CN117708222A - Association rule mining method for customer segmentation - Google Patents

Abstract

The invention provides a client subdivision-oriented association rule mining method, which comprises the following steps: carrying out client subdivision portraits on the collected client data sets, and dividing the client data sets into clients with different value categories; counting commodity data purchased by clients of each price class, and generating a commodity transaction data set of the clients of each price class; and screening out frequent item sets from each commodity transaction data set to obtain a target association rule for purchasing commodities by clients of each price class. The method improves the accuracy of client subdivision, results profit margin and achieves maximization of enterprise profit.

Description

Association rule mining method for customer segmentation

Technical field

The invention belongs to the technical field of data mining, and in particular relates to an association rule mining method for customer segmentation.

Background technique

Association rule mining for customer segmentation can discover the relationship between products from consumer transaction records, and then bring more sales through product bundling or related recommendations. Existing methods only discover correlations between commodities, without considering the importance of each attribute, and ignore the differences and diversity among different customers, and do not conduct correlation analysis between commodities and customer group characteristics. , resulting in the mined association rules lacking pertinence and effectiveness.

Contents of the invention

In view of the shortcomings of the existing technology, the present invention proposes an association rule mining method for customer segmentation. By segmenting customers and analyzing the correlation between products and customer group characteristics under the conditions of customer segmentation, the invention improves the efficiency of customer segmentation. It improves the efficiency of association rule mining and improves the profit margin of results.

In order to achieve the above object, on the one hand, the present invention provides an association rule mining method for customer segmentation, including:

Conduct customer segmentation portraits on the collected customer data sets and divide them into customers of different value categories;

Statistics of commodity data purchased by customers in each value category are generated, and a commodity transaction data set of customers in each value category is generated;

Frequent item sets are filtered out from each of the commodity transaction data sets, and target association rules for customers purchasing commodities in each value category are obtained.

In some embodiments, the time interval between the customer's last purchase date and the end date of the statistical period, the number of purchases made by the customer during the statistical period, and the total amount spent by the customer on purchases during the statistical period are calculated based on the time interval between the customer's last purchase date and the end date of the statistical period. Total profit is used as an evaluation indicator to construct a customer segmentation model and determine the value score of each customer in the customer data set.

score＝w _R ×R+w _F ×F+w _M ×M _p

Among them, score represents the customer value score, R represents the time interval between the customer's last purchase date and the end date of the statistical period, which is the first evaluation indicator; F represents the number of purchases by the customer during the statistical period, which is the second evaluation indicator; As the third evaluation index, M represents the total amount spent by customers on purchases during the statistical period, P represents the total profit generated by each customer's consumption, q1 and q2 are weights; w _R , w _F , and w _M represent R and F respectively. M _pThe weight of these three evaluation indicators.

In some embodiments, the K-means clustering algorithm is used to cluster customer value scores and classify customers into different value categories, including:

Calculate the value score of each customer in the customer data set and generate a value score data set;

Randomly select K value score data from the value score data set as the cluster centers of K clusters, where K represents the number of cluster groups;

Assign each remaining data in the value score data set to the cluster with the shortest distance between its centers;

Recalculate the new cluster center of each cluster, and when the new cluster center is the same as the cluster center obtained in the previous iteration, output the number of cluster groups.

In some embodiments, the optimal K value is determined by determining the silhouette coefficients under different clustering group numbers K.

In some embodiments, the entropy weight method combined with the analytic hierarchy process is used to determine the weight of each evaluation indicator in the customer segmentation model, including:

Construct a hierarchical model for the first evaluation index, the second evaluation index, and the third evaluation index,

Select an index from the first evaluation index, the second evaluation index, and the third evaluation index, and determine the first weight of the index using the entropy weight method;

The first weight is input into the hierarchical model, and the second weight of the first evaluation index, the second evaluation index, and the third evaluation index is obtained using the analytic hierarchy process.

In some embodiments, the entropy weight method is used to determine the first weight of the evaluation index, including:

Each evaluation indicator includes sub-indicators corresponding to each customer;

Normalize each sub-indicator and calculate the information entropy value of each sub-indicator as:

In the formula, The ratio of one sub-indicator/> X _ij represents the j-th sub-indicator of the i-th customer sample; i=1, 2,...,n; j=1, 2,...,m;

Calculate the information entropy difference coefficient as:

P _j ＝1-H _j (j＝1,2,…,m)

The weight of each sub-index is calculated based on the information entropy coefficient: Get the first weight.

In some embodiments, frequent item sets are filtered out from each commodity transaction data set, and association rules for commodities purchased by customers in each value category are obtained, including:

Scan each commodity transaction data set and find frequent item sets based on support,

Generate candidate association rules from the obtained frequent item sets, discard the candidate association rules whose confidence value is lower than the minimum confidence, and obtain the target association rules for the products purchased by customers in each value category.

In some embodiments, finding frequent itemsets based on support includes:

Count the number of occurrences of frequently occurring item sets in each commodity transaction data set, delete items whose support is lower than the first threshold, sort the remaining items in descending order of support and store them in the item header table;

Read the item header table, delete items whose support is lower than the first support threshold, sort the remaining items in descending order of support, and determine frequent item sets;

Insert the sorted frequent itemsets in each commodity transaction data set into the frequent pattern tree in order to construct an FP-tree.

In some embodiments, a hash table and an ordered linked list are added before the header table of the FP-tree to record the current last node of each data item.

In some embodiments, before performing customer segmentation profiling on the customer data set, it also includes:

Preprocess customer data sets, including:

Discard and delete data that meets preset cleaning conditions;

Perform neighborhood attribute reduction on the customer data set after data cleaning, calculate the neighborhood of each sample, analyze the consistency between the neighborhood and the sample, and generate positive domain samples among the remaining samples.

On the one hand, the present invention also provides a customer segmentation-oriented association rule mining device, adopting the above-mentioned customer segmentation-oriented association rule mining method, and the device at least includes:

The customer segmentation module is used to perform customer segmentation portraits on the collected customer data sets and divide them into customers of different value categories;

The association rule mining module is used to count the commodity data purchased by customers in each value category and generate a commodity transaction data set for customers in each value category;

Frequent item sets are filtered out from each of the commodity transaction data sets, and target association rules for customers purchasing commodities in each value category are obtained.

On the other hand, the present invention also provides a readable storage medium. A program or instruction is stored on the readable storage medium. When the program or instruction is executed by a processor, the steps of the above-mentioned association rule mining method for customer segmentation are implemented, and can achieve the same technical effect.

It can be seen from the above solutions that the advantages of the present invention are:

The association rule mining method for customer segmentation provided by the present invention performs customer segmentation portraits on the collected customer data sets and divides them into customers of different value categories; then, counts the product data purchased by customers of each value category, Generate commodity transaction data sets for customers in each value category; finally, filter out frequent item sets from each commodity transaction data set to obtain target association rules for products purchased by customers in each value category. This method introduces the profit attribute to improve the RFM model and improve the profit margin of the results; and uses the entropy weight method and the analytic hierarchy process to comprehensively calculate the weights to determine the weight of each indicator of the improved RFM model, and uses the contour coefficient to determine the optimal K value. Use the K-means algorithm for clustering, establish a customer segmentation model, improve the accuracy of customer segmentation, and maximize corporate profits. At the same time, the FP-tree is used to filter out frequent item sets, determine the target association rules for the products purchased by customers in each value category, and add a hash table and ordered linked list in front of the item header table to improve the association by avoiding repeated scanning of the data set. Rule mining efficiency greatly reduces search time.

Description of the drawings

Figure 1 shows the overall flow diagram of the association rule mining method for customer segmentation;

Figure 2 shows the FP-tree frequent pattern tree structure diagram;

Figure 3 shows the architecture diagram of the association rule mining device for customer segmentation;

in:

300 - Association rule mining device for customer segmentation;

301-Data preprocessing module;

302-Customer segmentation module;

303-Association rule mining module;

S1-S4: steps.

Detailed ways

In order to make the above-mentioned features and effects of the present invention more clear and understandable, examples are given below and are described in detail with reference to the accompanying drawings.

The association rule mining method for customer segmentation provided by the present invention will be described in detail below.

As shown in Figure 1, Figure 1 shows the overall flow chart of the association rule mining method for customer segmentation.

An association rule mining method for customer segmentation, including:

S1. Preprocess the collected customer data set.

In this embodiment, the attribute simplification algorithm based on neighborhood rough sets preprocesses the collected customer data to ensure that the most important features are selected. At the same time, the algorithm also takes into account the correlation between attributes. The attribute simplification algorithm based on neighborhood rough sets continues to iterate. Each iteration introduces an attribute that can maximize the potential of the positive domain into the attribute simplified set until the positive domain stops changing. At this time, this clarifies an attribute of the data set. simplify. It is worth noting that the optimal neighborhood radius in the attribute reduction algorithm based on neighborhood rough sets must be determined by the specific conditions of the data set.

In the data preprocessing stage, first, the collected initial customer data is cleaned to analyze data patterns and outliers. In specific implementation, Pandas can be used to discard and delete data that meets the cleaning conditions. Then, the attribute reduction algorithm based on neighborhood rough sets is used to implement attribute reduction on the customer data set after data cleaning, which can be implemented through Python language. Specifically, by calculating the neighborhood of each sample and analyzing the consistency between the neighborhood and the sample, positive domain samples are generated among the remaining samples. This can avoid a large number of repeated operations and ensure that the most important features are selected. Taking into account the correlation between attributes.

S2. Carry out customer segmentation portraits on the preprocessed customer data set and divide it into customers of different value categories.

Since the accuracy of the customer value classification results of the traditional RFM model is not high, the income of each type of value customers lacks differentiation. Not only that, the corporate income and per capita profit of important value customers after grouping are sometimes smaller than the average Customers, this makes the differentiation of high-yield customers not very high, and the grouping trend is not obvious. Therefore, in this embodiment, profit attributes are introduced, the RFM model is improved, and customer segmentation portraits are performed based on the improved RFM model and K-means clustering algorithm, and customers are divided into different value categories. details as follows:

In this embodiment, based on the traditional RFM model, profit attributes are introduced, the calculation method of the indicator M in the model is modified, and an improved customer segmentation model is constructed. Specifically, the evaluation is based on the time interval between the customer's last purchase date and the end date of the statistical period, the number of purchases made by the customer during the statistical period, the total amount spent by the customer on purchases during the statistical period and the total profit generated by each customer's consumption. Indicators, construct a customer segmentation model, and determine the value score of each customer in the customer data set, namely:

score＝w _R ×R+w _F ×F+w _M ×M _p

Among them, score represents the customer value score, R represents the time interval between the customer's last purchase date and the end date of the statistical period, which is the first evaluation indicator; F represents the number of purchases by the customer during the statistical period, which is the second evaluation indicator; As the third evaluation index, M represents the total amount spent by customers on purchases during the statistical period, P represents the total profit generated by each customer's consumption, q ₁ and q ₂ are weights; w _R , w _F , and w _M represent R respectively. The weights of the three evaluation indicators F and M _p .

Furthermore, for the weight values of each evaluation index in the customer segmentation model, in this embodiment, the entropy weight method combined with the analytic hierarchy process is used to determine the weight of each evaluation index in the customer segmentation model. Specifically:

First, for the third evaluation index The weights q ₁ and q ₂ are determined by principal component analysis of the analytic hierarchy process in this embodiment. Specifically, the total profit P generated by each customer's consumption is used as the fourth evaluation indicator, and the average payment amount of the customer within a period of time, that is, the ratio of the total amount spent by the customer on purchases during the statistical period to the number of purchases made by the customer during the statistical period, is used as the fifth evaluation indicator. For evaluation indicators, construct a hierarchical model for the fourth and fifth evaluation indicators, and use the analytic hierarchy process to obtain the weights q ₁ and q ₂ of the fourth and fifth evaluation indicators. Specifically: by constructing the principal components of the fourth evaluation index and the fifth evaluation index, using the variance contribution rate of the principal components as the weight, and then weighting the average of the coefficients of the index in the linear combination of each principal component and performing normalization processing. The unified index weight is the weight q ₁ and q ₂ of the fourth evaluation index and the fifth evaluation index.

Further, after determining the weights q ₁ and q2 of the fourth and fifth evaluation indicators, in this embodiment, the entropy weight method combined with the analytic hierarchy process is used to determine the first, second and third evaluation indicators. The weight of the three evaluation indicators. In specific implementation, the following methods can be used:

First, through the above-mentioned principal component analysis method, a hierarchical model is constructed for the first evaluation index, the second evaluation index, and the third evaluation index.

Then, at least one index is selected from the first evaluation index, the second evaluation index, and the third evaluation index, and the first weight of the index is determined using the entropy weight method.

Finally, the first weight is input into the hierarchical model, and the second weight of the first evaluation index, the second evaluation index, and the third evaluation index is further obtained through the analytic hierarchy process.

During specific implementation, this embodiment uses the entropy weight method to determine the first weight of the evaluation index. The entropy weight method is an objective weighting method. This method determines the weight of each indicator by calculating entropy and entropy weight. The greater the uncertainty, the greater the entropy value, and the smaller the corresponding entropy weight. If the entropy weight is zero, it indicates that it cannot provide any useful information to decision makers, so the indicator is deleted. Specifically includes:

At least one index is selected from the first evaluation index, the second evaluation index, and the third evaluation index, and each evaluation index includes sub-indexes corresponding to each customer. Normalize each sub-indicator. Since the indicators in the customer segmentation model only include positive indicators and negative indicators, if the selected indicator is a positive indicator, the normalization method is:

z _ij ＝x _ij -min(x _1j ,x _2j ,…,x _nj )/max(x _1j ,x _2j ,…,x _nj )-min(x _1j ,x _2j ,…,x _nj )

If the selected indicator is a negative indicator, the normalization method is:

z _ij ＝max(x _1j ,x _2j ,…,x _nj) -x _ij /max(x _1j ,x _2j ,…,x _nj )-min(x _1j ,x _2j ,…,x _nj )

Among _them ,

Further, after normalizing each sub-indicator, the information entropy value of each sub-indicator is calculated, and the information entropy difference coefficient is determined from the information entropy value, and then the weight of each sub-indicator of the cluster is calculated based on the information entropy coefficient. Specifically, the information entropy value of each sub-indicator is expressed as: In the formula,/> The proportion of one sub-indicator The information entropy difference coefficient is expressed as: P _j =1-H _j (j = 1,2,...,m); the weight of each sub-index is calculated as:/> Thus the first weight is obtained.

In this embodiment, after the first weight is determined using the entropy weight method, the first weight is input into the hierarchical model and converted into pairwise comparison data in the AHP judgment matrix. The AHP is further calculated through the analytic hierarchy process. The total ranking weight is to obtain the second weight of the first evaluation index, the second evaluation index, and the third evaluation index, and then construct a customer segmentation model to determine the value score of each customer in the customer data set, that is:

score＝w _R ×R+w _F ×F+w _M ×M _p

Further, in this embodiment, after constructing a customer segmentation model and determining the value score of each customer in the customer data set, the K-means clustering algorithm is used to cluster the customer value scores and divide them into different value categories. customers, specifically including:

Calculate the value score of each customer in the customer data set and generate a value score data set;

K value score data are randomly selected from the value score data set as the cluster centers of K clusters, where K represents the number of cluster groups.

Assign each remaining piece of data in the value score dataset to the cluster with the shortest distance between its centers. In specific implementation, Euclidean distance measurement is usually used, that is: Where m is the dimension of the data object, X _j and C _ij are the j-th attribute values of X and C _i .

Then, recalculate the new cluster center of each cluster. Specifically, the mean of all data in the group can be used as the new cluster center. The square error is And when the new clustering center is the same as the clustering center obtained in the previous iteration, the number of clustering groups K is output.

Regarding how to select the optimal clustering group number K, this embodiment introduces the silhouette coefficient. By calculating the silhouette coefficient under different clustering grouping numbers K, the optimal K value is determined to improve the effectiveness of K-means clustering. The silhouette coefficient S of a single data i is expressed as:

Among them, a(i) represents the average degree of dissimilarity between data i and other data points in the same cluster, b(i) represents the minimum value of the average degree of dissimilarity between data i and other clusters, -1＜S(i )<1, S(i) represents the tightness of grouping of data points in the cluster. When S(i) is closer to 1, the clustering effect is better. When S(i) is approximately equal to 0, then a(i) and b(i) are approximately equal.

S3. Statistics of commodity data purchased by customers in each value category, and generates a commodity transaction data set for customers in each value category.

In this embodiment, after the customer segmentation portrait is implemented through step S2, the data of goods purchased by customers of each value category is calculated, and a commodity transaction data set of customers of each value category is generated for further mining of association rules. .

S4. Filter out frequent item sets from each product transaction data set, and obtain the target association rules for customers purchasing products in each value category.

In specific implementations, low support and confidence are the main indicators to evaluate the effectiveness of generated rules. If the support of an itemset is greater than or equal to the user-defined minimum support, it is called a frequent itemset. Therefore, in this embodiment, by scanning the commodity transaction data set twice, frequent item sets are filtered out based on support and confidence, and then association rules are extracted, and association rules whose confidence value is lower than the minimum confidence are discarded. Specifically, by scanning each commodity transaction data set, find frequent item sets based on support, and then generate candidate association rules from the obtained frequent item sets, discarding candidate association rules with confidence values lower than the minimum confidence level, and obtain each value Target association rules for category customers to purchase items.

In this embodiment, an FP-tree is generated for the data set, and then frequent item sets are continuously mined from the generated FP-tree. That is: first, count the number of occurrences of frequently appearing item sets in each commodity transaction data set, delete items whose support is lower than the first threshold, sort the remaining items in descending order of support and store them in the item header table; read From the item header table, items whose support is lower than the first support threshold are deleted again, and the remaining items are sorted in descending order of support to determine frequent item sets. Then, the sorted frequent itemsets in each commodity transaction data set are inserted into the frequent pattern tree in turn to construct an FP-tree. After constructing the FP-tree, you can start mining frequent itemsets on the previously processed data sets. That is, first obtain the conditional pattern base from the constructed FP-tree, then construct the conditional FP-tree, and finally iteratively repeat the first two steps until the tree contains an element. Then generate candidate association rules from the obtained frequent item sets, set the minimum confidence level, discard the candidate association rules whose confidence value is lower than the minimum confidence level, and further filter the candidate association rules, thus obtaining the target of purchasing goods by customers in each value category. Association rules.

In addition, since in the above process of constructing FP-tree, the data set needs to be scanned twice, the search time is increased and the operating efficiency of the algorithm is affected. In view of this situation, this embodiment further adds a hash table and an ordered linked list before the item header table of the FP-tree to record the current last node of each data item and make full use of the transactions obtained from the first scan of the database. information, so that the FP-tree frequent pattern tree can be constructed by scanning the data set only once, reducing search time and eliminating the need to repeatedly scan the original data.

This step is explained below with a specific commodity transaction data set. The commodity transaction data set is shown in Table 1.

Table 1-1 Commodity transaction data

According to the data in Table 1, set the minimum support to 0.4 and conduct association rule algorithm mining. The steps are as follows:

(1) Scan the commodity transaction data set in Table 1 for the first time and establish a header table, as shown in Table 1-2.

Table 1-2 Data header table

(2) Through the sorted item header table, scan the data set for the second time and delete non-frequent 1-item sets. The sorted data set is shown in Table 1-3.

Table 1-3 Sorted data set

(3) Establish FP-Tree to mine association rules.

Based on the sorted data set, an FP-tree frequent pattern tree is established, and the results are shown in Figure 2.

First, find the lowest child node in the tree and dig from the bottom up to get the subtrees {B, A, D}, {B, A, C, D}, {B, C}, {C, D} .

For the lowest child node D, the FP subtree is {B:8, A:6, C:3, D:2}, {B:8, A:6, D:2}, {C:5, D :3}, according to support filtering, the conditional pattern base is {B:6, A:6, D:4}; for the lowest child node C, its FP subtree is {8:7, C:1}, Filtering based on support, {B, C} is a non-frequent item set.

To sum up, it can be concluded that the frequent 3-item set is {B, A, D}, the frequent 2-item set is {{B, A}, {B, D}, {A, D}}, and the frequent 1-item set is {{B}, {A}, {C}, {D}}.

Finally, the obtained frequent item sets are generated into candidate association rules, the minimum confidence is set, and candidate association rules whose confidence value is lower than the minimum confidence are discarded.

In addition, further, when the amount of data is large, the structure of the constructed FP-tree will also become very large, which greatly affects the performance of the algorithm, and even makes it difficult to construct the conditional pattern tree in the iteration in the machine memory, making it impossible to realize FP -Growth mining algorithm. To address this problem, we can specifically parallelize the association rule algorithm on the Hadoop platform, achieve load balancing through dynamic grouping, use multiple local FP-trees to replace the global FP-tree, perform FP mining in several partitions at the same time, and finally use The master node aggregates the statistical results and filters out data that is not lower than the threshold. This can solve the problem that huge FP-Tree cannot reside in memory under massive data, and effectively realize distributed mining of FP under the Hadoop framework.

In summary, the association rule mining method for customer segmentation provided by the present invention performs customer segmentation profiling on the collected customer data sets and divides them into customers of different value categories; and then counts the purchases of customers in each value category. Commodity data is used to generate commodity transaction data sets for customers in each value category; finally, frequent item sets are filtered out from each commodity transaction data set to obtain target association rules for products purchased by customers in each value category. During the data preprocessing process, this method uses neighborhood rough sets for attribute reduction to improve the operating efficiency of data processing. During the customer segmentation process, the RFM customer segmentation model has the problem of low segmentation accuracy. Consider using clustering algorithms to improve the quality of customer segmentation. In order to maximize corporate profits, the RFM model was improved by introducing profit attributes, and then the entropy weight method and the analytic hierarchy process were used to comprehensively calculate the weights to determine the weight of each indicator of the improved RFM model, and finally the contour coefficient was used to determine the optimal K value. Finally, the K-means algorithm is used for clustering, a customer segmentation model is established, and the effectiveness of the algorithm is verified. In the process of association rule mining, in order to solve the problem that the FP-Growth association rule mining algorithm needs to scan the data set twice, resulting in low efficiency, a hash table and an ordered linked list are added before the header table of FP-tree to avoid Repeatedly scan the data set to improve mining efficiency, greatly reducing search time. This customer segmentation-oriented association rule mining method is suitable for retail companies, e-commerce companies, and financial companies to conduct customer segmentation portraits and identify high-value customers for member users. Mining association rules for the products they purchase can guide accurate recommendation of products or products. Placement, etc.

In addition, the above-mentioned embodiments of the present invention can be applied to terminal devices with the function of the association rule mining method for customer segmentation. The terminal devices may include personal terminals, host computer terminals, etc., and the embodiments of the present invention are not limited thereto.

Referring to Figure 3, Figure 3 shows a customer segmentation-oriented association rule mining device 300, which can realize customer detail-oriented association rule mining as shown in Figure 1. The separate association rule mining device can realize each process of the above-mentioned customer segmentation-oriented association rule mining method. It contains at least:

The data preprocessing module 301 is used to preprocess the collected customer data sets.

The customer segmentation module 302 is used to perform customer segmentation portraits on the preprocessed customer data set and divide it into customers of different value categories;

The association rule mining module 303 is used to count the commodity data purchased by customers in each value category and generate a commodity transaction data set for customers in each value category;

Frequent item sets are filtered out from each product transaction data set, and the target association rules for customers purchasing products in each value category are obtained.

In addition, it should be understood that in the association rule mining device 300 for customer segmentation according to the embodiment of the present application, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be allocated according to needs. The functional modules are completed, that is, the association rule mining device 300 for customer segmentation can be divided into functional modules different from the modules illustrated above to complete all or part of the functions described above.

In addition, embodiments of the present application also provide an electronic device, including a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, the above-mentioned customer-oriented details can be implemented. The steps of the association rule mining method are divided into different steps, and can achieve the same technical effect.

In addition, embodiments of the present application also provide a readable storage medium. Programs or instructions are stored on the readable storage medium. When the program or instructions are executed by the processor, the steps of the above-mentioned association rule mining method for customer segmentation are implemented, and can achieve the same technical effect.

It should be noted that, in this document, the terms "comprising", "comprises" or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article or device that includes a series of elements not only includes those elements, It also includes other elements not expressly listed or inherent in the process, method, article or apparatus. Without further limitation, an element defined by the statement "comprises a..." does not exclude the presence of additional identical elements in a process, method, article or apparatus that includes that element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved. Functions may be performed, for example, the methods described may be performed in an order different from that described, and various steps may also be applied, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better. implementation.

The embodiments of the present application have been described above in conjunction with the accompanying drawings. However, the present application is not limited to the above-mentioned specific implementations. The above-mentioned specific implementations are only illustrative and not restrictive. Those of ordinary skill in the art will Inspired by this application, many forms can be made without departing from the purpose of this application and the scope protected by the claims, all of which fall within the protection of this application.

Claims (10)

1. The association rule mining method for client-oriented subdivision is characterized by comprising the following steps of:

carrying out client subdivision portraits on the collected client data sets, and dividing the client data sets into clients with different value categories;

counting commodity data purchased by clients of each price class, and generating a commodity transaction data set of the clients of each price class;

and screening out frequent item sets from each commodity transaction data set to obtain a target association rule for purchasing commodities by clients of each price class.

2. The method according to claim 1, characterized by comprising:

constructing a client subdivision model by taking the time interval between the last purchase date of the client and the end date of the statistics period, the number of purchases of the client during the statistics period, the total amount spent by the client during the statistics period in purchases and the total profit generated by consumption of each client as evaluation indexes, determining the value score of each client in the client data set,

score＝w _R ×R+w _F ×F+w _M ×M _p

wherein score represents a customer value score, R represents a time interval between a last purchase date of the customer and an end date of the statistical period, and is a first evaluation index; f represents the purchase times of the clients in the statistical period and is a second evaluation index;for the third evaluation index, M represents the total amount spent by the customer in purchasing during the statistics period, P represents the total profit generated by consumption of each customer, and q1 and q2 are weights; w (w) _R 、w _F 、w _M Respectively represent R, F, M _p Weights of the three evaluation indexes.

3. The method according to claim 2, characterized by comprising:

clustering the customer value scores by adopting a K-means clustering algorithm, and dividing the customers with different value categories, wherein the method comprises the following steps:

counting the value score of each client in the client data set to generate a value score data set;

randomly selecting K value score data from the value score data set as cluster centers of K clusters, wherein K represents the cluster grouping number;

distributing each data remaining in the value score data set to the cluster with the shortest central distance;

and recalculating a new cluster center of each cluster, and outputting the cluster grouping number when the new cluster center is the same as the cluster center obtained in the last iteration.

4. A method according to claim 3, comprising:

the optimal K value is determined by determining the contour coefficients under different cluster groupings K.

5. The method of claim 2, wherein the step of determining the position of the substrate comprises,

and determining the weight of each evaluation index in the client subdivision model by utilizing an entropy weight method and a analytic hierarchy process, wherein the weight comprises the following steps:

constructing a hierarchical model for the first, second and third evaluation indexes,

selecting an index from the first evaluation index, the second evaluation index and the third evaluation index, and determining a first weight of the index by using an entropy weight method;

and inputting the first weight into the hierarchical model, and obtaining the second weights of the first evaluation index, the second evaluation index and the third evaluation index by using a hierarchical analysis method.

6. The method of claim 5, wherein the step of determining the position of the probe is performed,

determining a first weight of the evaluation index by using an entropy weight method comprises the following steps:

each evaluation index comprises a sub-index corresponding to each customer;

normalizing each sub-index, and calculating the information entropy value of each sub-index as follows:

wherein,the ratio of one index to->X _ij A j-th sub-index representing an i-th customer sample; i=1, 2, …, n; j=1, 2, …, m;

the information entropy difference coefficient is calculated as follows:

P _j ＝1-H _j (j＝1,2,…,m)

calculating the weights of all the sub indexes according to the information entropy coefficient as follows:a first weight is obtained.

7. The method of claim 1, wherein screening the frequent item sets from each of the commodity transaction data sets to derive association rules for each value class customer to purchase commodity comprises:

scanning each commodity transaction data set, finding frequent item sets according to the support degree,

and generating candidate association rules from the obtained frequent item set, discarding candidate association rules with confidence values lower than the minimum confidence values, and obtaining target association rules of commodity purchase of clients with each value category.

8. The method of claim 7, wherein the step of determining the position of the probe is performed,

finding a frequent item set according to the support, including:

counting the occurrence times of the item sets frequently occurring in each commodity transaction data set, deleting the items with the support degree lower than a first threshold value, sorting the rest items according to the support degree descending order, and storing the rest items in an item header table;

reading the item header table, deleting items with the support degree lower than a first support degree threshold value, sorting the rest items in descending order of the support degree, and determining frequent item sets;

and sequentially inserting the ordered frequent item sets in each commodity transaction data set into a frequent pattern tree to construct an FP-tree.

9. The method as recited in claim 8, further comprising:

and adding a hash table and an ordered linked list in front of the head table of the FP-tree for recording the current last node of each data item.

10. The method of claim 1, further comprising, prior to client subdivision representation of the client data set:

preprocessing a customer data set, comprising:

discarding and deleting the data meeting the preset cleaning conditions;

and carrying out neighborhood attribute reduction on the client data set after data cleaning, calculating the neighborhood of each sample, analyzing the consistency of the neighborhood and the sample, and generating a positive domain sample in the residual samples.