Information determination method and device, computer readable storage medium and computer program product

Abstract

The embodiment of the invention discloses an information determination method. The method comprises the steps of obtaining asset data of assets possessed by a to-be-processed user and behavior data related to a transaction behavior of the to-be-processed user at the current moment; determining first features of the to-be-processed user based on the asset data and the behavior data, and determining a target feature based on the first features and the weight of each first feature; wherein the first feature is a feature associated with asset data and behavior data of the to-be-processed user; and obtaining first historical behavior data of the to-be-processed user, and determining a target value of the to-be-processed user based on the target feature, the target behavior prediction model and the first historical behavior data. The embodiment of the invention further discloses information determination equipment and device, a computer readable storage medium and a computer program product.

Description

Technical Field

[0001] The present application relates to information determination technology in the computer field, and in particular to an information determination method, device, computer-readable storage medium and computer program product. Background Art

[0002] In the securities industry, accurate assessment of user value is an important factor in improving service levels and optimizing decision-making. At present, the relevant technology uses the Target Group Index (TGI) model to determine representative target features from multiple features of the target user, and then based on the target features, determines the control user corresponding to the target user from multiple users through static matching. After that, a simple comparison method is used to compare and analyze the behavior data of the target user and the control user, and the value of the target user is evaluated based on the analysis results.

[0003] However, in the related technology, when screening target features through the traditional TGI model, it is easy to ignore certain features that account for a small proportion but are crucial to users. This will not only lead to inaccurate target features obtained by the final screening, but also lead to large differences between the determined control users and the target users, thereby affecting the evaluation results of the target user's value. Summary of the invention

[0004] To solve the above technical problems, the embodiments of the present application hope to provide an information determination method, device, computer-readable storage medium and computer program product, which solve the problem of inaccurate user value evaluation results in the process of evaluating user value in the related art.

[0005] The technical solution of this application is implemented as follows:

[0006] A method for determining information, the method comprising:

[0007] Acquire asset data of assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment;

[0008] Based on the asset data and the behavior data, determine a first feature of the user to be processed, and based on the first feature and a weight of each first feature, determine a target feature; wherein the first feature is a feature associated with the asset data and the behavior data of the user to be processed;

[0009] The first historical behavior data of the user to be processed is obtained, and the target value of the user to be processed is determined based on the target feature, the target behavior prediction model and the first historical behavior data.

[0010] In the above scheme, determining the first feature of the user to be processed based on the asset data and the behavior data, and determining the target feature based on the first feature and the weight of each first feature, includes:

[0011] Using a target feature extraction algorithm to process the asset data and the behavior data to obtain a first feature of the user to be processed;

[0012] Determine a first parameter corresponding to each first feature based on the first feature of the user to be processed, the second feature of the benchmark user, the first number of the benchmark users, and the second number of the users to be processed; wherein the first parameter represents the degree of attention paid to the first feature;

[0013] The target feature is determined from the plurality of first features based on the plurality of first parameters and the weight of each of the first features.

[0014] In the above solution, determining the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data includes:

[0015] Determine a reference user from a plurality of users to be screened based on the target feature and the target nearest neighbor algorithm;

[0016] The second historical behavior data of the reference user is obtained, and the target value is determined based on the target behavior prediction model, the first historical behavior data and the second historical behavior data.

[0017] In the above solution, the step of determining a reference user from a plurality of users to be screened based on the target feature and the target nearest neighbor algorithm includes:

[0018] Determining a first vector based on a first feature value of each target feature of the user to be processed;

[0019] Determining a plurality of second vectors based on the second feature value of each target feature of each to-be-screened user;

[0020] determining a Mahalanobis distance between the first vector and each second vector;

[0021] The reference user is determined from the multiple users to be screened by performing processing based on the target nearest neighbor algorithm and the multiple Mahalanobis distances.

[0022] In the above solution, determining the target value based on the target behavior prediction model, the first historical behavior data and the second historical behavior data includes:

[0023] The target behavior prediction model is used to process the first historical behavior data and the second historical behavior data respectively to obtain the first target behavior data of the user to be processed and the second target behavior data of the reference user;

[0024] Based on the first target behavior data and the second target behavior data, the target value of the user to be processed is determined.

[0025] In the above scheme, the method further includes:

[0026] Determine a second parameter based on the first historical behavior data and the behavior data; wherein the second parameter represents a behavior change of the user to be processed;

[0027] If the second parameter is greater than or equal to the target threshold, the target behavior prediction model is updated based on the behavior data.

[0028] In the above scheme, the method further includes:

[0029] Acquire historical asset data of assets owned by the user to be processed;

[0030] Determine a third parameter based on the asset data and the historical asset data; wherein the third parameter represents the value fluctuation of the asset;

[0031] The weight is adjusted based on the second parameter, the third parameter and a target adjustment coefficient.

[0032] An information determination device, the device comprising: a processor, a memory, and a communication bus;

[0033] The communication bus is used to realize the communication connection between the processor and the memory;

[0034] The processor is used to execute the information determination program stored in the memory to implement the steps of the above-mentioned information determination method.

[0035] A computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more determiners to implement the steps of the above-mentioned information determination method.

[0036] A computer program product, comprising a computer program, which implements the steps of the above-mentioned information determination method when executed by a determiner.

[0037] The information determination method, device, computer-readable storage medium and computer program product provided in the embodiments of the present application can obtain the asset data of the assets of the user to be processed and the behavior data related to the transaction behavior of the user to be processed at the current moment, and then determine the first feature associated with the asset data and behavior data of the user to be processed based on the asset data and behavior data, and determine the target feature based on the first feature and the weight of each first feature, and further obtain the first historical behavior data of the user to be processed, and determine the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data; in this way, the target feature can be determined from multiple first features based on the weight of each first feature of the user to be processed, instead of directly determining the target feature through the target group index model as in the related art, but ignoring the importance of each first feature, thereby improving the accuracy of determining the target feature; and by combining the target behavior prediction model, the target feature and the historical behavior data, the target value of the user to be processed can be accurately determined, instead of only using a simple comparison method to determine the value of the target user as in the related art, thus solving the problem of inaccurate evaluation results of user value in the process of evaluating user value in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flowchart of an information determination method provided in an embodiment of the present application;

[0039] Figure 2 A flowchart of another information determination method provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of the structure of an information determination device provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the structure of an information determination device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In the field of financial technology and big data analysis, especially in the securities industry, accurately evaluating users' values ​​and behavior patterns is an important factor in improving service levels and optimizing decision-making. Securities companies provide users with market insights and trading advice through information subscription services. The effectiveness of such services directly affects users' investment decisions and satisfaction. However, traditional user value evaluation methods usually rely only on static data analysis and simple comparison methods, which makes it difficult to capture the multi-dimensional characteristics of user groups and the complexity of their behaviors.

[0043] Existing user analysis methods are mostly based on basic statistical models, such as average asset changes, transaction frequency statistics, etc. These methods are often limited to the surface analysis of user historical data and fail to effectively reflect the differences in user behavior in different market environments. For example, a simple comparison of net asset value growth rate or transaction volume can provide basic user information, but it cannot identify potential complex characteristics, such as different users' risk tolerance in market fluctuations, trading preferences, and reliance on information services.

[0044] In recent years, with the rapid development of big data analysis technology and artificial intelligence algorithms, the securities industry has gradually begun to introduce more complex models to analyze and predict user behavior. These technologies can not only process huge data sets, but also reveal the deep-seated reasons behind user behavior patterns through the interactive relationship analysis of multi-dimensional features. In particular, the target group index (TGI) model has become an effective tool for comparative analysis of user group characteristics. It helps identify significant differences between target users and overall market users by calculating the relative performance of specific groups on certain characteristics. However, the TGI model has certain limitations when facing complex multi-dimensional characteristics, especially when certain characteristics account for too low a proportion in the sample, the traditional TGI model often cannot fully reflect the importance of these characteristics to the business.

[0045] In order to better deal with this problem, the present invention assigns higher weights to important but low-proportion features, so that the analysis results can be more comprehensive and accurate. Moreover, this weighting mechanism can not only quantify the significant differences between user groups, but also flexibly adjust the weights of features according to business needs to enhance the flexibility and operability of the model. However, although this method can improve the understanding of user behavior, in practical applications, how to effectively screen the benchmark reference group and ensure the similarity between the reference group and the target user group in multidimensional features is still a difficult problem to be solved. Traditional reference user selection often relies on simple matching methods, such as dividing users based on the similarity of a single feature, which makes it difficult to find accurate reference users in multidimensional space.

[0046] As the market environment becomes increasingly complex, the user groups of securities companies show highly differentiated behavior patterns in different market conditions and time periods. Therefore, simple static comparison is no longer sufficient to support accurate user value assessment and decision-making. In order to better identify the actual impact of information subscription services on user behavior, it is necessary to take into account the dynamic changes of multi-dimensional features and perform more intelligent reference group matching based on these features to improve the accuracy and operability of the evaluation results.

[0047] In view of the limitations of the above-mentioned prior art, the present invention introduces a weight mechanism to perform multi-dimensional feature screening, which can effectively identify the significant feature differences between the target user and the benchmark user, and screen reference users with similar features to the target user through an intelligent algorithm. In addition, advanced matching technologies such as the K-nearest neighbor algorithm and Mahalanobis distance are used to ensure that the reference user and the subscribed user are highly similar in multi-dimensional space.

[0048] In addition, the present invention further uses time series analysis technology to dynamically track and compare users' asset changes and transaction behaviors, and quantify the impact of information services on users' long-term value. This intelligent matching and dynamic analysis method can effectively solve the problems of inaccurate reference user selection and inaccurate feature screening in the prior art, and provide securities companies with more scientific user behavior insights and decision support.

[0049] Through the present invention, securities firms can comprehensively evaluate the actual value of information subscription services in a dynamic market environment, optimize user service strategies, and enhance market competitiveness.

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0051] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0052] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.

[0053] 1) TGI (Target Group Index): Target group index, which is used to measure the relative performance of a specific user group on a certain feature and compare it with the overall group.

[0054] 2) KNN Algorithm (K-Nearest Neighbors Algorithm): A classification algorithm that matches the most similar users based on the nearest neighbor distance in the feature space.

[0055] 3) Mahalanobis Distance: A metric for measuring the distance between data points, taking into account the covariance of data features, and is used for multidimensional data similarity analysis.

[0056] 4) Control Group: A user group with similar characteristics to the target user group but not subscribed, used for difference analysis and effect evaluation.

[0057] 5) ARIMA (AutoRegressive Integrated Moving Average): Autoregressive differential moving average model, commonly used in time series analysis to predict the trend of data changes over time.

[0058] 6) GARCH model (Generalized Autoregressive ConditionalHeteroskedasticity): Generalized autoregressive conditional heteroskedasticity model, predicting volatility and uncertainty in time series.

[0059] 7) Weighting: Assign different weights to specific features in feature analysis to reflect the relative importance of the features to the business results.

[0060] 8) Dynamic screening function: A function used to adjust feature weights and screening criteria in real time according to market changes to maintain the similarity between the control group and the target user group.

[0061] 9) Feedback Mechanism: Based on real-time data and market changes, the system will dynamically adjust the model and screening criteria to improve the accuracy and adaptability of the forecast.

[0062] 10) Loss Function: A function used to measure the prediction error of the model. By minimizing the loss function, the performance and prediction accuracy of the model are optimized.

[0063] 11) LSTM (Long Short-Term Memory): Long short-term memory network, a recurrent neural network (RNN) used to process and predict long-term and short-term dependencies in time series data.

[0064] 12) Adaptive Model: A model that can dynamically adjust parameters and weights according to changes in the market environment to ensure the timeliness and accuracy of the forecast results.

[0065] 13) LIME (Local Interpretable Model-agnostic Explanations): A local interpretability model method that explains the prediction results of complex models by generating local linear models.

[0066] 14) SHAP (SHapley Additive exPlanations): A game theory-based interpretability method that provides the marginal contribution of each feature to the model’s prediction results.

[0067] The present application embodiment provides an information determination method, which can be applied to an information determination device, referring to Figure 1 As shown, the method comprises the following steps:

[0068] Step 101: Obtain asset data of the assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment.

[0069] In the embodiment of the present application, the user to be processed may refer to a user who has subscribed to the information service; the asset data may include the asset size data of the user to be processed; the behavior data may refer to the data generated after the user to be processed performs a transaction, and may include transaction frequency data, transaction amount data, and data of the branch where the user is located, etc. Among them, the branch structure where the user is located is the transaction organization where the user's transaction behavior occurs.

[0070] It should be noted that the asset data and behavior data can be obtained from a database.

[0071] Step 102: Determine the first feature of the user to be processed based on the asset data and the behavior data, and determine the target feature based on the first feature and the weight of each first feature.

[0072] Among them, the first feature is a feature associated with the asset data and behavior data of the user to be processed.

[0073] In the embodiment of the present application, the first feature of the user to be processed may include the asset size, transaction frequency, transaction quantity, risk preference, and branch structure of the user to be processed; the target feature may refer to the feature used for reference user screening. Specifically, the target feature extraction algorithm may be used to extract features from the asset data and behavior data to obtain multiple first features of the user to be processed, and then the target feature is determined from the multiple first features based on the weight of each first feature.

[0074] It should be noted that the weight of the first feature is set based on three factors: the impact of the first feature on the business goal, the attention paid to the first feature, and the proportion of the first feature in the user group. Among them, the business goals may include user conversion, asset growth, and transaction activity, and the greater the impact of the first feature on the business goal, the higher the weight; accordingly, the attention paid to each feature can be determined by means of a chi-square test or a T test, and the higher the attention paid, the higher the weight of the first feature; in addition, the higher the proportion of the first feature in the user group, the higher the weight.

[0075] In the embodiment of the present application, the weight of the first feature can be calculated according to the following formula (1).

[0076] W i =α*Ii +β*S i +γ*R i Formula (1)

[0077] Wherein, i represents the serial number of the first feature; W i represents the weight of the i-th first feature; I i Indicates the impact of the first feature on the business goal; S i Indicates the degree of attention paid to the first feature; R i It represents the proportion of the first feature in the user group; α, β, and γ are control coefficients, which are empirical values.

[0078] Step 103: Obtain the first historical behavior data of the user to be processed, and determine the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data.

[0079] In an embodiment of the present application, the first historical behavior data may refer to the behavior data generated after the user to be processed performs a transaction within a period of time in the past; the target behavior prediction model may refer to a long short-term memory network (hereinafter referred to as: LSTM) model.

[0080] In an embodiment of the present application, a reference user can be determined from multiple users to be screened based on target characteristics, and the historical behavior data of the user to be processed and the historical behavior data of the reference user can be processed separately through a target behavior prediction model to predict the future behavior of the user to be processed and the reference user, and then the value of the user to be processed, that is, the target value, is determined based on the obtained future behavior.

[0081] It should be noted that in the related art, ARIMA model or GARCH model can usually be used to predict the future behavior of users. However, ARIMA model or GARCH model can only process stable data (that is, data does not change over time), but cannot process nonlinear trends and emergencies, which results in the inability of existing prediction methods to cope with complex market environments and user behaviors; in addition, ARIMA model or GARCH model only focuses on data changes in a single dimension, but ignores the dynamic interaction between multiple features. For example, the transaction behavior of a user may be related to factors such as his asset status and market fluctuations, and ARIMA model or GARCH model cannot simultaneously consider the feature changes of multiple dimensions, which leads to the inability to accurately predict the future behavior of the user. However, in the embodiment of the present application, the LSTM model can be used to simultaneously process the feature data of multiple dimensions such as the user's transaction behavior, asset data and transaction frequency, and accurately predict the user's future behavior, thereby solving the problem that the user's future behavior cannot be accurately predicted in the related art.

[0082] The information determination method provided in the embodiment of the present application can determine the target feature from multiple first features based on the weight of each first feature of the user to be processed, instead of directly determining the target feature through a target group index model as in the related art, which ignores the importance of each first feature, thereby improving the accuracy of determining the target feature; and by combining the target behavior prediction model, target features and historical behavior data, the target value of the user to be processed can be accurately determined, instead of just using a simple comparison method to determine the value of the target user as in the related art, thus solving the problem of inaccurate evaluation results of user value in the process of evaluating user value in the related art.

[0083] Based on the above embodiments, the embodiments of the present application provide an information determination method, which can be applied to an information determination device, referring to Figure 2 As shown, the method may include the following steps:

[0084] Step 201: The information determination device obtains asset data of the assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment.

[0085] In the embodiment of the present application, assets may refer to financial products such as securities and stocks owned by the user to be processed. Specifically, the required asset data and behavior data may be directly obtained from the database.

[0086] In the embodiments of the present application, distributed database technology (such as Apache HBase, Cassandra) is usually used to store data, which can support concurrent processing and horizontal expansion of large amounts of data, ensuring rapid response to queries and analysis requests for user data; and this distributed storage architecture not only improves data processing efficiency, but also enhances data fault tolerance, ensuring high availability and reliability of large-scale user data.

[0087] In addition, this application can process the acquired real-time asset data and user behavior data in real time and efficiently through real-time stream data processing technology (such as Apache Kafka, Apache Flink). Through this stream processing framework, it is possible to process and analyze massive amounts of user transaction behavior and market data at a millisecond response speed to ensure the real-time and accuracy of user value assessment and improve the operating efficiency of the overall system.

[0088] Step 202: The information determination device processes the asset data and behavior data using a target feature extraction algorithm to obtain a first feature of the user to be processed.

[0089] In the embodiment of the present application, the target feature extraction algorithm may refer to a feature extraction (FeatureExtraction, Extrac) algorithm. Specifically, the asset data and the behavior data may be cleaned, denoised and standardized, and then multiple first features of the user may be extracted from the processed data according to the following formula (2).

[0090] F i =Extract(D i ) Formula (2)

[0091] Among them, F i represents the i-th first feature, D i Represents asset data and behavior data obtained from the database.

[0092] Step 203: The information determination device determines a first parameter corresponding to each first feature based on the first feature of the user to be processed, the second feature of the benchmark user, the first number of the benchmark users, and the second number of the users to be processed.

[0093] The first parameter represents the degree of attention paid to the first feature.

[0094] In the embodiment of the present application, the benchmark user includes the user to be processed; the first parameter may refer to the target group index, i.e., the TGI index. Specifically, for a first feature, the third number b of users with the first feature among the users to be processed may be determined based on the first feature of the users to be processed and the second number a of the users to be processed, and then the first proportion of the users to be processed on the first feature may be determined based on the second number and the third number. For example, the third number of users with 10 transaction behaviors among the users to be processed may be determined, and then the third number and the second number of users to be processed are calculated to obtain the proportion of the users to be processed with the feature of having 10 transaction behaviors.

[0095] Afterwards, the fourth number d of users with a certain first feature among the benchmark users can be determined based on the first feature of the user to be processed, the second feature of the benchmark user, and the first number c. Further, the second proportion of the benchmark users in the certain first feature can be determined based on the first number and the fourth number.

[0096] Then, the target group model (TGI model) is used to process the first proportion and the second proportion, that is, the first proportion and the second proportion are used as input parameters of the TGI model, so as to obtain the TGI index of each first feature, that is, the first parameter. It should be noted that the calculation formula of the TGI model can be shown as the following formula (3):

[0097]

[0098] Among them, TGI represents the first parameter, that is, the TGI index; A represents the first proportion; B represents the second proportion. It should be noted that if the first parameter is higher than 120 or lower than 80, it means that there is a significant difference between the target user and the benchmark user in this feature.

[0099] Step 204: The information determination device determines a target feature from the multiple first features based on the multiple first parameters and the weight of each first feature.

[0100] In the embodiment of the present application, for each first feature, the weight of each first feature and the first parameter of each first feature are multiplied to obtain a target value, and then the first feature corresponding to the target value greater than the first threshold or less than the second threshold is used as the target feature. The first threshold and the second threshold are preset empirical values.

[0101] It should be noted that the calculation formula of the target value can be shown as the following formula (4):

[0102] C=TGI*W i Formula (4)

[0103] Wherein, C represents the target value; TGI represents the first parameter of each first feature; W i Represents the weight of each first feature.

[0104] In the embodiment of the present application, by weighting the features according to the importance of each feature, it can be ensured that certain features that have a low proportion in the user group but a high degree of impact on the business objectives can be fully considered. For example, by assigning different weights to the risk preferences and trading behaviors of a specific user group (such as high net worth users) in a specific market environment, the significant differences between user groups can be better captured, thereby accurately screening out the target features from multiple first features.

[0105] It should be noted that the weight of the first feature is not fixed. When the market value of the assets owned by the user or the user behavior fluctuates greatly, the weight value of the feature can be updated to obtain a more accurate target feature, and then obtain a reference user that is closer to the target user. Specifically, the weight can be adjusted according to steps A1 to A3.

[0106] A1. The information determination device obtains historical asset data of the assets owned by the user to be processed.

[0107] In an embodiment of the present application, data such as the asset size of the assets owned by the user to be processed over a period of time in the past may be obtained.

[0108] A2. The information determination device determines a third parameter based on the asset data and the historical asset data.

[0109] Among them, the third parameter represents the value fluctuation of the asset.

[0110] In the embodiment of the present application, the asset data at the current moment and the historical asset data can be calculated to obtain the value fluctuation of the assets owned by the user to be processed during the period (i.e., the third parameter). The calculation formula of the third parameter can be shown as formula (5):

[0111]

[0112] Wherein, D represents the third parameter; a 1 Represents historical asset data; a 2 Represents asset data.

[0113] A3. The information determining device adjusts the weight based on the second parameter, the third parameter and the target adjustment coefficient.

[0114] In the embodiment of the present application, the second parameter represents the behavior change of the user to be processed. Specifically, a dynamic adjustment function can be used to adjust the weight of the first feature according to the change of asset value (i.e., the third parameter), the change of user behavior (i.e., the second parameter) and the target adjustment coefficient. The formula of the dynamic adjustment function can be shown as the following formula (6):

[0115] W i (t+1)=W i (t)*(1+α*f(ΔM(t))) Formula (6)

[0116] Among them, W i (t+1) represents the adjusted weight of the first feature; W i (t) represents the current weight of the first feature; α is the target adjustment coefficient, which is an empirical value and can be adjusted according to the change in the asset or behavior; ΔM(t) represents the change in asset value and user behavior, that is, it includes the second parameter and the third parameter; f(ΔM(t)) is the feedback function. It should be noted that ΔM(t) = λ 1 *g 1 +λ 2 *g 2 Among them, g 1 Represents the third parameter, g 2 represents the second parameter, λ 1 , 2 is an adjustment coefficient, which can be a preset empirical value.

[0117] In a feasible manner, data related to the market environment, data related to the user churn rate, and the analysis result of the first feature can also be obtained, and a dynamic adjustment function is used to adjust the weights according to these three parameters. In this case, ΔM(t)=λ 3 *g 3 +λ 4 *g 4 +λ 5 *g 5 Among them, g 3 is a function related to market environment data, which is used to reflect the impact of market fluctuations on weights, and g 4 is a function related to changes in user behavior, and g 5 is a function related to changes in user behavior, and λ 3 , 4 , 5 It is also the weight adjustment coefficient, which is an artificially set empirical value, and the weight adjustment coefficient can be changed according to actual needs.

[0118] In an embodiment of the present application, after the weights are adjusted, the latest asset data and behavior data can be obtained in real time through the Application Programming Interface (API), and the weights can be adjusted again based on the real-time asset data and behavior data. In this way, necessary data related to the user can be obtained only through the API interface, thereby reducing the amount of message transmission between the user end and the server, reducing the delay in data transmission, improving the overall throughput and response speed, and effectively solving the problem of low processing efficiency caused by data redundancy in related technologies, thereby completing large-scale, multi-dimensional user data analysis in a shorter time, greatly improving business processing efficiency and system performance.

[0119] In another feasible way, when adjusting the weight of the first feature, the interaction relationship between the first features can be captured by a multivariate regression model, and the weight can be adjusted based on the interaction relationship. For example, the asset size and transaction frequency of a user may have an interactive effect, so the interaction relationship between the asset size and transaction frequency can be captured by a multivariate regression model, and the weight can be adjusted based on the obtained interaction relationship. The multivariate regression model can be expressed by the following formula (7):

[0120]

[0121] Where: Y represents the user's target behavior (such as user conversion rate and asset increment, etc.); X iThe vector representing the first target feature of the target user; X z The vector representing the second target feature of the target user; β ij Represents the interaction effect between the first target feature and the second target feature.

[0122] In an embodiment of the present application, by adjusting the weights of features according to actual business needs, that is, providing a flexible weight adjustment mechanism, it can be ensured that in certain specific business scenarios, certain key features can be fully considered, that is, the key features are given higher weights to ensure that these features are given priority in the analysis. For example: when conducting user behavior analysis for branches in different regions, their weights can be adjusted according to the contribution of the branch to the business. This adaptive feature weight adjustment mechanism significantly improves the flexibility and practicality of weight setting, and solves the problem of rigid user value evaluation results due to the fixed nature of feature weights in the prior art. Especially in a changing market environment, by dynamically adjusting the weight distribution, it is possible to quickly adapt to the latest market changes, thereby providing more accurate user value evaluation results.

[0123] In the embodiment of the present application, after adjusting the weight of the first feature, the target feature will be re-determined from the multiple first features, so the reference users also need to be re-screened. At this time, in order to improve the screening efficiency and accuracy of the reference users, the adjustment of the reference users can be completed in a shorter time through the iterative optimization function. Specifically, the optimization function can be defined as the following formula (8):

[0124]

[0125] Among them, L(W) is the optimization function, which is the distance between the feature vectors; X i A vector representing the target features of the target user; Y i The vector representing the target features of the users to be screened; TGI(X i ,Y i ) represents the group difference between the target users and the users to be screened. It should be noted that

[0126] In an embodiment of the present application, by minimizing the optimization function, the weight of the first feature and the screening rules of the reference users can be optimized, so that the group difference between the screened reference users and the target users is minimized. In this way, in periods of large market fluctuations, the screening criteria for reference users can be automatically adjusted, thereby improving the accuracy of the screened reference users.

[0127] Step 205: The information determination device obtains the first historical behavior data of the user to be processed.

[0128] Step 206: The information determination device determines a reference user from a plurality of users to be screened based on the target feature and the target nearest neighbor algorithm.

[0129] In the embodiment of the present application, the target nearest neighbor algorithm refers to the K nearest neighbor algorithm (hereinafter referred to as: KNN algorithm); the to-be-screened user may refer to a user who has not subscribed to the information service. Specifically, the Mahalanobis distance between the target feature of the to-be-processed user and the target feature of each to-be-screened user may be determined, and then the KNN algorithm may be used to process the multiple Mahalanobis distances to determine the reference user from the multiple to-be-screened users.

[0130] It should be noted that, in another feasible method, each first feature of the user to be processed can be divided into several areas according to its distribution; for example: the user's asset size can be divided into three intervals: low, medium and high; the user's transaction frequency can also be divided into three intervals: low, medium and high.

[0131] For example, if the transaction frequency is less than 5 times, it belongs to the low range; if the transaction frequency is greater than or equal to 5 times and less than or equal to 10 times, it belongs to the middle range; and if the transaction frequency is greater than 10 times, it belongs to the high range.

[0132] Afterwards, different feature intervals are combined into several categories, namely C ij =F i1 ∩F j1 ,F i2 ∩F j2 Among them, F i1 ,F j1 Represent different intervals of the first feature i and the first feature j respectively.

[0133] Furthermore, after combining the features, different combinations can be screened by hierarchical screening to obtain a target feature combination, and then the KNN algorithm and the Mahalanobis distance algorithm are used to accurately screen the users to be screened based on the obtained target feature combination, thereby obtaining the final reference users.

[0134] In the embodiment of the present application, step 206 can be implemented through steps 206a to 206d.

[0135] Step 206a: The information determination device determines a first vector based on a first feature value of each target feature of the user to be processed.

[0136] In the embodiment of the present application, a first vector X corresponding to the user to be processed can be constructed based on the first feature value of each target feature of the user to be processed. i , that is, X i =[x 1 , x 2 , ..., x n ]. Among them, x1 represents the feature value of the first target feature of the user to be processed; x n Indicates the feature value of the nth target feature of the user to be processed.

[0137] Step 206b: The information determination device determines a plurality of second vectors based on the second feature value of each target feature of each user to be screened.

[0138] In the embodiment of the present application, for each user to be screened, based on the second feature value of the target feature of each user to be screened, a second vector Y corresponding to each user to be screened is constructed. i =[y 1 ,y 2 , ..., y n ]. It should be noted that one user to be screened corresponds to one second vector.

[0139] Step 206c: The information determination device determines the Mahalanobis distance between the first vector and each second vector.

[0140] In the embodiment of the present application, the Mahalanobis distance between the first vector and each second vector can be calculated according to the following formula (9).

[0141]

[0142] Among them, d(X i , Y i ) represents the Mahalanobis distance between the first vector and the second vector; X i Represents the first vector; Y i represents the second vector; S -1 represents the covariance matrix.

[0143] It should be noted that the Mahalanobis distance can represent the similarity between the first vector and the second vector, that is, the similarity between the target user and the reference user. In other words, the smaller the Mahalanobis distance, the higher the similarity between the target user and the reference user (i.e., the control user), that is, the more similar the target user and the reference user are.

[0144] Step 206d: The information determination device determines a reference user from a plurality of users to be screened based on a target nearest neighbor algorithm and a plurality of Mahalanobis distances.

[0145] In an embodiment of the present application, a KNN algorithm may be used to process multiple Mahalanobis distances to select k Mahalanobis distances with the smallest distance values ​​from the multiple Mahalanobis distances according to the size of the distances, and then the user to be screened corresponding to each selected Mahalanobis distance is used as the final reference user.

[0146] For each target user X i , we can define the set N of reference users closest to the target userk (X i )={X j |d(X i ,Y i ) is the first k smallest distances}.

[0147] It should be noted that, in a feasible method, the Euclidean distance between the first vector and each second vector can be calculated first, and then the reference user can be determined from multiple users to be screened based on the KNN algorithm and multiple Euclidean distances; however, the Euclidean distance is only applicable to scenarios with relatively uniform feature distribution, while for scenarios with unbalanced weights between features, using the Euclidean distance to screen reference users will result in inaccurate screened reference users.

[0148] In the embodiment of the present application, by determining the Mahalanobis distance between the first vector and each second vector, not only external factors such as the market environment are dynamically considered, but also the correlation between different characteristics of the user is considered. In this way, it is possible to ensure that the reference user is highly consistent with the target user in terms of significant characteristics such as assets and transaction behaviors, rather than simply matching the user's asset size or transaction frequency and other characteristics to determine the reference user through a simple feature matching method as in the related art. This solves the problem of large differences between the reference users screened out in the related art and the target users, and significantly improves the comparability of the reference users.

[0149] In another feasible way, if the market is relatively stable and user behavior changes are relatively regular, you can manually set the screening criteria based on the user's characteristics (such as transaction frequency or branch, etc.) based on the preset business rules to select reference users similar to the target user. This method is simple to implement and easy to operate, and is suitable for scenarios where the accuracy of analysis is not high.

[0150] Step 207: The information determination device obtains the second historical behavior data of the reference user, and determines the target value based on the target behavior prediction model, the first historical behavior data, and the second historical behavior data.

[0151] In an embodiment of the present application, behavioral data generated after a user's transaction behavior in the past period of time, i.e., second historical behavior data, can be obtained from a database. Thereafter, a target behavior prediction model can be used to process the first historical behavior data and the second historical behavior data, respectively, to predict the future behavior of the user to be processed and the target user, and then evaluate the target value of the target user based on the future behavior.

[0152] In an embodiment of the present application, the target behavior prediction model may refer to an LSTM model, and LSTM is a deep learning model specifically used to process time series data. It can capture long-term and short-term time dependencies and is particularly suitable for dynamic prediction of user behavior.

[0153] Specifically, the structure of the LSTM unit contains three gating mechanisms: input gate, forget gate and output gate; among them, the input gate i t Used to control the impact of the current input, and i k =σ(W i *[h t-1 ,x t ]+b i ); forget gate f t It is used to control the forgetting ratio of the previous memory, and f t =σ(W f *[h t-1 ,x t ]+b f ); output gate o t Determines the output content, and o t =σ(W o *[h t-1 ,x t ]+b o ), and the output of the LSTM model is h t =o t *tanh(C t ).

[0154] In addition, the memory state update formula of the LSTM model can be shown as follows:

[0155]

[0156] Among them, C t Indicates the current memory status. represents the candidate memory state, σ represents the activation function, [h t-1 ,x t ] represents the concatenation result of the hidden layer state at the previous moment and the current input.

[0157] In the embodiment of the present application, step 207 can be implemented through steps 207a to 207b.

[0158] Step 207a: The information determination device uses the target behavior prediction model to process the first historical behavior data and the second historical behavior data respectively to obtain the first target behavior data of the user to be processed and the second target behavior data of the reference user.

[0159] In the embodiment of the present application, for the user to be processed and the reference user, a time series matrix can be constructed according to the first historical behavior data and the second historical behavior data, respectively. For example, for the user to be processed, the following time series matrix can be constructed according to the first historical behavior data: Among them, t 1 ,t 2 ,...,t T represents the time step; x i1 ,x i2 ,...,x in Indicates the characteristics of users in different time periods, such as transaction frequency, asset size, etc.

[0160] Afterwards, before inputting the time series matrix into the target behavior prediction model, the exponentially weighted moving average (EWMA) function can be used to smooth the time series data, and then the smoothed time series data is subjected to differential processing to remove the trend item in the time series and make the time series more stable. Further, the differentially processed time series data is subjected to standardized processing to obtain the standardized time series data. Among them, the formula of the EWMA function can be shown in the following formula (12), the differential processing formula can be shown in the following formula (13), and the standardized processing formula can be shown in the following formula (14):

[0161]

[0162] Δx i (t) = x i (t)-x i (t-1) Formula (13)

[0163]

[0164] Among them, α represents the smoothing coefficient; is the smoothed time series; μ i is feature x i The mean of i is feature x i The standard deviation of .

[0165] In the embodiment of the present application, the standardized time series data can be used as input parameters and input into the LSTM model. After that, the LSTM will process the input parameters to obtain the first target behavior data of the user to be processed in the future, and the second target behavior data of the reference user in the future. Specifically, the behavior prediction problem can be expressed as the following formula (15):

[0166]

[0167] in, is the predicted value of the user's behavior at the future time t, f(X i ) is the prediction function of the LSTM model after training.

[0168] In the embodiment of the present application, since the behavior data of the user to be processed is dynamically changing, the latest user behavior data can be obtained from the database in real time through the API interface, and the LSTM model can be updated according to the latest user behavior data. Specifically, the target behavior prediction model can be updated according to steps B1 to B2.

[0169] B1. The information determination device determines a second parameter based on the first historical behavior data and the behavior data.

[0170] The second parameter represents the behavior change of the user to be processed.

[0171] In the embodiment of the present application, the first historical behavior data and the latest behavior data of the user to be processed can be calculated to obtain the second parameter, that is, the behavior change of the user to be processed. The calculation formula of the second parameter ΔH(t) can be shown as the following formula (16):

[0172]

[0173] In an embodiment of the present application, by combining the user's historical behavior data and the latest behavior data to determine the second parameter, the target behavior prediction model can be updated in a timely manner when the user's behavior fluctuates greatly, so as to accurately predict the user's future behavior and obtain an accurate assessment value of the user's value.

[0174] B2. If the second parameter is greater than or equal to the target threshold, the information determination device updates the target behavior prediction model based on the behavior data.

[0175] In the embodiment of the present application, if it is determined that the second parameter is greater than or equal to the target threshold, it means that the current user's behavior has changed significantly. At this time, the parameters of the LSTM model (such as the number of hidden layer units, learning rate and time step, etc.) can be adjusted by cross-validation, grid search and random search. It should be noted that the target behavior prediction model is usually adjusted by minimizing the loss function, that is, updating the target behavior prediction model. In this way, the performance of the target behavior prediction model can be continuously optimized through real-time parameter tuning, thereby ensuring the accuracy of the prediction results. Among them, the loss function can be expressed by the following formula (17):

[0176]

[0177] Among them, L(θ) is the loss function; is the predicted data of the behavior; N is the number of samples; O i The actual behavior data.

[0178] It should be noted that in the embodiments of the present application, cloud computing resources may be used for model training, especially containerization technology (such as Docker, Kubernetes) to build an elastic computing environment, that is, through cloud model training and containerization technology, large-scale user data can be efficiently processed to achieve rapid iteration and upgrade of the LSTM model. This not only improves the speed of model training, but also significantly reduces infrastructure and maintenance costs.

[0179] In addition, the model can be deployed through model compression and quantization technology (such as TensorFlow Lite, ONNX), which can effectively reduce the computing resource usage of the LSTM model on the computer device, thereby reducing the consumption of computing resources while ensuring the accuracy of behavior prediction, which enables the model to adapt to different computing environments, including the high-performance computing environment of the server and the low-computing computing environment of the computer device. This flexible deployment method improves the portability of the information determination system, enabling it to be widely used in various financial institutions and corporate environments.

[0180] In the embodiment of the present application, the long short-term memory network (LSTM) model in deep learning is used. It can accurately process the multi-dimensional time series data of users, analyze the asset changes, trading behaviors and risk preferences of users in different market environments, and then capture the dynamic changes of user trading behaviors and asset changes to help brokers accurately predict the future behavior patterns of users; and, through regular training and automated data updates, the LSTM model can be optimized in real time according to changes in the market environment and user behavior. This adaptive mechanism ensures the stability and efficiency of the LSTM model in long-term use, especially for rapidly changing market environments.

[0181] It should be noted that in order to solve the "black box" problem in traditional models and enable users to understand and trust the prediction results, local interpretable models (Local Interpretable Model-agnostic Explanations, LIME) and game theory-based interpretability methods (SHapley Additive exPlanations, SHAP) can be used to deeply analyze the output results of the LSTM model and provide clear feature contributions and decision paths, so that users can intuitively understand the impact of each feature on the final behavior prediction results.

[0182] LIME generates a set of samples close to the original input by locally perturbing the input samples, and fits these samples with a simple linear model. In this way, LIME can provide a linear approximate explanation for the local prediction of a complex model. Specifically, sample fitting can be performed according to steps C1 to C3.

[0183] C1: Perturbation generation. For a given input feature vector X i =[x 1 , x 2 , ..., x n ], LIME first generates i Similar perturbation sample sets It should be noted that each By transforming the original input feature X i Obtained by random sampling.

[0184] C2: Linear model fitting. For the generated perturbation sample set, LIME uses the original complex model f to calculate the prediction result of each perturbation sample, and trains a simple linear model g to approximate the behavior of f in a local area, as shown in the following formula (18):

[0185]

[0186] Among them, β i is feature x i The linear regression coefficient indicates the contribution of each feature to the model prediction results.

[0187] C3: Feature contribution explanation. Through the linear model g(x), LIME generates the contribution explanation of each feature in the local area. Among them, the feature contribution (x i )=β i x i .

[0188] In this way, LIME provides an explanation of complex models in local areas through a simple linear model, which can help users understand the impact of each of the user's first features in the current behavior prediction.

[0189] In addition, SHAP can measure the marginal contribution of each feature to the behavior prediction result by calculating the Shapley value of each feature. The Shapley value is derived from cooperative game theory and can provide a fair contribution distribution scheme for each feature, which is particularly suitable for complex model interpretation of multi-dimensional features. Specifically, the marginal contribution of a feature can be measured according to steps D1 to D3.

[0190] D1: Feature combination generation. Generate all possible feature subsets And calculate the behavior prediction value under these feature combinations.

[0191] D2: The marginal contribution φ can be calculated by the following formula (19): i :

[0192]

[0193] Among them, f(S) represents the predicted value of feature subset S, φ i is feature x i The marginal contribution to the model prediction.

[0194] D3: Feature contribution explanation. SHAP provides a global explanation of the model and a feature contribution graph by calculating the Shapley value of each feature, where the contribution of each feature is based on its average marginal contribution in all feature combinations, thus ensuring fairness and consistency in contribution distribution.

[0195] In a feasible way, in order to help users understand the prediction results of the model more intuitively, the influence of each feature on the behavior prediction results can be intuitively displayed through feature contribution bar charts, force charts and trend charts. Among them, the feature contribution bar chart shows the contribution of each feature through a bar chart, so that users can clearly see which feature has the greatest impact on the prediction results, and the contribution of the feature is determined by the explanatory value provided by SHAP or LIME. The bar chart can significantly reflect the importance of the feature; correspondingly, the force chart is used to show the interaction between features and their impact on the final prediction, which is particularly suitable for capturing the complex relationship of feature interaction. That is, the force chart can show how the interaction of features jointly affects the model output, so that users can understand which combination of features has an important impact on the results; the trend chart is used to show the changes in time series data, and is particularly suitable for showing the trend of user behavior over time. By combining the prediction results of the time dimension, users can understand the expected changes in user behavior in the future.

[0196] In an embodiment of the present application, LIME is used to perturb the model output to generate samples to show the contribution value of each feature of the user; SHAP is used to calculate the marginal contribution of each feature based on game theory, which can not only help users understand the decision-making basis of the behavior prediction model, but also further improve the transparency of the model, and effectively solve the "black box" problem existing in complex models in related technologies; and by providing a feature contribution chart, the specific impact of each feature on the prediction result can be clearly seen, making the output of the model not only transparent but also easier to understand.

[0197] Step 207b: The information determination device determines the target value of the user to be processed based on the first target behavior data and the second target behavior data.

[0198] In the embodiment of the present application, after obtaining the first target behavior data and the second target behavior data, the first value of the user to be processed and the second value of the reference user can be determined according to the first target behavior data and the second target behavior data, respectively. After that, the target value is determined according to the error threshold between the first value and the second value, that is, if the error threshold between the first value and the second value meets the preset threshold, the first value is directly determined as the target value, otherwise, the second value is determined as the target value. It should be noted that the first value and the second value can be calculated according to the following formula (20).

[0199] N=num*e*p Formula (20)

[0200] Among them, N represents the first value or the second value; num represents the number of users to be processed or the number of reference users; e represents the first target behavior data or the second target behavior data; p is the average conversion value of the users to be processed or the average conversion value of the reference users; it should be noted that the average conversion value may refer to the historical average conversion value, which can be directly obtained from the database through the API interface.

[0201] It should be noted that in order to further enhance the user experience, the present invention also designs an interactive query function, which allows users to interactively query data and adjust models through a graphical interface. Users can manually select different feature combinations, adjust model parameters, and view prediction results under different parameter settings according to their business needs.

[0202] Specifically, you can select the feature combination of interest by clicking or dragging. For example, you can select the two features of transaction frequency and user risk preference as the feature combination. Then, you can predict the user's future behavior through the behavior prediction model based on the feature combination selected by the user, and update the visualization chart at the same time. In addition, you can also dynamically adjust the hyperparameters of the LSTM model (such as learning rate and number of hidden layer units, etc.). Then, after the parameter adjustment is completed, you can retrain the LSTM model based on the adjusted parameters to obtain updated prediction results. In this way, through this flexible interactive function, users can have a deep understanding of the working mechanism of the LSTM model and adjust the behavior of the LSTM model in different scenarios.

[0203] In one feasible way, in order to improve the user experience on different devices, an adaptive user interface (UI) can also be set, that is, automatically adjusting the layout according to the user device type (such as desktop, mobile) and screen size to ensure that each user can get the best experience on different devices. In other words, the UI interface will automatically adjust according to the size of the screen to ensure that it can be displayed reasonably on both desktop and mobile devices; accordingly, feature contribution charts, force graphs and trend graphs can be scaled and rearranged according to the screen size to ensure the readability and interactivity of the information.

[0204] In the embodiment of the present application, by designing an adaptive user interface, users can adjust and query analysis results in real time through an interactive interface, which greatly improves user participation and satisfaction; and by providing detailed analysis reports and personalized suggestions, users can better understand the impact of subscription services on user behavior, which not only improves user trust in the service, but also enhances loyalty to the service.

[0205] In the embodiments of the present application, by combining the interpretability framework, dynamic visualization tools and adaptive interactive interface of LIME and SHAP, the transparency of the LSTM model and the user experience are significantly improved. Users can not only clearly understand the behavior prediction logic of the LSTM model, but also independently explore the behavior and output results of the model through interactive query functions. This enables complex machine learning models to be more widely used in actual business scenarios, especially in the fields of finance and user analysis that have high requirements for transparency.

[0206] In summary, in the embodiments of the present application, the accuracy of user value assessment, the flexibility and adaptability of the system are significantly improved through weighted TGI model, multi-dimensional feature analysis, intelligent control group matching and dynamic market adaptation technology, and multiple bottleneck problems in the prior art are solved. By optimizing data processing efficiency, enhancing model transparency, reducing system load and other technical improvements, the present invention provides brokers with more accurate and efficient user value assessment and business optimization solutions.

[0207] The information determination method provided in the embodiment of the present application can determine the target feature from multiple first features based on the weight of each first feature of the user to be processed, instead of directly determining the target feature through a target group index model as in the related art, which ignores the importance of each first feature, thereby improving the accuracy of determining the target feature; and by combining the target behavior prediction model, target features and historical behavior data, the target value of the user to be processed can be accurately determined, instead of just using a simple comparison method to determine the value of the target user as in the related art, thus solving the problem of inaccurate evaluation results of user value in the process of evaluating user value in the related art.

[0208] Based on the above embodiments, the embodiments of the present application provide an information determination device, which can be applied to Figure 1 and 2 In the information determination method provided in the corresponding embodiment, refer to Figure 3 As shown, the information determination device 3 may include: a processor 31, a memory 32 and a communication bus 33, wherein:

[0209] The communication bus 33 is used to realize the communication connection between the processor 31 and the memory 32;

[0210] The processor 31 is used to execute the information determination program in the memory 32 to implement the following steps:

[0211] Obtaining asset data of the assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment;

[0212] Determine a first feature of the user to be processed based on the asset data and the behavior data, and determine a target feature based on the first feature and a weight of each first feature; wherein the first feature is a feature associated with the asset data and the behavior data of the user to be processed;

[0213] The first historical behavior data of the user to be processed is obtained, and the target value of the user to be processed is determined based on the target feature, the target behavior prediction model and the first historical behavior data.

[0214] In other embodiments of the present application, the processor 31 is used to execute the information determination program in the memory 32 to determine the first feature of the user to be processed based on the asset data and the behavior data, and determine the target feature based on the first feature and the weight of each first feature, so as to implement the following steps:

[0215] The target feature extraction algorithm is used to process the asset data and behavior data to obtain the first feature of the user to be processed;

[0216] Based on the first feature of the user to be processed, the second feature of the benchmark user, the first number of benchmark users and the second number of users to be processed, determining a first parameter corresponding to each first feature; wherein the first parameter represents the degree of attention paid to the first feature;

[0217] A target feature is determined from the plurality of first features based on the plurality of first parameters and a weight of each first feature.

[0218] In other embodiments of the present application, the processor 31 is used to execute the information determination program in the memory 32 to determine the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data, so as to implement the following steps:

[0219] Based on the target features and the target nearest neighbor algorithm, a reference user is determined from multiple users to be screened;

[0220] The second historical behavior data of the reference user is obtained, and the target value is determined based on the target behavior prediction model, the first historical behavior data and the second historical behavior data.

[0221] In other embodiments of the present application, the processor 31 is used to execute the target feature and target neighbor algorithm based on the information determination program in the memory 32 to determine the reference user from the multiple users to be screened, so as to implement the following steps:

[0222] Determine a first vector based on a first feature value of each target feature of the user to be processed;

[0223] Determining a plurality of second vectors based on the second feature value of each target feature of each to-be-screened user;

[0224] determining a Mahalanobis distance between the first vector and each second vector;

[0225] Based on the target nearest neighbor algorithm and multiple Mahalanobis distances, a reference user is determined from multiple users to be screened.

[0226] In other embodiments of the present application, the processor 31 is used to execute the information determination program in the memory 32 to determine the target value based on the target behavior prediction model, the first historical behavior data, and the second historical behavior data to implement the following steps:

[0227] The target behavior prediction model is used to process the first historical behavior data and the second historical behavior data respectively to obtain the first target behavior data of the user to be processed and the second target behavior data of the reference user;

[0228] Based on the first target behavior data and the second target behavior data, a target value of the user to be processed is determined.

[0229] In other embodiments of the present application, the processor 31 is used to execute the information determination program in the memory 32 to implement the following steps:

[0230] Determine a second parameter based on the first historical behavior data and the behavior data; wherein the second parameter represents a behavior change of the user to be processed;

[0231] If the second parameter is greater than or equal to the target threshold, the target behavior prediction model is updated based on the behavior data.

[0232] In other embodiments of the present application, the processor 31 is used to execute the information determination program in the memory 32 to implement the following steps:

[0233] Obtain historical asset data of assets owned by the user to be processed;

[0234] Determine a third parameter based on the asset data and the historical asset data; wherein the third parameter represents the value fluctuation of the asset;

[0235] The weight is adjusted based on the second parameter, the third parameter and the target adjustment coefficient.

[0236] It should be noted that the specific description of the steps executed by the processor can be referred to Figure 1 and 2 The information determination method provided in the corresponding embodiment will not be repeated here.

[0237] The information determination device provided in the embodiment of the present application can determine the target feature from multiple first features based on the weight of each first feature of the user to be processed, instead of directly determining the target feature through a target group index model as in the related art, which ignores the importance of each first feature, thereby improving the accuracy of determining the target feature; and by combining the target behavior prediction model, target features and historical behavior data, the target value of the user to be processed can be accurately determined, instead of just using a simple comparison method to determine the value of the target user as in the related art, thus solving the problem of inaccurate evaluation results of user value in the process of evaluating user value in the related art.

[0238] Based on the above embodiments, the present invention provides an information determination device, which can be applied to Figure 1 and 2 In the information determination method provided in the corresponding embodiment, refer to Figure 4 As shown, the information determination device 4 may include: an acquisition unit 41, a processing unit 42 and a determination unit 43, wherein:

[0239] An acquisition unit 41 is used to acquire asset data of assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment;

[0240] The processing unit 42 is used to determine the first feature of the user to be processed based on the asset data and the behavior data, and determine the target feature based on the first feature and the weight of each first feature; wherein the first feature is a feature associated with the asset data and the behavior data of the user to be processed;

[0241] The determination unit 43 is used to obtain the first historical behavior data of the user to be processed, and determine the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data.

[0242] In other embodiments of the present application, the processing unit 42 is further configured to perform the following steps:

[0243] The target feature extraction algorithm is used to process the asset data and behavior data to obtain the first feature of the user to be processed;

[0244] Based on the first feature of the user to be processed, the second feature of the benchmark user, the first number of benchmark users and the second number of users to be processed, determining a first parameter corresponding to each first feature; wherein the first parameter represents the degree of attention paid to the first feature;

[0245] A target feature is determined from the plurality of first features based on the plurality of first parameters and a weight of each first feature.

[0246] In other embodiments of the present application, the determining unit 43 is further configured to perform the following steps:

[0247] Based on the target features and the target nearest neighbor algorithm, a reference user is determined from multiple users to be screened;

[0248] The second historical behavior data of the reference user is obtained, and the target value is determined based on the target behavior prediction model, the first historical behavior data and the second historical behavior data.

[0249] In other embodiments of the present application, the determining unit 43 is further configured to perform the following steps:

[0250] Determine a first vector based on a first feature value of each target feature of the user to be processed;

[0251] Determining a plurality of second vectors based on the second feature value of each target feature of each to-be-screened user;

[0252] determining a Mahalanobis distance between the first vector and each second vector;

[0253] Based on the target nearest neighbor algorithm and multiple Mahalanobis distances, a reference user is determined from multiple users to be screened.

[0254] In other embodiments of the present application, the determining unit 43 is further configured to perform the following steps:

[0255] The target behavior prediction model is used to process the first historical behavior data and the second historical behavior data respectively to obtain the first target behavior data of the user to be processed and the second target behavior data of the reference user;

[0256] Based on the first target behavior data and the second target behavior data, a target value of the user to be processed is determined.

[0257] In other embodiments of the present application, the determining unit 43 is further configured to perform the following steps:

[0258] Determine a second parameter based on the first historical behavior data and the behavior data; wherein the second parameter represents a behavior change of the user to be processed;

[0259] If the second parameter is greater than or equal to the target threshold, the target behavior prediction model is updated based on the behavior data.

[0260] In other embodiments of the present application, the determining unit 43 is further configured to perform the following steps:

[0261] Obtain historical asset data of assets owned by the user to be processed;

[0262] Determine a third parameter based on the asset data and the historical asset data; wherein the third parameter represents the value fluctuation of the asset;

[0263] The weight is adjusted based on the second parameter, the third parameter and the target adjustment coefficient.

[0264] It should be noted that the specific description of the steps performed by each unit can be found in Figure 1 and 2 The description of the information determination method provided in the corresponding embodiment will not be repeated here.

[0265] The information determination device provided in the embodiment of the present application can determine the target feature from multiple first features based on the weight of each first feature of the user to be processed, instead of directly determining the target feature through a target group index model as in the related art, which ignores the importance of each first feature, thereby improving the accuracy of determining the target feature; and by combining the target behavior prediction model, target features and historical behavior data, the target value of the user to be processed can be accurately determined, instead of just using a simple comparison method to determine the value of the target user as in the related art. In this way, the problem of inaccurate evaluation results of user value in the process of evaluating user value in the related art is solved.

[0266] Based on the foregoing embodiments, the embodiments of the present application provide a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement Figure 1 and 2 The corresponding embodiments provide information to determine the steps in the method.

[0267] Based on the above embodiments, the embodiments of the present application provide a computer program product, the computer program product includes a computer program, the computer program is executed by the processor 31 to implement Figure 1 and 2 The corresponding embodiments provide steps of the information processing method.

[0268] It should be noted that the computer-readable storage medium may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (Flash Memory), or a hard disk.

[0269] Memory), magnetic surface storage, optical disk, or Compact Disc Read-Only

[0270] Memory, CD-ROM) and other memories; it can also be various electronic devices including one or any combination of the above memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0271] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0272] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0273] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course, by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0274] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0275] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0276] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0277] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for determining information, characterized in that: The method comprises: Acquire asset data of assets owned by the user to be processed and behavior data related to the transaction behavior of the user to be processed at the current moment; Determine a first feature of the user to be processed based on the asset data and the behavior data, and determine a target feature based on the first feature and a weight of each first feature; wherein the first feature is a feature associated with the asset data and the behavior data of the user to be processed; The first historical behavior data of the user to be processed is obtained, and the target value of the user to be processed is determined based on the target feature, the target behavior prediction model and the first historical behavior data.

2. The method according to claim 1, characterized in that The determining the first feature of the user to be processed based on the asset data and the behavior data, and determining the target feature based on the first feature and the weight of each first feature, includes: Processing the asset data and the behavior data using a target feature extraction algorithm to obtain a first feature of the user to be processed; Based on the first feature of the user to be processed, the second feature of the benchmark user, the first number of the benchmark users and the second number of the users to be processed, determining a first parameter corresponding to each first feature; wherein the first parameter represents the degree of attention paid to the first feature; The target feature is determined from the plurality of first features based on the plurality of first parameters and the weight of each of the first features.

3. The method according to claim 1, characterized in that Determining the target value of the user to be processed based on the target feature, the target behavior prediction model and the first historical behavior data includes: Determine a reference user from a plurality of users to be screened based on the target feature and the target nearest neighbor algorithm; The second historical behavior data of the reference user is obtained, and the target value is determined based on the target behavior prediction model, the first historical behavior data and the second historical behavior data.

4. The method according to claim 3, characterized in that The step of determining a reference user from a plurality of users to be screened based on the target feature and the target nearest neighbor algorithm includes: Determining a first vector based on a first feature value of each target feature of the user to be processed; Determining a plurality of second vectors based on the second feature value of each target feature of each to-be-screened user; determining a Mahalanobis distance between the first vector and each second vector; Based on the target nearest neighbor algorithm and the plurality of Mahalanobis distances, the reference user is determined from the plurality of users to be screened.

5. The method according to claim 3, characterized in that: Determining the target value based on the target behavior prediction model, the first historical behavior data, and the second historical behavior data includes: The target behavior prediction model is used to process the first historical behavior data and the second historical behavior data respectively to obtain the first target behavior data of the user to be processed and the second target behavior data of the reference user; Based on the first target behavior data and the second target behavior data, the target value of the user to be processed is determined.

6. The method according to claim 1, characterized in that The method further comprises: Determine a second parameter based on the first historical behavior data and the behavior data; wherein the second parameter represents a behavior change of the user to be processed; If the second parameter is greater than or equal to the target threshold, the target behavior prediction model is updated based on the behavior data.

7. The method according to claim 6, characterized in that The method further comprises: Acquire historical asset data of assets owned by the user to be processed; Determine a third parameter based on the asset data and the historical asset data; wherein the third parameter represents the value fluctuation of the asset; The weight is adjusted based on the second parameter, the third parameter and a target adjustment coefficient.

8. An information determination device, characterized in that The device comprises: a processor, a memory and a communication bus; The communication bus is used to realize the communication connection between the processor and the memory; The processor is used to execute the information determination program stored in the memory to implement the steps of the information determination method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The storage medium stores one or more programs, and the one or more programs can be executed by one or more determiners to implement the steps of the information determination method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a determiner, the computer program implements the steps of the information determination method according to any one of claims 1 to 7.

Images