A home-based elderly care fraud risk evolution prediction method and system based on old user behavior portrait and multi-modal event timing fusion

Abstract

The application discloses a kind of based on old user behavior portrait and multi-modal event timing fusion's home-based care involves fraud risk evolution prediction method and system, belong to artificial intelligence, intelligent care for the aged, information security and risk prediction technical field.The present application collects old user portrait data, voice interaction data, behavior trajectory data and home environment data, constructs multi-modal event sequence for home-based elderly people;MAPE static risk model evaluates fraud motive manipulation risk, old user vulnerability risk, fraud behavior trigger risk and home environment risk;T timing evolution model analyzes the occurrence sequence, time interval and behavior change trend of fraud-related events;Fusion multi-modal risk results, output risk level, current fraud stage and early warning strategy.Realize the dynamic prediction and hierarchical intervention of the fraud risk of home-based elderly people, improve the advance, accuracy and scene adaptability of risk identification of old fraud.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence, smart elderly care, information security and risk prediction technology, and in particular relates to a method and system for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series. Background Technology

[0002] With the development of smart devices and internet technology, telecommunications fraud, online fraud, and fraud targeting the elderly are on the rise. Existing fraud detection technologies primarily rely on phone number detection, fraud keyword matching, user reports, and analysis of abnormal financial transactions to identify unusual fraudulent behavior. However, these technologies mainly target ordinary users, typically focusing on communication content or financial transactions, lacking the ability to analyze risks specific to the living environments and behavioral characteristics of elderly users. In home-based elder care scenarios, elderly users are more vulnerable to fraud due to factors such as living alone, differences in digital literacy, health needs, emotional needs, and consumption habits. Current technologies do not incorporate these factors into risk analysis, making them unsuitable for predicting fraud risks in home-based elder care scenarios and resulting in insufficient accuracy in risk identification.

[0003] Fraudsters typically build trust gradually through prolonged communication, emotional manipulation, health marketing, and impersonating relatives. This leads to a continuous evolution of fraud targeting the elderly, progressing from risk exposure and unusual contact to relationship establishment, manipulation, and ultimately, financial loss. Current technologies often focus on identifying individual fraud incidents or the final financial loss, lacking the ability to model the continuous behavioral changes throughout the fraud process, particularly regarding the comprehensive analysis of the living conditions, behavioral patterns, and environmental changes of elderly people living at home. In practical applications, data on elderly users' daily life trajectories, changes in their home environment, voice interaction information, and consumption behavior characteristics are scattered and not effectively utilized. Current technologies cannot comprehensively utilize this multi-source information to predict fraud risks in advance, resulting in delayed warnings and hindering effective intervention before financial losses occur. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and system for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, thereby resolving the problems existing in the prior art.

[0005] Firstly, to achieve the above objectives, this invention provides a method for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, comprising the following steps: Collect profile data, multimodal perception data, and home environment data of elderly users; A behavioral profile of elderly users is constructed based on the aforementioned profile data; Standardized fraud-related events are generated based on the multimodal perception data and home environment data; Based on the behavioral profile and the standardized fraud-related events, static risk factors are calculated to obtain static risk values; Based on the standardized fraud-related events, a time-series risk factor is calculated to obtain a dynamic time-series risk value; Based on the multimodal perception data and home environment data, modal risk factors are calculated to obtain a multimodal fusion risk value; By integrating the static risk value, dynamic time-series risk value, and multimodal fusion risk value, a comprehensive risk value is obtained, and the risk level and intervention strategy are output.

[0006] Optionally, the process of collecting profile data includes: collecting age information, living alone status, family support status, digital device usage ability, consumption behavior characteristics, and historical risk behavior characteristics; the process of collecting multimodal perception data includes: collecting voice interaction data and behavioral trajectory data, with voice interaction data identifying semantics related to verification codes, money transfers, public security, procuratorate, and courts, requests for help from relatives and friends, free physical examinations, and elderly care projects, and behavioral trajectory data identifying events such as prolonged stays, abnormal movement, and proximity to valuable areas; the process of collecting home environment data includes: identifying events such as strangers entering, door locks being opened, and abnormal access.

[0007] Optionally, the process of constructing a behavioral profile of elderly users includes: constructing a profile vector based on their living alone status, digital ability, family support level, healthy consumption habits, and historical risk behaviors, wherein the profile vector uses living alone status, digital ability, family support level, healthy consumption habits, and historical risk behaviors as profile dimensions.

[0008] Optionally, the process of generating standardized fraud-related events includes: converting voice events, behavioral events, and environmental events into standardized fraud-related events, wherein the standardized fraud-related events consist of event type, event occurrence time, event occurrence location, event characteristics, and event credibility; and the standardized fraud-related events are arranged in chronological order to form a fraud-related event sequence.

[0009] Optionally, the process of calculating static risk factors includes: calculating the risk of fraud motive manipulation based on health anxiety, elderly care needs, emotional needs, and trust in authority; calculating the vulnerability risk of elderly users based on their living alone status, digital ability, family support, and historical risky behavior; calculating the risk of fraud triggering based on verification codes, transfers, QR code scanning, contract signing, and abnormal consumption events; calculating the risk of home environment based on events such as strangers entering, abnormal stays, changes in activity areas, and proximity of valuables; and integrating the aforementioned risks of fraud motive manipulation, the aforementioned vulnerability risk of elderly users, the aforementioned risk of fraud triggering, and the aforementioned risk of home environment to obtain a static risk value.

[0010] Optionally, the process of calculating the time-series risk factor includes: calculating the event sequence risk based on whether the event arrangement matches the chain of fraud involving the elderly; calculating the time interval risk based on whether high-risk events occur consecutively in a short period of time; calculating the behavioral change risk based on abnormal trajectories, proximity to property areas, and changes in behavior during long periods of stay; and integrating the event sequence risk, time interval risk, and behavioral change risk to obtain a dynamic time-series risk value.

[0011] Secondly, the present invention also provides a home-based elderly care fraud risk evolution prediction system based on elderly user behavior profiles and multimodal event time series fusion, used to implement a home-based elderly care fraud risk evolution prediction method based on elderly user behavior profiles and multimodal event time series fusion, the system comprising: The multi-dimensional data acquisition module is used to collect profile data, multimodal perception data, and home environment data of elderly users; The user profile building module is used to build a behavioral profile of elderly users based on the profile data. The fraud incident standardization module is used to generate standardized fraud incidents based on the multimodal perception data and home environment data. The static risk calculation module is used to calculate static risk factors based on the behavioral profile and the standardized fraud-related events to obtain static risk values. The dynamic time-series risk calculation module is used to calculate time-series risk factors based on the standardized fraud-related events to obtain dynamic time-series risk values; The multimodal risk fusion module is used to calculate modal risk factors based on the multimodal perception data and home environment data to obtain a multimodal fusion risk value; The integrated risk fusion and output module is used to fuse the static risk value, dynamic time-series risk value, and multimodal fusion risk value to obtain an integrated risk value and output the risk level and intervention strategy.

[0012] Thirdly, the present invention also provides a computer terminal device, comprising: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the method for predicting the evolution of home-based elderly care fraud risks based on the fusion of elderly user behavior profiles and multimodal event time series in the first aspect above.

[0013] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the method for predicting the evolution of home-based elderly care fraud risks based on elderly user behavior profiles and multimodal event time-series fusion in the first aspect described above.

[0014] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the method for predicting the evolution of home-based elderly care fraud risks based on elderly user behavior profiles and multimodal event time-series fusion in the first aspect described above.

[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention provides a method and system for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series. This improves the accuracy of fraud risk identification for elderly people living at home by incorporating elderly-specific characteristics such as their solitary status, consumption habits, lifestyle, and digital literacy into risk calculation. It shifts from detecting fraud outcomes to predicting the fraud process, identifying risks in advance through event time series evolution analysis. It integrates multi-source heterogeneous data such as voice, behavior, environment, and user profiles to enhance identification capabilities in complex elderly-related fraud scenarios. It employs non-camera-dependent behavior sensing methods such as RFID, WiFi sensing, and millimeter wave to reduce privacy intrusion at home. Different intervention strategies are matched according to risk levels, forming a closed loop of risk prediction, tiered early warning, and collaborative blocking. Attached Figure Description

[0016] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall system architecture according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, according to an embodiment of the present invention. Figure 3 This is a diagram illustrating the construction of elderly user behavior profiles and the calculation of basic vulnerability risks in an embodiment of the present invention. Figure 4 This is a schematic diagram illustrating the temporal evolution of a fraud case involving the elderly, as described in an embodiment of the present invention. Figure 5 This is a closed-loop diagram of risk classification and early warning, collaborative intervention and feedback optimization in an embodiment of the present invention; Figure 6 This is a risk value change curve representing a representative extremely high-risk scenario in an embodiment of the present invention; Figure 7 This is a comparison chart of risk level prediction indicators for six comparative models in this invention. Figure 8 This is a comparison chart of the early warning results of six comparative models in an embodiment of the present invention. Detailed Implementation

[0017] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0019] like Figures 1-5 As shown, this embodiment provides a method for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, including: Collect profile data, multimodal perception data, and home environment data of elderly users; A behavioral profile of elderly users is constructed based on the aforementioned profile data; Standardized fraud-related events are generated based on the multimodal perception data and home environment data; Based on the behavioral profile and the standardized fraud-related events, static risk factors are calculated to obtain static risk values; Based on the standardized fraud-related events, a time-series risk factor is calculated to obtain a dynamic time-series risk value; Based on the multimodal perception data and home environment data, modal risk factors are calculated to obtain a multimodal fusion risk value; By integrating the static risk value, dynamic time-series risk value, and multimodal fusion risk value, a comprehensive risk value is obtained, and the risk level and intervention strategy are output.

[0020] Furthermore, the process of collecting profile data includes: collecting age information, living alone status, family support status, digital device usage ability, consumption behavior characteristics, and historical risk behavior characteristics; the process of collecting multimodal perception data includes: collecting voice interaction data and behavioral trajectory data, with voice interaction data identifying semantics related to verification codes, money transfers, public security, procuratorate, and courts, requests for help from relatives and friends, free physical examinations, and elderly care projects, and behavioral trajectory data identifying events such as prolonged stays, abnormal movement, and proximity to valuable areas; the process of collecting home environment data includes: identifying events such as strangers entering, door locks being opened, and abnormal access.

[0021] Specifically, the implementation process of this embodiment includes: Step 1: Obtain multidimensional data of elderly users The system collects user profile data, multimodal perception data, and home environment data for the elderly. User profile data includes age information, status of living alone, family support status, digital device usage ability, consumption behavior characteristics, and historical risk behavior characteristics. Multimodal perception data includes voice interaction data, behavioral trajectory data, and environmental access data.

[0022] Voice data: obtained from voice sensors, call text, or ASR transcribed text, used to identify fraudulent semantics such as verification codes, money transfers, public security, procuratorate and court matters, requests for help from relatives and friends, free physical examinations, and elderly care projects.

[0023] Behavioral data: obtained by RFID, WiFi sensing, and millimeter-wave sensing devices, used to identify events such as activity trajectories, location changes, prolonged stays, and proximity to valuable areas.

[0024] Environmental data: obtained from door locks, doorbells, and visitor detection devices, used to identify events such as unfamiliar personnel entering, dwell time, and abnormal visits.

[0025] Furthermore, the process of constructing a behavioral profile of elderly users includes: constructing a profile vector based on their living alone status, digital ability, family support level, healthy consumption habits, and historical risk behaviors. The profile vector uses living alone status, digital ability, family support level, healthy consumption habits, and historical risk behaviors as profile dimensions.

[0026] Specifically, the implementation process of this embodiment includes: Step 2: Build a behavioral profile of elderly users An elderly user profile vector is constructed based on their living alone status, digital capabilities, consumption habits, family support status, and daily behavior patterns: U = (u1, u2, u3, u4, u5); Among them, u1 represents living alone, u2 represents digital ability, u3 represents the level of family support, u4 represents healthy consumption habits, and u5 represents historical risk behavior. The aforementioned elderly user behavior profile is used to differentiate it from ordinary financial anti-fraud models, reflecting the unique risk characteristics of elderly people living at home in terms of living alone, consumption preferences, daily routines, and prolonged voice interaction.

[0027] Furthermore, the process of calculating static risk factors includes: calculating the risk of fraud motive manipulation based on health anxiety, elderly care needs, emotional needs, and trust in authority; calculating the vulnerability risk of elderly users based on their living alone status, digital capabilities, family support, and historical risky behaviors; calculating the risk of fraud triggering based on verification codes, transfers, QR code scanning, contract signing, and abnormal consumption events; calculating the home environment risk based on events such as strangers entering, unusual stays, changes in activity areas, and proximity of valuables; and integrating the aforementioned risks of fraud motive manipulation, the vulnerability risk of elderly users, the risk of fraud triggering, and the home environment risk to obtain a static risk value.

[0028] Specifically, the implementation process of this embodiment includes: Step 3: Construct the MAPE static risk model MAPE static risk model includes the risk of manipulating fraud motives. Vulnerability risks for elderly users Risks triggered by fraudulent activities and home environment risks The static risk value is expressed as: ; in, These are the weights of the corresponding risk factors; This indicates that fraudsters exploit motivations stemming from factors such as health anxiety, retirement needs, emotional needs, and trust in authority to manipulate risk. This indicates the vulnerability risks faced by elderly users due to factors such as living alone, digital literacy, family support, and historical risky behaviors. This indicates that fraudulent activities triggered by events such as verification codes, transfers, QR code scanning, contract signing, and unusual spending are risky. This refers to the risks to the home environment caused by events such as strangers entering, unusual stays, changes in activity areas, or proximity of valuables.

[0029] Furthermore, the process of generating standardized fraud-related events includes: converting voice events, behavioral events, and environmental events into standardized fraud-related events, wherein the standardized fraud-related events consist of event type, event occurrence time, event occurrence location, event characteristics, and event credibility; and the standardized fraud-related events are arranged in chronological order to form a fraud-related event sequence.

[0030] Specifically, the implementation process of this embodiment includes: Step 4: Establish a temporal evolution model T for fraud events Transform voice events, behavioral events, and environmental events into standardized fraud-related events: Xi = (Ci, Ti, Li, Fi, Qi); Where Ci represents the event type, Ti represents the event time, Li represents the event location, Fi represents the event characteristics, and Qi represents the event credibility. Multiple events arranged chronologically constitute a fraud event sequence: S = {X1, X2, ..., Xn}; The fraud event time-series evolution model T is used to analyze the sequence of events, the time interval between events, and the trend of behavioral changes to determine whether fraudulent behavior has gradually evolved from risk exposure, abnormal contact, trust establishment, and induced operation to the stage of approaching loss.

[0031] Furthermore, the process of calculating the time-series risk factor includes: calculating the event sequence risk based on whether the event arrangement matches the chain of fraud involving the elderly; calculating the time interval risk based on whether high-risk events occur consecutively in a short period of time; calculating the behavioral change risk based on abnormal trajectories, proximity to property areas, and changes in behavior over long periods of time; and integrating the event sequence risk, time interval risk, and behavioral change risk to obtain the dynamic time-series risk value.

[0032] Specifically, the implementation process of this embodiment includes: Step 5: Calculate dynamic time series risk Calculate dynamic time-series risk based on event sequence, time interval, and behavioral change trends: ; in, This indicates the risk of the sequence of events, used to evaluate whether the arrangement of events conforms to a chain of fraud involving the elderly; This indicates time interval risk, used to assess whether high-risk events occur consecutively within a short period of time; It indicates the risk of behavioral changes and is used to assess whether elderly people exhibit abnormal behavior, such as moving close to areas with valuables, or staying there for extended periods. , , These are the weighting coefficients.

[0033] Step 6: Establish a multimodal risk fusion model Calculate voice risk Behavioral risks and environmental risks And the multimodal fusion risk value is obtained by fusing: ; in, Indicates speech semantic risk, Indicates the risk of behavioral trajectory. The risk level represents the risk in the home environment, where ρv, ρb, and ρe are the fusion weights for different modalities. The final comprehensive risk value is expressed as: ; in, represents the MAPE static risk value, represents the dynamic time-series risk value, represents the multimodal fusion risk value, , , are fusion weights.

[0034] Step 7: Perform risk graded warning Risk levels are divided according to the comprehensive risk value R: 0<R<30 is low risk, and the system performs background recording and continuous observation.

[0035] 30<= R<60 is medium risk, and the system sends an anti-fraud reminder to the elderly end.

[0036] 60<= R<80 is high risk, and the system notifies family members or guardians and strengthens monitoring.

[0037] R>= 80 is extremely high risk, the system issues a strong reminder, and links family members and the community for collaborative intervention.

[0038] Step 8: Execute graded intervention strategies According to the risk level and fraud stage, the system outputs a corresponding intervention strategy. In the low-risk state, the system only records the event; in the medium-risk state, the system sends voice or text prompts to the elderly end; in the high-risk state, the system sends the risk cause to the family member end, including suspicious calls, fraud keywords, abnormal behavior trajectories, etc.; in the extremely high-risk state, the system triggers a strong reminder, and promotes family members, the community or other guardians to carry out collaborative intervention.

[0039] Compared with the prior art, the present invention has at least the following beneficial effects: The accuracy of fraud risk identification for home-based elderly people is improved, and the elderly-specific characteristics such as living alone status, consumption habits, life trajectories and digital capabilities are incorporated into risk calculation.

[0040] The transformation from fraud result detection to fraud process prediction is realized, and risks are identified in advance through event time-series evolution analysis.

[0041] Multi-source heterogeneous data such as voice, behavior, environment and user portraits are fused to improve the recognition ability in complex elderly-related fraud scenarios.

[0042] Non-camera-dependent behavior perception methods such as RFID, WiFi sensing, and millimeter wave are adopted to reduce home privacy intrusion.

[0043] Different intervention strategies are matched according to the risk level, forming a closed loop of risk prediction, graded warning and collaborative blocking.

[0044] In a set of simulation experiments, the MAPE-TM model achieved an accuracy of 68.52%, a macro-average F1 score of 65.73%, and a macro-average recall of 65.87% on the independent test set. Its macro-average F1 score was 4.59 and 11.76 percentage points higher than that of the MAPE-T model and the MAPE model, respectively. The false alarm rate was 8.70%, which was lower than that of the MAPE-T model (13.04%) and the MAPE model (26.09%).

[0045] Based on the same general inventive concept, this invention also provides a home-based elderly care fraud risk evolution prediction system based on the fusion of elderly user behavior profiles and multimodal event time series. The following describes the home-based elderly care fraud risk evolution prediction system based on the fusion of elderly user behavior profiles and multimodal event time series provided by this invention. The home-based elderly care fraud risk evolution prediction system described below can be referred to in conjunction with the home-based elderly care fraud risk evolution prediction method described above. The system includes: The multi-dimensional data acquisition module is used to collect profile data, multimodal perception data, and home environment data of elderly users; The user profile building module is used to build a behavioral profile of elderly users based on the profile data. The fraud incident standardization module is used to generate standardized fraud incidents based on the multimodal perception data and home environment data. The static risk calculation module is used to calculate static risk factors based on the behavioral profile and the standardized fraud-related events to obtain static risk values. The dynamic time-series risk calculation module is used to calculate time-series risk factors based on the standardized fraud-related events to obtain dynamic time-series risk values; The multimodal risk fusion module is used to calculate modal risk factors based on the multimodal perception data and home environment data to obtain a multimodal fusion risk value; The integrated risk fusion and output module is used to fuse the static risk value, dynamic time-series risk value, and multimodal fusion risk value to obtain an integrated risk value and output the risk level and intervention strategy.

[0046] In this embodiment, a computer terminal device is provided, including: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of the above-mentioned method for predicting the evolution of home-based elderly care fraud risks based on the fusion of elderly user behavior profiles and multimodal event time series.

[0047] In this embodiment, a computer-readable storage medium is also provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the above-mentioned method for predicting the evolution of home-based elderly care fraud risks based on elderly user behavior profiles and multimodal event time-series fusion.

[0048] In this embodiment, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the above-mentioned method for predicting the evolution of home-based elderly care fraud risks based on elderly user behavior profiles and multimodal event time-series fusion.

[0049] Example 1: Risk Prediction of Telephone Refund Fraud In this embodiment, an elderly user living alone receives a call from an unknown number. The system acquires the call semantics through the voice device and identifies potentially fraudulent phrases such as "refund," "verification code," and "don't tell family members." Simultaneously, a passive sensing device detects that the elderly person has moved from the living room to the bedroom and is near valuables. The system encodes these events into a continuous event sequence, calculates the MAPE static risk and T dynamic temporal risk, and determines that the event chain has entered the near-loss stage. The system outputs a high-risk or extremely high-risk level and alerts the elderly user, while simultaneously sending the reason for the risk to their family members.

[0050] Example 2: Risk Prediction of Health Product Scams Committed by Strangers Entering Homes In this embodiment, the doorbell and door lock devices detect that an unfamiliar person has entered the home environment and stayed for more than a preset threshold. The voice device recognizes health-inducing semantics such as "free physical examination," "expert recommendation," and "health products"; the passive sensing device recognizes that the elderly person has been with an unfamiliar person for a long time and exhibits behavior related to approaching valuables or unusual spending. Based on the elderly user's health consumption habits and living alone, the system increases the vulnerability risk weight and, combined with an event sequence model, determines whether the scenario has moved from the risk exposure stage to the trust-building or inducement operation stage, thereby triggering a family notification.

[0051] Example 3: AI-powered risk prediction for scams involving requests for help from relatives and friends In this embodiment, fraudsters impersonate elderly users' relatives via voice or video, using phrases such as "Something's happened," "I urgently need money," and "Don't tell anyone" to induce the elderly to transfer money. The system extracts the risk from the voice and semantics, and combines this with the elderly's behavioral patterns to identify potential losses, such as searching for bank cards, opening payment devices, or approaching areas containing valuables. The MAPE-TM model calculates a comprehensive risk value based on the risk of being induced by family ties, the vulnerability of the elderly user, fraud-triggered behaviors, and the risk of temporal evolution. If the comprehensive risk value reaches an extremely high risk threshold, the system immediately triggers a strong alert and notifies a pre-set family member to verify the identity.

[0052] As can be seen from the above embodiments, the present invention can dynamically predict and tieredly intervene in fraud risks by integrating elderly user behavior profiles, multimodal perception data, and fraud event time-series evolution models for typical fraud scenarios involving the elderly, such as telephone fraud, stranger intrusion fraud, and AI-powered family and friends fraud.

[0053] Example 4: Algorithm Validation on Simulated Datasets To verify the feasibility of the MAPE-TM model and the technical effectiveness of its components, this embodiment uses Python to construct a simulation experiment system. The simulation data does not contain real personal information; its risk level is generated from the severity of independent potential scenarios, not directly derived from the MAPE-TM risk formula, thus avoiding circular verification caused by using model output as the truth value.

[0054] The simulated data covers scenarios such as telephone refunds, impersonation of public security, procuratorate, and court officials, strangers entering homes to sell health products, AI-generated requests for help from relatives and friends, and investment inducements. Simultaneously, it generates profiles of elderly users, time series of fraudulent events, voice risk, behavioral risk, environmental risk, and confidence levels for each modality. The dataset composition is shown in Table 1.

[0055] Table 1 Training, validation, and test sets are partitioned according to user identifiers, ensuring that the same elderly user does not appear in different sets simultaneously. Logistic Regression, XGBoost, and LSTM are all fitted using the training set; the risk thresholds for MAPE, MAPE-T, and MAPE-TM are calibrated only using the validation set; the test set is used only for final metric calculations. Evaluation metrics include accuracy, macro-average F1, macro-average recall, high-risk precision, and high-risk recall.

[0056] In this embodiment, the weights of M, A, P, and E in the MAPE static risk model are set to 0.25, 0.25, 0.30, and 0.20, respectively; the weights of event sequence, time interval, and behavioral change in the time series model are set to 0.45, 0.25, and 0.30, respectively; and the MAPE-T model uses 0.45Rs + 0.55RT. The fusion coefficients of the MAPE-TM model are determined by grid search on the validation set as λs = 0.05, λt = 0.15, and λm = 0.80. The multimodal risk RMM uses a confidence-weighted window of the most recent four time steps and incorporates a cross-modal co-occurrence term to reduce the impact of missing single-point perceptions.

[0057] Table 2 From Table 2 and Figure 7It can be seen that XGBoost achieved the highest macro-average F1 score of 68.12%; MAPE-TM achieved a macro-average F1 score of 65.73%, ranking second among the six methods and higher than LSTM's 64.39%. In the ablation comparison of the patented algorithms, MAPE-T improved the macro-average F1 score from 53.97% to 61.14% compared to MAPE, indicating that information on event sequence, time interval, and behavioral changes can improve risk assessment; after adding multimodal fusion, MAPE-TM further improved to 65.73%, indicating that the fusion of confidence in speech, behavior, and environmental risks has a gain. Figure 6 Examples of how the risk values ​​of three patent models change over time in a set of extremely high-risk scenarios are shown.

[0058] Table 3 The early warning time is based on high-risk reference events; missed or late reports are averaged as 0 minutes. See Table 3 and... Figure 8 As shown, MAPE-TM provides an average advance warning of 15.54 minutes, with a success rate of 76.24% and a false alarm rate of 8.70%, the lowest among the six models. While Logistic Regression and MAPE have longer average advance warning times, their false alarm rates reach 20.00% and 26.09%, respectively. Therefore, the MAPE-TM in this embodiment adopts a relatively conservative warning strategy, maintaining risk identification capabilities while minimizing unnecessary disruption to elderly users.

[0059] The above data represents the results of a simulation experiment with random seed 42, used to illustrate the feasibility and module gain of the technical solution of this invention. It does not represent fixed performance in a real deployment environment, nor does it constitute a limitation on the scope of the claims. In practical applications, parameters can be retrained and risk thresholds adjusted based on device type, user group, regional fraud patterns, and false alarm costs.

[0060] This invention provides a method and system for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series. This improves the accuracy of fraud risk identification for elderly people living at home by incorporating elderly-specific characteristics such as their solitary status, consumption habits, lifestyle, and digital literacy into risk calculation. It shifts from detecting fraud outcomes to predicting the fraud process, identifying risks in advance through event time series evolution analysis. It integrates multi-source heterogeneous data such as voice, behavior, environment, and user profiles to enhance identification capabilities in complex elderly-related fraud scenarios. It employs non-camera-dependent behavior sensing methods such as RFID, WiFi sensing, and millimeter wave to reduce privacy intrusion at home. Different intervention strategies are matched according to risk levels, forming a closed loop of risk prediction, tiered early warning, and collaborative blocking.

[0061] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for predicting the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, characterized in that, Includes the following steps: Collect profile data, multimodal perception data, and home environment data of elderly users; A behavioral profile of elderly users is constructed based on the aforementioned profile data; Standardized fraud-related events are generated based on the multimodal perception data and home environment data; Based on the behavioral profile and the standardized fraud-related events, static risk factors are calculated to obtain static risk values; Based on the standardized fraud-related events, a time-series risk factor is calculated to obtain a dynamic time-series risk value; Based on the multimodal perception data and home environment data, modal risk factors are calculated to obtain a multimodal fusion risk value; By integrating the static risk value, dynamic time-series risk value, and multimodal fusion risk value, a comprehensive risk value is obtained, and the risk level and intervention strategy are output.

2. The method according to claim 1, characterized in that, The process of collecting profile data includes: collecting age information, living alone status, family support status, digital device usage ability, consumption behavior characteristics, and historical risk behavior characteristics; the process of collecting multimodal perception data includes: collecting voice interaction data and behavioral trajectory data, with voice interaction data identifying semantics related to verification codes, money transfers, public security, procuratorate, and courts, requests for help from relatives and friends, free physical examinations, and elderly care projects, and behavioral trajectory data identifying events such as prolonged stays, abnormal movement, and proximity to valuable areas; the process of collecting home environment data includes: identifying events such as strangers entering, door locks being opened, and abnormal access.

3. The method according to claim 1, characterized in that, The process of constructing a behavioral profile of elderly users includes: constructing a profile vector based on their living status, digital ability, family support level, healthy consumption habits, and historical risk behaviors. The profile vector uses living status, digital ability, family support level, healthy consumption habits, and historical risk behaviors as profile dimensions.

4. The method according to claim 1, characterized in that, The process of generating standardized fraud-related events includes: converting voice events, behavioral events, and environmental events into standardized fraud-related events, which are composed of event type, event occurrence time, event occurrence location, event characteristics, and event credibility; and the standardized fraud-related events are arranged in chronological order to form a fraud-related event sequence.

5. The method according to claim 1, characterized in that, The process of calculating static risk factors includes: calculating the risk of fraud motive manipulation based on health anxiety, elderly care needs, emotional needs, and trust in authority; calculating the vulnerability risk of elderly users based on their living alone status, digital ability, family support, and historical risky behaviors; calculating the risk of fraud triggering based on verification codes, transfers, QR code scanning, contract signing, and abnormal consumption events; calculating the home environment risk based on events such as strangers entering, unusual stays, changes in activity areas, and proximity of valuables; and integrating the aforementioned risks of fraud motive manipulation, the vulnerability risk of elderly users, the risk of fraud triggering, and the home environment risk to obtain a static risk value.

6. The method according to claim 1, characterized in that, The process of calculating time-series risk factors includes: calculating the event sequence risk based on whether the event arrangement matches the chain of fraud involving the elderly; calculating the time interval risk based on whether high-risk events occur consecutively in a short period of time; calculating the behavioral change risk based on abnormal trajectories, proximity to property areas, and changes in behavior during long periods of stay; and integrating the event sequence risk, time interval risk, and behavioral change risk to obtain a dynamic time-series risk value.

7. A predictive system for the evolution of fraud risks in home-based elderly care based on the fusion of elderly user behavior profiles and multimodal event time series, characterized in that, The system for implementing the method of any one of claims 1-6 comprises: The multi-dimensional data acquisition module is used to collect profile data, multimodal perception data, and home environment data of elderly users; The user profile building module is used to build a behavioral profile of elderly users based on the profile data. The fraud incident standardization module is used to generate standardized fraud incidents based on the multimodal perception data and home environment data. The static risk calculation module is used to calculate static risk factors based on the behavioral profile and the standardized fraud-related events to obtain static risk values. The dynamic time-series risk calculation module is used to calculate time-series risk factors based on the standardized fraud-related events to obtain dynamic time-series risk values; The multimodal risk fusion module is used to calculate modal risk factors based on the multimodal perception data and home environment data to obtain a multimodal fusion risk value; The integrated risk fusion and output module is used to fuse the static risk value, dynamic time-series risk value, and multimodal fusion risk value to obtain an integrated risk value and output the risk level and intervention strategy.

8. A computer terminal device, characterized in that, include: One or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-6.

Images