E-commerce market trend prediction system based on big data analysis
By employing diversified distance quantization and adaptive cluster center updates, combined with search weight optimization, the e-commerce market trend prediction algorithm was improved, solving the problem of inappropriate cluster center and hyperparameter selection and achieving more accurate and stable market trend prediction.
Patent Information
- Application Number
- CN202410313720.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-19
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-03-19
AI Technical Summary
In existing e-commerce market trend prediction systems, inappropriate selection of cluster centers and hyperparameters leads to inaccurate prediction results, resulting in overfitting and low precision.
We employ diverse distance quantification methods to describe the similarity between data points, adaptively select the number of cluster centers, design search weights and gravity weights, update individual positions by combining individual and social experience, and optimize the prediction algorithm through adaptive clustering and movement strategies.
It improves the accuracy and stability of clustering results, enhances the robustness and reliability of the algorithm, and improves the accuracy and efficiency of market trend prediction.
Smart Images

Figure CN118212001B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to an e-commerce market trend prediction system based on big data analysis. Background Art
[0002] E-commerce market trend prediction is usually a complex task that requires a combination of multiple technical means and algorithms to achieve accurate prediction results. At the same time, with the continuous development of artificial intelligence and big data technology, the accuracy and efficiency of e-commerce market trend prediction will continue to improve. However, general market trend prediction systems have problems such as improper selection of cluster centers and improper selection of the number of cluster centers during clustering, resulting in inaccurate prediction results and overfitting; general prediction systems have problems such as improper selection of hyperparameters and poor convergence of hyperparameter searches, and low accuracy, resulting in large errors in prediction results. Summary of the invention
[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an e-commerce market trend prediction system based on big data analysis. In view of the fact that the general market trend prediction system has improper selection of cluster centers and improper selection of the number of cluster centers during clustering, resulting in inaccurate prediction results and overfitting problems, this scheme describes the similarity between data points more comprehensively based on a variety of distance quantification methods; adaptively selects the number of cluster centers based on changes in clustering errors; realizes adaptive cluster center updates based on probability value generation; achieves accuracy and stability of clustering results, and is able to cope with complex market trend prediction data; in view of the fact that the general prediction system has improper selection of hyperparameters and poor convergence of hyperparameter searches, and low accuracy leads to large errors in prediction results, this scheme improves the convergence and accuracy of the algorithm by designing search weights and gravity weights; by combining individual experience and social experience, as well as the optimal position of the population, the speed and position of the individual are updated to design a movement strategy, which helps the individual to better explore and utilize information in the search space, improves the efficiency of the search and the ability to find the global optimal solution, and enhances the robustness and reliability of the algorithm.
[0004] The technical solution adopted by the present invention is as follows: the e-commerce market trend prediction system based on big data analysis provided by the present invention includes a data acquisition module, a data preprocessing module, a preset module, an adaptive clustering module, a distance weight selection module and an e-commerce market trend prediction module;
[0005] The data collection module collects historical supply chain transaction data, historical supplier evaluation data, historical product information, historical market data, historical macroeconomic data and historical supply chain transaction risks;
[0006] The data preprocessing module performs data cleaning, data conversion and data standardization on the collected data;
[0007] The preset module defines the mixing distance, implements the cluster design and generates the number of cluster centers;
[0008] The adaptive clustering module completes adaptive clustering by initializing a cluster center set, generating a probability value based on a mixed distance, determining a cluster center set, assigning data points, updating cluster centers, and updating clustering results;
[0009] The distance weight selection module completes the selection of distance weight by initializing the population, designing the movement weight and movement strategy, and performing search judgment based on the fitness threshold;
[0010] The e-commerce market trend prediction module realizes market trend prediction based on clustering results and cluster labels.
[0011] Furthermore, in the data collection module, the historical user behavior data includes the user's browsing history, search history, click behavior and purchase history on the e-commerce platform; the historical product data includes the product category, price, sales volume and evaluation data; the historical user evaluation data includes the user's evaluation, rating and comment content on products and services; the historical transaction data includes order quantity, order amount, payment method and delivery method; and historical market data is used as data labels.
[0012] Furthermore, in the data preprocessing module, the data cleaning is to process missing values, duplicate values and outliers on the collected data; the data conversion is to convert the cleaned data into a vector form; and the data standardization is to standardize the converted data based on Min-Max scaling.
[0013] Furthermore, the pre-setting module specifically includes the following contents:
[0014] Define the blend distance using the following formula:
[0015] dx,c=αd cb x,c+βd cos x,c+γd corr x,c+ηd ma (x,c);
[0016] Where d(x,c) is the mixed distance between the data point and the cluster center; d cd (x,c),d cos (x,c),d corr (x,c),d ma (x,c) are Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively; α, β, γ and η are the distance weights of Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively;
[0017] Clustering design: collect historical user behavior data, historical product data, historical user evaluation data, historical transaction data, and historical market data as historical data; collect user behavior data, product data, user evaluation data, and transaction data in real time as real-time data; cluster historical data and real-time data together, ignoring the label dimension of historical data during clustering;
[0018] Generate the number of cluster centers K, get the clustering error based on SSE, and select the clustering center parameter when the clustering error increases sharply as the number of cluster centers; if the curve decreases slowly and there is no obvious inflection point, set the SSE threshold to 300, and use the clustering center parameter less than 300 as the number of cluster centers.
[0019] Furthermore, the adaptive clustering module specifically includes the following contents:
[0020] Initialization, initialize the cluster center set and select the first point as the cluster center;
[0021] Calculate the mixed distance. For each non-cluster center point, calculate the sum of the squares of its distances from all points in the current cluster center set.
[0022] Generate probability values. A probability threshold is set in advance. For each non-cluster center point, a probability value p is generated based on the distance squared distribution, ranging from [0, 1]. If p is greater than the probability threshold, the point is added to the cluster center set M. The formula used to generate the probability value is: Among them, d(x i ,M) 2 is the distance from the ith data point to the points in the cluster center set M, ∑ i d(x i , M) 2 It is the sum of the distances from all non-cluster center points to the points in the cluster center set M;
[0023] Cluster center set determination, pre-set distance threshold, repeatedly select cluster centers until K, if the average distance square sum of all points in the cluster center set is less than the distance threshold, the cluster center set selection initialization ends; otherwise, remove the cluster center with the largest average distance square sum to other points and continue to select cluster centers;
[0024] Assign data points, for each non-cluster center point, assign it to the cluster subset corresponding to the nearest cluster center;
[0025] Update the cluster center, and update the cluster center to the center point of the cluster set sample;
[0026] Update the clustering results, set an error threshold in advance, and record the error of each cluster subset, that is, the sum of the squares of the distances from each data point to the center of the cluster to which it belongs; when the sum of the squares of the errors of all cluster subsets is less than the error threshold, clustering ends;
[0027] Define cluster labels and use the label with the largest number of historical data as the cluster label.
[0028] Furthermore, the distance weight selection module specifically includes the following contents:
[0029] Initialize the population, initialize the parameter search space based on the distance weight to be selected, initialize the search population based on the parameter search space, and for the entire search population, normalize the sum of cluster subset errors when clustering is iterated k times based on the position of the search individual, and use the normalized sum of errors as the fitness value of the search individual; the formula used to initialize the population is:
[0030] x i,j =UP j -rand(0,1)·(UP j -LOW j );
[0031] In the formula, x i,j is the position of the i-th search individual in the j-th dimension, UP j and LOW j are the upper and lower limits of the j-th dimension of the search space respectively;
[0032] Design the moving weight using the following formula:
[0033]
[0034]
[0035] In the formula, ω i ,ω max and ω min are the search weight, maximum search weight and minimum search weight of the i-th search individual respectively; f i is the fitness value of the i-th search individual; N t is the maximum number of iterations; t is the current number of iterations; λ i,j is the gravitational weight of the jth dimension of the i-th search individual; G is the gravitational constant; M pi (t) and M ai (t) are the contribution of attraction and repulsion respectively; R i,j (·) is the distance between the i-th search individual in the j-dimension and the average position of the population in the j-dimension; ε is a small positive number to prevent the denominator from being zero; x i,j (t) is the position at the tth iteration; is the average position of the population in the jth dimension;
[0036] Design a mobile strategy using the following formula:
[0037] v i,j (t+1)=ω i ·v i,j (t)+c 1 ·r 1 ·λ i,j ·f i (t)+c 2 ·r 2 ·(gbest j -x i,j (t));
[0038] x i,j (t+1)=x i,j (t)+v i,j (t+1);
[0039] In the formula, v i,j (t+1) and v i,j (t) are the speeds of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; x i,j (t+1) and x i,j (t) are the positions of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; r 1 and r 2 are mutually independent random numbers between 0 and 1; c 1 and c 2 are the parameters controlling individual experience and social experience respectively; f i (t) is the fitness value of the i-th individual at the t-th iteration; gbest j is the position of the jth dimension of the optimal individual in the population;
[0040] Judgment, a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the individual position is the selected distance weight; if the maximum number of iterations is reached, the population position is reinitialized; otherwise, the mobile search continues.
[0041] Furthermore, the e-commerce market trend prediction module selects the distance weight based on the distance weight selection module, and based on the clustering results of the historical data and the real-time data, uses the cluster label to which the real-time data belongs as the market trend prediction result of the commodity.
[0042] The beneficial effects achieved by the present invention using the above scheme are as follows:
[0043] (1) In view of the problem that the general market trend forecasting system has inaccurate prediction results and overfitting due to improper selection of cluster centers and improper selection of the number of cluster centers when clustering exists, this scheme uses a variety of distance quantification methods to more comprehensively describe the similarity between data points; adaptively selects the number of cluster centers based on the change of clustering error; and adaptively updates the cluster centers based on probability value generation; achieves the accuracy and stability of the clustering results, and can cope with complex market trend forecasting data.
[0044] (2) In view of the problems of improper hyperparameter selection and poor convergence of hyperparameter search in general prediction systems, which lead to large errors in prediction results due to low precision, this scheme improves the convergence and precision of the algorithm by designing search weights and gravity weights; by combining individual experience and social experience, as well as the optimal position of the population, the speed and position of the individual are updated to design a movement strategy, which helps the individual to better explore and utilize information in the search space, improves the efficiency of the search and the ability to find the global optimal solution, and enhances the robustness and reliability of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 A schematic diagram of an e-commerce market trend prediction system based on big data analysis provided by the present invention;
[0046] Figure 2 It is a flowchart of the adaptive clustering module;
[0047] Figure 3 Schematic diagram of the flow of the distance weight selection module.
[0048] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0050] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.
[0051] Example 1, see Figure 1 , the e-commerce market trend prediction system based on big data analysis provided by the present invention includes a data acquisition module, a data preprocessing module, a preset module, an adaptive clustering module, a distance weight selection module and an e-commerce market trend prediction module;
[0052] The data collection module collects historical supply chain transaction data, historical supplier evaluation data, historical product information, historical market data, historical macroeconomic data and historical supply chain transaction risks;
[0053] The data preprocessing module performs data cleaning, data conversion and data standardization on the collected data;
[0054] The preset module defines the mixing distance, implements the cluster design and generates the number of cluster centers;
[0055] The adaptive clustering module completes adaptive clustering by initializing a cluster center set, generating a probability value based on a mixed distance, determining a cluster center set, assigning data points, updating cluster centers, and updating clustering results;
[0056] The distance weight selection module completes the selection of distance weight by initializing the population, designing the movement weight and movement strategy, and performing search judgment based on the fitness threshold;
[0057] The e-commerce market trend prediction module realizes market trend prediction based on clustering results and cluster labels.
[0058] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the data collection module, the historical user behavior data includes the user's browsing history, search history, click behavior and purchase history on the e-commerce platform; the historical commodity data includes the category, price, sales volume and evaluation data of the commodity; the historical user evaluation data includes the user's evaluation, rating and comment content on the commodity and service; the historical transaction data includes the order quantity, order amount, payment method and delivery method; and the historical market data is used as the data label.
[0059] Example 3, see Figure 1 This embodiment is based on the above embodiment. In the data preprocessing module, the data cleaning is to process missing values, duplicate values and outliers on the collected data; the data conversion is to convert the cleaned data into a vector form; the data standardization is to standardize the converted data based on Min-Max scaling.
[0060] Example 4, see Figure 1This embodiment is based on the above embodiment, and the presetting module specifically includes the following contents:
[0061] Define the blend distance using the following formula:
[0062] dx,c=αd cb x,c+βd cos x,c+γd corr x,c+ηd ma (x,c);
[0063] Where d(x,c) is the mixed distance between the data point and the cluster center; d cd (x,c),d cos (x,c),d corr (x,c),d ma (x,c) are Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively; α, β, γ and η are the distance weights of Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively;
[0064] Clustering design: collect historical user behavior data, historical product data, historical user evaluation data, historical transaction data, and historical market data as historical data; collect user behavior data, product data, user evaluation data, and transaction data in real time as real-time data; cluster historical data and real-time data together, ignoring the label dimension of historical data during clustering;
[0065] Generate the number of cluster centers K, get the clustering error based on SSE, and select the clustering center parameter when the clustering error increases sharply as the number of cluster centers; if the curve decreases slowly and there is no obvious inflection point, set the SSE threshold to 300, and use the clustering center parameter less than 300 as the number of cluster centers.
[0066] Example 5, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the adaptive clustering module specifically includes the following contents:
[0067] Initialization, initialize the cluster center set and select the first point as the cluster center;
[0068] Calculate the mixed distance. For each non-cluster center point, calculate the sum of the squares of its distances from all points in the current cluster center set.
[0069] Generate probability values. A probability threshold is set in advance. For each non-cluster center point, a probability value p is generated based on the distance squared distribution, ranging from [0, 1]. If p is greater than the probability threshold, the point is added to the cluster center set M. The formula used to generate the probability value is: Among them, d(x i ,M)2 is the distance from the ith data point to the points in the cluster center set M, ∑ i d(x i ,M) 2 It is the sum of the distances from all non-cluster center points to the points in the cluster center set M;
[0070] Cluster center set determination, pre-set distance threshold, repeatedly select cluster centers until K, if the average distance square sum of all points in the cluster center set is less than the distance threshold, the cluster center set selection initialization ends; otherwise, remove the cluster center with the largest average distance square sum to other points and continue to select cluster centers;
[0071] Assign data points, for each non-cluster center point, assign it to the cluster subset corresponding to the nearest cluster center;
[0072] Update the cluster center, and update the cluster center to the center point of the cluster set sample;
[0073] Update the clustering results, set an error threshold in advance, and record the error of each cluster subset, that is, the sum of the squares of the distances from each data point to the center of the cluster to which it belongs; when the sum of the squares of the errors of all cluster subsets is less than the error threshold, clustering ends;
[0074] Define cluster labels and use the label with the largest number of historical data as the cluster label.
[0075] By performing the above operations, in order to solve the problems of low distinction between different clusters caused by improper selection of general cluster centers, insufficient randomness of the cluster center search method, and low adaptability of the search strategy, this scheme uses the average of the cluster result evaluation values obtained by using the data point as the initial cluster center three times as the fitness value, which can better capture the stability of the cluster center; the moving strategy takes into account the global optimal individual and random parameters to improve the search efficiency and the discovery of local optimal solutions; thereby improving the search efficiency and the quality of the clustering results.
[0076] Example 6, see Figure 1 and Figure 3 This embodiment is based on the above embodiment, and the distance weight selection module specifically includes the following contents:
[0077] Initialize the population, initialize the parameter search space based on the distance weight to be selected, initialize the search population based on the parameter search space, and for the entire search population, normalize the sum of cluster subset errors when clustering is iterated k times based on the position of the search individual, and use the normalized sum of errors as the fitness value of the search individual; the formula used to initialize the population is:
[0078] x i,j =UP j-rand(0,1)·(UP j -LOW j );
[0079] In the formula, x i,j is the position of the i-th search individual in the j-th dimension, UP j and LOW j are the upper and lower limits of the j-th dimension of the search space respectively;
[0080] Design the moving weight using the following formula:
[0081]
[0082]
[0083] In the formula, ω i ,ω max and (ω min are the search weight, maximum search weight and minimum search weight of the i-th search individual respectively; f i is the fitness value of the i-th search individual; N t is the maximum number of iterations; t is the current number of iterations; λ i,j is the gravitational weight of the jth dimension of the i-th search individual; G is the gravitational constant; M pi (t) and M ai (t) are the contribution of attraction and repulsion respectively; R i,j (·) is the distance between the i-th search individual in the j-dimension and the average position of the population in the j-dimension; ε is a small positive number to prevent the denominator from being zero; x i,j (t) is the position at the tth iteration; is the average position of the population in the jth dimension;
[0084] Design a mobile strategy using the following formula:
[0085] v i,j (t+1)=ω i ·v i,j (t)+c 1 ·r 1 ·λ i,j ·f i (t)+c 2 ·r 2 ·(gbest j -x i,j (t));
[0086] x i,j (t+1)=x i,j (t)+v i,j (t+1);
[0087] In the formula, vi,j (t+1) and v i,j (t) are the speeds of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; x i,j (t+1) and x i,j (t) are the positions of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; r 1 and r 2 are mutually independent random numbers between 0 and 1; c 1 and c 2 are the parameters controlling individual experience and social experience respectively; f i (t) is the fitness value of the i-th individual at the t-th iteration; gbest j is the position of the jth dimension of the optimal individual in the population;
[0088] Judgment, a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the individual position is the selected distance weight; if the maximum number of iterations is reached, the population position is reinitialized; otherwise, the mobile search continues.
[0089] By performing the above operations, this solution improves the convergence and accuracy of the algorithm by designing search weights and gravity weights to address the problems of improper hyperparameter selection and poor convergence of hyperparameter search in general prediction systems, as well as low precision leading to large errors in prediction results. This solution also updates the speed and position of individuals by combining individual experience and social experience, as well as the optimal position of the population, to design a movement strategy, which helps individuals better explore and utilize information in the search space, improves search efficiency and the ability to find the global optimal solution, and enhances the robustness and reliability of the algorithm.
[0090] Embodiment 7, see Figure 1 This embodiment is based on the above embodiment. The e-commerce market trend prediction module selects the distance weight based on the distance weight selection module, and based on the clustering results of historical data and real-time data, the cluster label to which the real-time data belongs is used as the market trend prediction result of the commodity.
[0091] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0092] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.
[0093] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.
Claims
1. E-commerce market trend prediction system based on big data analysis, Features: The system includes a data acquisition module, a data preprocessing module, a preset module, an adaptive clustering module, a distance weight selection module and an e-commerce market trend prediction module; The data collection module collects historical supply chain transaction data, historical supplier evaluation data, historical product information, historical market data, historical macroeconomic data and historical supply chain transaction risks; The data preprocessing module performs data cleaning, data conversion and data standardization on the collected data; The preset module defines the mixing distance, implements the cluster design and generates the number of cluster centers; The adaptive clustering module completes adaptive clustering by initializing a cluster center set, generating a probability value based on a mixed distance, determining a cluster center set, assigning data points, updating cluster centers, and updating clustering results; The distance weight selection module completes the selection of distance weight by initializing the population, designing the movement weight and movement strategy, and performing search judgment based on the fitness threshold; The e-commerce market trend prediction module realizes market trend prediction based on clustering results and cluster labels; The adaptive clustering module specifically includes the following contents: Initialization, initialize the cluster center set and select the first point as the cluster center; Calculate the mixed distance. For each non-cluster center point, calculate the sum of the squares of its distances from all points in the current cluster center set. Generate probability values, set a probability threshold in advance, and for each non-cluster center point, generate a probability value p based on the distance square sum distribution, ranging from [0, 1]; if p is greater than the probability threshold, add the point to the cluster center set M; The formula used to generate the probability value is Among them, d(x i ,M) 2 is the distance from the ith data point to the points in the cluster center set M, Σ i d(x i ,M) 2 It is the sum of the distances from all non-cluster center points to the points in the cluster center set M; Cluster center set determination, pre-set distance threshold, repeatedly select cluster centers until K, if the average distance square sum of all points in the cluster center set is less than the distance threshold, the cluster center set selection initialization ends; otherwise, remove the cluster center with the largest average distance square sum to other points and continue to select cluster centers; Assign data points, for each non-cluster center point, assign it to the cluster subset corresponding to the nearest cluster center; Update the cluster center, and update the cluster center to the center point of the cluster set sample; Update the clustering results, set an error threshold in advance, and record the error of each cluster subset, that is, the sum of the squares of the distances from each data point to the center of the cluster to which it belongs; when the sum of the squares of the errors of all cluster subsets is less than the error threshold, clustering ends; Define cluster labels, and use the label with the largest number of historical data as the cluster label; The distance weight selection module specifically includes the following contents: Initialize the population, initialize the parameter search space based on the distance weight to be selected, initialize the search population based on the parameter search space, and for the entire search population, normalize the sum of cluster subset errors when clustering is iterated k times based on the position of the search individual, and use the normalized sum of errors as the fitness value of the search individual; the formula used to initialize the population is: x i,j =UP j -rand(0,1)·(UP j -LOW j ); In the formula, x i,j is the position of the i-th search individual in the j-th dimension, UP j and LOW j are the upper and lower limits of the j-th dimension of the search space respectively; Design the moving weight using the following formula: In the formula, ω i ,ω max and ω min are the search weight, maximum search weight and minimum search weight of the i-th search individual respectively; f i is the fitness value of the i-th search individual; N t is the maximum number of iterations; t is the current number of iterations; λ i,j is the gravitational weight of the jth dimension of the i-th search individual; G is the gravitational constant; M pi (t) and M ai (t) are the contribution of attraction and repulsion respectively; R i,j (·) is the distance between the i-th search individual in the j-dimension and the average position of the population in the j-dimension; ε is a small positive number to prevent the denominator from being zero; x i,j (t) is the position at the tth iteration; is the average position of the population in the jth dimension; Design a mobile strategy using the following formula: v i,j (t+1)=ω i ·v i,j (t)+c 1 ·r 1 ·λ i,j ·f i (t)+c 2 ·r 2 ·(gbest j -x i,j (t)); x i,j(t+1) =x i,j (t)+v i,j (t+1); In the formula, v i,j (t+1) and v i,j (t) are the speeds of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; x i,j (t+1) and x i,j (t) are the positions of the i-th search individual at the t+1th iteration and the tth iteration of the j-th dimension; r 1 and r 2 are mutually independent random numbers between 0 and 1; c 1 and c 2 are the parameters controlling individual experience and social experience respectively; f i (t) is the fitness value of the i-th individual at the t-th iteration; gbest j is the position of the jth dimension of the optimal individual in the population; Judgment, a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the individual position is the selected distance weight; if the maximum number of iterations is reached, the population position is reinitialized; otherwise, the mobile search continues; The pre-setting module specifically includes the following contents: Define the blend distance using the following formula: d(x,c)=αd cb (x,c)+βd cos (x,c)+γd corr (x,c)+ηd ma (x,c); Where d(x,c) is the mixed distance between the data point and the cluster center; d cd (x,c),d cos (x,c),d corr (x,c),d ma (x,c) are Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively; α, β, γ and η are the distance weights of Manhattan distance, cosine distance, correlation distance and Mahalanobis distance respectively; Clustering design: collect historical user behavior data, historical product data, historical user evaluation data, historical transaction data, and historical market data as historical data; collect user behavior data, product data, user evaluation data, and transaction data in real time as real-time data; cluster historical data and real-time data together, ignoring the label dimension of historical data during clustering; Generate the number of cluster centers K, get the clustering error based on SSE, and select the clustering center parameter when the clustering error increases sharply as the number of cluster centers; if the curve decreases slowly and there is no obvious inflection point, set the SSE threshold to 300, and use the clustering center parameter less than 300 as the number of cluster centers.
2. According to the e-commerce market trend prediction system based on big data analysis according to claim 1, Features: In the data collection module, the historical user behavior data includes the user's browsing history, search history, click behavior and purchase history on the e-commerce platform; the historical product data includes the product category, price, sales volume and evaluation data; the historical user evaluation data includes the user's evaluation, rating and comment content on products and services; the historical transaction data includes order quantity, order amount, payment method and delivery method; and the historical market data is used as data labels.
3. According to the e-commerce market trend prediction system based on big data analysis according to claim 1, Features: In the data preprocessing module, the data cleaning is to process missing values, duplicate values and outliers on the collected data; the data conversion is to convert the cleaned data into a vector form; and the data standardization is to standardize the converted data based on Min-Max scaling.
4. The e-commerce market trend prediction system based on big data analysis according to claim 1, Features: The e-commerce market trend prediction module selects the distance weight based on the distance weight selection module, and uses the cluster label to which the real-time data belongs as the market trend prediction result of the commodity based on the clustering result of the historical data and the real-time data.
Citation Information
Patent Citations
Cascade reservoir optimal operation method based on adaptive particle swarm optimization algorithm
CN103942612A
Self-adaptation clustering method based on maximum distance, minimum distance and K-means
CN104376057A
Target assignment optimizing method based on particle swarm algorithm of population explosion
CN105005820A
Clustering system based on business data
CN107220841A
Improved spectral clustering and parallelization method
CN108520284A
