pool.

[0072] The cumulative distribution data of user contributions includes the user purchase ratio ranking, the cumulative purchase ratio curve of goods and the distribution mapping of goods amount. The product concentration screening results include the list of goods that meet the concentration threshold, the priority ranking score and the inventory status data. The product synergy combination data includes the product synergy strength matrix, the product clustering center identification data and the conversion rate correlation results. The potential unpopular product set includes the network centrality data of unpopular product nodes, the interaction path weight distribution data and the time series consumption preference characteristics. The live broadcast recommended product pool includes the sales distribution optimized product data, the product marking results after anomaly elimination and the recommended preferred product set.

[0073] See also Figure 2 and Figure 3 , the contribution distribution analysis module includes:

[0074] The user purchase data collection submodule is based on the user purchase data of commodities, including user purchase records and corresponding amount information, and groups the user purchase records by commodity, counts the distribution of the amount in each group, and performs data formatting to generate commodity user purchase records and amount distribution data;

[0075] First, the user's purchase record data is imported into a data processing tool, such as Python's Pandas library or Excel. After grouping the data by product category, the user's purchase amount in each group is extracted, and these amount data are preliminarily cleaned, including removing duplicate records and detecting outliers. For example, compare whether the purchase amount in each group of data is too high or too low, and determine the outlier standard based on the reasonable range of the actual retail price. After removing the outliers, use a histogram tool (such as Matplotlib) to visualize the distribution of each group of amount data. By analyzing the distribution characteristics of each amount range in the histogram, the normalization method is further used to standardize the data, and the final data is converted into a product purchase record and amount distribution data table, such as outputting a table by product category, including amount range, purchase frequency and standardized data fields.

[0076] The user purchase data accumulation submodule accumulates the purchase amount of each product user based on the product user purchase records and amount distribution data, calculates the cumulative proportion of user purchases, sorts the data by proportion, and generates the cumulative proportion data of product user purchases;

[0077] Arrange the amount data of each commodity in ascending order, for example, use Python's Numpy library to complete the data sorting, and then record the cumulative proportion of each amount to the total amount by step-by-step accumulation. The specific operation includes first calculating the absolute value proportion of each amount value, and then accumulating these proportions to generate the cumulative proportion. The result is output as an Excel or CSV table. In this process, the correctness of the sorting of the cumulative proportion data is verified by Excel's sorting function, and the final generated cumulative proportion data is presented in a visual form. For example, a line chart is used to show the relationship between the cumulative proportion of the amount and the sorted amount, and the correspondence between each cumulative proportion value and the actual amount range is marked in the chart.

[0078] The contribution curve generation submodule converts the cumulative proportion of commodity user purchases into a cumulative contribution curve of commodity user purchases, and generates the cumulative distribution data of user contributions based on the user purchase situation of each commodity;

[0079] First, by analyzing the changes in the cumulative proportion of user purchases in different amount ranges, key interval data are extracted, such as distinguishing the low, medium, and high contribution areas of the cumulative proportion, and using the Scipy library in Python to fit the curve for each interval data. When fitting, linear fitting or polynomial fitting methods are selected. The contribution distribution of user purchases is judged according to the rate of change of the slope of the fitting curve in different intervals. The result of the curve fitting is plotted as a cumulative contribution curve chart through the Matplotlib library. In actual scenarios, by distinguishing the density of user purchases in each interval, such as the distribution changes of the amount concentrated in a certain interval, the intervals with concentrated contributions can be highlighted with colors or marking points, and finally the user contribution cumulative distribution data table is output to support further analysis of user purchase behavior.

[0080] See also Figure 2 and Figure 4 , the commodity concentration optimization module includes:

[0081] The commodity concentration attribute analysis submodule calculates the integral value of the commodity user purchase cumulative contribution curve based on the user contribution cumulative distribution data, sums up the integral value to represent the concentration attribute of each commodity, and generates the commodity Gini coefficient;

[0082] To calculate the integral value of the cumulative contribution curve of commodity users’ purchases, the formula is used:

[0083]

[0084] Get the integral value I, which represents the commodity concentration attribute;

[0085] Among them, x represents the cumulative proportion of the purchase amount, ranging from 0 to 1, extracted from the purchase amount distribution data of the product, and the calculation formula is: F(x) represents the function of the user contribution accumulation curve, which reflects the cumulative relationship between user contribution and amount distribution. It is obtained by fitting the user purchase amount and contribution data. The function value is determined by linear fitting or polynomial fitting method of discrete data points. ∫ represents the integral symbol, which is used to calculate the area under the curve and measure the concentration of purchase distribution. It is calculated by numerical integration method, such as trapezoidal integration method to calculate the area of ​​discrete data points. The specific formula is: n is the total number of data points, x i and F(x i ) is the horizontal and vertical coordinates of each data point i, [0, 1] represents the range of the integral, from the minimum value 0 to the maximum value 1 of the cumulative proportion of the purchase amount, dx represents the small change value of the cumulative proportion x, and is the unit of continuous change in the integral calculation. In actual calculation, discrete integral is used (x i+1 -x i ) means, for example, x 1=0.1,x 2 =0.2, then dx=0.1.

[0086] Taking a certain product as an example, assuming that the user purchase amount data is 100, 200, 300, 400, and the total amount is 1000. Accumulate each amount and divide it by the total amount to obtain 0.1, 0.3, 0.6, and 1.0. According to the user contribution distribution of the purchase amount, assuming that the user contribution of each amount segment is 0.05, 0.15, 0.30, and 0.50, through linear fitting, we get the discrete data points of F(x): F(0.1)=0.05, F(0.3)=0.15, F(0.6)=0.30, and F(1.0)=0.50.

[0087] According to the formula Convert the integral to discrete integral form:

[0088] Example data point: (x 1 =0.1,F(x 1 )=0.05),(x 2 =0.3,F(x 2 )=0.15),(x 3 =0.6,F(x 3 )=0.30),(x 4 =1.0,F(x 4 )=0.50).

[0089] For the interval [x 1 ,x 2 Calculate the area:

[0090] Bring in data:

[0091] For the interval [x 2 ,x 3 ]、[x 3 ,x 4 ] The same calculation, the final sum is I = A 1 +A 2 +A 3 ; Calculated I≈0.06.

[0092] According to the integral value I, the formula is adopted:

[0093] G = 2·I-1;

[0094] Calculate the Gini coefficient G;

[0095] Among them, the value range of G is 0≤G≤1. When G is closer to 0, it means that the purchase distribution is more uniform. When it is closer to 1, it means that the purchase concentration is higher.

[0096] Assuming that I = 0.56 is calculated by integration, the Gini coefficient is:

[0097] G = 2 0.56-1 = 0.12;

[0098] The Gini coefficient indicates that the concentration of commodity purchases is low, which means that users' purchasing behavior is more dispersed.

[0099] The commodity concentration screening submodule compares the Gini coefficient with the set concentration threshold based on the commodity Gini coefficient, screens commodities with Gini coefficients within the threshold range, and generates concentration screening commodity data;

[0100] Taking an online retail platform as an example, first extract the Gini coefficient data of all commodities from the commodity analysis table. For example, the Gini coefficient of commodity A is 0.25, commodity B is 0.55, and commodity C is 0.15. Set the concentration threshold range to 0.2 to 0.5, filter the data through Python's Pandas library, and use conditional statements to extract data records within the range. The screening results include commodity A (Gini coefficient 0.25) and commodity B (Gini coefficient 0.55 is not met). Mark the commodity records that meet the threshold conditions and generate a new screening table. Output the filtered data as an Excel format file, and use Excel's conditional formatting tool to highlight the qualified commodities to complete the generation of concentration screening commodity data.

[0101] The priority sorting submodule filters product data based on concentration, performs sales statistics and analysis on the filtered products, obtains sales data, inventory data and user evaluation data of the products, sorts them by priority according to the set weights, and generates product concentration screening results;

[0102] Take an online retail platform as an example, extract the sales, inventory and user evaluation data of products A, B and C, obtain the monthly sales of products from the sales database, for example, the sales of product A is 1,000 pieces, product B is 750 pieces, and product C is 500 pieces, obtain the current inventory quantity through the inventory management system, for example, the inventory of product A is 200 pieces, product B is 150 pieces, and product C is 50 pieces, crawl the evaluation data of the corresponding products from the user review platform, for example, the average score of product A is 4.8, product B is 4.2, and product C is 3.9, and normalize the sales, inventory and user evaluation data Normalization processing, for example, sales volume is normalized to 0, 1 intervals are respectively commodity A (1.0), commodity B (0.75), commodity C (0.5), inventory is normalized to commodity A (1.0), commodity B (0.75), commodity C (0.25), user evaluation is normalized to commodity A (1.0), commodity B (0.88), commodity C (0.65), according to the importance of each factor, the weight is set, the linear weighted sorting model is used, the priority value of each commodity is calculated in turn, the result is output in the form of priority sorting, and the commodity concentration screening result data table is generated. Among them, the priority calculation formula is: P = w s ·S+w k ·K+w r ·R, P are the product priority values, which are used to comprehensively evaluate the importance of the product and are calculated using the formula. s 、w k 、w r They represent the weights of sales, inventory, and user reviews, respectively. They are set based on historical data analysis and business needs. For example, when sales are dominant, the weight w is set. s >w k ,w r . S is the normalized value of sales, which is calculated by dividing the sales data by the maximum sales. For example, if the sales of product A is 1,000 pieces and the maximum sales is 1,000 pieces, the normalized value is 1.0. K is the normalized value of inventory, which is calculated by dividing the inventory quantity by the maximum inventory. For example, if the inventory of product A is 200 pieces and the maximum inventory is 200 pieces, the normalized value is 1.0. R is the normalized value of user evaluation, which is calculated by dividing the average score by the highest score. For example, if the score of product A is 4.8 and the highest score is 5.0, the normalized value is 4.8 / 5.0=0.96. Taking product A as an example, its sales volume, inventory and user reviews are 1,000 pieces, 200 pieces and 4.8 respectively. Assuming the sales volume weight is 0.5, the inventory weight is 0.3, and the review weight is 0.2, the priority value of product A is: P = 0.5·1.0+0.3·1.0+0.2·0.96=0.992. By comparing the priority values ​​of other products, for example, the priority value of product B is 0.765 and that of product C is 0.465, it can be seen that product A has the highest priority.

[0103] See also Figure 2 and Figure 5 ,The commodity collaborative mining module includes:

[0104] The user purchase record extraction submodule extracts the user purchase record data of the filtered products based on the product concentration screening results, groups and organizes the user purchase information by product, and generates user purchase and evaluation data of the filtered products;

[0105] First, we access the commodity transaction database to obtain the filtered commodity list, and then retrieve the purchase records of each commodity one by one, extract the user ID, purchase time, purchase quantity and other information involved in the record, and organize these data by commodity ID. For example, the purchase record of commodity A may contain the purchase time and quantity records of user IDs 001, 002, and 003, and commodity B contains the relevant records of users 004 and 005. The organized data is stored as a user purchase data table classified by commodity. Subsequently, by crawling the evaluation data of the user review platform, the natural language processing tool is used to perform sentiment analysis on the comment text to generate a sentiment score associated with the user ID. For example, the analysis of the comment "very good" obtains a sentiment score of 0.9, and the analysis of the comment "average quality" obtains a sentiment score of 0.4. After associating the sentiment score with the user ID, it is merged with the previous purchase record table to form a complete commodity purchase and evaluation data table.

[0106] The similarity matrix construction submodule is based on filtering the user purchase and evaluation data of commodities, calculating the synergy strength value of each pair of commodities by counting the number of co-occurrence purchases and the overlap of user evaluations between commodities, and quantifying the correlation between each pair of commodities based on the synergy strength value to construct the commodity similarity matrix;

[0107] To calculate the synergy strength value of each pair of commodities, the formula is used:

[0108]

[0109] Get S ij Represents the synergy strength value S of product i and product j ij ;

[0110] Among them, C ij represents the number of co-occurrence purchases of product i and product j, R ij represents the number of overlapping user reviews for product i and product j, C i and C j Represents the total number of purchases of product i and product j, R i and R j Respectively represent the total number of evaluations of product i and product j, W 1 and W 2 are the co-occurrence purchase weight and user evaluation overlap weight respectively. ijIndicates the number of times the same user purchases these two items at the same time. The number of records of the same user ID purchasing item i and item j at the same time is counted through the item transaction database. For example, in a transaction record table, user 001 purchases item A and item B in one order, which is counted as a co-occurrence purchase. ij Indicates the overlap of users who purchased these two products in the evaluation records. The number of overlapping user IDs who purchased products i and j is counted through the evaluation data table. For example, in the evaluation records of products A and B, user 001 and user 002 both participated in the comments, so the number of overlaps is 2, C i , C j Indicates the cumulative purchase volume of the product in the entire transaction data. All purchase records of a product are counted through the product transaction database. For example, the purchase record of product A is that users 001, 002, and 003 purchased it once each, and the total number of purchases is 3 times. i , R j Indicates the cumulative number of reviews for the product. The number of user reviews for a product is counted through the review data table. For example, the review record of product A is 1 review by user 001 and 002, and the total number of reviews is 2. 1 , W 2 They represent the weight factors of co-occurrence purchase and user evaluation overlap, respectively, and adjust the relative importance of the two to the synergy strength value. If the analysis results show that the impact of user evaluation behavior on synergy strength is more significant, for example, the user overlap rate of evaluation is higher than the purchase co-occurrence rate, then W can be appropriately increased. 2 to reflect the dominance of evaluation behavior; if co-occurrence purchases can better reflect the synergy strength relationship between commodities, then W should be appropriately increased. 1 The weight can be determined based on actual data distribution, business needs or through model optimization experimental results.

[0111] Represents the normalization factor of products i and j, which is used to balance the differences in purchase and evaluation scales of different products.

[0112] For example, get the current transaction data table and evaluation data table:

[0113] The transaction data table contains the following records: User 001 purchased products A and B, User 002 purchased products A and B, and User 003 purchased only product A. The total number of purchases of product A is 3 times, the total number of purchases of product B is 2 times, and the number of co-occurrence purchases (users 001 and 002) is 2 times.

[0114] The evaluation data table contains the following records: users 001 and 002 evaluated product A and product B respectively, the total number of evaluations for product A is 2, the total number of evaluations for product B is 2, and the number of overlapping evaluations is 2 (users 001 and 002).

[0115] Bring in data: C ij =2, R ij =2, C i =3, C j =2, R i =2, R j =2,W 1 =0.5,W 2 =0.5, the formula is:

[0116]

[0117] The results show that the current synergy strength value S ij =0.447.

[0118] The synergistic combination generation submodule identifies the cluster center products and related products in the similarity matrix of the products, analyzes the cross-data of product purchase conversion rate and user behavior, determines the synergistic combination relationship between products, and obtains product synergistic combination data;

[0119] First, by analyzing the synergy strength values ​​in the similarity matrix, we extract product pairs whose synergy strength is higher than a certain threshold. For example, if the synergy strength threshold is set to 0.4, the synergy strength value between product A and product B is 0.447, the synergy strength value between product A and product C is 0.75, and the synergy strength value between product B and product C is 0.3. Then, the synergy strength values ​​between product A and product B and product C are higher than the threshold. Therefore, product A is identified as the cluster center product and is associated with products B and C. Further analysis is conducted by combining the purchase conversion rate of commodities and the cross-data of user behavior. The purchase conversion rate of commodities is extracted through transaction data. For example, the conversion rate of commodity A is 10%, commodity B is 15%, and commodity C is 5%. Through SQL query, the proportion of users who purchase commodity A and purchase commodities B and C at the same time is counted. It is found that the proportion of users who purchase commodity B is 60%, and the proportion of users who purchase commodity C is 30%. User behavior analysis tools are used to further quantify user preferences. For example, the frequency of keywords mentioned in the evaluation of users who purchase commodities A and B is high, such as keywords such as "high quality" and "frequent purchase", while the evaluation overlap of commodities A and C is low. Therefore, commodities A and B are more related and form a synergistic combination first, with commodities A and C as a secondary combination. Finally, the synergistic combination data of commodities is generated.

[0120] See also Figure 2 and Figure 6 , the unpopular product identification module includes:

[0121] The user interaction network construction submodule builds an interaction network between products and users based on the product collaborative combination data, extracts the click behavior data, collection behavior data and purchase behavior data of unpopular products, connects the interaction paths between nodes according to the weight of the behavior data, and generates product user interaction network data;

[0122] First, extract product pairs and corresponding user information from the product collaborative combination data, and obtain user click, collection and purchase behavior data related to these product pairs through the transaction database. For example, user 001 clicks on product A, collects product B, and purchases product C. Generate an initial relationship table between users and products based on these behaviors, and assign different weights to click behaviors, collection behaviors and purchase behaviors, such as 0.2 for clicks, 0.5 for collections, and 1.0 for purchases. Use the behavior weights as the basis for the strength of the interaction path between nodes, and use a matrix to represent the interaction relationship between users and products. For example, the path weight between user 001 and product A is 0.2, and the path weight between user 001 and product C is 1.0. After constructing a complete user-product interaction matrix, use network analysis tools (such as Gephi) to construct an interaction network between products and users, generate product-user interaction network data, and display the network as an interaction graph through network visualization tools to further analyze the interaction relationship between product nodes.

[0123] The unpopular product analysis submodule counts the interaction frequency between unpopular products and other products or user nodes based on the product-user interaction network data, analyzes the position relationship and proximity relationship of unpopular product nodes in the interaction network, and generates the network centrality data of unpopular products;

[0124] By analyzing the network position relationship of unpopular commodity nodes, the proximity relationship is extracted, and the interaction frequency of unpopular commodities is converted into a network centrality index. For example, for a certain unpopular commodity A, the sum of its interaction weights with user nodes is 10, and the sum of its interaction weights with other commodity nodes is 5. The degree centrality and closeness centrality methods are used to calculate the centrality value of commodity A respectively. The degree centrality indicates the number of nodes directly connected to commodity A, which is proportional to the node weight. The closeness centrality reflects its indirect connection with other nodes by calculating the sum of the shortest path weights from unpopular commodities to other network nodes. The network position of unpopular commodities is evaluated by combining these centrality indicators. For example, the centrality value of unpopular commodity A is high, indicating that it has certain potential value in the interactive network. Finally, the analysis results are converted into a network centrality data table of unpopular commodities.

[0125] The unpopular product marking submodule compares the centrality value of the unpopular product node with the preset centrality threshold based on the unpopular product network centrality data, marks the unpopular product nodes whose centrality value meets the threshold range, and generates a potential unpopular product set;

[0126] First, set the centrality threshold range, for example, the degree centrality threshold is 0.2 to 0.5, and the proximity centrality threshold is 0.3 to 0.6. By comparing the centrality data of each unpopular product, select the product nodes within the range. For example, the degree centrality value of unpopular product A is 0.35, and the proximity centrality value is 0.4, which are within the threshold range. The centrality value of unpopular product B exceeds the threshold range and is not included in the screening. The unpopular product nodes that meet the conditions are marked as potential unpopular products, and a potential unpopular product set is generated. The output data file includes the product node ID, centrality value and marking status.

[0127] See also Figure 2 and Figure 7 , the product recommendation pool optimization module includes:

[0128] The product sales data collection submodule extracts the sales time series data of each product based on the potential unpopular product set, organizes the original sales data by segment according to the time axis, draws the product sales distribution curve by connecting the sales data of the time periods, and generates the product sales distribution curve data;

[0129] First, obtain the sales records of potential unpopular products from the database, extract fields including product ID, sales date and sales quantity, and organize the original sales data into segments according to the time axis. For example, segment it by week, calculate the weekly sales data, and visualize the segmented data in the form of a time series. The sales trend of each product is displayed by drawing a time series curve. During the drawing process, connect the sales data points in the time period to form a complete sales distribution curve, which is further saved as a structured data table. The table contains the product ID, time period and corresponding sales field to generate the product sales distribution curve data.

[0130] The sales anomaly elimination submodule identifies abnormal peak points and deviation intervals in the sales distribution based on the sales distribution curve data of the products, extracts the sales records and user behavior feature data of the corresponding abnormal products, cross-validates and analyzes the sales records of abnormal products with the user behavior features, eliminates the sales records of products that are judged to be abnormal, and generates optimized product sales data;

[0131] First, statistical analysis is performed on the sales data in the time series. For example, the mean and standard deviation of sales in each product time period are calculated, and the threshold range for peak detection is set. For example, the time period when the sales exceed three times the standard deviation of the mean is marked as an abnormal peak point. The product sales records corresponding to each abnormal time period are extracted, and cross-validated with user behavior data such as clicks, favorites, and reviews. For example, by analyzing the difference in user click frequency in the abnormal time period and other time periods, abnormal records caused by abnormal purchasing behavior are identified, and these abnormal records are removed from the product sales time series data, including product ID, time period, and optimized sales data fields, to generate optimized product sales data.

[0132] The recommendation pool marking submodule integrates the optimized product sales records with the potential unpopular product set based on the optimized product sales data, marks the optimized product set as recommendable, and generates an optimized live broadcast recommendation product pool;

[0133] The optimized product sales records are integrated with the potential unpopular product set. By comparing the optimized sales data with the product list in the potential unpopular product set, the status of products that meet the recommendation conditions is marked. For example, the sales data after optimization is analyzed to see whether it reaches the preset recommendation threshold, such as the increase in sales data or the improvement in user behavior indicators. The qualified products are marked as recommendable. Finally, the optimized product sales data is integrated to generate a live broadcast recommendation product pool. The output data table contains product ID, recommendation status and related indicator fields to support the selection of live broadcast recommendation products and further strategy optimization.

[0134] See also Figure 8 The method for selecting products to be broadcasted based on data analysis is performed based on the above-mentioned system for selecting products to be broadcasted based on data analysis, and includes the following steps:

[0135] S1: Based on the product purchase data, generate the product user purchase cumulative contribution curve after sorting, and obtain the user contribution cumulative distribution data;

[0136] S2: Based on the cumulative distribution data of user contributions, the Gini coefficient is calculated using the cumulative contribution curve, and the products whose concentration meets the threshold range are screened to generate the product concentration screening results;

[0137] S3: Based on the product concentration screening results, analyze the co-occurrence purchases and evaluation overlaps, build a similarity matrix, extract cluster center products and their related products, and generate product synergy combination data;

[0138] S4: Build a user interaction network based on product collaborative combination data to identify potential unpopular products, optimize sales curves to screen products, and generate an optimized live broadcast recommendation product pool.

[0139] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0140] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0141] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0142] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0143] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0145] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0146] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0147] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0148] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A product screening system based on data analysis, characterized in that: The system comprises: The contribution distribution analysis module collects the user purchase data of the product, sorts the user purchase data of the product by proportion, converts it into the cumulative contribution curve of the user purchase of the product, and generates the user contribution cumulative distribution data; The commodity concentration optimization module calculates the Gini coefficient of the commodity based on the integral value of the commodity user purchase cumulative contribution curve in the user contribution cumulative distribution data, screens commodities with concentration within a threshold range, and generates a commodity concentration screening result; The product collaborative mining module extracts the user purchase records and user evaluation data of the screened products based on the product concentration screening results, identifies the cluster center products and related products in the product similarity matrix through the co-occurrence purchase times of the products and the overlap of user evaluations, and generates product collaborative combination data; The unpopular product identification module builds a user interaction network of products based on product collaborative combination data, adds unpopular products as nodes to the user interaction network, connects the interaction paths between users and products according to the behavioral data weights of unpopular products, and generates a set of potential unpopular products; The product recommendation pool optimization module obtains the current product sales time series data and constructs a product sales distribution curve based on the potential unpopular product set, identifies abnormal peaks and deviation intervals of the product sales distribution curve, and generates an optimized live broadcast recommendation product pool.

2. The commodity screening system based on data analysis according to claim 1 is characterized in that: The user contribution cumulative distribution data includes user purchase ratio ranking, commodity cumulative purchase ratio curve and commodity amount distribution mapping; the commodity concentration screening results include a list of commodities that meet the concentration threshold, priority ranking scores and inventory status data; the commodity synergy combination data includes a commodity synergy strength matrix, commodity cluster center identification data and conversion rate correlation results; the potential unpopular commodity set includes unpopular commodity node network centrality data, interaction path weight distribution data and time series consumption preference characteristics; the live broadcast recommended commodity pool includes sales distribution optimized commodity data, commodity marking results after anomaly elimination and a recommended preferred commodity set.

3. The commodity screening system based on data analysis according to claim 1 is characterized in that: The contribution distribution analysis module includes: The user purchase data collection submodule is based on the user purchase data of commodities, including user purchase records and corresponding amount information, and groups the user purchase records by commodity, counts the distribution of the amount in each group, and performs data formatting to generate commodity user purchase records and amount distribution data; The user purchase data accumulation submodule accumulates the purchase amount of each commodity user based on the commodity user purchase records and amount distribution data, calculates the cumulative proportion of user purchases, sorts the data by proportion, and generates the cumulative proportion data of commodity user purchases; The contribution curve generation submodule converts the cumulative proportion data of user purchases of the commodities into a cumulative contribution curve of user purchases of the commodities, and generates cumulative distribution data of user contributions in combination with the user purchase situation of each commodity.

4. The commodity screening system based on data analysis according to claim 1 is characterized in that: The commodity concentration optimization module includes: The commodity concentration attribute analysis submodule calculates the integral value of the commodity user purchase cumulative contribution curve based on the user contribution cumulative distribution data, sums up the integral value to represent the concentration attribute of each commodity, and generates the commodity Gini coefficient; The commodity concentration screening submodule compares the Gini coefficient with the set concentration threshold based on the commodity Gini coefficient, screens commodities whose Gini coefficients are within the threshold range, and generates concentration screening commodity data; The priority sorting submodule filters the commodity data based on the concentration, performs sales statistics analysis on the filtered commodities, obtains the sales data, inventory data and user evaluation data of the commodities, sorts the commodities according to the set weights, and generates commodity concentration screening results.

5. The commodity screening system based on data analysis according to claim 4 is characterized in that: To calculate the integral value of the cumulative contribution curve of commodity users’ purchases, the formula is used: Get the integral value I, which represents the commodity concentration attribute; Where x represents the cumulative proportion of the purchase amount, F(x) represents the function of the user contribution cumulative curve, ∫ represents the integral symbol, which is used to calculate the area under the curve, [0, 1] represents the range of the integral, and dx represents the change value of the cumulative proportion x; According to the integral value I, the formula is adopted: G = 2·I-1; Calculate the Gini coefficient G; Among them, the value range of G is 0≤G≤1. When G is closer to 0, it means that the purchase distribution is more uniform. When it is closer to 1, it means that the purchase concentration is higher.

6. The commodity screening system based on data analysis according to claim 1 is characterized in that: The commodity collaborative mining module includes: The user purchase record extraction submodule extracts the user purchase record data of the screened products based on the product concentration screening result, groups and organizes the user purchase information by product, and generates user purchase and evaluation data of the screened products; The similarity matrix construction submodule is based on the user purchase and evaluation data of the selected products, calculates the synergy strength value of each pair of products by counting the number of co-occurrence purchases and the overlap of user evaluations between the products, and quantifies the correlation between each pair of products based on the synergy strength value to construct a similarity matrix of the products; The synergistic combination generation submodule identifies the cluster center products and related products in the similarity matrix of the products, analyzes the cross data of product purchase conversion rate and user behavior, determines the synergistic combination relationship between products, and obtains product synergistic combination data.

7. The commodity screening system based on data analysis according to claim 6 is characterized in that: To calculate the synergy strength value of each pair of commodities, the formula is used: Get S ij Represents the synergy strength value S of product i and product j ij ; Among them, C ij represents the number of co-purchases of product i and product j, R ij represents the number of overlapping user reviews for product i and product j, C i and C j Represents the total number of purchases of product i and product j, R i and R j They represent the total number of evaluations for product i and product j respectively. W1 and W2 are the co-occurrence purchase weight and user evaluation overlap weight respectively.

8. The commodity screening system based on data analysis according to claim 1 is characterized in that: The unpopular product identification module includes: The user interaction network construction submodule establishes an interaction network between products and users based on the product collaborative combination data, extracts click behavior data, collection behavior data and purchase behavior data of unpopular products, connects the interaction paths between nodes according to the weight of the behavior data, and generates product user interaction network data; The unpopular product analysis submodule counts the interaction frequency between unpopular products and other products or user nodes based on the product user interaction network data, analyzes the position relationship and neighbor relationship of unpopular product nodes in the interaction network, and generates unpopular product network centrality data; The unpopular product marking submodule compares the centrality value of the unpopular product node with a preset centrality threshold based on the unpopular product network centrality data, marks the unpopular product nodes whose centrality values ​​meet the threshold range, and generates a potential unpopular product set.

9. The commodity screening system based on data analysis according to claim 1 is characterized in that: The commodity recommendation pool optimization module includes: The commodity sales data collection submodule extracts the sales time series data of each commodity based on the potential unpopular commodity set, organizes the original sales data by segment according to the time axis, draws the commodity sales distribution curve by connecting the sales data of the time periods, and generates commodity sales distribution curve data; The sales anomaly elimination submodule identifies the abnormal peak points and deviation intervals in the sales distribution based on the sales distribution curve data of the commodities, extracts the sales records and user behavior characteristic data of the corresponding abnormal commodities, cross-validates and analyzes the sales records of the abnormal commodities with the user behavior characteristics, eliminates the sales records of the commodities determined to be abnormal, and generates optimized commodity sales data; The recommendation pool marking submodule integrates the optimized product sales records with the potential unpopular product set based on the optimized product sales data, marks the optimized product set as being in a recommendable state, and generates an optimized live broadcast recommendation product pool.

10. A method for selecting products to be broadcasted based on data analysis, characterized in that: The commodity broadcast screening system based on data analysis according to any one of claims 1 to 9 is implemented, comprising the following steps: Based on the product purchase data, generate the product user purchase cumulative contribution curve after sorting, and obtain the user contribution cumulative distribution data; Based on the user contribution cumulative distribution data, the Gini coefficient is calculated using the cumulative contribution curve, and the products whose concentration meets the threshold range are screened to generate a product concentration screening result; Based on the commodity concentration screening results, co-occurrence purchases and evaluation overlaps are analyzed, a similarity matrix is ​​constructed, cluster center commodities and their related commodities are extracted, and commodity synergy combination data is generated; Combined with the synergistic combination data of the above commodities, a user interaction network is constructed to identify potential unpopular commodities, and at the same time, the sales curve is optimized to screen commodities, thereby generating an optimized live broadcast recommendation commodity pool.

Images