A consumption habit classification method and system based on big data

Abstract

The present invention relates to the field of data processing technology, and in particular to a method and system for classifying consumption habits based on big data. The method comprises: treating the consumption habit data points of each user obtained, which contain multiple dimensions, as a cluster; obtaining the distribution of each cluster using data distribution modeling; and obtaining the classification results reflecting the user's consumption habits using a hierarchical clustering algorithm to achieve the classification of consumption habits based on big data. In the iterative process of the hierarchical clustering algorithm, the method comprises: merging the clusters with the smallest inter-cluster W distance, constructing a minimization function of the quality vector of the merged cluster under constraints, determining the optimized distribution of the merged cluster, determining the homogeneity index within the cluster, and determining whether to proceed to the next round of iteration. The present invention measures the differences between clusters with W distance, constructs a quality vector minimization function, updates the optimized distribution of the merged cluster, and combines dual-threshold iterative classification to accurately reflect the characteristics of group consumption habits.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a consumption habit classification method and system based on big data. Background Art

[0002] With the rapid development of the digital economy, consumption scenarios are becoming increasingly diverse. Vast amounts of consumer data (such as transaction amounts, purchase frequency, and product category preferences) are being recorded and stored in real time, providing a data foundation for consumer profiling, precision marketing, and market segmentation. Currently, user segmentation based on consumer data has become a core requirement in industries such as retail, finance, and e-commerce. By accurately segmenting user groups with different consumption habits, companies can develop differentiated marketing strategies and optimize product recommendation mechanisms, thereby increasing user stickiness and conversion rates. However, consumer data is often high-dimensional and unstructured, and user consumption behavior is dynamic (such as seasonal fluctuations) and heterogeneous (such as significant differences in individual consumption preferences). This places higher demands on the accuracy of existing classification methods. Currently, the industry generally relies on clustering algorithms to process consumer data, but how to achieve efficient and accurate classification while preserving users' personalized consumption characteristics remains a key challenge that needs to be addressed.

[0003] In the field of user classification based on consumption data, hierarchical clustering algorithms are widely used because they can intuitively present the hierarchical relationships of user groups. Traditional hierarchical clustering algorithms use Euclidean distance as the core metric to measure differences in consumption habits among different consumers, and complete classification by presetting the number of clusters. However, when quantifying differences in consumption habits, Euclidean distance tends to overlook the uniqueness of individual consumption behaviors, resulting in users with significantly different consumption habits being mistakenly classified into the same category. Even when hierarchical clustering algorithms are improved using Wasserstein (W) distance, the simple weighted average approach used to update the distribution during cluster merging dilutes or even loses the original distribution characteristics of each cluster before the merger, making it difficult to accurately reflect the true consumption habits of the merged group. Summary of the Invention

[0004] In order to solve the problem that traditional hierarchical clustering algorithms use Euclidean distance as the core indicator to measure the differences in consumption habits of different consumers and complete classification by presetting the number of clusters, while Euclidean distance tends to ignore the uniqueness of individual consumption behavior when quantifying differences in consumption habits, resulting in users with significantly different consumption habits being mistakenly classified into the same category, even if the hierarchical clustering algorithm is improved by using W distance, when processing the distribution update during the cluster merging process, the original distribution characteristics of each cluster before the merger will be diluted or lost due to the use of a simple weighted average method, making it difficult to accurately reflect the true consumption habit characteristics of the merged group. The present invention provides a consumption habit classification method and system based on big data.

[0005] In a first aspect, the present invention provides a consumption habit classification method based on big data, which adopts the following technical solutions:

[0006] A consumption habit classification method based on big data, comprising: treating the consumption habit data points of each user containing multiple dimensions as a cluster, obtaining the distribution of each cluster by using data distribution modeling, and obtaining a classification result reflecting the consumption habits of the user by using a hierarchical clustering algorithm, so as to realize the classification of consumption habits based on big data; in each round of iterative process of obtaining the classification result by using the hierarchical clustering algorithm, comprising: for this round of iteration, obtaining the W distance between all clusters at the beginning of the iteration based on the distribution, merging the clusters with the smallest W distance, recording the proportion of the number of users in each cluster at the time of merging to the number of users in the cluster after merging as the weight of each cluster at the time of merging, and merging the clusters according to the weight, the distance between the clusters at the time of merging, and the weight of each cluster at the time of merging. The Euclidean distance from each user in each cluster to each user in the merged cluster and the transmission quality obtained by using a linear programming solver are used to construct a minimization function of the quality vector of the merged cluster under the constraint condition; the optimized distribution of the merged cluster is determined according to the quality of each user in the quality vector and the corresponding distribution; for the unmerged and merged clusters, the intra-cluster homogeneity index is determined according to the W distance from each user in each cluster to the centroid within the cluster; based on the intra-cluster homogeneity index and the minimum W distance between clusters, it is determined whether to perform the next round of iteration; in response to performing the next round of iteration, the optimized distribution is used as the distribution of the merged cluster, and the unmerged and merged clusters are used as the clusters at the beginning of the next round of iteration.

[0007] By replacing the traditional Euclidean distance with the W distance, the present invention can better measure the differences between the distributions of consumption habit data of different users, avoid the neglect of the uniqueness of individual behavior in the feature space by the Euclidean distance, and improve the clustering ability to capture actual consumption behavior patterns; in the cluster merging process, the Euclidean distance is combined with the transmission quality obtained by using the linear programming solver, and a quality vector minimization function based on constraint conditions is constructed, which effectively avoids the dilution of original cluster features caused by the simple weighted average method and maintains the diversity and representativeness of the consumption habits of users in the merged cluster; the W distance from the user in the cluster to the centroid is used as the intra-cluster homogeneity indicator, and combined with the minimum W distance between clusters to decide whether to perform further iterations, ensuring that the clustering process can not only finely distinguish different consumer groups but also avoid over-clustering, thereby improving the stability and rationality of the clustering results; a more refined and accurate consumption habit classification method helps companies to deeply understand the differences in user behavior, thereby formulating personalized marketing strategies, optimizing resource allocation, and improving user experience and corporate benefits.

[0008] Furthermore, the multiple dimensions include age, monthly consumption times, average single consumption amount, proportion of electronic product consumption, number of promotional activities participated in, repurchase rate, and consumption period concentration index.

[0009] Furthermore, the data distribution modeling adopts Dirac function as a modeling tool for data distribution modeling.

[0010] Furthermore, the W distance is obtained by using a Sinkhorn algorithm.

[0011] Furthermore, the minimization function satisfies:

[0012] Where, is the mass vector of the cluster after merging under the constraint conditions, for Minimize the independent variable function, is the mass vector to be solved, is the number of clusters when merging, The first The weight of a cluster, The first The number of users in a cluster, is the number of users in the merged cluster, The first In-cluster users to the merged cluster The Euclidean distance between users, The first In-cluster users to the merged cluster The transmission quality of each user obtained by using a linear programming solver is influence each other.

[0013] The present invention combines the Euclidean distance-weighted transmission quality between users to quantify the transportation cost between the consumption habit distributions of each user when clusters are merged, thereby making the clustering process more in line with the actual user distribution differences and enhancing the internal consistency of the merged clusters. The linear programming solver is used to dynamically adjust the transmission mode and quality distribution to achieve optimal matching of the mass distribution of the merged clusters, avoiding excessive smoothing of consumption characteristics and information loss caused by simple averaging. The minimization of the independent variable function is solved under constraints to ensure that the physical and statistical interpretation of the quality vector conforms to the actual application scenario, and the merging process is guaranteed to be legal and reliable, thereby improving the interpretability and practical value of the clustering results in consumption habit analysis.

[0014] Furthermore, the constraint condition is to ensure that the distribution is not lost: The quality of each user in the quality vector of the merged cluster is equal to the sum of the transmission quality of each user in each cluster at the time of merging to each user in the merged cluster: , is the mass vector of the merged cluster. The quality of each user; the sum of the quality of each user in the quality vector of the merged cluster: ; Specific practical meaning of parameters: .

[0015] Furthermore, the optimized distribution satisfies:

[0016] Where, is the optimal distribution of the merged clusters, is the number of users in the merged cluster, is the mass vector of the merged cluster. The quality of each user, The first The corresponding distribution of users.

[0017] The present invention introduces the original distribution as the distribution basis function of each user, and expresses the overall distribution of the merged cluster as a weighted combination of the quality vector and the individual user distribution, thereby accurately retaining the contribution of each user's characteristics and personalized information; the optimized distribution is not a simple average, but is weighted according to the solved quality vector. The unique consumption behavior of each user will not be weakened or lost due to the merging operation, thereby improving the representativeness and recognition within the cluster.

[0018] Furthermore, the intra-cluster homogeneity index satisfies:

[0019] Where, is the intra-cluster homogeneity index, is the total number of unmerged and merged clusters, The first cluster in the unmerged and merged clusters The number of users in a cluster, express The i-th sample in , The first cluster in the unmerged and merged clusters In-cluster Users to The W distance between the centroids of the clusters.

[0020] The intra-cluster homogeneity index of the present invention calculates the square average of the distribution differences between users in each cluster and the cluster centroid based on the W distance, which can carefully reflect the consistency of samples in the cluster and help evaluate and ensure the aggregation quality of clustering results. The use of W distance instead of traditional Euclidean distance can more effectively capture the distribution form and differences of consumption data, avoid the defect of traditional distance measurement that ignores the diversity of user behavior patterns, and make the homogeneity index more consistent with the statistical characteristics of actual consumption habits.

[0021] Furthermore, the determination of whether to proceed to the next round of iteration includes: in response to the intra-cluster homogeneity index being less than a preset first abnormality threshold, and the minimum W distance between the unmerged and merged clusters being greater than a preset second abnormality threshold, determining that the hierarchical clustering algorithm has reached the termination condition in this round of iteration, and using the unmerged and merged clusters of this round of iteration as the classification results of the hierarchical clustering algorithm; otherwise, proceeding to the next round of iteration.

[0022] The present invention combines the dual indicators of intra-cluster homogeneity and inter-cluster distance for judgment, ensuring early termination when the internal consistency of the cluster is high enough and the differences between clusters are obvious, avoiding classification errors caused by over-merging or under-merging; the termination condition is dynamically set based on the actual data characteristics, improving the adaptability of the algorithm to different data structures and distribution characteristics, avoiding the limitation of a simple fixed number of clusters, and making the clustering process more flexible and intelligent; by controlling the clustering results through dual thresholds, the fineness and granularity of the classification can be adjusted according to the threshold settings, which is conducive to targeted adjustment of clustering parameters in application scenarios and obtaining consumer group divisions that meet actual needs.

[0023] In a second aspect, the present invention provides a consumption habit classification system based on big data, which adopts the following technical solutions:

[0024] A consumption habit classification system based on big data includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned consumption habit classification method based on big data is implemented.

[0025] By adopting the above technical solution, the above-mentioned consumption habit classification method based on big data is generated into a computer program and stored in the memory to be loaded and executed by the processor, so that a terminal device is made based on the memory and the processor for easy use.

[0026] The present invention has the following technical effects:

[0027] (1) By breaking through the limitation of traditional hierarchical clustering that relies on Euclidean distance, the W distance is introduced and the distribution calculation method of cluster merging is optimized to effectively avoid users with significant differences in consumption habits from being mistakenly classified into the same category. At the same time, based on the weight design of the proportion of user numbers and the transmission quality constraint, the merged cluster distribution can retain the core characteristics of each original cluster, reduce the problem of distribution dilution or loss, and more accurately reflect the true form of group consumption habits.

[0028] (2) By constructing a minimization function of the independent variable of the quality vector and combining it with the original distribution to determine the optimized distribution after merging, the cluster merging process takes into account the contribution weight of each cluster and the distance relationship and transmission efficiency between users, avoiding the feature distortion caused by simple weighted averaging, making the clustering results more stable during the iteration process, and reducing the classification fluctuation caused by distribution deviation.

[0029] (3) The homogeneity index is calculated based on the W distance from each user in the cluster to the centroid, and the minimum W distance between clusters is combined to determine whether to continue iteration, so that the clustering process can be adaptively terminated according to the data characteristics, avoiding the subjective bias caused by the preset number of clusters, and improving the flexibility and applicability of classification.

[0030] (4) More accurate classification results of consumer habits can provide companies with more detailed user portraits, helping to identify the demand preferences, purchasing patterns and other characteristics of different consumer groups. Companies can then formulate targeted marketing strategies (such as personalized recommendations and targeted promotions), optimize product layout and service design, and improve user satisfaction and conversion rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of a method for classifying consumption habits based on big data in an embodiment of the present invention.

[0032] Figure 2 This is a method flow chart of step S3 in a consumption habit classification method based on big data in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0034] The embodiment of the present invention discloses a consumption habit classification method based on big data, referring to Figure 1 , including steps S1 to S3:

[0035] S1: The consumption habit data points of each user with multiple dimensions are taken as a cluster.

[0036] It should be noted that the consumption habit data points of each user are collected using methods such as the e-commerce platform order system, user historical order records, and order product classification labels. The values ​​of the consumption habit data points in each dimension are normalized by Z-score to reduce the impact of different dimensions. For data examples, see Table 1.

[0037] Table 1

[0038]

[0039] Specifically, the multiple dimensions include age, monthly consumption times, average single consumption amount, proportion of electronic product consumption, number of participation in promotional activities, repurchase rate, and consumption period concentration index.

[0040] S2: Use data distribution modeling to obtain the distribution of each cluster.

[0041] Specifically, the data distribution modeling uses the Dirac function as a modeling tool for data distribution modeling.

[0042] S3: Use the hierarchical clustering algorithm to obtain classification results that reflect user consumption habits, so as to realize the classification of consumption habits based on big data.

[0043] In each round of iteration of obtaining classification results using the hierarchical clustering algorithm, refer to Figure 2 , step S3 includes steps S301 to S304, which are specifically as follows:

[0044] S301: Merge clusters with the smallest inter-cluster W distance, and construct a function that minimizes the independent variable of the mass vector of the merged cluster under the constraint condition.

[0045] It should be noted that, taking the number of monthly purchases as an example, as shown in Table 2, when using traditional Euclidean distance to classify the four users (consumers), A and B are grouped together because the Euclidean distance between A and B is smaller than the Euclidean distance between A and C. However, in practice, it is more appropriate to group A and C together. W-distance is a metric used to measure the similarity between two probability distributions. It measures the minimum cost required to transform one distribution into another, that is, the minimum moving average distance. This metric can better extract comprehensive information about user consumption habits. Therefore, in this step, W-distance is selected as the inter-cluster similarity metric rather than Euclidean distance. This metric can capture the distribution characteristics of consumption habit data and avoid misclassification caused by close local data points but large differences in overall distribution.

[0046] Table 2

[0047]

[0048] For this iteration, based on the distribution, the W distance between all clusters at the start of the iteration is obtained. The clusters with the smallest W distance are merged. The proportion of the number of users in each cluster at the time of merging to the number of users in the merged cluster is recorded as the weight of each cluster at the time of merging. Based on the weight, the Euclidean distance from each user in each cluster at the time of merging to each user in the merged cluster, and the transmission quality obtained using the linear programming solver, a function is constructed to minimize the independent variable of the quality vector of the merged cluster under the constraints.

[0049] Specifically, the W distance is obtained using the Sinkhorn algorithm. Since the calculation of the Wasserstein distance of discrete data is computationally intensive, the Sinkhorn algorithm is used to transform the original problem into a convex optimization problem that can be efficiently solved iteratively by introducing an entropy regularization term, thereby greatly reducing the computational complexity while ensuring controllable accuracy.

[0050] Specifically, the minimization function satisfies:

[0051] ;

[0052] Where, is the mass vector of the cluster after merging under the constraint conditions, for Minimize the independent variable function, is the mass vector to be solved, is the number of clusters when merging, The first The weight of a cluster, The first The number of users in a cluster, is the number of users in the merged cluster, The first In-cluster users to the merged cluster The Euclidean distance between users, The first In-cluster users to the merged cluster The transmission quality of each user obtained by using a linear programming solver is Mutual influence ( for An element within which For the The state transfer matrix from the cluster to the merged cluster, the elements of the state transfer matrix correspond to the physical "force", the Euclidean distance corresponds to the "distance" of work done, and minimizing the independent variable function is to find the smallest distribution so that the "work done" of state transfer is minimized; the Euclidean distance is used here to reflect the transmission cost for computational efficiency. If the W distance is used to reflect the transmission cost, it will lead to the problem of "nested solution").

[0053] Specifically, the constraints are:

[0054] Ensure that the distribution is not lost: ;

[0055] The quality of each user in the quality vector of the merged cluster is equal to the sum of the transmission quality of each user in each cluster at the time of merging to each user in the merged cluster: , is the mass vector of the merged cluster. The quality of individual users;

[0056] The sum of the quality of each user in the quality vector of the merged cluster: ;

[0057] Specific practical meaning of the parameters: .

[0058] For example, the monthly consumption times in the dimension are selected as indicator 1, and the average word consumption amount is selected as indicator 2. Assuming that there are two clusters in this round of iterative merging, there are two users in cluster A and three users in cluster B, then there are five users in the merged cluster, as shown in Table 3. The normalized data is shown in Table 4.

[0059] Table 3

[0060]

[0061] Table 4

[0062]

[0063] The distance cost from users in cluster A to users in the merged cluster is calculated using the Euclidean distance:

[0064] A1 (cluster A) to A1 (merged cluster): ;

[0065] A1 (cluster A) to A2 (merged cluster): ;

[0066] A1 (cluster A) to B1 (merged cluster): ;

[0067] The remaining distance cost is calculated in the same way and will not be described here;

[0068] The distance cost from users in cluster B to users in the merged cluster is calculated using the Euclidean distance:

[0069] B1 (cluster B) to B1 (merged cluster): ;

[0070] B1 (cluster B) to B2 (merged cluster): ;

[0071] The remaining distance cost calculation is the same and will not be repeated here;

[0072] No. The state transition matrix from clusters to the merged cluster: , then the state transfer cost of cluster A is:

[0073]

[0074] State transition cost of cluster B The calculation method is the same as that of cluster A, so I will not go into details here;

[0075] but ,when When the minimum within They are: , satisfying the sum of the quality of each user in the quality vector of the merged cluster: .

[0076] S302: Determine the optimal distribution of the merged clusters.

[0077] It should be noted that the merging operation causes the distribution of clusters to change. Directly calculating the W distance of the merged clusters would greatly increase the complexity of the algorithm, while directly using the original distribution would lose the structural information contained in the merging process. Therefore, this step reweights the merged clusters using the mass vector. This optimizes the distribution to preserve the characteristics of the original data points while reflecting the transmission relationship between the clusters, thereby more accurately representing the overall consumption habits of the merged clusters.

[0078] The optimized distribution of the merged cluster is determined according to the quality of each user in the quality vector and the corresponding distribution.

[0079] Specifically, the optimized distribution satisfies:

[0080] ;

[0081] Where, is the optimal distribution of the merged clusters, is the number of users in the merged cluster, is the mass vector of the merged cluster. The quality of each user, The first The corresponding distribution of users.

[0082] S303: Determine the intra-cluster homogeneity index.

[0083] It's important to note that the goal of hierarchical clustering is not only to minimize inter-cluster distances but also to ensure high similarity within clusters. Traditional Euclidean distance cannot effectively measure similarity at the distribution level. Therefore, this step uses the W distance to calculate the distance from each user to the centroid within the cluster. This allows us to assess the compactness of the data within the cluster from a distribution perspective and avoid misjudgments of homogeneity due to local density differences.

[0084] For the unmerged and merged clusters, the intra-cluster homogeneity index is determined based on the W distance from each user in each cluster to the centroid of the cluster.

[0085] Specifically, the intra-cluster homogeneity index satisfies:

[0086] ;

[0087] Where, is the intra-cluster homogeneity index, is the total number of unmerged and merged clusters, The first cluster in the unmerged and merged clusters The number of users in a cluster, express The i-th sample in , The first cluster in the unmerged and merged clusters In-cluster Users to The W distance between the centroids of the clusters.

[0088] S304: Determine whether to proceed to the next round of iteration.

[0089] It should be noted that clustering quality depends on a balance between intra-cluster compactness and inter-cluster separation. Relying on a single metric can lead to overfitting or underfitting. Therefore, this step considers both intra-cluster homogeneity and inter-cluster separation, using a dual threshold to determine clustering stability and rationality. This ensures that the final classification results reflect the inherent structure of consumer habits while avoiding meaningless over-merging.

[0090] Based on the intra-cluster homogeneity index and the minimum W distance between clusters, determining whether to perform the next round of iteration, in response to performing the next round of iteration, using the optimized distribution as the distribution of the merged cluster, and using the unmerged and merged clusters as clusters at the beginning of the next round of iteration.

[0091] Specifically, determining whether to proceed to the next round of iteration includes:

[0092] In response to the intra-cluster homogeneity index being less than a preset first abnormal threshold, and the minimum W distance (inter-cluster separability index) between the unmerged and merged clusters being greater than a preset second abnormal threshold, it is determined that the hierarchical clustering algorithm has reached the termination condition in this round of iteration, and the unmerged and merged clusters of this round of iteration are used as the classification results of the hierarchical clustering algorithm. Otherwise, the next round of iteration is performed.

[0093] Implementers can set the first abnormality threshold and the second abnormality threshold according to specific implementation conditions. For example, the first abnormality threshold is 0.1 and the second abnormality threshold is 0.2.

[0094] An embodiment of the present invention also discloses a consumption habit classification system based on big data, including a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, a consumption habit classification method based on big data according to the present invention is implemented.

[0095] The above system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are known in the art and will not be described in detail here.

[0096] The above are all preferred embodiments of the present invention, and are not intended to limit the scope of protection of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A consumption habit classification method based on big data, characterized by: include: The acquired consumption habit data points containing multiple dimensions of each user are grouped as a cluster. The distribution of each cluster is obtained using data distribution modeling. A hierarchical clustering algorithm is then used to obtain classification results that reflect the user's consumption habits, thereby achieving the classification of consumption habits based on big data. Each round of iterative process of obtaining classification results using the hierarchical clustering algorithm includes: For this round of iteration, based on the distribution, the W distance between all clusters at the beginning of the iteration is obtained, the clusters with the smallest W distance are merged, and the proportion of the number of users in each cluster at the time of merging to the number of users in the merged cluster is recorded as the weight of each cluster at the time of merging. Based on the weight, the Euclidean distance from each user in each cluster at the time of merging to each user in the merged cluster, and the transmission quality obtained using the linear programming solver, a function that minimizes the independent variable of the quality vector of the merged cluster under the constraint condition is constructed; based on the quality of each user in the quality vector and the corresponding distribution, the optimized distribution of the merged cluster is determined; for the unmerged and merged clusters, the intra-cluster homogeneity index is determined based on the W distance from each user in each cluster to the cluster center of mass; based on the intra-cluster homogeneity index and the minimum W distance between clusters, it is determined whether to perform the next round of iteration. In response to performing the next round of iteration, the optimized distribution is used as the distribution of the merged cluster, and the unmerged and merged clusters are used as the clusters at the beginning of the next round of iteration.

2. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The multiple dimensions include age, monthly consumption times, average single consumption amount, proportion of electronic product consumption, number of promotional activities participated in, repurchase rate and consumption period concentration index.

3. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The data distribution modeling adopts Dirac function as a modeling tool for data distribution modeling.

4. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The W distance is obtained using the Sinkhorn algorithm.

5. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The minimization function of the independent variable satisfies: ； Where, is the mass vector of the cluster after merging under the constraint conditions, for Minimize the independent variable function, is the mass vector to be solved, is the number of clusters when merging, The first The weight of a cluster, The first The number of users in a cluster, is the number of users in the merged cluster, The first In-cluster users to the merged cluster The Euclidean distance between users, The first In-cluster users to the merged cluster The transmission quality of each user obtained by using a linear programming solver is influence each other.

6. The method for classifying consumption habits based on big data according to claim 5, characterized in that: The constraints are: Ensure that the distribution is not lost: ; The quality of each user in the quality vector of the merged cluster is equal to the sum of the transmission quality of each user in each cluster at the time of merging to each user in the merged cluster: , is the mass vector of the merged cluster. The quality of individual users; The sum of the quality of each user in the quality vector of the merged cluster: ; Specific practical meaning of the parameters: .

7. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The optimized distribution satisfies: ； Where, is the optimal distribution of the merged clusters, is the number of users in the cluster after merging, is the mass vector of the merged cluster. The quality of each user, The first The corresponding distribution of users.

8. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The intra-cluster homogeneity index satisfies: ； Where, is the intra-cluster homogeneity index, is the total number of unmerged and merged clusters, The first cluster in the unmerged and merged clusters The number of users in a cluster, express The i-th sample in , The first cluster in the unmerged and merged clusters In-cluster Users to The W distance between the centroids of the clusters.

9. The method for classifying consumption habits based on big data according to claim 1, characterized in that: The determining whether to proceed to the next round of iteration includes: In response to the fact that the intra-cluster homogeneity index is less than a preset first abnormality threshold, and the minimum W distance between the unmerged and merged clusters is greater than a preset second abnormality threshold, it is determined that the hierarchical clustering algorithm has reached the termination condition in this round of iteration, and the unmerged and merged clusters of this round of iteration are used as the classification results of the hierarchical clustering algorithm. Otherwise, the next round of iteration is performed.

10. A consumption habit classification system based on big data, characterized by: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a consumption habit classification method based on big data according to any one of claims 1 to 9 is implemented.

Images