Decision-making auxiliary method and device based on portrait data, equipment and medium

Abstract

The invention relates to the technical field of data processing, and discloses a decision-making auxiliary method, device and equipment based on portrait data and a medium, which can be applied to financial and medical scenarios, and the method comprises the following steps: processing an original data set to obtain standard data; respectively constructing a basic attribute portrait, a service behavior portrait and a consumption behavior portrait based on the standard data to extract multi-dimensional feature information, and converting the multi-dimensional feature information into a customer portrait vector; identifying customer behaviors according to the customer portrait vector, and constructing a portrait dynamic updating mechanism to obtain dynamic portrait data; and constructing a decision-making auxiliary model by using the dynamic portrait data, and operating the decision-making auxiliary model to output a decision-making auxiliary scheme. In the invention, aiming at the problem of unreasonable resource allocation caused by incomplete client portrait construction, the decision-making auxiliary model can be constructed and operated by utilizing the calculated dynamic portrait data, so as to output the decision-making auxiliary scheme. Therefore, resources can be reasonably distributed, and the working efficiency of personnel is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a decision-making assistance method, device, equipment and medium based on portrait data. Background Art

[0002] In the actual operation of customer service systems, service departments in fields such as finance and healthcare are increasingly accumulating vast amounts of data resources generated by interactions with customers, such as consultation records, complaint feedback, service satisfaction surveys, transaction logs, and historical medical records. However, existing technologies often use static analysis to process various data sources separately, lacking unified data integration mechanisms and dynamic modeling capabilities. This makes it difficult to achieve deep integration and real-time updates of multi-source heterogeneous data, resulting in fragmented customer information and difficulty unlocking the value of this data.

[0003] In this context, customer service departments are unable to accurately capture customers' behavioral preferences, risk profiles, and service needs at different stages of their lifecycles. This results in incomplete and inaccurate customer profiles, hindering the scientific and timely nature of key decisions such as resource allocation, personalized service strategy development, and customer relationship maintenance. In the financial sector, for example, the lack of customer profiles can lead to delayed identification of high-risk customers, potentially leading to defaults or wasted marketing resources. In healthcare, the lack of dynamic profiles can lead to unclear follow-up consultation paths and decreased service satisfaction. Summary of the Invention

[0004] The present invention provides a decision-making assistance method, device, equipment and medium based on portrait data to solve the technical problem in the prior art that customer portrait construction is incomplete and leads to unreasonable resource allocation.

[0005] First, a decision support method based on portrait data is provided, including:

[0006] Acquire customer data and integrate the customer data to obtain an original data set;

[0007] Cleaning and preprocessing the original data set, and performing standardization processing to obtain standard data;

[0008] Based on the standard data, a basic attribute profile, a service behavior profile, and a consumption behavior profile are respectively constructed to extract multi-dimensional feature information, and the multi-dimensional feature information is converted into a customer profile vector;

[0009] Identify customer behavior based on the customer portrait vector, and build a dynamic portrait update mechanism to obtain dynamic portrait data;

[0010] A decision support model is constructed using the dynamic portrait data, and the decision support model is run to output a decision support solution.

[0011] In a second aspect, a decision support device based on portrait data is provided, comprising:

[0012] A data acquisition module, used to acquire customer data and integrate the customer data to obtain an original data set;

[0013] A data processing module is used to clean and preprocess the original data set and perform standardization to obtain standard data;

[0014] A data extraction module is used to construct a basic attribute profile, a service behavior profile, and a consumption behavior profile based on the standard data to extract multi-dimensional feature information, and convert the multi-dimensional feature information into a customer profile vector;

[0015] A data update module is used to identify customer behavior based on the customer portrait vector and build a dynamic portrait update mechanism to obtain dynamic portrait data;

[0016] The data output module is used to construct a decision support model using the dynamic portrait data, and run the decision support model to output a decision support solution.

[0017] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned decision-making assistance method based on portrait data when executing the computer program.

[0018] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned decision-making assistance method based on portrait data are implemented.

[0019] The above-mentioned solution implemented by the decision-making assistance method, device, equipment and medium based on portrait data can be applied to financial and medical scenarios. The method includes obtaining customer data and integrating the customer data to obtain an original data set; cleaning and preprocessing the original data set, and standardizing it to obtain standard data; constructing basic attribute portraits, service behavior portraits and consumption behavior portraits based on the standard data to extract multi-dimensional feature information, and converting the multi-dimensional feature information into a customer portrait vector; identifying customer behavior according to the customer portrait vector, and constructing a dynamic portrait update mechanism to obtain dynamic portrait data; using the dynamic portrait data to construct a decision-making assistance model, and running the decision-making assistance model to output a decision-making assistance solution. In the present invention, in order to address the problem of incomplete customer portrait construction leading to unreasonable resource allocation, the calculated dynamic portrait data can be used to construct a decision-making assistance model and run it to output a decision-making assistance solution. In this way, resources can be reasonably allocated and the work efficiency of personnel can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0021] Figure 1 This is a flow chart of a decision support method based on portrait data in one embodiment of the present invention;

[0022] Figure 2 yes Figure 1 A specific implementation process diagram of step S30 Figure 1 ;

[0023] Figure 3 yes Figure 1 A specific implementation process diagram of step S30 Figure 2 ;

[0024] Figure 4 yes Figure 1 A specific implementation process diagram of step S30 Figure 3 ;

[0025] Figure 5 yes Figure 1 A schematic flow chart of a specific implementation of step S40;

[0026] Figure 6 yes Figure 1 A schematic flow chart of a specific implementation of step S50;

[0027] Figure 7 This is another flowchart of a decision support method based on portrait data in one embodiment of the present invention;

[0028] Figure 8 This is a schematic structural diagram of a decision support device based on portrait data in one embodiment of the present invention;

[0029] Figure 9 is a structural diagram of a computer device in one embodiment of the present invention;

[0030] Figure 10 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0032] See also Figure 1 As shown, Figure 1 A flowchart of a decision support method based on portrait data provided by an embodiment of the present invention includes the following steps:

[0033] S10: Obtain customer data and integrate the customer data to obtain an original data set.

[0034] Build a multi-source heterogeneous data collection architecture and configure data access channels covering financial and medical service scenarios. For multiple data sources such as customer service hotline call records, online customer service chat records, customer complaint tickets, service satisfaction survey results, customer purchase history, and customer basic information databases, configure voice recognition interfaces, API crawling interfaces, database query interfaces, and file import interfaces to achieve comprehensive collection of structured and unstructured data.

[0035] For voice data, such as recordings of financial customers' telephone consultations or remote medical consultations, voice recognition technology is used to transcribe them into text and uniformly encode them into a standard format. For structured data, such as customer account transaction records or medical history information, they are extracted according to field specifications through the database query interface and mapped into a unified data model. A data collection strategy that combines active and passive data can also be adopted. Passive collection includes automatically capturing updated data from various business systems at preset time intervals. Active collection includes proactively pushing questionnaires, forms, and other information collection requests to customers through SMS, App pop-ups, or IVR voice channels in specific business scenarios (such as the release of new financial products, vaccination launches, etc.).

[0036] Perform data fusion processing on the collected data. By building a unified data standard system and field mapping relationship library, complete the semantic alignment and type specification conversion of data fields between different systems, and reduce the data silos between financial business systems and medical information systems.

[0037] To ensure the security and traceability of the data collection process, blockchain technology can be introduced to record the collection logs on-chain. The records include the source identification of the data item, the collection timestamp, the collection method, and the responsible node information to prevent the tampering of key data during transmission or storage. This is particularly suitable for data scenarios involving customer privacy and compliance supervision.

[0038] Ultimately, multi-source, multi-format, and multi-timeliness customer data is integrated to form an original data set for the entire customer life cycle, providing a high-quality, highly complete data foundation for subsequent cleaning preprocessing, portrait construction, and decision modeling.

[0039] S20: Cleaning and preprocessing the original data set, and performing standardization processing to obtain standard data.

[0040] For different types of data in the original data set, a multi-dimensional cleaning process is constructed. First, noise data is identified and eliminated, including garbled text, advertising information, invalid symbols and other non-business related content; at the same time, duplicate records are detected and removed, and abnormal data with serious omissions or format errors are eliminated through field rule verification to ensure the integrity of the basic structure of the original data. For text data, especially financial consulting content in the financial industry and medical consultation records in the medical industry, natural language processing technology is used for word segmentation, part-of-speech tagging and stop word filtering to construct a text corpus with clear semantics and standardized structure; for data containing professional terms (such as insurance terms, disease names, etc.), domain dictionaries are introduced to identify proprietary words and normalize them to ensure terminology uniformity and semantic consistency. For numerical data, a unified dimension conversion operation is performed, and indicators such as account transaction amounts, medical expenses, and satisfaction scores are standardized and normalized to ensure that data from different sources and different magnitudes are comparable, which facilitates unified processing of subsequent models;

[0041] A deep learning-based abnormal data detection mechanism can be introduced, using a generative adversarial network (GAN) model to learn customer behavior patterns, automatically identifying abnormal financial behavior characteristics (such as abnormally high-frequency trading in a short period of time) or data anomalies in the medical system (such as extreme diagnosis and treatment scores), labeling the detected abnormal samples, and visualizing them for manual review. A data quality monitoring system can also be established to evaluate core indicators such as the completeness, accuracy, and consistency of cleaned data in real time. If any indicator falls below the set threshold (such as a field missing rate >3% or a timestamp anomaly ratio >5%), the data backtracking process is automatically triggered to re-collect or complete the missing fields to ensure that the data received by the downstream processing module is stable and reliable.

[0042] During the data processing process, a distributed parallel computing framework is built, and the Apache Spark platform can be used to shard large-scale data for processing, thereby improving processing efficiency. This framework is suitable for processing millions of transaction logs in the financial industry and for the simultaneous cleaning of medical records from multiple institutions in the medical industry. It ultimately outputs standard data with a unified structure and controllable quality, providing a data foundation for customer portrait modeling and trend forecasting.

[0043] S30: Based on the standard data, a basic attribute portrait, a service behavior portrait and a consumption behavior portrait are respectively constructed to extract multi-dimensional feature information, and the multi-dimensional feature information is converted into a customer portrait vector.

[0044] Standard data can be extracted to include fields such as customer name, age, gender, region, occupation, income level, and health status. In the financial sector, this can include risk level and account type, while in the medical sector, this can include disease label and frequency of medical visits. For missing fields, inference algorithms based on association rules can be used to fill in the missing fields. For example, using spending power and product preferences to infer income level, or medical expenditure frequency to infer disease risk level. After classification and numerical coding, all attributes are uniformly converted into basic attribute profiles.

[0045] Furthermore, based on service interaction data such as call logs, customer service chat logs, complaint tickets, and satisfaction surveys, behavioral characteristics such as consultation frequency, problem type distribution, complaint type and resolution time, and satisfaction ratings can be extracted. Sequence modeling and clustering methods can be used to identify high-frequency behavioral paths and unusual interaction patterns. In financial scenarios, this can identify the path from high-frequency financial consultations to complaints to declining satisfaction, and in medical scenarios, the path from multiple unresolved visits to negative feedback. Behavioral labels are mapped to discrete events to form a service behavior profile.

[0046] In financial scenarios, we extract information from wealth management product purchase records, including purchase frequency, average transaction amount, channel preferences (e.g., online banking, apps, self-service terminals), and brand loyalty (e.g., the proportion of customers holding a particular financial brand's products long-term). In healthcare scenarios, we extract information such as number of appointments, drug categories, payment amounts, medical insurance usage, and preferred hospital departments. Using dynamic window analysis and survival models, we calculate customer spending cycles, spending trends, and brand preference stability, ultimately creating a consumer behavior profile.

[0047] The above three types of portraits are subjected to feature extraction. After extracting multi-dimensional feature information, it is fused and processed and converted into customer portrait vectors, which serve as the core input data for subsequent modeling and decision-making.

[0048] Combine Figure 2 As shown, the basic attribute portrait, service behavior portrait and consumption behavior portrait are constructed based on the standard data to extract multi-dimensional feature information. The specific steps include:

[0049] S311: extracting basic attributes of customers from the standard data to construct an initial attribute set.

[0050] Extract fields containing basic identity information and static features from standardized customer data to construct an initial set of customer attributes. Extracted attributes include: customer name, gender, age, region, occupation, income level, account registration channel, and initial service time. For financial services, further attributes such as customer risk appetite, credit score category, and asset size can be extracted. For medical services, health attribute indicators such as visit frequency, chronic disease history, and medical insurance type can be added. This initial set of attributes provides raw data support for the subsequent construction of a basic customer profile.

[0051] S312: Perform prediction and filling processing on the missing data in the initial attribute set based on a data mining algorithm to obtain customer filling attributes.

[0052] For fields with missing values ​​in the initial attribute set, data mining algorithms based on association rules and statistical learning can be applied for automatic filling, and customer filling attributes can be obtained. The filling logic includes: in the financial scenario, if the customer has not clearly declared his income level, the income level can be inferred based on the amount of wealth management product purchases, transaction frequency and product risk level in the past six months, combined with the characteristics of other similar customer groups, using a rule model built based on the Apriori algorithm. In the medical scenario, if the customer's occupation or health status is missing, the customer's registration department, historical prescription category, and interval time between visits can be used to perform feature association analysis to identify the population characteristic category to which he belongs, and complete the filling of occupational type and chronic disease risk level. The filling process is controlled by the confidence threshold to ensure the rationality and interpretability of the inference results.

[0053] S313: Classify and encode the customer-filled attributes, and convert the obtained classification encoding results into recognizable feature vectors.

[0054] After missing data is filled in, the customer attributes are subjected to a standardized conversion process. Discrete attributes (such as gender, region, occupation, and health insurance type) are discretized using one-hot encoding or label encoding. For ordinal attributes (such as income level, credit score, and health risk classification), piecewise normalization is used to ensure the relative magnitude of the numerical representation. All encoded attributes are arranged in a predefined order and dimensionalized to construct a recognizable feature vector representation.

[0055] S314: Constructing a basic attribute profile based on the feature vector to support customer behavior identification and dynamic updating of the profile.

[0056] The generated feature vector is persistently stored in the portrait database and marked as the basic attribute portrait of the customer. As the static feature part of the multimodal portrait system, this portrait is used to collaborate with the service behavior portrait and the consumption behavior portrait to build a holographic portrait of the customer. Later, in the customer behavior identification stage, customers can be grouped according to the basic attribute portrait. For example, in financial scenarios, it can be used to accurately identify high-net-worth customers and high-credit customer groups, and then formulate differentiated product recommendations and risk management strategies; in medical scenarios, it can be used to identify key groups, the elderly and other high-attention objects, optimize service language style, medical route planning and follow-up arrangements. The portrait also supports a dynamic portrait update mechanism. When the basic attributes of the customer change (such as an increase in income level or a change in medical insurance type), it can automatically trigger feature updates and portrait reconstruction to ensure that the portrait data continues to reflect the customer's true status.

[0057] Combine Figure 3 As shown, the basic attribute portrait, service behavior portrait and consumption behavior portrait are constructed based on the standard data to extract multi-dimensional feature information. The specific steps also include:

[0058] S321: Extracting service behavior features from the data related to customer service behaviors in the standard data to obtain a service behavior feature set.

[0059] Customer service-related data, including customer service hotline call logs, online customer service chat logs, complaint ticket information, and service satisfaction survey results, is extracted from standard data. Through data preprocessing, core behavioral indicators such as customer consultation frequency, distribution of consultation question types, complaint frequency, complaint question types, complaint resolution time, and service satisfaction scores are extracted. For example, in financial services scenarios, the frequency of customer inquiries regarding financial product terms and conditions, complaints regarding insurance policy processing timelines, and feedback scores are extracted. In medical services scenarios, information such as customer registration consultations, wait times for appointments, and satisfaction with medical services is extracted to form a structured set of service behavior characteristics.

[0060] S322: Constructing a time series sequence based on the service behavior feature set, and sorting the service behavior events corresponding to the preset customer identifiers by timestamp according to the time series sequence to obtain original sequence data.

[0061] Based on the service behavior feature set, a time series sequence is constructed. All service behavior events are aggregated according to the customer ID and sorted in ascending timestamp order to construct customer-level raw sequence data. For example, a customer's service interaction data might form the following sequence: {Auto Insurance Terms Consultation @ 2023-07-01, Claim Complaint @ 2023-07-05, Complaint Handling Time 3 Days @ 2023-07-08, Satisfaction Rating 85\ @ 2023-07-10}. This sequence reflects the customer's complete service journey, from consultation to complaint to evaluation, and is suitable for subsequent behavior modeling and analysis.

[0062] S323: Perform feature label mapping processing on the original sequence data, and discretize events related to customer service behaviors into event labels to obtain labeled behavior sequences.

[0063] The raw sequence data is mapped to feature labels, and customer service behaviors are discretized into standardized event labels based on pre-set rules. For example, in the financial industry, "auto insurance policy consultation" is labeled "CQ01," "claim settlement time complaint" is labeled "CP03," "2-3 days complaint resolution time" is classified as "T2," and "80-90" satisfaction scores are classified as "S4." In the healthcare industry, "registration consultation" can be mapped to "MQ01," and "drug shortage complaint" can be labeled "MP02." This process constructs structured, parseable, and labeled behavior sequences, providing semantically consistent input for subsequent behavior mining.

[0064] S324: Mining the behavior sequence using a sequence mining algorithm to obtain a customer behavior pattern.

[0065] The PrefixSpan sequential pattern mining algorithm can be used to identify frequent subsequence patterns in customer behavior tag sequences. By combining time window constraints (e.g., intervals between adjacent behaviors ≤ 30 days) with a minimum support setting (e.g., 5%), typical customer service paths can be identified. For example, a frequently occurring behavioral pattern among financial customers is identified as: "<Product Inquiry → Terms Inquiry → Complaint>," while a common pattern among medical customers is: "<Registration → Consultation → Follow-up Complaint → Low-Rating Feedback>." By constructing a projection database and setting time-decay weights, high-value, behaviorally meaningful customer behavior patterns can be effectively discovered.

[0066] S325: Perform risk assessment and behavioral intention identification on the customer behavior pattern, and output a service risk index.

[0067] For each frequent customer behavior pattern, a risk assessment model is constructed based on variables such as complaint duration and satisfaction score. For example, for the pattern "\<CQ01→CP03→S2> ", using the weighted formula R = 0.6 × T3 weight + 0.4 × (1-S2 weight) to calculate the risk index, identifying potential risk signals such as service quality issues and declining customer satisfaction. At the same time, combined with behavioral intention recognition logic, it determines the motivations of customer behavior, such as whether the complaint stems from a misunderstanding of the terms or a delayed service response, and ultimately outputs the service risk index.

[0068] S326: Based on the service risk index and using business strategy rules, a service behavior profile is constructed to express the customer's behavioral tendencies, response patterns, and potential demands during the service process.

[0069] Service behavior profiles are used to express customers' behavioral habits (such as a tendency to repeat consultations), emotional tendencies (such as susceptibility to complaints), service response sensitivity (such as how quickly ratings change), and potential service demands (such as preference for self-service or manual service). In the financial sector, this can be used to identify high-risk customer groups and adjust service priorities; in medical scenarios, users with multiple low ratings can be matched with experienced doctors or receive early intervention services to improve satisfaction and treatment efficiency. The resulting service behavior profile will serve as an important component of the customer profile system, continuously participating in dynamic profile updates and input into intelligent decision-making models.

[0070] Combine Figure 4 As shown, the basic attribute portrait, service behavior portrait and consumption behavior portrait are constructed based on the standard data to extract multi-dimensional feature information. The specific steps also include:

[0071] S331: extracting customer purchase data from the standard data, and preprocessing the customer purchase data to obtain a purchase behavior dataset.

[0072] Extract original records related to customer purchasing behavior from standard datasets, including purchase orders from e-commerce platforms, offline store transaction records, official website self-service order data, and financial product purchase records from financial systems. In medical service scenarios, purchase data may include registration payment records, drug purchase orders, medical insurance settlement details, etc. To solve the problem of asynchronous timestamps across different channels, a dynamic time warping algorithm (DTW) is used to align time information; for data with large price differences across different product categories, a Box-Cox transformation can be used to normalize the amount; a hierarchical labeling system is established for purchase channels to achieve standardization of channel coding for primary (online / offline) and secondary (official website / app / third-party platform), ultimately constructing a purchase behavior dataset with a unified structure and consistent format.

[0073] S332: Based on the purchase behavior dataset, construct a customer-product bipartite graph and a time series feature storage architecture to extract consumption behavior features.

[0074] A customer-product bipartite graph is constructed based on customer and product IDs. Node attributes in the graph include basic indicators such as purchase frequency, amount, and purchase interval. Edges represent the purchase relationship between customers and products, and edge weights can be weighted by amount or frequency. On this basis, a time series feature storage system (e.g., based on Apache Pinot) is constructed to support time window queries, enabling millisecond-level access queries on customer behavior. Consumer behavior feature extraction includes: detecting the main frequency of the cycle through FFT, adaptively setting the observation window, and calculating the frequency index; inputting the channel purchase frequency vector and calculating the channel preference index through a multi-channel attention model; introducing survival analysis modeling to calculate the brand switching probability and derive the loyalty index, combining the time decay factor to strengthen the weight of recent behavior; and using the Poisson-Gamma model to identify low-frequency / high-frequency customers whose consumption behavior deviates from the normal distribution.

[0075] S333: Construct a cross-modal feature matrix based on the consumption behavior characteristics to obtain cross-modal behavior data.

[0076] By integrating consumer behavior characteristics with service behavior characteristics (such as consultation frequency, complaint types, and service satisfaction), a cross-modal feature matrix was constructed. The matrix was indexed by customer ID, and the column dimension covered heterogeneous data such as consultation behavior, purchase behavior, and brand preference. To explore the potential relationships between the features, a canonical correlation analysis (CCA) was further conducted. For example, it was found that consultation frequency was positively correlated with purchase frequency (ρ = 0.32, p < 0.01), while claims complaints were negatively correlated with purchases of high-value products (ρ = -0.27).

[0077] In addition, key consumption events (such as "high-priced product purchase" and "frequent purchase transfer") are embedded in the service behavior sequence, and the semantics of behavior labels are expanded to form cross-modal behavior data with time series characteristics, which serves as the input basis for trend prediction.

[0078] S334: Utilize the cross-modal behavior data to build a consumption trend prediction model, perform consumption time series modeling, and output trend prediction results.

[0079] The ARIMA model can be used to model customer consumption behavior time series. Specifically, the following methods are used: using the customer ID as the primary key, aggregating daily purchase amounts, number of consultations, and number of complaints to form a multivariate time series; applying STL decomposition to seasonal interpolation for missing values; assessing series stationarity using the ADF test and KPSS test, and selecting the optimal (p, d, q) parameter combination through EACF grid search and the AICc criterion; integrating exogenous variables into the ARIMA model framework to improve forecasting accuracy; using a rolling window approach, conducting a joint forecast of consumption frequency, amount, and channel shifts every seven days to support trend warnings and identify evolving customer behavior; and using Granger causality tests to identify important factors influencing forecast results, such as "consultation frequency significantly affects purchase amount after a 3-day lag" (p < 0.01), ensuring the explanatory power of the forecast model. Ultimately, trend forecast results are obtained.

[0080] S335: Generate a consumer behavior portrait for subsequent dynamic updates and decision-making assistance calls based on the trend prediction results.

[0081] The trend prediction results are integrated with the current consumer behavior characteristics to generate a consumer behavior portrait that includes static attributes (such as purchase frequency, channel preference, and loyalty) and dynamic trends (such as amount growth rate and changes in behavior cycle). The consumer portrait data structure includes: current state feature vector (number of purchases, average amount, etc.), trend label (growth, decline, fluctuation), prediction results (consumption probability in the next 7 days, expected amount, etc.), risk identification and recommended actions (such as high-potential customers, push upgrade packages). Ultimately, the consumer behavior portrait is synchronously written into the customer portrait database and serves as the input basis for the dynamic portrait update mechanism and decision-making assistance module. It is used for intelligent marketing triggering and quota adjustment in financial scenarios, and for matching health product push and follow-up reminder strategies in medical scenarios.

[0082] S40: Identify customer behavior based on the customer portrait vector, and build a dynamic portrait update mechanism to obtain dynamic portrait data.

[0083] Sliding window analysis and threshold determination are performed on time-sensitive indicators in customer profile vectors (such as consultation frequency, purchase frequency, and complaint resolution cycle) to identify customer behavior change events, such as increased risk, decreased satisfaction, and brand switching. In financial services, this can identify, for example, a shift from high-frequency trading to account stagnation; in healthcare, it can identify sudden increases in visits or department changes. An event-driven update engine can be introduced, connected to a message queue system such as Kafka, to trigger the profile update process in real time for each customer event (such as a new transaction, new complaint, or new visit). Incremental update strategies can also be used to achieve self-adjustment of the profile. Online learning models use new data to adjust parameters (such as weights for financial product purchase preferences or satisfaction rating models), transitioning customer profiles from static representation to dynamic evolution (i.e., a dynamic profile update mechanism). The final dynamic profile output includes the updated feature vector as well as additional metadata such as behavioral trend labels and profile evolution logs, ensuring that the customer profile always reflects its current state.

[0084] Combine Figure 5 As shown, the customer behavior is identified based on the customer portrait vector, and a dynamic portrait update mechanism is constructed to obtain dynamic portrait data. The specific steps include:

[0085] S41: Construct a real-time feature calculation layer based on the customer portrait vector, and incrementally update various time-sensitive features to obtain updated feature values.

[0086] Based on the existing customer portrait vectors, a real-time feature calculation layer with low-latency processing capabilities is constructed to monitor and incrementally update time-sensitive dynamic features such as purchase frequency, consultation frequency, and complaint frequency to obtain updated feature values. For example, a sliding window mechanism can be adopted, using a ring buffer structure to store feature event window data, maintaining only the head and tail pointers to reduce memory access and computing costs. In addition, for indicators that are easily affected by time fluctuations, such as service response time and consultation activity, an exponential decay mechanism can be introduced for time-weighted processing, which is suitable for the dynamic trade-off between behavioral patterns such as financial transaction activity and medical follow-up frequency.

[0087] S42: Synchronously modify the associated items of the updated feature value to obtain updated features.

[0088] Based on the correlation dependency between the updated feature values, a composite feature update strategy is constructed to ensure that local feature changes can drive the consistency update of the overall profile content to obtain updated features. For example, in the insurance business, when the time it takes to handle a customer complaint changes, the customer satisfaction index (CSI) should be revised at the same time. Here, the system implements mapping corrections through the structural equation model (SEM) coefficients. In the medical scenario, if the prescription drugs recently purchased by the user change, the corresponding preference profile, the possibility of follow-up visits, and the health risk level can be revised simultaneously. In addition, for behavioral sequence features, a skip table structure can be used to maintain the event sequence relationship, realize time-complex data insertion and adjustment, and adapt to the online processing needs of high-frequency customer interactions.

[0089] S43: Inputting the updated features into a learning model based on a stochastic gradient descent algorithm for prediction and identification to obtain a customer behavior prediction result.

[0090] The updated eigenvalues ​​are fed into the system's built-in online learning model, which uses the stochastic gradient descent (SGD) algorithm for continuous training and inference to generate customer behavior predictions. For time series models (such as ARIMA), the incremental QR decomposition mechanism can be used to rapidly update the covariance matrix, maintaining the effectiveness of predictive models for short-term purchasing trends and medical drug risk assessments.

[0091] S44: Build a portrait update engine and update the customer behavior prediction results in real time to build a dynamic portrait update mechanism.

[0092] Build an event-driven profile update engine and implement partitioned streaming access to customer behavior events through Kafka. Consumer nodes perform hash routing by customer ID to ensure that behavioral data is processed in order. State management can use Redis to store customer profile vectors, combined with the WAL log mechanism to ensure that the update process has ACID properties. In the update logic, when a new behavioral event (such as a purchase, complaint, or consultation) occurs, the behavioral content is immediately parsed and the following operations are performed to build a dynamic profile update mechanism: update feature values ​​(such as purchase frequency and loyalty), call the predictive model interface in real time (such as the probability of purchase in the next 30 days), and write the updated customer profile vector to the HBase columnar database for subsequent module calls.

[0093] S45: Perform fault-tolerant detection according to the dynamic portrait update mechanism. When it is determined that the key features have drifted, trigger the portrait version rollback operation to obtain dynamic portrait data.

[0094] To ensure operational stability and accuracy, a feature drift detection mechanism can be deployed to monitor fluctuations in key features daily. Drift thresholds are set for consumer behavior features (such as amount and frequency) to detect outliers, and relative rates of change are calculated for model explanatory metrics (such as SHAP values). A version management mechanism is also introduced, using a two-phase commit protocol and a checkpoint mechanism to save snapshots of the profile status to HDFS every 5 minutes. When drastic fluctuations in model prediction error or unusual changes in customer behavior are detected (such as when triggered by the Page-Hinkley test), the system can roll back to the most recent stable version to obtain dynamic profile data, ensuring the business availability of the profile data and the sustainability of the prediction model.

[0095] S50: constructing a decision support model using the dynamic portrait data, and running the decision support model to output a decision support solution.

[0096] A decision-support model is built, running independently within a microservices architecture and outputting corresponding decision-support solutions to provide guidance to personnel. In financial scenarios, this can identify high-value users nearing churn in real time and automatically assign them dedicated consultants. In healthcare, it can predict a user's return risk and schedule follow-up appointments in advance. The continuous operation of this model assists customer service departments in making dynamic, accurate, and efficient decisions under complex business conditions.

[0097] Combine Figure 6 As shown, the method of constructing a decision support model using the dynamic portrait data and running the decision support model to output a decision support solution includes the following specific steps:

[0098] S51: Extract key feature parameters from the dynamic portrait data as an input feature set for decision assistance.

[0099] Extract decision-related core features from real-time updated dynamic portrait data to construct the input feature set. The core features include:

[0100] Customer value characteristics: spending amount, purchase frequency, brand loyalty, credit rating, account contribution cycle, etc.

[0101] Service demand characteristics: consultation frequency, complaint frequency, complaint level, service satisfaction score, etc.;

[0102] Behavioral trend features: purchase probability in the next 7 days, predicted probability of follow-up visits, behavioral path risk index, etc.

[0103] In financial scenarios, feature sets can reflect changes in customer investment preferences; in medical scenarios, they can indicate the urgency of user service requests and their willingness to participate in health management. These feature sets are standardized and fed into various decision-making models as input feature sets.

[0104] S52: Constructing a service resource allocation model based on the input feature set and calculating the customer priority score to output an optimal resource allocation plan and a service channel recommendation strategy.

[0105] By building a service resource allocation model, a hierarchical weighted scoring method can be used to prioritize customers and generate output. These outputs include resource allocation plans, assigning experienced customer service representatives, doctors, or dedicated consultants to high-priority customers; and service channel recommendation strategies, identifying customers who require VIP service channels, such as dedicated telephone lines, video consultations, and secure medical access. In the financial sector, for example, if a customer's credit is fluctuating but their trading activity is high, customer service can be immediately deployed. In the healthcare sector, if a customer has low satisfaction and a history of repeated visits, the system assigns them to a dedicated medical coordinator queue.

[0106] S53: Constructing a service strategy formulation model based on the service resource allocation model to push product information and activity information.

[0107] Further build a service strategy formulation model. Based on the customer priority calculation results, combined with the basic attributes of the customer profile (such as age, occupation, region), consumption behavior characteristics and service preference information, formulate personalized service strategies. The output of the strategy includes: pushing innovative financial products, health management services or personalized protection packages to young high-spending customers; for elderly customers, recommending a telephone appointment + manual service model in medical scenarios; recommending door-to-door service by financial advisors or customized investment reminders in financial scenarios; combining customer profile preferences and behavior prediction results, proactively push event coupons, exclusive physical examination invitations, product trial links, etc. to enhance customer engagement. The service strategy formulation model combines LightGBM and rule engine to achieve a hybrid implementation. It can be compatible with real-time online reasoning and manual rule interpretation to ensure the efficiency and compliance of model application.

[0108] S54: Using the dynamic portrait data to identify potential lost customers and high-value customer groups, so as to build a customer relationship maintenance model.

[0109] Based on the portrait trend indicators in dynamic portrait data (such as the rate of decline in consumption amount, continuous decline in service satisfaction, and predicted activity decay), we can identify lost customers and build a high-value customer identification mechanism in combination with the customer lifetime value (CLV). The specific process of building a customer relationship maintenance model is as follows: If the purchase frequency in the past three cycles has dropped by ≥30%, and the satisfaction score is less than 70 points, the customer is marked as a potential lost customer; if the contribution is stable for 6 consecutive months, the brand loyalty is higher than 0.8, and the credit rating is Class A, the customer is included in the high-value customer management plan. Differentiated maintenance strategies can also be formulated for different customer groups: for potential lost customers, customer service priority processing channels are automatically assigned, care reminders are sent, or limited-time discounts are issued; for high-value customers, a VIP service system is established to provide value-added services such as exclusive doctors, financial advisors, and one-on-one service representatives.

[0110] S55: respectively running the service resource allocation model, the service strategy formulation model and the customer relationship maintenance model, and outputting a comprehensive decision-making support solution.

[0111] The aforementioned service resource allocation model, service strategy development model, and customer relationship maintenance model are invoked and run in parallel to generate an integrated decision-making support solution. Outputs include: a customer service priority level and service resource allocation table, a list of recommended product packages and service strategies (including delivery channels and timing recommendations), and a customer retention plan (including early warning indicators, maintenance plans, and responsible person assignments). This decision-making support solution can be integrated into the business operations platform, providing visual display and manual intervention support through the decision-making support interface.

[0112] Combine Figure 7 As shown, the process of constructing a decision support model using the dynamic portrait data and running the decision support model to output a decision support solution includes:

[0113] S60: Construct a systematic evaluation process, conduct a multi-dimensional evaluation of the decision support solution, and obtain an evaluation result.

[0114] To build a systematic evaluation process for dynamic customer profiles, we can adopt a PDCA (Plan-Do-Check-Act) closed-loop mechanism to conduct comprehensive evaluations from three aspects: data quality, profile accuracy, and decision effectiveness:

[0115] In terms of data quality assessment, metadata management tools (such as Apache Atlas) are deployed to continuously monitor the missing rate of key fields, and business rule engines (such as Drools) are combined to implement logical verification, such as cross-field logical consistency verification of bank account transaction records in financial scenarios, and time sequence verification of medical records in medical settings.

[0116] In terms of consistency, data lineage analysis tools (such as Informatica CLAIRE) are introduced to identify data deviations that may be introduced during the ETL process;

[0117] In terms of timeliness, an SLA indicator system is built based on data pipelines such as Kafka, and a T+1 data synchronization limit is set (the default limit is no more than 15 minutes).

[0118] Furthermore, during the portrait accuracy verification phase, a test case matrix was developed for rule-driven portrait construction to ensure coverage across different risk levels and customer lifecycle stages (e.g., the application stage for financial customers, the recovery stage for medical users, etc.). For model-driven portraits, a KS value stability standard was set for the classification model (e.g., quarterly fluctuations not exceeding 0.2). Clustering models were evaluated for cluster clarity based on the silhouette coefficient (which must be maintained above 0.6). Real-time portraits were evaluated for input distribution drift based on the PSI (Population Stability Index).

[0119] During the decision-making effectiveness analysis phase, a visual mapping relationship between business strategies and actual results is constructed. For example, in the financial sector, approval strategies are linked to post-loan default rates, and customer segmentation strategies are linked to repurchase rates. In the medical sector, correlations are assessed between treatment path recommendation strategies, recovery cycles, and post-diagnosis follow-up rates. At the same time, the Shapley value method is introduced to explain the contributions of the model's key features, thereby supporting the rationality analysis of the strategy. In terms of economic benefit evaluation, by constructing a complete ROI calculation model, the compound interest effect needs to be considered in the financial sector, and the balance between medical insurance cost control and user satisfaction needs to be considered in medical scenarios. Ultimately, the evaluation results are obtained by integrating all the above results.

[0120] S70: Execute a modular optimization solution according to the evaluation result, and simultaneously perform structural enhancement on the decision support model to obtain an optimization strategy execution result.

[0121] After obtaining the evaluation results, the modular optimization mechanism is triggered based on the problems found in the evaluation, specifically including the following three aspects:

[0122] In terms of data layer optimization, collection strategies can implement differentiated sampling based on customer risk level. For example, in the financial industry, 100% transaction behavior collection is implemented for high-risk credit customers, and in the medical industry, comprehensive medical record collection is implemented for suspected rare diseases. At the same time, advanced data cleaning technologies are introduced, using generative adversarial networks (GANs) to repair abnormal data, and the STL (Seasonal and Trend Decomposition using Loess) method can be used to remove periodic noise from time series data.

[0123] In terms of image layer iteration, feature space optimization is performed. In the financial sector, user transaction records can be converted into multidimensional vector embeddings to identify credit behavior characteristics. In medical scenarios, user diagnosis and treatment path maps can be constructed, encoding multidimensional medical histories into vectors for symptom identification. In terms of model structure, the Dynamic Time Warping (DTW) algorithm is introduced to enhance the time series modeling capabilities based on behavioral trajectories.

[0124] To enhance the decision-making layer, a reinforcement learning reward function oriented towards target performance is developed. For example, in financial credit, a 5:3:2 weighting is set based on default rate, customer satisfaction, and approval speed; in medical services, a 4:4:2 weighting is set based on treatment effectiveness, user satisfaction, and medical insurance compliance. Furthermore, a multi-expert model fusion strategy is implemented, integrating XGBoost, LightGBM, and Transformer architectures into the pricing model. A difference threshold is set to determine whether to trigger automatic retraining, thereby improving model robustness and generalization.

[0125] Finally, structural enhancement of the decision support model means introducing a multi-expert model fusion mechanism and a time series modeling module on the basis of maintaining the original model framework, such as integrating XGBoost and Transformer structures, and integrating dynamic time warping algorithms to enhance the model's ability to express and generalize complex behavior patterns, thereby outputting better optimization strategy execution results.

[0126] S80: Continuously monitor the execution results of the optimization strategy, and trigger a special evaluation process when an early warning signal is detected to optimize the output decision-making assistance plan.

[0127] Establish a three-level monitoring system covering the entire process:

[0128] Real-time monitoring layer: tracks data pipeline latency (P99 less than 800 milliseconds), model inference time, and monitors the real-time change curves of key feature inputs;

[0129] Hourly monitoring layer: Analyzes the PSI change trend of the profile model and issues alerts for abnormal policy triggering frequencies.

[0130] Daily monitoring layer: Outputs the clustering model stability index and strategy ROI trend chart for regular review by the business team.

[0131] When an indicator is detected at any level exceeding the warning threshold (such as a model performance degradation exceeding 15%, a cumulative strategy execution deviation exceeding 10%, etc.), a special assessment process is triggered to continuously optimize the decision support system in a closed-loop manner to ensure its stable operation and decision-making reliability in high-risk, high-value scenarios such as finance and healthcare.

[0132] Specifically, the special evaluation process is triggered when abnormal performance of the decision-making support model or system is monitored. A more frequent and refined evaluation mechanism is adopted. The evaluation content includes: focusing on analyzing the distribution changes of the model input features (through PSI and KS values), the consistency of the model output, and whether there is significant performance degradation. If necessary, backtesting is performed to identify the source of the deviation. The recently output decision-making support solution is compared with the historical best strategy. The Shapley value method is used to evaluate the changes in the contribution of each key variable to determine whether the strategy deviation is reasonable. By constructing simulated samples in specific scenarios (such as different customer types and extreme behavior patterns), the performance of the model in the new scenario is tested to verify its generalization ability. The strategy ROI is recalculated based on actual business feedback. In the financial field, it is reviewed in combination with the default cost and customer acquisition cost. In the medical scenario, the use of medical resources and changes in user satisfaction are evaluated. Check the coordination of various links such as data sources, portrait modules, and model call logic to identify potential data delays, cleaning errors, or model call anomalies.

[0133] It can be seen that in the above scheme, the present invention realizes the rational allocation of customer service resources and the automatic formulation of personalized service strategies by accurately constructing customer portraits, thereby improving service response speed and problem handling efficiency, thereby effectively improving customer satisfaction and overall service efficiency. Secondly, it can identify potential lost customers and high-value customers, support the implementation of differentiated customer relationship management measures, effectively reduce customer churn rate and enhance customer loyalty. Thirdly, relying on comprehensive and accurate portrait information, a scientific decision-making support model is constructed to provide reliable data support for various business decisions of the customer service department, thereby improving the scientificity, rationality and execution effect of the decision. Finally, in terms of data security and quality assurance, blockchain technology is also introduced to realize the secure storage and traceable management of data, and at the same time establish a complete data quality monitoring system and quantitative evaluation indicators to improve the stability, accuracy and overall performance of data processing.

[0134] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0135] In one embodiment, a decision support device based on portrait data is provided, which corresponds one-to-one to the decision support method based on portrait data in the above embodiment. Figure 8 As shown, the decision support device based on portrait data includes: a data acquisition module 101, a data processing module 102, a data extraction module 103, a data update module 104, and a data output module 105. The functional modules are described in detail as follows:

[0136] The data acquisition module 101 is used to acquire customer data and integrate the customer data to obtain an original data set;

[0137] The data processing module 102 is used to clean and preprocess the original data set, and perform standardization processing to obtain standard data;

[0138] A data extraction module 103 is configured to construct a basic attribute profile, a service behavior profile, and a consumption behavior profile based on the standard data to extract multi-dimensional feature information, and convert the multi-dimensional feature information into a customer profile vector;

[0139] The data updating module 104 is used to identify customer behavior based on the customer portrait vector and establish a dynamic portrait updating mechanism to obtain dynamic portrait data;

[0140] The data output module 105 is used to construct a decision support model using the dynamic portrait data, and run the decision support model to output a decision support solution.

[0141] In one embodiment, the data extraction module 103 is specifically configured to:

[0142] Extracting basic attributes of customers from the standard data to construct an initial attribute set;

[0143] Performing prediction and filling processing on the missing data in the initial attribute set based on a data mining algorithm to obtain customer filling attributes;

[0144] Classify and encode the customer-filled attributes, and convert the obtained classification encoding results into recognizable feature vectors;

[0145] Based on the feature vector, a basic attribute profile is constructed to support customer behavior identification and dynamic updating of the profile.

[0146] In one embodiment, the data extraction module 103 is further configured to:

[0147] Extracting service behavior features from data related to customer service behavior in the standard data to obtain a service behavior feature set;

[0148] Constructing a time series sequence based on the service behavior feature set, and sorting the service behavior events corresponding to the preset customer identifiers by timestamp according to the time series sequence to obtain original sequence data;

[0149] Performing feature label mapping on the original sequence data, and discretizing events related to customer service behaviors into event labels to obtain labeled behavior sequences;

[0150] Mining the behavior sequence through a sequence mining algorithm to obtain a customer behavior pattern;

[0151] Conduct risk assessment and behavioral intention identification on the customer behavior pattern, and output a service risk index;

[0152] Based on the service risk index and using business strategy rules, a service behavior profile is constructed to express the customer's behavioral tendencies, response patterns and potential demands during the service process.

[0153] In one embodiment, the data extraction module 103 is further configured to:

[0154] Extracting customer purchase data from the standard data and preprocessing the customer purchase data to obtain a purchase behavior dataset;

[0155] Based on the purchase behavior dataset, a customer-product bipartite graph and a time series feature storage architecture are constructed to extract consumption behavior features;

[0156] Constructing a cross-modal feature matrix based on the consumption behavior characteristics to obtain cross-modal behavior data;

[0157] Utilizing the cross-modal behavior data to build a consumption trend prediction model, perform consumption time series modeling, and output trend prediction results;

[0158] A consumer behavior profile is generated based on the trend prediction results for subsequent dynamic updates and decision-making assistance calls.

[0159] In one embodiment, the data updating module 104 is specifically configured to:

[0160] Constructing a real-time feature calculation layer based on the customer portrait vector, and incrementally updating various time-sensitive features to obtain updated feature values;

[0161] Synchronously correcting the associated items of the updated feature values ​​to obtain updated features;

[0162] Inputting the updated features into a learning model based on a stochastic gradient descent algorithm for prediction and identification to obtain a customer behavior prediction result;

[0163] Build a profile update engine and update the customer behavior prediction results in real time to build a dynamic profile update mechanism;

[0164] Fault-tolerant detection is performed according to the dynamic portrait update mechanism. When it is determined that key features have drifted, a portrait version rollback operation is triggered to obtain dynamic portrait data.

[0165] In one embodiment, the data output module 105 is specifically configured to:

[0166] Extracting key feature parameters from the dynamic portrait data as an input feature set for decision assistance;

[0167] Building a service resource allocation model based on the input feature set and calculating customer priority scores to output an optimal resource allocation plan and a service channel recommendation strategy;

[0168] Building a service strategy formulation model based on the service resource allocation model to push product information and activity information;

[0169] Using the dynamic profile data to identify potential churn customers and high-value customer groups to build a customer relationship maintenance model;

[0170] The service resource allocation model, service strategy formulation model and customer relationship maintenance model are respectively run to output a comprehensive decision-making support solution.

[0171] In one embodiment, the decision support device based on portrait data further includes:

[0172] The data evaluation module 106 is used to build a systematic evaluation process, perform a multi-dimensional evaluation on the decision support scheme, and obtain an evaluation result;

[0173] A data enhancement module 107 is configured to execute a modular optimization solution based on the evaluation results and simultaneously perform structural enhancement on the decision support model to obtain an optimization strategy execution result;

[0174] The data monitoring module 108 is used to continuously monitor the execution results of the optimization strategy and trigger a special evaluation process when an early warning signal is detected to optimize the output decision-making auxiliary plan.

[0175] For the specific definition of the decision-making assistance device based on portrait data, please refer to the definition of the decision-making assistance method based on portrait data above, which will not be repeated here. The various modules in the above-mentioned decision-making assistance device based on portrait data can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0176] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 9As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the server side of a decision-making assistance method based on portrait data.

[0177] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 10 As shown. The computer device includes a processor, memory, network interface, display screen and input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a decision-making assistance method based on portrait data.

[0178] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed, the steps provided in the above embodiment can be implemented.

[0179] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can be found in the relevant descriptions of the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0180] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0181] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0182] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A decision support method based on portrait data, characterized in that: include: Acquire customer data and integrate the customer data to obtain an original data set; Cleaning and preprocessing the original data set, and performing standardization processing to obtain standard data; Based on the standard data, a basic attribute profile, a service behavior profile, and a consumption behavior profile are respectively constructed to extract multi-dimensional feature information, and the multi-dimensional feature information is converted into a customer profile vector; Identify customer behavior based on the customer portrait vector, and build a dynamic portrait update mechanism to obtain dynamic portrait data; A decision support model is constructed using the dynamic portrait data, and the decision support model is run to output a decision support solution.

2. The decision-making assistance method based on portrait data according to claim 1, characterized in that: The basic attribute portrait, service behavior portrait and consumption behavior portrait are constructed based on the standard data to extract multi-dimensional feature information, including: Extracting basic attributes of customers from the standard data to construct an initial attribute set; Performing prediction and filling processing on the missing data in the initial attribute set based on a data mining algorithm to obtain customer filling attributes; Classify and encode the customer-filled attributes, and convert the obtained classification encoding results into recognizable feature vectors; Based on the feature vector, a basic attribute profile is constructed to support customer behavior identification and dynamic updating of the profile.

3. The decision-making assistance method based on portrait data according to claim 1, characterized in that: The step of constructing a basic attribute portrait, a service behavior portrait, and a consumption behavior portrait based on the standard data to extract multi-dimensional feature information further includes: Extracting service behavior features from data related to customer service behavior in the standard data to obtain a service behavior feature set; Constructing a time series sequence based on the service behavior feature set, and sorting the service behavior events corresponding to the preset customer identifiers by timestamp according to the time series sequence to obtain original sequence data; Performing feature label mapping on the original sequence data, and discretizing events related to customer service behaviors into event labels to obtain labeled behavior sequences; Mining the behavior sequence through a sequence mining algorithm to obtain a customer behavior pattern; Conduct risk assessment and behavioral intention identification on the customer behavior pattern, and output a service risk index; Based on the service risk index and using business strategy rules, a service behavior profile is constructed to express the customer's behavioral tendencies, response patterns and potential demands during the service process.

4. The decision-making assistance method based on portrait data according to claim 1, characterized in that: The step of constructing a basic attribute portrait, a service behavior portrait, and a consumption behavior portrait based on the standard data to extract multi-dimensional feature information further includes: Extracting customer purchase data from the standard data and preprocessing the customer purchase data to obtain a purchase behavior dataset; Based on the purchase behavior dataset, a customer-product bipartite graph and a time series feature storage architecture are constructed to extract consumption behavior features; Constructing a cross-modal feature matrix based on the consumption behavior characteristics to obtain cross-modal behavior data; Utilizing the cross-modal behavior data to build a consumption trend prediction model, perform consumption time series modeling, and output trend prediction results; A consumer behavior profile is generated based on the trend prediction results for subsequent dynamic updates and decision-making assistance calls.

5. The decision-making assistance method based on portrait data according to claim 1, characterized in that: The identifying of customer behavior based on the customer portrait vector and building a dynamic portrait update mechanism to obtain dynamic portrait data include: Constructing a real-time feature calculation layer based on the customer portrait vector, and incrementally updating various time-sensitive features to obtain updated feature values; Synchronously correcting the associated items of the updated feature values ​​to obtain updated features; Inputting the updated features into a learning model based on a stochastic gradient descent algorithm for prediction and identification to obtain a customer behavior prediction result; Build a profile update engine and update the customer behavior prediction results in real time to build a dynamic profile update mechanism; Fault-tolerant detection is performed according to the dynamic portrait update mechanism. When it is determined that key features have drifted, a portrait version rollback operation is triggered to obtain dynamic portrait data.

6. The decision-making assistance method based on portrait data according to claim 1, characterized in that: The method of constructing a decision support model using the dynamic portrait data and running the decision support model to output a decision support solution includes: Extracting key feature parameters from the dynamic portrait data as an input feature set for decision assistance; Building a service resource allocation model based on the input feature set and calculating customer priority scores to output an optimal resource allocation plan and a service channel recommendation strategy; Building a service strategy formulation model based on the service resource allocation model to push product information and activity information; Using the dynamic profile data to identify potential churn customers and high-value customer groups to build a customer relationship maintenance model; The service resource allocation model, service strategy formulation model and customer relationship maintenance model are respectively run to output a comprehensive decision-making support solution.

7. The decision-making assistance method based on portrait data according to claim 1, characterized in that: After constructing a decision support model using the dynamic portrait data and running the decision support model to output a decision support solution, the method further includes: Establish a systematic evaluation process to conduct a multi-dimensional evaluation of the decision support program and obtain evaluation results; Executing a modular optimization scheme according to the evaluation results, and simultaneously performing structural enhancement on the decision support model to obtain an optimization strategy execution result; The execution results of the optimization strategy are continuously monitored, and when an early warning signal is detected, a special evaluation process is triggered to optimize the output decision-making support plan.

8. A decision support device based on portrait data, characterized in that: include: A data acquisition module, used to acquire customer data and integrate the customer data to obtain an original data set; A data processing module is used to clean and preprocess the original data set and perform standardization to obtain standard data; A data extraction module is used to construct a basic attribute profile, a service behavior profile, and a consumption behavior profile based on the standard data to extract multi-dimensional feature information, and convert the multi-dimensional feature information into a customer profile vector; A data update module is used to identify customer behavior based on the customer portrait vector and build a dynamic portrait update mechanism to obtain dynamic portrait data; The data output module is used to construct a decision support model using the dynamic portrait data, and run the decision support model to output a decision support solution.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the decision-making assistance method based on portrait data as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the decision-making assistance method based on portrait data as claimed in any one of claims 1 to 7 are implemented.

Images