Association mining method, equipment, medium and product for lost customer behavior characteristic mode

Abstract

The invention discloses a lost customer behavior characteristic mode association mining method and device, a medium and a product, and relates to the technical field of financial intelligent prediction, and the method comprises the steps: carrying out the preprocessing of a preset lost customer data set, and obtaining a transaction data table, the lost customer data set comprises at least one group of lost customer behavior characteristics and at least one lost mode; determining a frequent item set according to the transaction data table, extracting association rules in the frequent item set, and determining a key association rule from the association rules; determining a mapping relationship between the lost customer behavior feature combination and the lost mode according to the key association rule; and querying a target contact loss mode corresponding to the to-be-predicted contact loss customer behavior feature combination in the mapping relationship, and generating a risk early warning strategy according to the target contact loss mode. According to the method, the complex association between the lost customer behavior characteristics and the lost mode can be analyzed.

Description

Technical Field

[0001] This application relates to the field of financial intelligent prediction technology, and particularly to a method, device, medium and product for mining the association between the behavior characteristics and patterns of lost customers. Background Art

[0002] With the development of the financial market, the problem of lost loan customers has become increasingly prominent, bringing huge risks to financial institutions. In today's financial environment, the phenomenon of lost loan customers not only affects the loan recovery rate, resulting in direct economic losses to financial institutions, but may also trigger a chain reaction, affecting the stability and healthy development of the financial market. Existing research mostly analyzes the behavior patterns of lost loan customers from data resources such as customer social communication, daily consumption transactions, and financial platform lending, such as information matching models, path association index models, etc., but few literatures deeply explore the association between lost characteristics and lost patterns. Most research focuses on using customers' social relationships, daily behavior trajectories, and multi-platform information to identify lost customers. For example, lost customer contact information is identified through mobile social big data, or the transaction behavior patterns of loan customers within a specific time period are mined. Although these studies have enriched the understanding of the behavior patterns of lost loan customers to a certain extent, they mainly focus on the analysis of single data sources or single behavior dimensions, lacking in-depth exploration of the complex association between lost characteristics and lost patterns. This limitation makes it difficult for financial institutions to accurately identify and predict lost risks in actual operations, thus affecting the effectiveness of risk management strategies. Summary of the Invention

[0003] The main purpose of this application is to provide a method for mining the association between the behavior characteristics and patterns of lost customers, aiming to analyze the complex association between the behavior characteristics and lost patterns of lost customers.

[0004] To achieve the above purpose, this application proposes a method for mining the association between the behavior characteristics and patterns of lost customers, including:

[0005] Preprocess a preset lost customer data set to obtain a transaction data table, where the lost customer data set includes at least one set of lost customer behavior characteristics and at least one lost pattern;

[0006] Determine frequent item sets according to the transaction data table, extract association rules from the frequent item sets, and determine key association rules from the association rules;

[0007] Determine the mapping relationship between the lost customer behavior characteristic combinations and lost patterns according to the key association rules;

[0008] Query the target lost pattern corresponding to the lost customer behavior characteristic combination to be predicted in the mapping relationship, and generate a risk warning strategy according to the target lost pattern.

[0009] In one embodiment, the steps of preprocessing a preset dataset of lost - contact customers to obtain a transaction data table include:

[0010] Determine the discrete feature data in the lost - contact customer dataset, and fill the missing values in the discrete feature data according to the mode filling strategy;

[0011] Convert the character - form data in the discrete features into numerical - form data to obtain the transaction data corresponding to the discrete feature data;

[0012] Determine the continuous feature data in the lost - contact customer dataset, and fill the missing values in the continuous feature data according to the median filling strategy;

[0013] According to the value range of the continuous feature data, divide the continuous feature data into multiple interval data, and set identifiers for each interval data to obtain the transaction data corresponding to the continuous feature data;

[0014] Merge the transaction data corresponding to the discrete feature data and the transaction data corresponding to the continuous feature data, and convert them into string - type to obtain string - type transaction data;

[0015] Perform encoding processing on the string - type transaction data, and convert the encoded string - type transaction data into transaction data in a two - dimensional table - type data structure to obtain a transaction data table.

[0016] In one embodiment, the steps of determining frequent item sets according to the transaction data table include:

[0017] Count the frequency of each item of data in the transaction data table, and determine the item support of each item of data as the frequency;

[0018] Delete the data in the transaction data table whose item support is lower than the preset item support threshold to obtain the remaining data in the transaction data table;

[0019] For the remaining data in the transaction data table, sort them in descending order according to the item support to obtain a transaction list;

[0020] Construct a frequent tree based on the data in the transaction list;

[0021] For each target item data in the frequent tree, extract the conditional pattern base of the target item data, and mine the conditional frequent item sets of the conditional pattern base;

[0022] Combine the conditional frequent item sets with the target item data to form frequent item sets.

[0023] In one embodiment, before the step of deleting the data in the transaction data table whose item support is lower than the preset item support threshold, it includes:

[0024] The items whose item support is higher than the preset initial item support threshold are determined as the initial frequent itemsets;

[0025] If the frequent item data coverage of the initial frequent item set is less than the preset coverage threshold, the initial item support threshold is lowered;

[0026] If the noise item data of the initial frequent item set is more than the preset number of rules, the initial item support threshold is increased, where the noise item data refers to data that has no practical meaning or represents contingency;

[0027] The adjusted initial item support threshold is determined as the preset item support threshold.

[0028] In one embodiment, the step of extracting association rules from frequent item sets includes:

[0029] Divide each frequent item set into multiple different combinations of antecedents and consequents, where each combination includes a antecedent and a consequent;

[0030] Each combination is determined as an association rule, wherein the combination corresponds to the association rule one by one.

[0031] In one embodiment, the step of determining the key association rule from the association rules includes:

[0032] Calculate the rule score of each association rule, where the rule score includes the rule confidence and the lift index. The lift index is the ratio of the rule confidence to the item support of the subsequent item.

[0033] Determine an association rule whose rule confidence is higher than a preset rule confidence threshold and whose lift index is higher than a preset lift index threshold as a key association rule; or

[0034] An association rule tree diagram is generated according to the rule score of each association rule, and an association rule with an area greater than a preset area in the association rule tree diagram is determined as a key association rule, wherein the occupied area is calculated according to the product of the rule confidence and the lift index.

[0035] In one embodiment, before determining an association rule having a rule confidence higher than a preset rule confidence threshold and a boost index higher than a preset boost index threshold as a key association rule, the step includes:

[0036] Determine the items whose rule confidence is higher than the preset initial rule confidence threshold as initial association rules;

[0037] If the initial association rules have redundancy, the confidence threshold of the initial rules is increased;

[0038] If the number of initial association rules is less than the preset number of rules, the confidence threshold of the initial rules is lowered;

[0039] Determine the adjusted initial rule confidence threshold as the preset rule confidence threshold.

[0040] In addition, to achieve the above object, the present application also proposes an associated mining device for the behavior feature pattern of lost customers, which includes:

[0041] A preprocessing module that preprocesses a preset lost customer data set to obtain a transaction data table, where the lost customer data set includes at least one set of lost customer behavior features and at least one lost mode;

[0042] An association rule module that determines frequent item sets according to the transaction data table, extracts association rules from the frequent item sets, and determines key association rules from the association rules;

[0043] An association conclusion module that determines the mapping relationship between the lost customer behavior feature combination and the lost mode according to the key association rules;

[0044] A risk strategy module that queries the target lost mode corresponding to the lost customer behavior feature combination to be predicted in the mapping relationship, and generates a risk warning strategy according to the target lost mode.

[0045] In addition, to achieve the above object, the present application also proposes a device for the associated mining of the behavior feature pattern of lost customers, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the method for the associated mining of the behavior feature pattern of lost customers as described above.

[0046] In addition, to achieve the above object, the present application also proposes a medium, which is a computer-readable storage medium, and a computer program is stored on the medium, and when the computer program is executed by a processor, it implements the steps of the method for the associated mining of the behavior feature pattern of lost customers as described above.

[0047] In addition, to achieve the above object, the present application also provides a product, which is a computer program product, and the computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps of the method for the associated mining of the behavior feature pattern of lost customers as described above.

[0048] One or more technical solutions proposed by the present application have at least the following technical effects:

[0049] This application preprocesses a preset dataset of lost-contact customers to obtain a transaction data table. Among them, the dataset of lost-contact customers includes at least one set of lost-contact customer behavior characteristics and at least one lost-contact mode. This process ensures the consistency and analyzability of the data, and helps to improve the accuracy and efficiency of association rule mining. Determine frequent item sets according to the transaction data table, extract the association rules in the frequent item sets, and determine the key association rules from the association rules. This step reveals the potential patterns and rules hidden in the data through in-depth analysis of the transaction data, and identifies the key association rules crucial for risk management. Determine the mapping relationship between the combination of lost-contact customer behavior characteristics and the lost-contact mode according to the key association rules, establish the connection between specific behavior characteristic combinations and specific lost-contact modes, so that financial institutions can more clearly understand the risk factors behind different characteristic combinations. Query the target lost-contact mode corresponding to the combination of lost-contact customer behavior characteristics to be predicted in the mapping relationship, and generate a risk warning strategy according to the target lost-contact mode. Using the established mapping relationship, the potential lost-contact mode of the customer to be predicted can be quickly located, so as to formulate corresponding risk warning strategies in a timely manner. This method analyzes the complex association between the behavior characteristics of lost-contact customers and the lost-contact mode, improves the ability of financial institutions to deal with the problem of loan loss of contact, enables them to identify and intervene in possible risks at an early stage, effectively reduces economic losses, and protects the stability and healthy development of the financial market. Brief Description of the Drawings

[0050] The accompanying drawings here are incorporated into the description and form a part of this description, showing embodiments consistent with this application, and are used together with the description to explain the principles of this application.

[0051] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the accompanying drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0052] Figure 1 It is a schematic flowchart of the first embodiment of the method for mining the association between the behavior characteristics and patterns of lost-contact customers in this application;

[0053] Figure 2 It is a heat map of frequent item sets of the method for mining the association between the behavior characteristics and patterns of lost-contact customers in this application;

[0054] Figure 3 It is a tree diagram of association rules of the method for mining the association between the behavior characteristics and patterns of lost-contact customers in this application;

[0055] Figure 4 It is a flowchart of the work for mining frequent item sets and association rules of the device for mining the association between the behavior characteristics and patterns of lost-contact customers in the embodiment of this application;

[0056] Figure 5 It is a schematic diagram of the device structure of the hardware operating environment involved in the method for mining the association of the behavior characteristic patterns of lost customers in the embodiments of the present application.

[0057] The realization of the purpose, functional characteristics and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0058] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0059] In order to better understand the technical solutions of the present application, the following will be described in detail with reference to the accompanying drawings of the specification and specific embodiments.

[0060] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device, a terminal system, etc. that can implement the above functions. The following takes the system as an example to illustrate this embodiment and the following embodiments.

[0061] Based on this, this embodiment provides a method for mining the association of the behavior characteristic patterns of lost customers, referring to Figure 1 , Figure 1 It is a flowchart of the method for mining the association of the behavior characteristic patterns of lost customers in the present application. The method for mining the association of the behavior characteristic patterns of lost customers includes steps S10 to S30:

[0062] Step S10, preprocess the preset lost customer data set to obtain a transaction data table, where the lost customer data set includes at least one set of lost customer behavior characteristics and at least one lost mode;

[0063] Step S20, determine the frequent item sets according to the transaction data table, extract the association rules in the frequent item sets, and determine the key association rules from the association rules;

[0064] Step S30, determine the mapping relationship between the lost customer behavior characteristic combinations and the lost modes according to the key association rules;

[0065] Step S40, query the target lost mode corresponding to the lost customer behavior characteristic combination to be predicted in the mapping relationship, and generate a risk warning strategy according to the target lost mode.

[0066] It should be noted that in this embodiment, preprocessing the preset dataset of lost-contact customers refers to converting the original data (such as the behavioral characteristics and lost-contact patterns of lost-contact customers) into structured transaction data through operations such as cleaning, standardization, and discretization. Each transaction represents a set of all relevant characteristics and patterns of a lost-contact customer. Subsequently, by mining the frequent item sets in the transaction data (i.e., the combinations of characteristics that appear simultaneously with a frequency exceeding a threshold), and combining association rule extraction techniques (such as support and confidence calculations), important rules are screened out. Finally, these rules are mapped to the relationship between the behavioral characteristics and lost-contact patterns of lost-contact customers, and specific risk warning strategies are formulated based on this relationship.

[0067] Specifically, the whole process starts with preprocessing the preset dataset of lost-contact customers. This step includes operations such as cleaning data, handling missing values, discretizing continuous features, and encoding discrete features, aiming to ensure the consistency and usability of the data and obtain a structured transaction data table. Next, frequent item sets are mined from the transaction data table. For example, the FP-Growth algorithm (a data mining algorithm), which can effectively reduce the number of data reads and improve the algorithm efficiency by constructing the FP-Tree (frequent tree) data structure and only scanning the database twice. When constructing the FP-Tree, first count the support count of each item, then generate a frequent item list sorted in descending order of support, and finally construct a tree structure according to this list. By recursively traversing the FP-Tree, all frequent item sets can be mined. Subsequently, association rules are extracted from the frequent item sets, and the support and confidence of each rule are calculated to evaluate the strength and reliability of the rule. Support represents the frequency of transactions containing all items in the association rule in the lost-contact customer dataset, and confidence represents the probability that the consequent is also included in the transactions containing the antecedent. According to the set minimum support and confidence thresholds, key association rules are screened out. Then, based on these key association rules, it is analyzed which combinations of lost-contact customer behavioral characteristics are significantly associated with specific lost-contact patterns, thus obtaining an association conclusion. For example, it is found that the feature combination {x2 = on-net but unavailable, x12 = 180 - 270 days, x13 = 300000 - 500000 yuan} has a strong association with the hide-and-seek lost-contact pattern. Finally, based on the association conclusion, that is, the obtained lost-contact pattern, risk warning strategies are formulated, such as strengthening collection efforts and adjusting the repayment plan.

[0068] In this embodiment, in order to further improve the accuracy and timeliness of risk warning, automated feature engineering can be introduced in the data preprocessing stage. By using machine learning algorithms to automatically screen the features that are most influential in the identification of the loss of contact pattern, manual intervention is reduced and the accuracy of feature selection is improved. At the same time, in the process of frequent item set mining, combined with parallel computing technology, a multi-core processor is used to process different data subsets simultaneously, significantly accelerating the mining speed, enabling the system to more quickly respond to the analysis requirements of large-scale loss of contact customer data sets.

[0069] In this embodiment, the preset loss of contact customer data set is preprocessed to obtain a transaction data table. Among them, the loss of contact customer data set includes at least one set of loss of contact customer behavior characteristics and at least one loss of contact pattern. This process ensures the consistency and analyzability of the data, which helps to improve the accuracy and efficiency of association rule mining. Determine the frequent item sets according to the transaction data table, extract the association rules in the frequent item sets, and determine the key association rules from the association rules. This step reveals the potential patterns and rules hidden in the data through in-depth analysis of the transaction data, and identifies the key association rules that are crucial for risk management. Determine the mapping relationship between the loss of contact customer behavior characteristic combinations and the loss of contact patterns according to the key association rules, establish the connection between specific behavior characteristic combinations and specific loss of contact patterns, so that financial institutions can more clearly understand the risk factors behind different characteristic combinations. Query the target loss of contact pattern corresponding to the loss of contact customer behavior characteristic combination to be predicted in the mapping relationship, and generate a risk warning strategy according to the target loss of contact pattern. Using the established mapping relationship, the potential loss of contact pattern of the customer to be predicted can be quickly located, so as to timely formulate corresponding risk warning strategies. This method analyzes the complex association between the loss of contact customer behavior characteristics and the loss of contact patterns, improves the ability of financial institutions to deal with the problem of loan loss of contact, enables them to identify and intervene possible risks at an early stage, effectively reduces economic losses, and protects the stability and healthy development of the financial market.

[0070] In a feasible implementation manner, the steps in step S10 may include steps T10 to T40:

[0071] Step T10, determine the discrete feature data in the loss of contact customer data set, and fill the missing values in the discrete feature data according to the mode filling strategy;

[0072] Step T20, convert the character-form data in the discrete features into numerical-form data to obtain the transaction data corresponding to the discrete feature data;

[0073] Step T30, determine the continuous feature data in the loss of contact customer data set, and fill the missing values in the continuous feature data according to the median filling strategy;

[0074] Step T40: Divide the continuous feature data into multiple interval data according to the value range of the continuous feature data, and set identifiers for each interval data to obtain the transaction data corresponding to the continuous feature data.

[0075] Step T50: Merge the transaction data corresponding to the discrete feature data and the transaction data corresponding to the continuous feature data, and convert them into string type to obtain the string type transaction data.

[0076] Step T60: Perform encoding processing on the string type transaction data, and convert the encoded string type transaction data into transaction data with a two-dimensional table data structure to obtain a transaction data table.

[0077] It should be noted that in this embodiment, discrete feature data refers to those data types with a finite number of possible values, such as gender, region, etc.; the mode filling strategy refers to the method of using the value with the highest frequency in the lost connection customer dataset to replace the missing value, and this method is suitable for filling the gaps in discrete feature data. Continuous feature data refers to those data types that can take any numerical value, such as age, income, etc.; the median filling strategy uses the median value in the lost connection customer dataset to fill the missing value and is applicable to the processing of continuous feature data. Interval data is obtained by dividing the continuous feature data into several parts according to its value range and assigning an identifier to each part for subsequent analysis. Numerical form data refers to numerical encoding. Since the algorithm used in this embodiment cannot recognize character form data (such as "A", "B"), it is necessary to convert it into numerical form encoding according to a preset mapping table. For example, referring to the rule of A = 1, B = 2 in the specified mapping table, convert "A" and "B" into 1 and 2 respectively. In this embodiment, the two-dimensional table data structure specifically refers to the DataFrame (a table data structure) data structure.

[0078] Exemplarily, in the credit field, the default records of loan customers constitute a typical imbalanced lost connection customer dataset. The characteristic of such a lost connection customer dataset is that it contains a large number of samples of repaid loans, while the default (i.e., unpaid loan) samples are relatively few. Particularly noteworthy is that among these limited default cases, the lost connection customer samples only account for a very small proportion. Given that the lost connection samples involve the personal privacy information of loan customers and are difficult to obtain, it is challenging to conduct direct research. Therefore, this study relies on the lost connection customer dataset and uses the FP-Growth algorithm to deeply explore the lost connection characteristics of loan lost connection customers and the underlying association rules and patterns.

[0079] The dataset of lost-contact customers contains 60 samples, including the identity characteristics of loan lost-contact customers, as well as the characteristics of permanent residence addresses, the in-network status of mobile phone numbers, the call record characteristics of mobile phone numbers, the short message signaling characteristics, the email characteristics, the other signaling characteristics, the in-network status of the mobile phone numbers of emergency contacts, the call record characteristics of mobile phone numbers, the short message signaling characteristics, the email characteristics, and 17 lost-contact characteristics of loan lost-contact customers, such as loan type, overdue days, overdue loans, estimated recoverable amount, lost-contact duration, number of valid contacts, and relationship with valid contacts, and a total of 18 characteristics including lost-contact patterns. The feature codes, names, value ranges, and feature attributes of the dataset of lost-contact customers are shown in the following table.

[0080] Table 1 Feature Codes, Names, Value Ranges, and Attributes of the Dataset of Lost-Contact Customers

[0081]

[0082]

[0083] Next, data processing is performed on the dataset of lost-contact customers. First, features irrelevant to the research are removed. The dataset of lost-contact customers has a total of 18 features. Among them, the identity characteristics of loan lost-contact customers have no direct effect on mining the association rules between lost-contact characteristics and lost-contact patterns and will interfere with the subsequent data processing process. Therefore, in this paper, the identity characteristic is deleted in the data processing link, making the data more focused on the features related to lost-contact patterns. In addition, according to the research, seven features in the dataset of lost-contact customers, namely the short message signaling characteristics, the email characteristics, the other signaling characteristics, the in-network status of the mobile phone numbers of emergency contacts, the short message signaling characteristics of emergency contacts, the email characteristics of emergency contacts, and the relationship between loan lost-contact customers and valid contacts, have no impact on lost-contact pattern recognition. In order to enhance the robustness of the model, in some application scenarios, these features with low contribution degrees can be considered for removal or reduction. Therefore, we further remove the seven features x4, x5, x6, x7, x9, x10, and x17.

[0084] [[ID=IS]]Secondly, missing values are filled. The discrete feature columns x1, x2, x11, and x18 in the dataset of lost-contact customers are identified. For the missing values in these columns, the mode filling strategy is adopted. This strategy is based on statistical principles and can effectively maintain the integrity of the data and the accuracy of the analysis. Further, the continuous feature columns x3, x8, x12, x13, x14, x15, and x16 are identified. For the missing values in these columns, the median filling strategy is adopted. This method is widely considered an effective means of dealing with missing continuous data in statistics because it can better maintain the distribution characteristics of the data.

[0085] Then, segment and encode the continuous features. Since the FP-Growth algorithm mainly processes discrete data, continuous data will result in the generation of too many frequent itemsets and cannot reflect meaningful association patterns. For example, different ranges of overdue days may have stronger associations with different disconnection patterns, but it is difficult to discover such relationships directly using the original continuous overdue day values. Therefore, we need to further discretize the data to ensure that all input data is discrete. Based on the value ranges of x3, x8, x12, x13, x14, x15, and x16 in this paper, the data's belonging intervals are segmented, and a label is assigned to each segment. For example, x15 is divided into 5 interval segments according to [0, 30), [31, 60), [61, 90), [91, 120), [120, M) (M represents the maximum value of x15), and the values are assigned as 1, 2, 3, 4, and 5 respectively.

[0086] Finally, encode the discrete features. Since the original data of the two discrete features x1 and x18 are already numerically encoded, there is no need to process these two features. The value range of x2 is A, B, C, D, E, F, and the value range of x11 is P, Q, R, S. The algorithm cannot recognize the categorical information in these character forms and needs to convert them into numerically encoded forms for association rule mining. Therefore, in this paper, different categories in x2 are replaced with numerical encodings, and different loan types in x11 are replaced with numerical encodings.

[0087] After processing, the data is shown in Table 2:

[0088] Table 2 Data after data processing

[0089] x1 x2 x3 x8 x11 x12 x13 x14 x15 x16 x18 1 1 5 3 2 2 5 5 5 5 1 1 6 5 3 4 1 5 5 5 5 1 1 4 5 5 3 5 3 0 5 5 2 1 4 2 3 4 3 5 5 5 5 1 1 4 2 2 1 3 5 5 5 5 1 0 1 2 3 3 1 2 0 3 5 0 1 4 5 2 2 5 4 3 5 5 2 1 4 4 1 3 5 2 0 5 5 2 1 4 2 2 2 4 5 5 4 5 1 0 1 3 4 3 3 1 0 5 5 0 ... ... ... ... ... ... ... ... ... ... ... 0 4 1 2 1 4 5 5 5 5 2

[0090] Next, to construct transaction data, all features need to be converted to string type. In view of the requirement for data type consistency when constructing transaction data later, all data in this paper is converted to string type.

[0091] Here, we regard each row of data as a transaction and each feature as an item. First, create an empty list transactions to store transaction data. Then, by traversing each row of the data frame, extract the data of the specified columns (['x1', 'x2', 'x3', 'x8', 'x11', 'x12', 'x13', 'x14', 'x15', 'x16', 'x18']) for each row and convert them into a list form, which is used as a transaction and added to the transactions list.

[0092] In this embodiment, we use a transaction encoder to preprocess transaction data for subsequent frequent itemset mining. The specific steps are as follows: First, initialize the TransactionEncoder, create a TransactionEncoder object to encode the transaction data; Second, use the fit method of the transaction encoder to learn the items in the transaction data and convert the data into one-hot encoding form. One-hot encoding is a method of converting categorical variables into numerical form, and assigns a unique binary vector to each category; Then, convert the one-hot encoded array into a DataFrame (a format), and specify the column names when constructing the data frame.

[0093] Furthermore, this embodiment can also consider an optimization scheme, that is, introducing an adaptive interval partitioning technique in the interval data partitioning process, automatically adjusting the size and number of intervals according to the data distribution, rather than manually setting a fixed interval range. This improvement is especially effective when dealing with data with highly skewed distributions, because it can more accurately capture the changing trend of the data and improve the prediction accuracy of the model. For example, when analyzing the loan default patterns in different regions, the interval partitioning of the overdue days can be adaptively adjusted according to the economic level differences in each region, so as to achieve more accurate risk assessment.

[0094] This embodiment improves the quality and usability of the data by accurately filling missing values, reasonably partitioning intervals, and converting data formats, providing a solid foundation for subsequent association rule mining.

[0095] Based on Embodiment 1 of this application, in Embodiment 2 of this application, the same or similar content as the above Embodiment 1 can be referred to the above introduction and will not be repeated hereinafter. On this basis, the steps of step S20 further include steps A10 to A60:

[0096] Step A10, count the frequency of each item of data in the transaction data table, and determine the item support of each item of data as the frequency;

[0097] Step A20, delete the data in the transaction data table whose item support is lower than the preset item support threshold, and obtain the remaining data in the transaction data table;

[0098] Step A30, for the remaining data in the transaction data table, sort them in descending order according to the item support to obtain a transaction list;

[0099] Step A40, construct a frequent tree based on the data in the transaction list;

[0100] Step A50, for each target item data in the frequent tree, extract the conditional pattern base of the target item data and mine the conditional frequent itemset of the conditional pattern base;

[0101] Step A60: Combine the conditional frequent item sets with the target item data to form frequent item sets.

[0102] It should be noted that in this embodiment, item support refers to the frequency of each item's occurrence in the transaction lost - contact customer dataset, which reflects the prevalence of this item in all transactions; the preset item support threshold is a user - defined minimum support standard, and items below this standard will be removed to reduce computational complexity and improve mining efficiency; the transaction list is an ordered list obtained by sorting the support counts of the remaining items in descending order, and is used to construct the frequent pattern tree (FP - tree) subsequently; the frequent pattern tree is a data structure for efficiently storing and retrieving frequent item sets; the conditional pattern base is a set of prefix path collections extracted from the frequent pattern tree for the target item data, and these paths represent the context information when the target item data appears in different transactions; the conditional frequent item sets are the frequent item sets mined from the conditional pattern base.

[0103] Exemplarily, if the FP - Growth algorithm is used to mine frequent item sets from transaction data, first scan the transaction data table to count the frequency of each item's occurrence in the transaction data table, and this frequency is the support count of this item. Let the lost - contact customer dataset be D. For item i, its support count support(i) = |{t ∈ D|i ∈ t}|, where t represents the transactions in the lost - contact customer dataset. Set the minimum support threshold (determined by the user according to actual requirements and data characteristics), and remove the items with support counts lower than this threshold from the lost - contact customer dataset. That is, if the support count of item i, support(i) < min_sup (min_sup is the minimum support threshold), then item i does not participate in subsequent analysis. Sort the remaining frequent items in descending order according to their support counts to obtain the frequent item list L. Each element in this list contains a frequent item and its corresponding support count, specifically represented as

[0104] L = {(i1, support(i1)), (i2, support(i2)),...,(i n , support(i n ))}

[0105] and satisfy support(i1) ≥ support(i2)... ≥ support(i n ).

[0106] Scan the lost - contact customer dataset for the second time, re - sort the items in each transaction according to the order of the frequent item list L, and delete the items not in L. In this way, each transaction is transformed into an ordered frequent item list. For example, the original transaction t = {i1, i2,…, i m},

[0107] The frequent item list L = {(j1, support(j1)), (j2, support(j2)),..., (j n , support(j n ))}, and the adjusted transaction where are the frequent items in t arranged in the order of L.

[0108] Next, construct the FP-tree. First, create the root node, labeled as null (empty). For each sorted transaction, start from the root node and traverse the FP-tree downward item by item in the transaction. If the current item already exists on the path, increment the count of that node by 1; if it does not exist, create a new node, set the count to 1, and link it to the parent node.

[0109] Suppose the transaction The current node is node. If already exists on the path, then and node.count += 1; if it does not exist, create a new node new_node, make new_node.count = 1, set new_node.parent = node, then update node to new_node, and continue to process the next item.

[0110] For each frequent item i, construct its conditional pattern base. The conditional pattern base is a subtree composed of the transaction paths containing item i, where the count of each node represents the support count of that node in the transactions containing item i.

[0111] For the frequent item i, find all paths in the FP-tree that end with i from the FP-tree. These paths form the conditional pattern base. Traverse each path from the bottom up, and set the count of each node to the count of its parent node until reaching the root node.

[0112] For example, for the path node1 → node2 → node3 → i, if initially node3.count = 3, node2.count = 2, node1.count = 1, then after adjustment node3.count = 2, node2.count = 1, node1.count = 1.

[0113] Construct the conditional FP-tree based on the conditional pattern base and perform recursive mining until the conditional FP-tree is empty or contains only one node.

[0114] Let the conditional pattern base be CPB (a custom variable). The process of constructing the conditional FP-tree is similar to that of constructing the initial FP-tree, except that the dataset of lost customers becomes CPB. Then, recursively mine the conditional FP-tree to obtain the frequent item sets containing item i.

[0115] Finally, combine the frequent item sets containing item i with item i to obtain the complete frequent item sets. The results are shown in Table 3 (the "frozenset" in the table is an immutable set type):

[0116] Table 3

[0117]

[0118] Furthermore, this embodiment can also consider an optimization scheme by introducing a method of dynamically adjusting the item support threshold during the process of constructing the frequent tree. Specifically, the item support threshold can be automatically adjusted according to the data distribution instead of being fixed. For example, when processing financial transaction data, if the trading volume significantly increases during certain time periods, the item support threshold can be correspondingly reduced to capture more possible frequent item sets, thereby providing a more detailed risk assessment. This adaptive adjustment can better adapt to different types of datasets of lost customers and improve the flexibility and practicality of the algorithm.

[0119] This embodiment simplifies the original dataset of lost customers and improves the efficiency and accuracy of mining frequent item sets through precise calculation, support screening, reasonable sorting, construction of the frequent tree, and effective mining of the conditional pattern base.

[0120] In a feasible implementation manner, steps A201 to A204 are also included before step A20:

[0121] Step A201: Determine the items with item support higher than the preset initial item support threshold as the initial frequent item sets;

[0122] Step A202: If the data coverage rate of the frequent items in the initial frequent item sets is less than the preset coverage rate threshold, then reduce the initial item support threshold;

[0123] Step A203: If the number of noise item data in the initial frequent item sets is more than the preset rule number, then increase the initial item support threshold, where the noise item data refers to the data that has no practical meaning or represents randomness;

[0124] Step A204: Determine the adjusted initial item support threshold as the preset item support threshold.

[0125] It should be noted that the initial item support threshold refers to a minimum support standard set at the beginning of frequent item set mining, which is used to filter out items that are frequent enough; the initial frequent item set is a set composed of those items whose support is higher than the initial item support threshold. These items are considered to appear frequently enough in the dataset of lost contact customers and may contain valuable information; the omission of frequent item data refers to the situation where some data that are actually frequent but do not reach the current threshold are wrongly excluded due to the too high setting of the initial item support threshold; the noise item data refers to those data that, although their support is higher than the threshold, actually do not have practical significance or represent chance; the preset item support threshold is the final item support threshold after adjustment, which is used for subsequent data mining processes.

[0126] First of all, in this embodiment, an initial item support threshold is determined as the preliminary standard for screening frequent items. Then, the dataset of lost contact customers is scanned, the support of each item is counted, and the data with support higher than the initial item support threshold are determined as the initial frequent item set. If it is found that there are omissions of valid item sets in the initial frequent item set, that is, some actually frequent items are not included, then the initial item support threshold needs to be lowered to include more potential frequent items. On the contrary, if there are too many noise item sets in the initial frequent item set, exceeding the preset rule quantity, it means that the current threshold is too low and too many item sets without practical significance are introduced. At this time, the initial item support threshold should be raised to reduce the noise. Through such repeated adjustments, until a suitable threshold that can both avoid the omission of valid item sets and control the quantity of noise item sets is found, and it is determined as the preset item support threshold for subsequent analysis. In this embodiment, through multiple experiments, it is found that when the minimum support threshold is 0.1, the algorithm can mine the most statistically significant and practically valuable frequent item sets.

[0127] Furthermore, this embodiment can consider an optimization scheme, and introduce a machine learning algorithm to automatically adjust the threshold during the process of adjusting the initial item support threshold. For example, use the supervised learning method to train a model to predict the best item support threshold, and the model input can include parameters such as the characteristics of the dataset of lost contact customers and the expected size of the frequent item set. This method is especially suitable for the association mining of the behavior feature patterns of loan lost contact customers in the financial field, because the data distributions of different financial institutions may have significant differences, and automatic adjustment can more accurately adapt to the needs of specific scenarios and improve the mining efficiency and accuracy.

[0128] By reasonably setting and dynamically adjusting the initial item support threshold, this embodiment can not only ensure capturing all important frequent item sets, but also effectively filter out noise item sets, thereby improving the quality of the mining results.

[0129] Based on Embodiment 1 or Embodiment 2 of the present application, in Embodiment 3 of the present application, the same or similar content as that in Embodiment 1 or Embodiment 2 above can be referred to the above introduction and will not be elaborated hereinafter. After the steps of Step S20, the following steps B10 to B40 are further included:

[0130] Step B10, dividing each frequent item set into multiple combinations of different antecedents and consequents, where each combination includes one antecedent and one consequent;

[0131] Step B20, determining each combination as an association rule, where each combination corresponds to an association rule one by one.

[0132] Step B30, calculating the rule score of each association rule, where the rule score includes rule confidence and lift index, and the lift index is the ratio of the rule confidence to the item support of the consequent;

[0133] Step B40, determining the association rules with rule confidence higher than the preset rule confidence threshold and lift index higher than the preset lift index threshold as key association rules.

[0134] It should be noted that in this embodiment, a frequent item set refers to an item combination that appears more frequently than a set threshold in the dataset of lost customers; the combination of antecedent and consequent divides the items in the frequent item set into two parts, one part as a precondition (antecedent) and the other part as a result (consequent); an association rule is an inferential relationship composed of one antecedent and one consequent, used to describe the potential relationship between data; rule confidence measures the strength of the association rule, that is, the probability of also including the consequent when the antecedent is included; the lift index is the ratio of the rule confidence to the support of the consequent, used to evaluate the actual effect of the association rule, indicating to what extent this rule can improve the prediction accuracy compared to the random case; the preset rule confidence threshold and the preset lift index threshold are user-defined criteria for screening out association rules with high reliability and practical significance.

[0135] First, in this embodiment, each mined frequent item set is segmented, and all possible combinations of antecedents and consequents are tried. For example, for the frequent item set {x1, x2, x3}, it can be segmented into various combination ways such as (x1 -> x2, x3), (x1, x2 -> x3), etc. Each combination is regarded as an association rule. Next, the rule scores of each association rule are calculated, including rule confidence and lift index. The rule confidence is calculated by dividing the number of transactions that contain both the antecedent and the consequent by the total number of transactions that contain the antecedent, and the lift index is obtained by dividing the rule confidence by the support of the consequent, so as to evaluate the effectiveness of the association rule. Finally, the association rules with rule confidence higher than the preset rule confidence threshold and lift index higher than the preset lift index threshold are determined as key association rules, and these rules are considered to be both accurate and of practical value.

[0136] Exemplarily, the mined association rules are shown in Table 4:

[0137] Table 4

[0138] This embodiment can also introduce an optimization scheme, and combine time series analysis technology when calculating the rule scores

[0139] Rule number Previous term Next term 1 frozenset({'5'}) frozenset({'1'}) 2 frozenset({'1'}) frozenset({'5'}) 3 frozenset({'5'}) frozenset({'2'}) 4 frozenset({'2'}) frozenset({'5'}) 5 frozenset({'1'}) frozenset({'2'}) 6 frozenset({'2'}) frozenset({'1'}) 7 frozenset({'5','1'}) frozenset({'2'}) 8 frozenset({'5','2'}) frozenset({'1'}) 9 frozenset({'1','2'}) frozenset({'5'}) 10 frozenset({'5'}) frozenset({'1','2'}) ... ... ... 228 frozenset({'4','0','2','3'}) frozenset({'5','1'})

[0140] to consider the change trend of association rules in different time periods. For example, when analyzing the behavioral characteristics of loan defaulters, by adding the time dimension, it can be analyzed whether the association rules of certain feature combinations have changed within a specific time period.

[0141] By systematically dividing the frequent item sets and calculating the rule scores of the association rules in this embodiment, the key association rules that have both strong correlation and practical application value can be effectively identified.

[0142] In a feasible implementation manner, the steps of determining the key association rules from the association rules further include steps C10 - C20:

[0143] Step C10, generating a tree diagram of association rules according to the rule scores of the association rules;

[0144] Step C20, determining the association rules with an area occupied in the association rule tree diagram greater than the preset area as the key association rules, where the occupied area is calculated according to the product of the rule confidence and the lift index.

[0145] It should be noted that the association rule tree diagram is a visualization tool that graphically displays the different degrees of importance of association rules, wherein each block represents an association rule, and the size or color of the block reflects the importance of the rule; the area occupied in the association rule tree diagram can be calculated based on the product of the rule confidence and the lift index, that is, the higher the rule confidence and the lift index, the larger the area occupied by the block. The specific occupied area can be determined based on the ratio of the product of the rule confidence and the lift index to the total area, which is not limited in this embodiment. The preset area refers to a user-defined threshold used to filter out association rules that occupy a larger area in the association rule tree diagram, that is, association rules with higher importance. The preset area threshold can be determined based on a specific percentage of the total area, or a specific amount of area can be directly specified as the preset area threshold.

[0146] For example, this embodiment not only visualizes the association through the association rule tree diagram, but also uses heat map to visualize the data in order to observe the association between the data more intuitively after mining the frequent item sets. Figure 2 As shown in the heat map, the colors of the squares vary, with higher values indicating darker colors, from 0 being light yellow to 6 being dark blue. Dark blue squares in the heat map represent high-frequency feature combinations. When these overlap with the columns / rows encoding the loss pattern, this indicates a strong correlation between the combination and the pattern. Correlated regions have a high degree of overlap and a darker color, reflecting frequent interactions between these features.

[0147] Combined with the frequent item set heat map, we found that

[0148] (1) Features x2 (mobile phone number online status C: online but unavailable), x12 (overdue days 180-270 days), and x13 (overdue loan amount 300,000-500,000 yuan) are associated with the Hide-and-Seek pattern. When these three features appear simultaneously, they are associated with a high frequency of Hide-and-Seek patterns in the missing customer dataset.

[0149] (2) Features x1 (permanent address is either the household registration address or the business address), x2 (mobile phone number online status is C: online but unavailable), and x11 (loan type is Q: mortgage loan) are associated with the absconding with funds model. When these three features appear in combination, there is a strong correlation with the absconding with funds model.

[0150] (3) Features x8 (the emergency contact's mobile phone call record feature is a certain value, assumed to be a code value corresponding to a high number of outages), x12 (the number of overdue days is 180-270 days), and x15 (the length of time lost contact is a certain interval value, assumed to be 30-60 days) are associated with the false disappearance and loss of contact pattern.

[0151] (4) Feature `x2` (mobile phone number in-network status is C: in-network but unavailable), `x11` (loan type is Q: mortgage loan) and `x14` (expected recoverable amount is in a certain range, assumed to be 0 - 100,000 yuan) are associated with the hide-and-seek loss-of-contact mode.

[0152] (5) Feature `x1` (permanent address is the household registration address, business address), `x12` (overdue days are 180 - 270 days) and `x16` (number of valid contacts is a certain value, assumed to be 0 - 1 person) are associated with the absconding with funds mode.

[0153] (6) Feature `x2` (mobile phone number in-network status is C: in-network but unavailable), `x8` (emergency contact mobile phone call record feature is a certain specific value, assumed to be the encoded value corresponding to a relatively high number of out-of-service etc. times) and `x13` (overdue loan amount is 300,000 - 500,000 yuan) are associated with the fake disappearance and loss-of-contact mode.

[0154] On this basis, we use a tree diagram to visualize the association rules data, referring to Figure 3 , Figure 3 In it, the larger squares correspond to rules with high confidence and high lift index. The color depth of the squares intuitively reflects core indicators such as the confidence and lift index of the association rules. Dark-colored squares need to be processed first, and light-colored squares can be selectively ignored. (The lift index is an important indicator in association rule mining, used to measure the effectiveness and relevance of the rules. It represents the ratio of the confidence of the rule to the support of the consequent (result item), reflecting whether the predictive ability of the rule exceeds the random level. In the analysis of loan loss-of-contact customers, if a certain rule has a high lift index (such as the lift index of the rule "x2 = 1 (C status), x11 = 2 (mortgage loan) → x18 = 1 (absconding with funds)") > 1, it indicates that this feature combination's predictive ability for the absconding with funds mode is significantly better than random, and can be used as an important basis for risk warning.).

[0155] In addition, the tree diagram clearly shows the importance ranking of the association rules. For example, the association rules related to `x1` (permanent address) being the household registration address or business address are in a relatively prominent position in the tree diagram, indicating that the association of this feature combination with the fake disappearance and loss-of-contact mode is relatively crucial in the overall analysis and should be given priority attention.

[0156] It should be noted that the heat map and the tree map can jointly assist in the comprehensive analysis of multiple features and thus serve as the basis for risk early warning. For example, when dark squares frequently appear in the relevant areas where 'x2' is 'online but unavailable', 'x12' is '180 - 270 days', and 'x13' is '300,000 - 500,000 yuan' in the heat map, and the squares corresponding to the association rules in the tree map are relatively large, the lending institution can timely conduct risk early warning on the customers based on this and take corresponding management measures, which may be related to the mode of disappearing like playing hide-and-seek.

[0157] This embodiment can intuitively and efficiently identify the rules with the most practical application value.

[0158] In a feasible implementation manner, steps D10 to D40 are further included before step B40:

[0159] Step D10: Determine the items with rule confidence levels higher than the preset initial rule confidence level threshold as the initial association rules;

[0160] Step D20: If there is a situation of rule redundancy in the initial association rules, then increase the initial rule confidence level threshold;

[0161] Step D30: If the number of initial association rules is less than the preset number of rules, then decrease the initial rule confidence level threshold;

[0162] Step D40: Determine the adjusted initial rule confidence level threshold as the preset rule confidence level threshold.

[0163] It should be noted that the initial rule confidence level threshold refers to a minimum confidence level standard set at the beginning of mining association rules, which is used to screen out sufficiently reliable association rules; the initial association rules are a set composed of those items with rule confidence levels higher than the initial rule confidence level threshold, and these rules are considered to have relatively high credibility; rule redundancy refers to the phenomenon that due to the too low setting of the initial rule confidence level threshold, some rules that actually do not provide additional information or are repetitive are included; the preset number of rules is a standard of the number of association rules that the user defines and expects to obtain, which is used to guide the adjustment of the rule confidence level threshold; the preset rule confidence level threshold is the final rule confidence level threshold after adjustment, which is used for subsequent mining of association rules.

[0164] First, in this embodiment, an initial rule confidence threshold is determined as the preliminary criterion for screening association rules. Then, the rule confidence of each association rule is calculated, and the rules with a rule confidence higher than the initial rule confidence threshold are determined as the initial association rules. If it is found that there is rule redundancy in the initial association rules, that is, the information provided by some rules is repeated or has no practical significance, it is necessary to increase the initial rule confidence threshold to reduce the number of redundant rules. On the contrary, if the number of initial association rules is less than the preset number of rules, it means that the current threshold is too high, excluding too many potentially useful rules. At this time, the initial rule confidence threshold should be decreased to increase the number of rules. Through such repeated adjustments, until a suitable threshold that can avoid rule redundancy and meet the requirement of the preset number of rules is found, it is determined as the preset rule confidence threshold for subsequent analysis. In this embodiment, through multiple experiments, it is found that when the minimum confidence threshold is 0.7, the algorithm can mine the association rules with the most statistical significance and practical application value.

[0165] Furthermore, this embodiment can consider an optimization scheme by introducing an adaptive learning mechanism during the process of adjusting the initial rule confidence threshold to dynamically adjust the threshold according to historical data and real-time feedback. For example, in the association mining of the behavior feature patterns of loan-loss customers, the rule confidence threshold can be automatically adjusted according to the change trend of customer behavior in different time periods. This method can more accurately capture the high-risk behavior patterns in a specific period, thereby improving the effectiveness and accuracy of risk management.

[0166] By reasonably setting and dynamically adjusting the initial rule confidence threshold, this embodiment can not only ensure that all important association rules are screened out, but also effectively remove redundant rules, improving the quality of the mining results.

[0167] All of the above embodiments can be freely combined. Exemplarily, considering all the embodiments together, the full process of this application can refer to Figure 4First, the data set of lost customers is processed, and features irrelevant to the data mining effect are eliminated. Missing values, continuous features and discrete features are processed. The mode is used to fill the missing values of discrete features and convert them into digital codes. The median is used to fill the missing values of continuous features, and the data is discretized and converted into digital codes. Secondly, the data is converted into string type to construct transaction data. Then, the key parameters of the FP-Growth algorithm are set, and the frequent item sets of lost features and lost patterns are mined, and the mining results are evaluated. If the mining effect does not meet the requirements, the preset item support threshold and other parameters are readjusted, and the frequent item sets of lost features and lost patterns are continued to be mined. If the mining effect meets the requirements, the association rules of lost features and lost patterns are further mined, and the correlation between the lost features of loan lost customers and the lost patterns is analyzed. Then, it is evaluated whether the mining results meet the requirements. If they meet the requirements, the process is terminated. Otherwise, the preset rule confidence threshold is adjusted and mining continues.

[0168] It should be noted that the above examples are only used to understand this application and do not constitute a limitation on the method for mining associations of behavioral feature patterns of lost customers in this application. More simple transformations based on this technical concept are all within the scope of protection of this application.

[0169] The present application also provides a device for mining associations of behavioral feature patterns of lost customers, which includes: a preprocessing module for preprocessing a preset lost customer data set to obtain a transaction data table, wherein the lost customer data set includes at least one set of lost customer behavioral features and at least one lost connection pattern; an association rule module for determining frequent item sets based on the transaction data table, extracting association rules from the frequent item sets, and determining key association rules from the association rules; and an association conclusion module for determining a mapping relationship between a combination of lost customer behavioral features and a lost connection pattern based on the key association rules;

[0170] The risk strategy module queries the target loss pattern corresponding to the combination of lost customer behavior characteristics to be predicted in the mapping relationship, and generates a risk warning strategy based on the target loss pattern.

[0171] The device for mining the association of behavioral characteristic patterns of lost customers provided in this application utilizes the method for mining the association of behavioral characteristic patterns of lost customers in the above-mentioned embodiments, capable of analyzing the complex associations between behavioral characteristics of lost customers and loss patterns. Compared to the prior art, the beneficial effects of the device for mining the association of behavioral characteristic patterns of lost customers provided in this application are the same as those of the method for mining the association of behavioral characteristic patterns of lost customers provided in the above-mentioned embodiments. Other technical features of the device for mining the association of behavioral characteristic patterns of lost customers are the same as those disclosed in the above-mentioned embodiments and are not further elaborated here.

[0172] The present application provides a device for mining the association of behavior feature patterns of lost customers. The device for mining the association of behavior feature patterns of lost customers includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method for mining the association of behavior feature patterns of lost customers in the first embodiment above.

[0173] Reference is made below Figure 5 , which shows a schematic structural diagram of a device for mining the association of behavior feature patterns of lost customers suitable for implementing the embodiments of the present application. The device for mining the association of behavior feature patterns of lost customers in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The shown device for mining the association of behavior feature patterns of lost customers is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0174] As Figure 5As shown in the figure, the device for mining the association of behavior feature patterns of lost customers may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to the program stored in the read-only memory 1002 or the program loaded from the storage device 1003 into the random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the device for mining the association of behavior feature patterns of lost customers are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. The input / output interface 1006 is also connected to the bus. Generally, the following systems may be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the device for mining the association of behavior feature patterns of lost customers to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a device for mining the association of behavior feature patterns of lost customers with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or had alternatively.

[0175] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are executed.

[0176] The device for mining the association of behavior feature patterns of lost customers provided by the present application adopts the method for mining the association of behavior feature patterns of lost customers in the above embodiments, and can analyze the complex association between the behavior features of lost customers and the lost patterns. Compared with the prior art, the beneficial effects of the device for mining the association of behavior feature patterns of lost customers provided by the present application are the same as those of the method for mining the association of behavior feature patterns of lost customers provided by the above embodiments, and other technical features in the device for mining the association of behavior feature patterns of lost customers are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0177] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0178] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all of them should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0179] This application provides a medium, which is a computer-readable storage medium and has computer-readable program instructions (i.e., computer programs) stored thereon. The computer-readable program instructions are used to execute the method for mining the association pattern of the behavior characteristics of lost customers in the above embodiments.

[0180] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0181] The above computer-readable storage medium can be included in the device for mining the association pattern of the behavior characteristics of lost customers; it can also exist separately without being assembled into the device for mining the association pattern of the behavior characteristics of lost customers.

[0182] The above computer-readable storage medium carries one or more programs, which, when executed by the device for mining the correlation of behavior characteristics of lost customers, cause the device for mining the correlation of behavior characteristics of lost customers to:

[0183] Preprocess a preset data set of lost customers to obtain a transaction data table, where the data set of lost customers includes at least one set of behavior characteristics of lost customers and at least one type of lost mode; determine frequent item sets according to the transaction data table, extract association rules in the frequent item sets, and determine key association rules from the association rules; determine the mapping relationship between the combination of behavior characteristics of lost customers and the lost mode according to the key association rules; query the target lost mode corresponding to the combination of behavior characteristics of the to-be-predicted lost customers in the mapping relationship, and generate a risk warning strategy according to the target lost mode.

[0184] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0185] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0186] The modules involved in the embodiments of the present application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation to the unit itself in some cases.

[0187] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned method for mining the association of the behavior characteristics of lost customers, and can analyze the complex association between the behavior characteristics of lost customers and the lost patterns. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the method for mining the association of the behavior characteristics of lost customers provided by the above embodiments, and will not be elaborated here.

[0188] The present application also provides a product, which is a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the method for mining the association of the behavior characteristics of lost customers as described above are implemented.

[0189] The computer program product provided by the present application can analyze the complex association between the behavior characteristics of lost customers and the lost patterns. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the method for mining the association of the behavior characteristics of lost customers provided by the above embodiments, and will not be elaborated here.

[0190] The above are only some embodiments of the present application, and do not limit the patent scope of the present application. All equivalent structural transformations made under the technical concept of the present application by using the content of the specification and drawings of the present application, or directly / indirectly applied to other related technical fields, are included in the patent protection scope of the present application.

Claims

1. A method for mining the association of behavior feature patterns of lost customers, characterized in that, The method for mining the association of the behavior feature patterns of lost customers includes: Preprocessing a preset lost customer data set to obtain a transaction data table, where the lost customer data set includes at least one set of lost customer behavior features and at least one lost mode; Determining frequent item sets according to the transaction data table, extracting association rules from the frequent item sets, and determining key association rules from the association rules; Determining the mapping relationship between the lost customer behavior feature combinations and the lost modes according to the key association rules; Querying the target lost mode corresponding to the to-be-predicted lost customer behavior feature combination in the mapping relationship, and generating a risk warning strategy according to the target lost mode.

2. The method for mining the association of the behavior feature patterns of lost customers according to claim 1, wherein, The step of preprocessing a preset lost customer data set to obtain a transaction data table includes: Determining the discrete feature data in the lost customer data set, and filling the missing values in the discrete feature data according to the mode filling strategy; Converting the character-form data in the discrete features into numerical-form data to obtain the transaction data corresponding to the discrete feature data; Determining the continuous feature data in the lost customer data set, and filling the missing values in the continuous feature data according to the median filling strategy; Dividing the continuous feature data into multiple interval data according to the value range of the continuous feature data, and setting identifiers for each interval data to obtain the transaction data corresponding to the continuous feature data; Merging the transaction data corresponding to the discrete feature data and the transaction data corresponding to the continuous feature data, and converting them into a string type to obtain string type transaction data; Performing encoding processing on the string type transaction data, and converting the encoded string type transaction data into transaction data in a two-dimensional table data structure to obtain a transaction data table.

3. The method for mining the association of the behavior feature patterns of lost customers according to claim 1, wherein, The step of determining frequent item sets according to the transaction data table includes: Counting the frequency of occurrence of each data item in the transaction data table, and determining the item support of each data item as the frequency; Deleting the data in the transaction data table with an item support lower than a preset item support threshold to obtain the remaining data in the transaction data table; For the remaining data in the transaction data table, sorting them in descending order according to the item support to obtain a transaction list; Constructing a frequent tree based on the data in the transaction list; For each target item data in the frequent tree, extracting the conditional pattern base of the target item data and mining the conditional frequent item sets of the conditional pattern base; Combining the conditional frequent item sets with the target item data to form frequent item sets.

4. The method for mining the association of the behavior feature patterns of lost customers according to claim 3, wherein, Before the step of deleting the data in the transaction data table with an item support lower than a preset item support threshold includes: Determining the items with an item support higher than a preset initial item support threshold as initial frequent item sets; If the coverage rate of the frequent item data in the initial frequent item set is less than a preset coverage rate threshold, then reducing the initial item support threshold; If the number of noise item data in the initial frequent item set is more than a preset rule number, then increasing the initial item support threshold, where the noise item data refers to data that has no practical significance or represents randomness; Determine the adjusted initial item support threshold as the preset item support threshold.

5. The method for mining the association of the behavior feature patterns of lost customers according to claim 1, wherein, The step of extracting association rules from the frequent item sets includes: Divide each frequent item set into multiple combinations of different antecedents and consequents, where each combination includes one antecedent and one consequent; Determine each of the combinations as an association rule, where each combination corresponds one-to-one to the association rule.

6. The method for mining the association of the behavior feature patterns of lost customers according to claim 5, wherein, The step of determining key association rules from the association rules includes: Calculate the rule score for each association rule, where the rule score includes rule confidence and lift index, and the lift index is the ratio of the rule confidence to the item support of the consequent; Determine as key association rules those association rules whose rule confidence is higher than the preset rule confidence threshold and whose lift index is higher than the preset lift index threshold; or, Generate an association rule tree diagram based on the rule scores of each of the association rules, and determine as key association rules those association rules in the association rule tree diagram whose occupied area is larger than the preset area, where the occupied area is calculated based on the product of the rule confidence and the lift index.

7. The method for mining the association of the behavior feature patterns of lost customers according to claim 6, wherein Before the step of determining as key association rules those association rules whose rule confidence is higher than the preset rule confidence threshold and whose lift index is higher than the preset lift index threshold, include: Determine as initial association rules those items whose rule confidence is higher than the preset initial rule confidence threshold; If there is a situation of rule redundancy in the initial association rules, then increase the initial rule confidence threshold; If the number of initial association rules is less than the preset rule quantity, then decrease the initial rule confidence threshold; Determine the adjusted initial rule confidence threshold as the preset rule confidence threshold.

8. An equipment for mining the association of behavior feature patterns of lost customers, characterized in that The device for mining the association of the behavior feature patterns of lost customers includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the method for mining the association of the behavior feature patterns of lost customers according to any one of claims 1 to 7.

9. A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the method for mining the association of the behavior feature patterns of lost customers according to any one of claims 1 to 7.

10. A computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the method for mining the association of the behavior feature patterns of lost customers according to any one of claims 1 to 7.

Images