Intelligent allocation of financial user service policies

Abstract

The application provides a kind of intelligent allocation method of financial user service strategy, it is related to the technical field of financial technology, including the following steps: obtaining the multidimensional feature value of the multidimensional feature of target user in current business cycle, call the historical reaction database of target user belonging to customer group grouping, the first K users in the same group with the highest historical repayment success rate are used as reference users, and a reference user set is formed;Identify at least two kinds of multidimensional features with the highest frequency in the reference user set, and use them as the common characteristic set, and calculate the data dispersion of the characteristic value of each multidimensional feature in the common characteristic set;The two multidimensional features with the lowest data dispersion in the common characteristic set are combined to obtain the target cross feature of the target user;The target cross feature value or the target cross feature value and the multidimensional feature value are input into the service strategy decision model to determine the corresponding target service strategy, and the target service strategy is allocated to the target user.

Description

Technical Field

[0001] This application relates to the field of financial technology, specifically to an intelligent allocation method for financial user service strategies. Background Technology

[0002] With the rapid development of financial technology (FinTech), traditional financial institutions (such as banks, consumer finance companies, and internet finance platforms) are accelerating their digital transformation. Among these transformations, post-loan management (i.e., customer service and risk control after loan disbursement) is one of the core aspects of financial operations.

[0003] In existing technologies, such as the invention patent with publication number CN119313449B, an intelligent repayment strategy generation system and method driven by multi-source information analysis is proposed. This reveals that the user's post-loan repayment rate is improved by dynamically adjusting the repayment strategy. However, it has the following drawbacks: multi-dimensional features are directly input into the decision model as independent input variables. Since the cross-effects and synergistic laws that may exist between different features and have a relevant impact on the user's final repayment behavior are ignored, the quality of the input features of the decision model is low. This not only increases the computational complexity and resource consumption of the model, but also introduces noise from a large number of weakly correlated or unrelated features, which interferes with the model's learning process and reduces the accuracy and efficiency of strategy allocation. Summary of the Invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide an intelligent allocation method for financial user service strategies to improve the decision-making efficiency and accuracy of repayment strategy allocation;

[0005] The method includes the following steps:

[0006] Obtain the multidimensional feature values ​​corresponding to the multidimensional features of the target user within the current business cycle. The multidimensional features include at least two of the following: repayment ability fluctuation index, repayment willingness score, service sensitivity, user difficulty level, and user value level.

[0007] The historical response database of the target user's customer group is retrieved. The historical response database includes multiple users in the same group, the historical multidimensional feature values ​​corresponding to the multidimensional features of each user in the same group, and the historical repayment success rate corresponding to each user in the same group.

[0008] The top K users with the highest historical repayment success rates in the same group are selected as reference users and form a reference user set.

[0009] Identify at least two of the most frequently occurring multidimensional features in the reference user set and use them as a common feature set; wherein, the frequency of occurrence of a multidimensional feature is the number of reference users who possess the historical multidimensional feature value corresponding to that multidimensional feature;

[0010] For each of the multidimensional features in the set of common features, calculate the data dispersion of its historical multidimensional feature values ​​within the reference user set;

[0011] From the set of common features, the two multidimensional features with the lowest data dispersion are combined to obtain the target cross features of the target user;

[0012] The target cross feature value corresponding to the target cross feature is input into the service strategy decision model to determine the corresponding target service strategy; or, the target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature are input into the service strategy decision model to determine the corresponding target service strategy.

[0013] The target service policy is assigned to the target user.

[0014] According to the technical solution provided in this application, the target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature are input into the service strategy decision model to determine the corresponding target service strategy, including the following steps:

[0015] Based on the historical multidimensional feature values ​​of the reference user set and their corresponding historical repayment success rates, the feature importance weight of each multidimensional feature in predicting the repayment success rate is calculated using a supervised machine learning model.

[0016] Based on the feature importance weights corresponding to each of the multidimensional features, the multidimensional feature values ​​of the target user are weighted and summed to obtain a basic score value;

[0017] Based on the historical performance data of the customer group to which the target user belongs, a dynamic scaling factor is generated to scale the target cross feature value of the target cross feature, so as to obtain the scaled target cross feature value.

[0018] The service strategy score is obtained by adding the base score to the scaled target cross feature value.

[0019] The service strategy score is matched with a preset strategy threshold range to determine the corresponding target service strategy.

[0020] According to the technical solution provided in this application, the target cross feature value corresponding to the target cross feature is input into the service strategy decision model to determine the corresponding target service strategy, including the following steps:

[0021] Retrieve and traverse the preset strategy mapping table to obtain the target feature value interval, wherein the target feature value interval is the feature value interval into which the target cross feature value falls;

[0022] The service policy mapped to the target feature value range is determined as the target service policy;

[0023] The strategy mapping table is established by analyzing the historical response database. The strategy mapping table represents the mapping relationship between different feature value ranges of the target cross features and the service strategy with the highest historical repayment success rate.

[0024] According to the technical solution provided in this application, after allocating the target service policy to the target user, the method further includes the following steps:

[0025] Real-time monitoring of the target user's response behavior within a preset time window;

[0026] Extract the dynamic feedback feature set from the response behavior, the dynamic feedback feature set including strategy response delay duration, operation completion degree, and secondary consultation frequency;

[0027] Based on the dynamic feedback feature set, the target service strategy is dynamically adjusted.

[0028] According to the technical solution provided in this application, the step of dynamically adjusting the target service strategy based on the dynamic feedback feature set includes the following steps:

[0029] Based on the dynamic feedback feature set, the real-time feedback valence of the target service strategy is calculated; the real-time feedback valence is used to quantify the degree of positive response of the target user to the target service strategy.

[0030] If the real-time feedback valence is less than a first preset threshold, the reference user set is updated to redetermine the target cross feature.

[0031] According to the technical solution provided in this application, obtaining the multidimensional feature values ​​of the target user's multidimensional features within the current business cycle includes the following steps:

[0032] Retrieve the target user's historical behavior sequence data within the current business cycle; the historical behavior sequence data includes behavioral events recorded in chronological order, each behavioral event including a behavior type and an occurrence timestamp; the behavior type includes repayment operation, communication contact, interface interaction, or document processing;

[0033] Based on the historical behavior sequence data, a timestamped behavior chain is generated, which represents the time trajectory and type sequence of user behavior within the current business cycle;

[0034] The multidimensional feature value is calculated based on the behavior type, timestamp, and sequence information contained in the behavior chain.

[0035] According to the technical solution provided in this application, before calculating the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain, the following steps are also included:

[0036] For each of the multidimensional features, determine whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value.

[0037] According to the technical solution provided in this application, after determining whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value for each of the multidimensional features, the method further includes the following steps:

[0038] If the target behavior type of at least two of the multidimensional features is covered, then the at least two multidimensional features are taken as the first feature type;

[0039] The calculation of the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain includes the following steps:

[0040] Based on the target behavior type, timestamp, and sequence information corresponding to the first feature type contained in the behavior chain, the feature value corresponding to the first feature type is calculated.

[0041] According to the technical solution provided in this application, the first feature type includes the repayment ability fluctuation index;

[0042] The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps:

[0043] The characteristic value corresponding to the repayment ability fluctuation index is obtained by measuring the fluctuation of the time interval between repayment failure operations and combining the frequency and urgency weight of the target user's access to income verification documents.

[0044] According to the technical solution provided in this application, the first feature type includes the repayment willingness score;

[0045] The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps:

[0046] The characteristic value corresponding to the repayment willingness score is obtained by measuring the ratio of the frequency of target users actively contacting customer service versus the frequency of the system actively contacting users, the density of the keywords of repayment commitment during the call, and the proportion of users skipping the agreement page.

[0047] Compared with existing technologies, the beneficial effects of this application are as follows: Based on the data science principle that the stable common feature combination of successful user groups is the key objective law of their successful repayment behavior, this application first screens users with high repayment success rates from historical user groups. By identifying the common feature set and verifying its data dispersion, target cross features are discovered. By mining target cross features from historical high-value user groups, a crucial data purification is achieved. This cross feature is a strong signal strongly correlated with high repayment success rate, verified by both commonality and stability. Using it as model input is equivalent to providing the decision-making model with a core driving factor with extremely high information density and strong discriminative power, thereby significantly reducing the model's learning difficulty and improving the accuracy and efficiency of service strategy allocation. Through the pre-process of cross feature mining, efficient feature dimensionality reduction is achieved for the downstream strategy decision-making model. The model no longer needs to perform complex calculations on all possible feature combinations, but instead focuses on the few most effective cross features. This directly reduces the computational complexity of the model and the consumption of hardware resources, enabling the system to achieve faster response speeds when processing massive amounts of user data. It is particularly suitable for financial technology business scenarios with high real-time requirements. Attached Figure Description

[0048] Figure 1 A flowchart illustrating the steps of the intelligent allocation method for financial user service strategies provided in this application. Detailed Implementation

[0049] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0050] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0051] Example 1

[0052] As mentioned in the background section, this application proposes an intelligent allocation method for financial user service strategies to address the problems in existing technologies. It should be noted that this method aims to provide users with a better service experience and more suitable repayment plans, rather than forcibly bundling services. Users always have the right to accept, refuse, or negotiate service strategies. The target service strategy recommended by the system is not a mandatory command, but rather an intelligent decision-making assistance suggestion provided to customer service representatives or the user. The user data collection, monitoring, and analysis involved in this method are all completed with the explicit authorization of the user and under the premise of ensuring privacy protection; Figure 1 As shown, the method includes the following steps:

[0053] S1. Obtain the multidimensional feature values ​​corresponding to the multidimensional features of the target user in the current business cycle. The multidimensional features include at least two of the following: repayment ability fluctuation index, repayment willingness score, service sensitivity, user difficulty level, and user value level.

[0054] Furthermore, obtaining the multidimensional feature values ​​of the target user's multidimensional features within the current business cycle includes the following steps:

[0055] Retrieve the target user's historical behavior sequence data within the current business cycle; the historical behavior sequence data includes behavioral events recorded in chronological order, each behavioral event including a behavior type and an occurrence timestamp; the behavior type includes repayment operation, communication contact, interface interaction, or document processing;

[0056] Based on the historical behavior sequence data, a timestamped behavior chain is generated, which represents the time trajectory and type sequence of user behavior within the current business cycle;

[0057] The multidimensional feature value is calculated based on the behavior type, timestamp, and sequence information contained in the behavior chain.

[0058] Specifically, the data source for historical behavior sequence data is: extracting a sequence of behavioral events sorted by timestamp within the current business cycle (the current business cycle is a collection period after the target user completes financial business, such as 30 days) from the user behavior log database. Behavior types include: repayment operations (successful / failed repayment records, repayment amount), communication contacts (customer service call recordings, SMS interaction records), interface interactions (APP page dwell time, button click paths), and document processing (access records of documents such as income certificates and credit authorization letters uploaded by users). The operation method for generating behavior chains is to concatenate behavioral events into a continuous sequence by timestamp to form a structured behavior chain. For example: [08:00 Repayment failed] → [10:15 View income certificate document] → [14:30 Connect to customer service call] → [16:20 Click on the deferred repayment agreement].

[0059] Furthermore, before calculating the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain, the following steps are also included:

[0060] For each of the multidimensional features, determine whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value.

[0061] Specifically, in real business environments, due to differences in the completeness of data collection, the scope of user authorization, and the stage of the user's lifecycle, the historical behavior sequence data of a specific target user may not cover all the behavior types required to calculate all multidimensional features.

[0062] Furthermore, after determining whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value for each of the multidimensional features, the method further includes the following steps:

[0063] If the target behavior type of at least two of the multidimensional features is covered, then the at least two multidimensional features are taken as the first feature type;

[0064] The calculation of the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain includes the following steps:

[0065] Based on the target behavior type, timestamp, and sequence information corresponding to the first feature type contained in the behavior chain, the feature value corresponding to the first feature type is calculated.

[0066] Specifically, the target behavior types required for the repayment willingness score are communication contact and interface interaction. Communication contact can be used to distinguish whether the user initiates contact with customer service or the system initiates contact with the user, calculate the frequency ratio, and obtain the text of customer service call recordings to analyze the density of repayment commitment keywords. Interface interaction can characterize the interaction behavior on key agreement pages such as repayment agreement and extension agreement, and is used to calculate the proportion of agreement pages skipped.

[0067] Specifically, the target behavior types required for service sensitivity are communication contact and interface interaction. Communication contact can be used to obtain the response speed of the customer service channel, that is, the time from when the system sends a notification to when the user connects or responds. Interface interaction can be used to obtain the user's operation delay time in the APP, such as the response delay of clicking a button. By combining interface interaction and communication contact, it is possible to determine whether the user has adopted the service suggestion, such as whether they clicked the installment plan link pushed to them, or agreed to a proposal during a call.

[0068] Specifically, the target behavior types required for the user's difficulty level are repayment operation, document processing, interface interaction, and communication contact. Among them, repayment operation, document processing, and interface interaction are used together to check whether a series of preset key behavior nodes, such as "on-time repayment confirmation", "income certificate update", and "emergency contact information maintenance", are missing. Communication contact can obtain the records of contacting emergency contacts and analyze the distribution concentration of the contact types.

[0069] Specifically, the target behavior types required for user value level are repayment operation, communication contact, interface interaction and document processing. Among them, these four can be used to assess long-term stability and the degree of current abnormal deviation. Interface interaction can obtain the depth of use of value-added service functions (such as financial management and insurance), such as different levels of behavior such as browsing, collecting, and purchasing.

[0070] For example, assuming that all target behavior types required for the repayment ability volatility index are covered, that is, the behavior types included in the behavior chain include repayment operations and income verification access in document processing, then the repayment ability volatility index will be used as one of the first feature types.

[0071] Furthermore, the first feature type includes the repayment ability fluctuation index;

[0072] The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps:

[0073] The characteristic value corresponding to the repayment ability fluctuation index is obtained by measuring the fluctuation of the time interval between repayment failure operations and combining the frequency and urgency weight of the target user's access to income verification documents.

[0074] Furthermore, the first feature type includes the repayment willingness score;

[0075] The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps:

[0076] The characteristic value corresponding to the repayment willingness score is obtained by measuring the ratio of the frequency of target users actively contacting customer service versus the frequency of the system actively contacting users, the density of the keywords of repayment commitment during the call, and the proportion of users skipping the agreement page.

[0077] Specifically, service sensitivity, user difficulty level, and user value level are calculated using the following methods:

[0078] The service sensitivity is obtained by the ratio of customer service response speed to user interface operation latency, and the proportion of target users who adopt service suggestions.

[0079] The user's difficulty level is obtained by calculating the missing proportion of key behavioral nodes and the distribution concentration of the target user's contact emergency contact types.

[0080] The user value level is obtained by analyzing the stability of the target user's historical behavior chain, the degree of abnormal deviation of the current behavior chain from the historical behavior chain, and the depth of the target user's use of value-added service functions.

[0081] Specifically, the repayment ability volatility index is an indicator that measures the stability of a user's repayment ability; the higher the value, the greater the volatility of repayment ability. Calculation method: Analyze the time interval between all failed repayment operations of a user, calculate its standard deviation as the base volatility value, count the frequency of user access to income verification documents, and assign weights based on the urgency of the documents (such as "urgent" tags). The base volatility value and the document access weighted value are then linearly combined and normalized to an index of 0-100.

[0082] Specifically, the repayment willingness score is an indicator that assesses the likelihood of a user actively repaying their loan; a higher score indicates a stronger willingness to repay. Calculation method: The ratio of the number of times a user proactively contacts customer service to the number of times the system proactively contacts the user is calculated. The frequency of keywords such as "will definitely repay" and "promise to pay" in the call recordings is analyzed. The percentage of users who skip the repayment agreement page is also counted. These three indicators are weighted, summed, and normalized to a score of 0-100.

[0083] Specifically, service sensitivity reflects the degree to which users respond to the services of financial institutions. Calculation method: Calculate the ratio of the average customer service response time to the user interface operation delay time, and statistically analyze the proportion of service suggestions actually adopted by users out of the total number of push suggestions. The two indicators are then weighted and normalized to a sensitivity value of 0-100.

[0084] Specifically, the user hardship level is used to assess the degree of financial hardship currently faced by the user. Calculation method: The missing percentage of preset key behavioral nodes (such as timely repayment confirmation, income verification updates, etc.) is checked, and the distribution concentration of the user's emergency contact types is analyzed (e.g., if 80% of contacts are the same person (e.g., spouse), it indicates weak social support). The two indicators are weighted and divided into 1-5 levels. Then, according to the first mapping rule, the original level is mapped to a hardship level score of 0-100. The first mapping rule is as follows: Level 1 corresponds to a hardship level score range of 0-20, Level 2 to 21-40, Level 3 to 41-60, Level 4 to 61-80, and Level 5 to 81-100. This normalizes the user's hardship level to a feature value of 0-100.

[0085] Specifically, user value levels are used to assess a user's long-term value potential. The calculation method includes: historical behavior chain stability (standard deviation of repayment time over the past 6 months, the smaller the deviation, the more stable); current behavior chain deviation (number of abnormal occurrences compared to historical behavior patterns, such as sudden frequent contact with customer service); and value-added service depth (level of use of additional functions such as wealth management / insurance, e.g., browsing only = level 1, purchasing = level 3). These three indicators are combined to form five levels: A, E, and E. Subsequently, the original levels are mapped to user value level scores of 0-100 according to a second mapping rule. The second mapping rule is as follows: Level A corresponds to a user value level score range of 81-100; Level B corresponds to a user value level score range of 61-80; Level C corresponds to a user value level score range of 41-60; Level D corresponds to a user value level score range of 21-40; and Level E corresponds to a user value level score range of 0-20. This normalizes the user value level to a feature value of 0-100.

[0086] S2. Retrieve the historical response database of the customer group to which the target user belongs. The historical response database includes multiple users in the same group, the historical multidimensional feature values ​​corresponding to the multidimensional features of each user in the same group, and the historical repayment success rate corresponding to each user in the same group.

[0087] Specifically, customer grouping refers to the division of user groups according to business rules (such as loan product type, user geographic distribution, credit rating, age range, etc.). In practice, the system first determines the identifier of the group to which the target user belongs based on the user's attributes, and then queries the corresponding historical data set from the data warehouse based on the identifier.

[0088] S3. Select the top K users in the same group with the highest historical repayment success rate as reference users and form a reference user set;

[0089] S4. Identify at least two of the most frequently occurring multidimensional features in the reference user set and use them as a common feature set; wherein, the frequency of occurrence of one of the multidimensional features is the number of reference users who possess the historical multidimensional feature value corresponding to the multidimensional feature.

[0090] S5. For each of the multidimensional features in the common feature set, calculate the data dispersion of its historical multidimensional feature values ​​within the reference user set;

[0091] S6. From the common features, the two multidimensional features with the lowest data dispersion are combined to obtain the target cross features of the target user;

[0092] S7. Input the target cross feature value corresponding to the target cross feature into the service strategy decision model to determine the corresponding target service strategy; or, input the target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature into the service strategy decision model to determine the corresponding target service strategy.

[0093] S8. Assign the target service policy to the target user.

[0094] For example, Step 1: Target User: The system successfully calculated four characteristic values ​​for the target user: repayment ability fluctuation index: 72; repayment willingness score: 85; user hardship level: 41; user value level: 78. Since this user is a new user and has not yet generated sufficient customer service interaction, the service sensitivity characteristic cannot be calculated; therefore, the service sensitivity characteristic is missing for this user. Step 2: Retrieve Historical Database and Form Reference User Set: The system retrieves 500 historical user data points from the target user's customer group group, "First-Tier City White-Collar Customer Group." Step 3: Set parameter K=50 and filter out the 50 users with the highest historical repayment success rate to form a reference user set. These 50 high-value users also have different sets of characteristic categories due to their different historical circumstances. For example: User A has the following feature set: {Repayment ability, repayment willingness, service sensitivity, user value level} (lacking "difficulty level"); User B has the following feature set: {Repayment ability, repayment willingness, user difficulty level, user value level} (lacking "service sensitivity"); User C has the following feature set: {Repayment ability, repayment willingness, service sensitivity, user difficulty level, user value level} (all 5 features are present); Step 4: Identify common feature sets: The system counts the actual frequency of each multidimensional feature in the entire reference user set (50 people): Repayment ability fluctuation index: appears 50 times (100%); Repayment willingness score: appears 50 times (100%); Service sensitivity: appears 35 times (70%); User difficulty level: appears 48 times (96%); User value level: appears 49 times (98%); Filtering logic: The four multidimensional features with the highest frequency of occurrence. According to statistics, the following features meet the requirements: "Repayment Ability Fluctuation Index" (occurring 50 times), "Repayment Willingness Score" (occurring 50 times), "User Difficulty Level" (occurring 48 times), and "User Value Level" (occurring 49 times). Therefore, the common feature set is determined as: {Repayment Ability Fluctuation Index, Repayment Willingness Score, User Difficulty Level, User Value Level}. Step 5: Calculate Data Dispersion. The system calculates the data dispersion (coefficient of variation) of each feature within the common feature set and within the subset of reference users possessing that feature: Repayment Ability Fluctuation Index: 0.15, Repayment Willingness Score: 0.08, User Difficulty Level: 0.12, User Value Level: 0.09. The two features with the lowest dispersion are selected and combined: Repayment Willingness Score (0.08) and User Value Level (0.09) to generate the target cross feature for the target user: Repayment Willingness Score × User Value Level. The cross feature value = 85 × 78 = 6630. Step 7: Input the target cross-feature values ​​and the multi-dimensional feature values ​​actually possessed by the target user into the service strategy decision model. The model outputs the target service strategy and assigns it.

[0095] In a preferred embodiment, the target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature are input into the service strategy decision model to determine the corresponding target service strategy, including the following steps:

[0096] Based on the historical multidimensional feature values ​​of the reference user set and their corresponding historical repayment success rates, the feature importance weight of each multidimensional feature in predicting the repayment success rate is calculated using a supervised machine learning model.

[0097] Based on the feature importance weights corresponding to each of the multidimensional features, the multidimensional feature values ​​of the target user are weighted and summed to obtain a basic score value;

[0098] Based on the historical performance data of the target user's customer group, a dynamic scaling factor is generated to scale and adjust the target cross feature value.

[0099] The service strategy score is obtained by adding the base score to the scaled target cross feature value.

[0100] The service strategy score is matched with a preset strategy threshold range to determine the corresponding target service strategy.

[0101] Specifically, supervised machine learning model training involves: using the "historical multidimensional feature values" of the reference user set as input and their corresponding "historical repayment success rate" (converted into high / low success rate labels) as the prediction target, training a random forest model. Weight extraction: directly extracting the "feature importance" index for each multidimensional feature from the trained model and normalizing it; the result is the importance weight of each feature. Calculating the base score: numerical normalization: normalizing the multidimensional feature values ​​of the target user to a uniform numerical range to eliminate the influence of unit weights. Weighted summation: multiplying each normalized feature value by its corresponding feature importance weight, and summing all products; the final sum is the user's base score.

[0102] Optionally, the system divides user groups by age range, retrieves the historical recovery rate data (i.e. the percentage of users in that group who ultimately succeed in repaying) of the target user's customer group and the average historical recovery rate data of all users in the sample, calculates the ratio between the two, and obtains the dynamic scaling factor of that customer group.

[0103] For example, if the target user is classified as a "youth group," with a historical repayment rate of 72.3% and a full-sample average repayment rate of 68.5%, the calculated dynamic scaling factor is 1.055 (0.723 / 0.685). This factor indicates that the overall repayment performance of the youth group users is better than the average level, thus amplifying the positive signal implied by their cross-features by 5.5%. After scaling the target cross-feature values, the service strategy score is calculated as follows: Service Strategy Score = Base Score + Scaled Cross-Feature Value. Finally, the calculated service strategy score is matched with a preset strategy threshold range to determine the corresponding target service strategy.

[0104] For example, the system can preset the following policy mapping rules: if the service policy score If the score is 0.9, then a "priority retention strategy" (such as offering discounted rates or dedicated customer service) will be assigned; if it is 0.7... If the service strategy score is less than 0.9, a "standard auxiliary strategy" (such as sending repayment reminders and providing flexible installment plans) will be assigned.

[0105] In a preferred embodiment, the target cross feature value corresponding to the target cross feature is input into the service strategy decision model to determine the corresponding target service strategy, including the following steps:

[0106] Retrieve and traverse the preset strategy mapping table to obtain the target feature value interval, wherein the target feature value interval is the feature value interval into which the target cross feature value falls;

[0107] The service policy mapped to the target feature value range is determined as the target service policy;

[0108] The strategy mapping table is established by analyzing the historical response database. The strategy mapping table represents the mapping relationship between different feature value ranges of the target cross features and the service strategy with the highest historical repayment success rate.

[0109] Specifically, the construction logic of the preset strategy mapping table is as follows: For each user in the historical response database, the system extracts the service strategy actually adopted (i.e., the historical fact strategy) and backtracks to calculate the target cross-feature value within the business cycle at that time (the calculation method is the same as above). Subsequently, the system divides users into several consecutive feature value intervals according to their target cross-feature values ​​(e.g., 0-0.3, 0.3-0.6, 0.6-0.8, 0.8-1.0). Determining the mapping relationship: Within each feature value interval, the system statistically analyzes all service strategies used by all users and calculates their corresponding average historical repayment success rate. Finally, the service strategy with the highest historical repayment success rate within that interval is established as the service strategy mapped to that feature value interval. The preset strategy mapping table is shown in Table 1:

[0110] Table 1

[0111]

[0112] Specifically, when assigning a strategy to a target user, the system first calculates the user's target cross-feature value (e.g., 0.663). Then, the system retrieves a pre-built strategy mapping table applicable to this customer group from storage. The system iterates through this mapping table, comparing the target user's target cross-feature value with the predefined feature value intervals in Table 1 to determine which interval it falls into. Based on the example value 0.663, it falls into the interval [0.6, 0.8), and the system directly locks the service strategy (i.e., the "proactive care strategy") mapped to this interval in the strategy mapping table as the final service strategy to be assigned to the target user.

[0113] In a preferred embodiment, after assigning the target service policy to the target user, the method further includes the following steps:

[0114] Real-time monitoring of the target user's response behavior within a preset time window;

[0115] Extract the dynamic feedback feature set from the response behavior, the dynamic feedback feature set including strategy response delay duration, operation completion degree, and secondary consultation frequency;

[0116] Based on the dynamic feedback feature set, the target service strategy is dynamically adjusted.

[0117] Specifically, from the moment a target service strategy (e.g., a "proactive care strategy") is assigned and executed, the system initiates a preset time window (e.g., 24 hours, 3 days, or a complete billing cycle). This time window is a critical period for observing user responses. Monitoring content: Within this time window, the system captures and records in real-time the target user's strategy-related response behaviors across all touchpoints, including the application, customer service channels, and repayment pages. These behaviors are recorded as timestamped event streams. Extracting dynamic feedback feature sets: From the captured response behavior event streams, the system extracts three core dynamic feedback features to form a dynamic feedback feature set, used to quantify the user's immediate response to the strategy:

[0118] Policy response delay: This refers to the time interval between policy push and the user's first positive action (such as clicking the policy link or viewing installment details). For example, if a user clicks the link within 2 hours of receiving the SMS, the delay is 2 hours. The shorter this value, the more proactive the user response is generally.

[0119] Operation Completion Rate: This measures whether a user has completed the entire operation process guided by the strategy. For example, if the strategy guides a user to complete an installment application, the operation completion rate can be defined as: (Number of completed steps / Total number of steps) * 100%. If a user browses the installment page but does not submit an application, the completion rate may only be 50%.

[0120] Secondary consultation frequency: This refers to the frequency at which users proactively initiate a new round of customer service inquiries (including phone and online customer service) after receiving the strategy. This characteristic has a dual meaning: an increased frequency may indicate user confusion or dissatisfaction (negative feedback), or it may indicate strong user interest and a desire to learn more details (positive feedback). It needs to be judged in conjunction with sentiment analysis of the consultation content.

[0121] The system dynamically adjusts the target service strategy based on the dynamic feedback feature set: the extracted dynamic feedback feature set is input into a strategy adjustment rule engine. This rule engine has a set of pre-defined business rules used to determine how to adjust the strategy based on the values ​​of the feedback features.

[0122] Example adjustment logic:

[0123] Scenario 1 (Positive Feedback): If the user's strategy response delay is very short (e.g., <1 hour) and the operation completion rate is high (e.g., >90%), it indicates that the current "proactive care strategy" is effective. The system not only maintains this strategy but may also strengthen it, for example, by offering more attractive installment rates in the next interaction.

[0124] Scenario 2 (Negative Feedback): If the user's strategy response delay is very long (e.g., >3 days) or the operation completion rate is extremely low (e.g., <20%), and the frequency of secondary consultations increases significantly with negative sentiment analysis of the consultation content, it indicates that the current strategy is ineffective or has caused resistance. The system will downgrade or switch strategies, for example, from a "proactive care strategy" to a more direct "standard support strategy" or a gentler "educational guidance strategy".

[0125] Scenario 3 (Intervention Required): If the frequency of follow-up inquiries is unusually high, but the completion rate of operations also increases simultaneously, it may indicate an obstacle in the process. In addition to potentially adjusting its strategy, the system will trigger a manual service ticket, prompting customer service personnel to proactively intervene and assist the user in resolving the issue.

[0126] Furthermore, the step of dynamically adjusting the target service strategy based on the dynamic feedback feature set includes the following steps:

[0127] Based on the dynamic feedback feature set, the real-time feedback valence of the target service strategy is calculated; the real-time feedback valence is used to quantify the degree of positive response of the target user to the target service strategy.

[0128] If the real-time feedback valence is less than a first preset threshold, the reference user set is updated to redetermine the target cross feature.

[0129] Specifically, real-time feedback valence is a comprehensive numerical indicator used to quantify the degree of positive response from target users to the currently assigned target service strategy. A higher valence indicates a more successful strategy. Real-time feedback valence is obtained through a weighted summation model. First, each feature in the dynamic feedback feature set is normalized and oriented (ensuring that a larger value represents a more positive response): the reciprocal of the strategy response delay is taken to convert it into "response speed." Operation completion rate itself is a positive indicator. For the frequency of secondary consultations, combined with sentiment analysis results, positive consultations are assigned positive weights, and negative consultations are assigned negative weights. Then, based on business experience, weights are assigned to these three processed features, a weighted sum is calculated, and the result is normalized to the 0-1 range to obtain the final real-time feedback valence.

[0130] Specifically, if the real-time feedback valence is low, it is determined that the current strategy has triggered a significant negative or ineffective response, and the user's behavioral pattern differs greatly from that of the successful role model group represented by the initial reference user set. At this point, simple strategy fine-tuning is insufficient, and it is necessary to find a more similar "role model" group. The reference user set is updated: the system excludes the currently used reference user set. It returns to the historical response database of the target user's customer group, but this time the selection criteria are modified. For example, it may no longer focus solely on users with the "highest historical repayment success rate," but instead look for users initially assigned the same strategy and with high real-time feedback valence, or users with similar dynamic feedback characteristic patterns to the target user. Based on this new reference user set, which places greater emphasis on "immediate response similarity," the system re-executes steps S4 to S6, namely, re-identifying the common feature set, calculating the dispersion, and generating a new target cross-feature that is more adapted to the current user's response pattern.

[0131] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for intelligent allocation of financial user service strategies, characterized in that, Includes the following steps: Obtain the multidimensional feature values ​​corresponding to the multidimensional features of the target user within the current business cycle. The multidimensional features include at least two of the following: repayment ability fluctuation index, repayment willingness score, service sensitivity, user difficulty level, and user value level. The historical response database of the target user's customer group is retrieved. The historical response database includes multiple users in the same group, the historical multidimensional feature values ​​corresponding to the multidimensional features of each user in the same group, and the historical repayment success rate corresponding to each user in the same group. The top K users with the highest historical repayment success rates in the same group are selected as reference users and form a reference user set. Identify at least two of the most frequently occurring multidimensional features in the reference user set and use them as a common feature set; wherein, the frequency of occurrence of a multidimensional feature is the number of reference users who possess the historical multidimensional feature value corresponding to that multidimensional feature; For each of the multidimensional features in the set of common features, calculate the data dispersion of its historical multidimensional feature values ​​within the reference user set; From the set of common features, the two multidimensional features with the lowest data dispersion are combined to obtain the target cross features of the target user; The target cross feature value corresponding to the target cross feature is input into the service strategy decision model to determine the corresponding target service strategy; or, the target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature of the target user are input into the service strategy decision model to determine the corresponding target service strategy. Assign the target service policy to the target users; The target cross feature value corresponding to the target cross feature and the multidimensional feature value corresponding to the multidimensional feature of the target user are input into the service strategy decision model to determine the corresponding target service strategy, including the following steps: Based on the historical multidimensional feature values ​​of the reference user set and their corresponding historical repayment success rates, the feature importance weight of each multidimensional feature in predicting the repayment success rate is calculated using a supervised machine learning model. Based on the feature importance weights corresponding to each of the multidimensional features, the multidimensional feature values ​​of the target user are weighted and summed to obtain a basic score value; Based on the historical performance data of the customer group to which the target user belongs, a dynamic scaling factor is generated to scale the target cross feature value of the target cross feature, so as to obtain the scaled target cross feature value. The service strategy score is obtained by adding the base score to the scaled target cross feature value. The service strategy score is matched with a preset strategy threshold range to determine the corresponding target service strategy; The target cross feature value corresponding to the target cross feature is input into the service strategy decision model to determine the corresponding target service strategy, including the following steps: Retrieve and traverse the preset strategy mapping table to obtain the target feature value interval, wherein the target feature value interval is the feature value interval into which the target cross feature value falls; The service policy mapped to the target feature value range is determined as the target service policy; The strategy mapping table is established by analyzing the historical response database. The strategy mapping table represents the mapping relationship between different feature value ranges of the target cross features and the service strategy with the highest historical repayment success rate. After assigning the target service policy to the target user, the process further includes the following steps: Real-time monitoring of the target user's response behavior within a preset time window; Extract the dynamic feedback feature set from the response behavior, the dynamic feedback feature set including strategy response delay duration, operation completion degree, and secondary consultation frequency; Based on the dynamic feedback feature set, the target service strategy is dynamically adjusted; The process of obtaining the multidimensional feature values ​​of the target user's multidimensional features within the current business cycle includes the following steps: Retrieve the target user's historical behavior sequence data within the current business cycle; the historical behavior sequence data includes behavioral events recorded in chronological order, and each behavioral event includes a behavior type and an occurrence timestamp; the behavior type includes repayment operation, communication contact, interface interaction, or document processing. Based on the historical behavior sequence data, a timestamped behavior chain is generated, which represents the time trajectory and type sequence of user behavior within the current business cycle; The multidimensional feature value is calculated based on the behavior type, timestamp, and sequence information contained in the behavior chain.

2. The intelligent allocation method for financial user service strategies according to claim 1, characterized in that: The step of dynamically adjusting the target service strategy based on the dynamic feedback feature set includes the following steps: Based on the dynamic feedback feature set, the real-time feedback valence of the target service strategy is calculated; the real-time feedback valence is used to quantify the degree of positive response of the target user to the target service strategy. If the real-time feedback valence is less than a first preset threshold, the reference user set is updated to redetermine the target cross feature.

3. The intelligent allocation method for financial user service strategies according to claim 1, characterized in that: Before calculating the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain, the following steps are also included: For each of the multidimensional features, determine whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value.

4. The intelligent allocation method for financial user service strategies according to claim 3, characterized in that: After determining whether the behavior types included in the behavior chain cover the target behavior type required to calculate the multidimensional feature value for each of the multidimensional features, the method further includes the following steps: If the target behavior type of at least two of the multidimensional features is covered, then the at least two multidimensional features are taken as the first feature type; The calculation of the multidimensional feature value based on the behavior type, timestamp, and sequence information contained in the behavior chain includes the following steps: Based on the target behavior type, timestamp, and sequence information corresponding to the first feature type contained in the behavior chain, the feature value corresponding to the first feature type is calculated.

5. The intelligent allocation method for financial user service strategies according to claim 4, characterized in that: The first feature type includes the repayment ability fluctuation index; The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps: The characteristic value corresponding to the repayment ability fluctuation index is obtained by measuring the fluctuation of the time interval between repayment failure operations and combining the frequency and urgency weight of the target user's access to income verification documents.

6. The intelligent allocation method for financial user service strategies according to claim 5, characterized in that: The first feature type includes the repayment willingness score; The step of calculating the feature value corresponding to the first feature type based on the target behavior type, the timestamp, and the sequence information contained in the behavior chain includes the following steps: The characteristic value corresponding to the repayment willingness score is obtained by measuring the ratio of the frequency of target users actively contacting customer service versus the frequency of the system actively contacting users, the density of the keywords of repayment commitment during the call, and the proportion of users skipping the agreement page.

Images