An AI and IoT-based multi-source data-driven home service process compliance evaluation method

Abstract

The application belongs to the technical field of wisdom pension service, and discloses a multi-source data driven home service process compliance evaluation method based on AI and IoT, which comprises the following steps: collecting actual basic data, the actual basic data comprising IOT data and service duration; setting standard data, the standard data comprising standard service geographic range and standard service duration; calculating overall trajectory non-compliance degree Pout and stay duration proportion Pkey according to the IOT data and the standard service geographic range; calculating deviation rate Dtime according to the service duration and the standard service duration; constructing feature vector F according to Pout, Pkey, Dtime; calculating dynamic weight W, and calculating basic score Sbasic according to F and W; calculating confidence C, and calculating correction score Scorrected according to confidence C and basic score Sbasic. In the prior art, it is difficult to effectively supervise and judge the service effect. Compared with the prior art, the application provides an objective judgment standard and is suitable for private space, so that effective supervision can be performed.

Description

Technical Field

[0001] This invention relates to the field of smart elderly care service technology, and in particular to a compliance assessment method for home service processes based on multi-source data driven by AI and IoT. Background Technology

[0002] As the population ages, the demand for professional home-based elderly care services is increasing, especially for high-value-added and highly skilled services. For example, in-home bathing assistance services.

[0003] Currently, the quality supervision of home-based elderly care services mainly relies on the following two existing technological means: (1) Basic Task Management System: Service personnel typically use mobile applications (Apps) to clock in and out of services at their location (GPS or LBS) to record the time and approximate location of the service. Some systems require service personnel to manually check in or take photos to document key task milestones.

[0004] (2) Post-service user feedback mechanism: After the service is completed, the system will collect subjective evaluations from the elderly or their families through questionnaires or star ratings, which will serve as the main reference indicators for service quality.

[0005] Therefore, the current supervision of the quality of home care services has the following prominent problems: (1) Reliance on subjective feedback and lack of objectivity: The current evaluation results are essentially a reflection of the emotional experience of the service, which is easily affected by the user's emotions, cultural level or information asymmetry. Its structure determines that it cannot objectively and in real time reflect the professional process compliance of service personnel.

[0006] (2) Difficult to supervise and lack of transparency: The service process is carried out in the user's private space. Existing technology can only record whether the service personnel are present, but cannot confirm whether the service personnel have completed the key operations within the specified time (such as the "disinfection-transfer-bathing-cleaning" steps in the bathing service). This results in a lack of supervision of the service process and makes it difficult to guarantee the quality of service.

[0007] (3) Lack of standardization and difficulty in scaling: Existing technologies lack quantitative indicators to measure service efficiency and professionalism. Relying solely on time points and subjective ratings makes it difficult to extract objective quality characteristics that can be used for AI training, which restricts the standardization and large-scale expansion of high-value services.

[0008] (4) High cost and privacy risks: If companies adopt video surveillance or send on-site supervisors in pursuit of quality, the increased equipment and manpower costs will lead to uncontrolled costs of high value-added services. This makes it impossible for existing technologies to effectively reduce regulatory costs without infringing on user privacy.

[0009] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0010] The main objective of this invention is to provide a compliance assessment method for home service processes based on multi-source data driven by AI and IoT, aiming to solve the technical problem of the difficulty in regulating existing home services.

[0011] To achieve the above objectives, this invention provides a compliance assessment method for home service processes driven by multi-source data based on AI and IoT, comprising: collecting actual basic data, including IoT data and service duration; setting standard data, including standard service geographical range and standard service duration; calculating the overall trajectory non-compliance score Pout and the dwell time percentage Pkey based on IoT data and standard service geographical range; calculating the deviation rate Dtime based on service duration and standard service duration; constructing a feature vector F based on Pout, Pkey, and Dtime; calculating the dynamic weight W, and calculating the basic score Sbase based on F and W; calculating the confidence score C, and calculating the corrected score Scorrected based on the confidence score C and the basic score Sbase.

[0012] Furthermore, setting the standard service duration also includes the following steps: based on the service type, preset the basic standard duration Tstandard_base; the standard service duration is: Tstandard = Tstandard_base × (1 + α × Chealth + β × Cenv); Chealth is the health level adjustment coefficient; Cenv is the environmental complexity coefficient; α and β are weighting coefficients.

[0013] Furthermore, calculating the overall trajectory non-compliance score Pout also includes the following steps: based on IoT data, calculate the off-duty time of service personnel outside the standard service geographic area; calculate the percentage of off-duty time to service time to obtain Pout.

[0014] Furthermore, calculating the percentage of dwell time (Pkey) also includes the following steps: based on IoT data, calculate the operation time of service personnel within the standard service geographic area; calculate the percentage of operation time to service time to obtain the Pkey.

[0015] Furthermore, the deviation rate Dtime is:

[0016] Tactual refers to the service duration.

[0017] Furthermore, calculating the dynamic weights W also includes the following steps: Preprocess Pout, Pkey, and Dtime to obtain Pout', Pkey', and Dtime'; Calculate the pairwise correlation coefficients r12, r13, and r23 for Pout', Pkey', and Dtime';

[0018]

[0019]

[0020] Calculate W1, W2, and W3 based on r12, r13, and r23;

[0021]

[0022]

[0023] S is based on: .

[0024] Furthermore, the confidence level C includes a feature reasonableness check C1: when At that time, C1=1; when hour,

[0025] In other cases

[0026] K is the scene adaptation coefficient; Based on C1 and S, the adjusted score S1 is calculated as follows: S1 = Sbase × C1.

[0027] Furthermore, the confidence level C also includes a data consistency check C2:

[0028] When C2 < 0, take C2 = 0; Calculate the adjusted fraction S2 based on C2 and S1; S2 = S1 × (1 + C2) / 2.

[0029] Furthermore, the confidence level C also includes scenario adaptability verification C3: When the lower limit of the reasonable score for the scenario is less than or equal to S2 and the upper limit of the reasonable score for the scenario is less than or equal to the upper limit of the reasonable score for the scenario, C3 = 1; When S2 is greater than the upper limit of the reasonable score for the scene and (S2 - upper limit of the reasonable score for the scene) ≤ 10

[0030] In other cases C3= ; The corrected score S is calculated based on C3 and S2. Scorrection = S2 × C3.

[0031] Compared with the prior art, the present invention has the following advantages: This invention clarifies an objective scoring standard and provides objective quality features that can be used for AI training, thereby enabling standardization, real-time processing, and scalability in conjunction with AI models. Simultaneously, it eliminates the need for monitoring equipment such as cameras. On one hand, this aligns with the application scenario where the service process takes place in the user's private space, thus reducing the risk of infringing on user privacy. On the other hand, it effectively reduces monitoring costs.

[0032] This invention calculates dynamic weights based on the pairwise correlation of three features, so that the weight allocation fits the actual data features of each service, rather than being mechanically fixed, thereby effectively preventing unreasonable weight allocation and distorted evaluation results.

[0033] Compared to the single model of "weighted summation to obtain a total score, and then determining the grade based on the total score," this invention adopts a "basic score + hierarchical confidence verification" model. First, a basic service quality score is calculated, and then the score is corrected through a three-layer confidence verification process: feature rationality verification, data consistency verification, and scenario adaptability verification. The final output is a quality grade with confidence levels. On the one hand, this improves the reliability of the evaluation results, thereby reducing the impact of abnormal data on the final result. On the other hand, it effectively improves interpretability, thus providing a clear explanation for why a particular final score was given.

[0034] Compared to common technologies that only output a quality rating, this solution, after tiered verification, can trace the source of anomalies for services with a confidence level below 0.8 or a score below 70, identify the anomaly type and cause, provide corrective suggestions, and adjust the final score. This effectively explains the reasons for low scores, thus facilitating service improvement. Attached Figure Description

[0035] Figure 1 Overall process diagram.

[0036] Figure 2 Overall hardware configuration diagram.

[0037] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0038] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0039] This invention provides a method for compliance assessment of home service processes based on multi-source data driven by AI and IoT, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the process of the present invention. It includes: collecting actual basic data, including IoT data and service duration; setting standard data, including standard service geographical range and standard service duration; calculating the overall trajectory non-compliance score Pout and the dwell time percentage Pkey based on the IoT data and standard service geographical range; calculating the deviation rate Dtime based on the service duration and standard service duration; constructing a feature vector F based on Pout, Pkey, and Dtime; calculating the dynamic weight W, and calculating the basic score Sbase based on F and W; calculating the confidence level C, and calculating the corrected score Scorrected based on the confidence level C and the basic score Sbase.

[0040] Specifically, in practice, service personnel collect their real-time geographical location by wearing IoT devices with GPS or Bluetooth beacon positioning capabilities or using a mobile terminal app with integrated positioning functionality while providing services. This generates IoT data. On the other hand, service personnel clock in and out via the app at the start and end of the service. They also clock in at key task points, such as disinfection preparation, initiating bathing assistance, and completing cleaning. This generates the service duration (Tactual). At this point, the collection of basic data is complete.

[0041] For different service types, such as assisted bathing, rehabilitation care, and home cleaning, a preset base service duration Tstandard_base is established. Simultaneously, a health level adjustment coefficient Chealth is set based on the user's health record, such as mild or moderate disability. The value range of Chealth is 0.0 to 0.3. An environmental complexity coefficient Cenv is set based on the complexity of the service environment, such as bathroom size and facilities. The value range of Cenv is 0.0 to 0.2. Furthermore, weighting coefficients α and β are set. α and β can be adjusted according to actual needs. Therefore, the standard service duration is Tstandard = Tstandard_base × (1 + α × Chealth + β × Cenv).

[0042] On the other hand, the standard service geographic area is determined using one of the following two methods. The first method involves the user or service personnel uploading a floor plan of their home via an app before the initial service, marking key functional areas such as the bathroom, bedroom, and kitchen. The second method involves the service personnel using a mobile device to pair with Bluetooth beacons pre-placed in key areas before the service begins, automatically identifying and recording the area's coordinates. In summary, the aforementioned ranges are combined and extended outwards by a 1-meter buffer zone. This determines the standard service geographic area.

[0043] Once the standard service duration and standard service geographic range are determined, the off-duty time of service personnel outside the standard service geographic range is calculated based on IoT data. The percentage of off-duty time to service duration is calculated to obtain Pout. The operation time of service personnel within the standard service geographic range is calculated based on IoT data. The percentage of operation time to service duration is calculated to obtain Pkey. Simultaneously, the deviation rate Dtime is calculated based on the service duration Tactual and the standard service duration Tstandard. The deviation rate Dtime is:

[0044] At this point, the feature vector F can be constructed based on Pout, Pkey, and Dtime.

[0045] A scenario adaptation coefficient K is set, assigned based on the type of home-based elderly care service, with different K values ​​for different service types. Simultaneously, an AI model is used for dynamic iteration combined with actual operational data. For example, home-based elderly care services are categorized into basic daily care, health care, and emotional support. The K values ​​are set as follows: Basic daily care (e.g., cleaning, meal delivery): K=0.8 (high requirements for trajectory compliance, moderate tolerance for duration deviation). Health care (e.g., blood pressure measurement, medication guidance): K=1.2 (strict requirements for dwell time percentage and duration deviation, moderate requirements for trajectory compliance). Emotional support (e.g., counseling, companionship): K=0.6 (lower requirements for trajectory compliance, higher tolerance for duration deviation). A trajectory weight correction coefficient α1 is set, specifically a fixed value of 0.95, to correct minor errors in IoT trajectory data. A dwell time weight correction coefficient β1 is set, specifically a fixed value of 1.05, to strengthen the weight of dwell time at key locations, aligning with the requirement of "key services being provided effectively" in home-based elderly care services. The duration deviation weighting correction coefficient γ is set to a fixed value of 1.0 to ensure the objectivity of the duration deviation and avoid excessive amplification or reduction of the deviation. Preferably, the above value can be adjusted using an AI model combined with actual operating data. Therefore, Pout, Pkey, and Dtime are preprocessed to obtain Pout', Pkey', and Dtime'. Specifically: . .

[0046] For ease of understanding, we define the service type as healthcare and assign the following values: Pout = 0.15, Pkey = 0.78, and Dtime = -0.12. The negative sign of Dtime indicates that the service duration Tactual is shorter than the standard service duration Tstandard. Therefore, the feature vector F is F = [0.15, 0.78, -0.12]. After the aforementioned preprocessing calibration, Pout' = 0.15 × 1.2 × 0.95 = 0.171. Pkey' = 0.78 × 1.2 × 1.05 = 0.9828. Dtime' = |-0.12| × 1.2 × 1.0 = 0.144. Therefore, the calibrated feature vector F' = [0.171, 0.9828, 0.144].

[0047] Calculate the pairwise correlation coefficients r12, r13, and r23 for Pout', Pkey', and Dtime'. Details are as follows:

[0048]

[0049]

[0050] Among them, r12, r13, and r23 are the correlation coefficients between the three features, with values ​​ranging from [0,1]. The closer the coefficient is to 1, the stronger the correlation between the two features. Adding 0.01 to the denominator is to avoid the denominator being 0 and to improve the stability of the formula.

[0051] Calculate W1, W2, and W3 based on r12, r13, and r23. Details are as follows:

[0052]

[0053]

[0054] W1, W2, and W3 have values ​​ranging from [0,1], satisfying W1+W2+W3=1. The core logic is: the weaker the correlation between a feature and the other two features, the higher its weight, because features with weak correlation can more independently reflect a certain dimension of service quality and need to be focused on. Conversely, the stronger the correlation, the lower the weight, to avoid double scoring.

[0055] Then the basis of S is:

[0056] In summary, this invention fully considers the correlation between the three features: Pout, Pkey, and Dtime. For example, non-compliant trajectories may lead to abnormal dwell time percentages, and dwell time deviations may be related to trajectory compliance. Therefore, this invention calculates dynamic weights based on the pairwise correlations of the three features, ensuring that the weight allocation aligns with the actual data characteristics of each service, rather than being mechanically fixed. This effectively prevents unreasonable weight allocation and distorted evaluation results.

[0057] The following example will still use F'=[0.171, 0.9828, 0.144].

[0058]

[0059]

[0060]

[0061] The numerator of the dynamic weight W is:

[0062]

[0063]

[0064] The denominator is: 0.4835 + 0.8315 + 0.497 = 1.812. Therefore, the final weight value is:

[0065]

[0066]

[0067] Verification: 0.267 + 0.459 + 0.274 = 1.0, which meets the weighting requirements. Therefore, the dynamic weight W for this service is [0.267, 0.459, 0.274]. It is evident that Pkey' (percentage of time spent after calibration) has the highest weight (0.459) because it has the weakest correlation with the other two features and can independently reflect whether service personnel spend sufficient time at key locations (such as the elderly person's bedside or medication station), aligning with the core needs of healthcare services.

[0068] On the other hand, since Pout' and Dtime' are "negative indicators," the larger the value, the worse the service quality; while Pkey' is a "positive indicator," the larger the value, the better the service quality. Therefore, the S base adopts a "reverse scoring + weighted summation" model. Specifically, the S base value ranges from [0, 100], and the higher the score, the better the service quality. (1 - Pout') is the reverse scoring of Pout', which transforms the negative indicator into a positive indicator. The closer 1 - Pout' is to 1, the better the trajectory compliance. Pkey' is the direct scoring of the positive indicator; the larger the value, the stronger the rationality of the stay. 1 - Dtime' is the reverse scoring of Dtime', which transforms the negative indicator into a positive indicator. The closer 1 - Dtime' is to 1, the smaller the duration deviation. Based on the previous calculation results, F'=[0.171, 0.9828, 0.144], W=[0.267, 0.459, 0.274], K=1.2. The value of S_base is as follows: S_base = [(1-0.171)×0.267+0.9828×0.459+(1-0.144)×0.274]×100=90.7.

[0069] Feature rationality verification C1 is used to verify whether a single feature meets the scenario expectations. Details are as follows: when At that time, C1=1; when hour,

[0070] In other cases

[0071] Based on the scenario adaptation coefficient K, a reasonable range is set for each feature. The greater the deviation from the reasonable range, the lower the confidence level C1. Therefore, S1 = S_base × C1. If C1 ≥ 0.8, no correction is made; if 0.5 ≤ C1 < 0.8, the score is corrected; if C1 < 0.5, it is marked as "pending review".

[0072] Data consistency check C2 is used to verify the consistency between IoT data and service duration data. The overall trajectory non-compliance rate Pout', the dwell time percentage Pkey', and the deviation rate Dtime' of service personnel should be logically consistent. For example, when the trajectory is severely non-compliant, the dwell time percentage should be low, and the duration deviation should be large. If a logical contradiction occurs, such as a non-compliant trajectory but an extremely high dwell time percentage, the confidence level should be lowered. Specifically:

[0073] If C2 < 0, then C2 = 0. The score correction formula is then S2 = S1 × (1 + C2) / 2. This balances the base score with the consistency check result, preventing score distortion caused by a single feature anomaly.

[0074] Scenario suitability verification (C3) is used to verify the degree of matching between the score and the service scenario. Different types of home-based elderly care services have different reasonable ranges for quality scores. For example, health care services should generally score higher than basic daily living care services. If the score exceeds the reasonable range for the scenario, the confidence level is lowered. Details are as follows: When the lower limit of the reasonable score for the scenario is less than or equal to S2 and the upper limit of the reasonable score for the scenario is less than or equal to the upper limit of the reasonable score for the scenario, C3 = 1; When S2 is greater than the upper limit of the reasonable score for the scene and (S2 - upper limit of the reasonable score for the scene) ≤ 10

[0075] In other cases C3= ; For ease of understanding, the following provides examples of scenario scoring parameters. Basic daily care: Reasonable range [70, 90], average score 80, standard deviation 5. Health care: Reasonable range [75, 95], average score 85, standard deviation 4. Emotional support: Reasonable range [65, 85], average score 75, standard deviation 6. The score correction formula is: Scorrected = S2 × C3. This score is the final corrected score, with a value range of [0, 100].

[0076] The following calculation example uses the aforementioned calculation results. The specific calculation process is as follows: Perform feature validity verification: For the health care category, K=1.2, and the reasonable ranges for each feature are: Pout' ≤ 1.2×0.3=0.36; Pkey' ≥ 1.2×0.6=0.72; Dtime' ≤ 1.2×0.2=0.24. Currently, F'=[0.171, 0.9828, 0.144], all of which meet the reasonable range. Therefore, C1=1, S1=90.7×1=90.7 points.

[0077] Perform data consistency verification: C2=1-|(1-0.171)-0.9828|-|(1-0.171)-(1-0.144)|=0.8192.

[0078] S2 = 90.7 × (1 + 0.8192) / 2 ≈ 90.7 × 0.9096 ≈ 82.6 points.

[0079] Perform scene adaptability verification: Health Care Category: Reasonable range [75, 95], average score 85, standard deviation 4; S2 = 82.6 points, within the reasonable range, therefore C3 = 1. S correction = 82.6 × 1 = 82.6 points. Subsequently, the average of C1, C2, and C3 is verified: C = (C1 + C2 + C3) / 3 ≈ (1 + 0.8192 + 1) / 3 ≈ 0.94. The confidence level is high, indicating that the evaluation results are reliable.

[0080] In summary, compared to the single model of "weighted summation to obtain a total score, and then determining the grade based on the total score," this invention adopts a "basic score + hierarchical confidence verification" model. It first calculates a basic service quality score, and then corrects the score through a three-layer confidence verification process: feature rationality verification, data consistency verification, and scenario adaptability verification. The final output is a quality grade with confidence. On the one hand, this improves the reliability of the evaluation results, thereby reducing the impact of abnormal data on the final result. On the other hand, it effectively improves interpretability, thus providing a clear explanation for why a particular final score was given. Simultaneously, it facilitates anomaly tracing when data anomalies occur. The specific tracing method is as follows: If C1 < 0.8: The anomaly type is "feature anomaly," and the cause is that a certain feature exceeds the reasonable range of the scenario. For example, if Pout' is too high, it indicates that the trajectory is not compliant; if Dtime' is too high, it indicates that the duration deviation is too large.

[0081] If C2 < 0.8: the anomaly type is "data contradiction," and the root cause is logical inconsistency between features. For example, if Pout' is too high but Pkey' is too high, there may be IoT data forgery or false alarms.

[0082] If C3 < 0.8: The exception type is "scenario mismatch," and the reason is that the score exceeds the reasonable range for the current service type. For example, if the score for the health care category is too low, there may be missing service items.

[0083] Based on the above source tracing results, the S-correction is adjusted for the final result. For example, if the data is contradictory, deduct 5-10 points; if the features are abnormal, deduct 3-8 points, and output "Final Score + Quality Level + Confidence Score + Reason for Anomaly + Improvement Suggestions". The specific quality level classification is as follows: Excellent: S-correction ≥ 85 points, confidence level ≥ 0.9.

[0084] Good: 75 ≤ S-correction < 85 points, confidence level ≥ 0.8.

[0085] Acceptable: 65 ≤ S-correction < 75 points, confidence level ≥ 0.7.

[0086] Unacceptable: S-correction < 65 points, or confidence level < 0.7 (requires manual review).

[0087] Therefore, compared to common technologies that only output a quality rating, this solution, after tiered verification, can trace the source of anomalies for services with a confidence level below 0.8 or a score below 70, identify the anomaly type and cause, provide corrective suggestions, and adjust the final score. This effectively explains the reasons for low scores, thus facilitating service improvement.

[0088] In summary, this invention clarifies an objective scoring standard and provides objective quality features that can be used for AI training, thereby enabling standardization, real-time processing, and scalability in conjunction with AI models. Simultaneously, it eliminates the need for monitoring equipment such as cameras. On one hand, this aligns with the application scenario where the service process takes place in the user's private space, thus reducing the risk of infringing on user privacy. On the other hand, it effectively reduces monitoring costs.

[0089] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0090] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0092] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

[0093] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

Claims

1. A compliance assessment method for home service processes based on multi-source data driven by AI and IoT, characterized in that, include: Collect actual basic data, including IoT data and service duration; Set standard data, which includes the standard service geographical range and standard service duration; Based on the IoT data and the geographical range of the standard service, calculate the overall trajectory non-compliance score Pout and the dwell time percentage Pkey. The deviation rate Dtime is calculated based on the service duration and the standard service duration. Based on Pout, Pkey, and Dtime, construct the feature vector F; Calculate the dynamic weight W, and calculate the basic score S based on F and W; Calculate the confidence level C, and based on the confidence level C and the base score S, calculate the corrected score S.

2. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 1, characterized in that, Setting the standard service duration also includes the following steps: Based on the service type, a preset base standard duration Tstandard_base is defined; The standard service duration is: Tstandard=Tstandard_base ×(1+α×Chealth+β×Cenv); Chealth is the health level adjustment factor; Cenv is the environment complexity coefficient; α and β are weighting coefficients.

3. The compliance assessment method for home service processes based on multi-source data driven by AI and IoT as described in claim 1, characterized in that, Calculating the overall trajectory non-compliance Pout also includes the following steps: Based on the IoT data, the standard service geographic range is used to calculate the off-duty time of service personnel outside the standard service geographic range; Calculate the percentage of the disconnection time relative to the service time to obtain the Pout.

4. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 1, characterized in that, Calculating the percentage of dwell time Pkey also includes the following steps: Based on the IoT data, the standard service geographic range is used to calculate the operation time of service personnel within the standard service geographic range; Calculate the percentage of the operation time to the service time to obtain the Pkey.

5. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 1, characterized in that, The deviation rate Dtime is: The term "Tactual" refers to the service duration.

6. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 1, characterized in that, Calculating the dynamic weight W further includes the following steps: Preprocess the Pout, Pkey, and Dtime to obtain Pout', Pkey', and Dtime'; Calculate the pairwise correlation coefficients r12, r13, and r23 of Pout', Pkey', and Dtime'; Calculate W1, W2, and W3 based on r12, r13, and r23; The S-based foundation is: .

7. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 6, characterized in that, The confidence level C includes feature rationality verification C1: when At that time, C1=1; when hour, In other cases K is the scene adaptation coefficient; Based on C1 and S, the adjusted score S1 is calculated: S1 = Sbase × C1.

8. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 7, characterized in that, The confidence level C also includes a data consistency check C2: When C2 < 0, take C2 = 0; Calculate the adjustment score S2 based on C2 and S1; S2 = S1 × (1 + C2) / 2.

9. The method for compliance assessment of home service processes based on multi-source data driven by AI and IoT as described in claim 8, characterized in that, The confidence level C also includes a scene adaptability check C3: When the lower limit of the reasonable score for the scenario is less than or equal to S2 and the upper limit of the reasonable score for the scenario is less than or equal to the upper limit of the reasonable score for the scenario, C3 = 1; When S2 is greater than the upper limit of the reasonable score for the scene and (S2 - upper limit of the reasonable score for the scene) ≤ 10 In other cases C3= ； The correction score S is calculated based on C3 and S2. Scorrection = S2 × C3.

Images