Household service quality evaluation model construction method

Abstract

The invention discloses a housekeeping service quality evaluation model construction method. The method comprises the following steps: optimizing an evaluation index system: establishing a full-scene index framework containing a basic security dimension (20% weight), a process-result double-track dimension and a service suitability dimension; according to the dynamic weight optimization mechanism, weights of a client side (a CRITIC algorithm), a server side (a skill authentication level) and a platform side (an AHP method and an expert database) are fused, and weight floating is automatically triggered based on a service scene (including old people / infants); multi-source data acquisition and privacy protection: replacing silent customer feedback with Internet of Things equipment (a smart bracelet track), voice interaction records and behavior data, and performing data desensitization by adopting regional coding conversion and hash + salt value encryption; the scene adaptation model architecture is combined with an improved fuzzy comprehensive evaluation method (three-level membership function), a random forest regression device, an attention mechanism neural network and a transfer learning module; and dynamic optimization and anti-cheating are realized.

Description

Technical Field

[0001] This invention relates to the field of domestic service technology, specifically a method for constructing a domestic service quality assessment model. Background Technology

[0002] There are many significant pain points in the current field of domestic service quality assessment, which restrict the standardized development of the industry. The existing assessment system generally suffers from single and static dimensions, relying heavily on subjective customer ratings or fixed indicators, and lacks systematic consideration of basic safety dimensions. Key indicators such as background checks of service personnel and compliance of tool disinfection are often overlooked. For special service scenarios such as the elderly and infants, a dynamic weight adjustment mechanism has not been established, resulting in insufficient safety assessment weights for high-risk scenarios, making it difficult to guarantee service safety.

[0003] There are two dilemmas in data collection: on the one hand, there is an over-reliance on proactive customer feedback, and the lack of data on silent customers leads to sample bias, especially for low-frequency service types such as villa cleaning, which are difficult to accurately assess due to insufficient data; on the other hand, the ability to integrate multi-source data is weak, behavioral data such as IoT device trajectories and voice interaction records are not effectively utilized, and privacy protection measures are rudimentary, with the risk of leakage when raw data is used directly, which restricts the breadth of data application.

[0004] Insufficient model adaptability and timeliness have become technical bottlenecks. Traditional model architectures are rigid, and adding new service types requires tens of thousands of labeled data to support retraining. The response cycle to new industry regulations can take several months. Anti-fraud mechanisms are generally weak, and the identification of fake positive reviews is based solely on simple text duplication rate detection. There is a lack of cross-validation of multi-dimensional behaviors such as service trajectory and order rejection patterns, which leads to frequent problems such as false reviews and abnormal operations, seriously affecting the credibility of the evaluation.

[0005] At the commercial implementation level, there are shortcomings in user experience, insufficient transparency in the scoring system, and customers' inability to obtain specific details of deductions, which reduces trust. At the same time, the model algorithm is highly closed and lacks standardized API interfaces to support on-demand calls, which restricts the release of commercial value. These problems together make it difficult for the existing assessment system to meet the development needs of refined, safe and dynamic domestic services, and there is an urgent need to build a more comprehensive assessment solution. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method for constructing a domestic service quality assessment model. By building a dynamic assessment system covering all scenarios, integrating multi-source data while protecting privacy, and optimizing model adaptability and anti-cheating mechanisms, it solves the problems of existing assessments such as single dimensions, insufficient data, rigid models, and frequent cheating, achieving comprehensive, accurate, safe, efficient, and commercially viable assessments.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a domestic service quality assessment model, comprising:

[0010] Optimize the evaluation index system: Establish a full-scenario index framework that includes a basic security dimension (20% weight), a process-result dual-track dimension, and a service adaptability dimension;

[0011] Dynamic weight optimization mechanism: It integrates the weights of the client (CRITIC algorithm), the server (skill certification level), and the platform (AHP method + expert database), and automatically triggers weight increase based on the service scenario (including the elderly / infants);

[0012] Multi-source data collection and privacy protection: Replace silent customer feedback with IoT devices (smart bracelet tracks), voice interaction records, and behavioral data, and use regional coding conversion and hash + salt value encryption for data anonymization;

[0013] Scene adaptation model architecture: combining improved fuzzy comprehensive evaluation method (three-level membership function), random forest regressor, attention mechanism neural network and transfer learning module;

[0014] Dynamic optimization and anti-fraud: The model is updated based on the complaint rate / score threshold, and fraudulent behavior is identified through anomaly detection and cross-validation of third-party data.

[0015] Preferably, the basic security dimensions include:

[0016] Service personnel background check pass rate (integration with the public security system);

[0017] Service tool disinfection compliance rate (regular disinfection records);

[0018] Customer privacy protection compliance (prohibition of environmental surveillance filming).

[0019] Preferably, in the dynamic weight optimization mechanism:

[0020] When the service scenario includes "elderly people over 70 years old" or "infants aged 0-3 years", the weight of "safety compliance" will automatically increase by 15%.

[0021] When service personnel hold advanced skills certificates, the weight of related indicators increases by 10%.

[0022] Preferably, the data acquisition includes:

[0023] The service personnel's mobile app automatically records their operation trajectory (GPS movement trajectory is used to determine cleaning coverage);

[0024] The SMOTE algorithm is used to expand the sample of low-frequency service types (such as villa cleaning).

[0025] Preferably, in the model architecture:

[0026] The transfer learning module is based on a pre-trained model, and adding a new service type only requires 5,000 fine-tuning data points;

[0027] Daily incremental learning incorporates the previous day's data (updates are completed within 1 hour), and quarterly full-scale retraining responds to new industry regulations.

[0028] Preferably, the anti-cheating mechanism includes:

[0029] Identify repeated positive reviews from the same device within a short period of time (text repetition rate ≥ 80%);

[0030] Testing service personnel abnormally refuse orders (the proportion of orders refused over a distance of ≥3 kilometers is abnormal);

[0031] The platform conducts triangulation of self-assessment / customer ratings based on random data sampling.

[0032] Preferred options also include localization and adaptation solutions:

[0033] The client-side APP displays a breakdown of the scoring (e.g., "Safety compliance +10 points, efficiency -5 points") and records of specific deductions.

[0034] The core algorithm of the model is encapsulated as an API interface, supporting pay-per-use (0.1 yuan / call);

[0035] Service personnel upload photos of the cleaned areas via a WeChat mini-program, which automatically identifies the level of cleanliness.

[0036] A domestic service quality assessment system, characterized in that it includes:

[0037] Dynamic indicator configuration module: Automatically adjusts weights and thresholds according to the scenario;

[0038] Multi-source data acquisition engine: integrates IoT devices, voice interaction, and behavioral substitution data;

[0039] Federated learning interface: Supports privacy-preserving training with cross-institutional data sharing;

[0040] Anti-cheating verification module: Real-time analysis and evaluation of the compliance of data and behavioral patterns.

[0041] (III) Beneficial Effects

[0042] Compared with existing technologies, this invention provides a method for constructing a domestic service quality assessment model, which has the following beneficial effects:

[0043] 1. The method for constructing the domestic service quality assessment model improves the comprehensiveness and accuracy of the assessment: It constructs a multi-dimensional, full-scenario indicator framework, combined with a dynamic weight optimization mechanism, and adjusts the weights for special service scenarios (such as elderly and infant care) and highly skilled service personnel, making the assessment more in line with actual needs; multi-source data collection and sample expansion technology enhances data coverage and assessment accuracy.

[0044] 2. The construction method of this domestic service quality assessment model strengthens security and privacy protection: the basic security dimension clearly defines multiple security indicators, combined with strict data anonymization, privacy protection implementation measures and federated learning interface, to effectively protect customer privacy and cross-institutional data security while ensuring service security.

[0045] 3. The construction method of this domestic service quality assessment model enhances the model's adaptability and timeliness: The scenario-adaptive model architecture uses a variety of advanced algorithms, and the transfer learning and incremental learning mechanisms enable the model to quickly adapt to new service types, respond promptly to new industry regulations, and ensure the timeliness and industry adaptability of the assessment.

[0046] 4. The construction method of this domestic service quality assessment model ensures the fairness and reliability of the assessment: a multi-layered anti-cheating mechanism effectively prevents cheating by identifying abnormal behavior and performing data cross-validation, ensuring the fairness of the assessment results; triangulation and other methods further improve the reliability of the evaluation data. Attached Figure Description

[0047] Figure 1 This is a flowchart of the main system deployment of the method for constructing a domestic service quality assessment model proposed in this invention.

[0048] Figure 2 The flowchart of the dynamic weight optimization system for the method of constructing a domestic service quality assessment model proposed in this invention is shown below.

[0049] Figure 3 This invention presents a flowchart of a multi-source data acquisition and de-identification system for a method of constructing a domestic service quality assessment model.

[0050] Figure 4 The flowchart of the anti-cheating and dynamic optimization system for the domestic service quality assessment model construction method proposed in this invention is shown below.

[0051] Figure 5 This is an API gateway flowchart for a method of constructing a domestic service quality assessment model proposed in this invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figures 1 to 5 The present invention provides a technical solution: a method for constructing a domestic service quality assessment model.

[0054] (I) Establishment of the evaluation indicator system:

[0055] 1. Refinement of basic security dimensions:

[0056] We connect with the public security system to build a background check database for service personnel, accurately screening for criminal records through ID cards and fingerprint recognition; we install IoT disinfection equipment to upload disinfection time and concentration data in real time, ensuring accurate statistics on tool disinfection compliance rates; and we deploy privacy-protecting surveillance in key areas of customers' homes to prohibit service personnel from taking photos, with violators receiving direct point deductions.

[0057] 2. Process-Result Dual-Track Design:

[0058] 2.1 Process dimension: Service personnel's real-time location trajectory via mobile APP, combined with service duration to determine cleaning coverage; voice interaction records are analyzed for tone and keywords to identify service attitude issues.

[0059] 2.2 Results Dimension: Photos of the cleaned area are taken before and after the service. Image recognition is used to compare stain residue and item placement to quantify service effectiveness.

[0060] 3. Service adaptability dimension construction:

[0061] Based on the client's family structure (ratio of elderly to children), house size, and decoration style, we match suitable service packages; we analyze historical service data to recommend value-added services for high-demand clients, such as regular deep disinfection for elderly households.

[0062] (II) Dynamic Weight Optimization

[0063] 1. Client weight calculation:

[0064] We collect data such as customer historical ratings, service type preferences, and complaint records, and use the CRITIC algorithm to calculate the weight of different indicators on customer satisfaction. For example, younger customers value efficiency more, so the weight allocation is tilted towards cleaning speed.

[0065] 2. Server-side weight allocation:

[0066] Establish a service personnel skills certification database, with those holding advanced certificates receiving a 10% increase in weighting for professional service indicators; dynamically adjust weightings based on service personnel's monthly performance and customer complaint rates, with a 5% increase in weighting for three consecutive months of zero complaints.

[0067] 3. Platform-side weight determination:

[0068] A team of experts in the domestic service industry was organized to construct a hierarchical analysis model using the AHP method. Service quality, safety standards, and customer satisfaction were used as the criteria layer, while specific service indicators were used as the solution layer. The weights were determined through questionnaires and expert scoring. When the service scenario involves special groups, an automatic weighting mechanism for safety standards was triggered.

[0069] (III) Multi-source data collection and privacy protection

[0070] 1. Data Acquisition Expansion

[0071] The service personnel's mobile app integrates GPS and accelerometer sensors to collect cleaning movement trajectories and dwell time, accurately determining cleaning coverage; intelligent voice collection devices are deployed, and key information from service exchanges is extracted after anonymization to identify responses to special customer needs; and water and electricity meter data are accessed with customer authorization to verify whether water and electricity usage has returned to normal after cleaning.

[0072] 2. Enhanced privacy protection

[0073] Sensitive information such as customer home address and contact information is encrypted and stored using the national cryptographic algorithm SM4; video and audio data collected during the service process are converted into feature vectors in real time, the original files are discarded, and only feature data that cannot be restored to privacy is retained for analysis; when sharing data across platforms, federated learning technology is used to jointly train models without exposing the original data.

[0074] (iv) Scenario Adaptation Model Architecture

[0075] 1. Application of the improved fuzzy comprehensive evaluation method

[0076] To assess the cleaning effectiveness of key areas such as kitchens and bathrooms, a three-level membership function is constructed: Level 1 is the degree of stain residue (0-30%), Level 2 is the neatness of item placement (30%-70%), and Level 3 is the absence of obvious odor (70%-100%). The cleaning quality level is comprehensively evaluated through membership function calculation.

[0077] 2. Random Forest Regressor Tuning

[0078] Historical service data, including service duration, customer ratings, and cleaning area, are collected as feature variables, and the complaint rate is used as the target variable to train a random forest regression model. The dataset is updated regularly, and hyperparameters such as the number and depth of trees are optimized to improve the model's accuracy in predicting service complaints.

[0079] 3. Attention Mechanism Neural Network Design

[0080] A Transformer-based neural network model is constructed, which takes the key behavioral sequences in the service process (such as disinfection steps and cleaning order) as input. During the model training process, the model adaptively learns the weights of the impact of different behaviors on service quality, focusing on key factors affecting service quality.

[0081] (V) Dynamic Optimization and Anti-Cheating Mechanism

[0082] 1. Model update strategy

[0083] A complaint rate threshold of 5% is set. If the complaint rate exceeds the threshold for a consecutive week, the model will be automatically retrained. Service data from the previous day is extracted from the database at midnight every day. Through incremental learning algorithms, the model parameters are fine-tuned within 1 hour to ensure that the model can adapt to changes in service scenarios in a timely manner.

[0084] 2. Enhanced anti-cheating algorithm

[0085] Integrating device fingerprint recognition technology, it accurately identifies repeated positive reviews from the same device within a short period of time; it constructs service personnel behavior profiles, analyzes the characteristics of order rejection time, location, and frequency, and marks abnormal order rejection patterns; it regularly obtains real service feedback from public security and communities, cross-verifies it with platform data, and eliminates false reviews.

[0086] (vi) Commercial Implementation

[0087] 1. Client-side APP development

[0088] The system features a visual scoring breakdown interface, displaying scores for various service quality indicators in radar chart format. A new "Service Report" function has been added, allowing customers to view details of past services and improvement trends. Online customer service is also integrated, enabling real-time communication when customers have questions about the scoring.

[0089] 2. API Interface Service Deployment

[0090] The model algorithm is encapsulated using Docker containers and deployed on Alibaba Cloud servers to support high-concurrency API calls; an API usage monitoring system is developed to collect real-time statistics on call volume, response time, and other metrics to ensure service quality; and a data security agreement is signed with the housekeeping platform to ensure data security during the call process.

[0091] 3. Optimization of the service personnel mini-program

[0092] An image recognition module is integrated into the WeChat mini program. After service personnel upload cleaning photos, stains and item locations are automatically marked, and a cleanliness assessment report is generated. An incentive mechanism is designed to reward service personnel with points based on the photo approval rate, which can be redeemed for training courses.

[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus 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 apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a domestic service quality assessment model, characterized in that, include: Optimize the evaluation index system: Establish a full-scenario index framework that includes a basic security dimension (20% weight), a process-result dual-track dimension, and a service adaptability dimension; Dynamic weight optimization mechanism: It integrates the weights of the client (CRITIC algorithm), the server (skill certification level), and the platform (AHP method + expert database), and automatically triggers weight increase based on the service scenario (including the elderly / infants); Multi-source data collection and privacy protection: Replace silent customer feedback with IoT devices (smart bracelet tracks), voice interaction records, and behavioral data, and use regional coding conversion and hash + salt value encryption for data anonymization; Scene adaptation model architecture: combining improved fuzzy comprehensive evaluation method (three-level membership function), random forest regressor, attention mechanism neural network and transfer learning module; Dynamic optimization and anti-fraud: The model is updated based on the complaint rate / score threshold, and fraudulent behavior is identified through anomaly detection and cross-validation of third-party data.

2. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: The basic security dimensions include: Service personnel background check pass rate (integration with the public security system); Service tool disinfection compliance rate (regular disinfection records); Customer privacy protection compliance (prohibition of environmental surveillance filming).

3. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: In the dynamic weight optimization mechanism: When the service scenario includes "elderly people over 70 years old" or "infants aged 0-3 years", the weight of "safety compliance" will automatically increase by 15%. When service personnel hold advanced skills certificates, the weight of related indicators increases by 10%.

4. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: The data collection includes: The service personnel's mobile app automatically records their operation trajectory (GPS movement trajectory is used to determine cleaning coverage); The SMOTE algorithm is used to expand the sample of low-frequency service types (such as villa cleaning).

5. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: In the model architecture: The transfer learning module is based on a pre-trained model, and adding a new service type only requires 5,000 fine-tuning data points; Daily incremental learning incorporates the previous day's data (updates are completed within 1 hour), and quarterly full-scale retraining responds to new industry regulations.

6. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: The anti-cheating mechanism includes: Identify repeated positive reviews from the same device within a short period of time (text repetition rate ≥ 80%); Testing service personnel abnormally refuse orders (the proportion of orders refused over a distance of ≥3 kilometers is abnormal); The platform conducts triangulation of self-assessment / customer ratings based on random data sampling.

7. The method for constructing a domestic service quality assessment model according to claim 1, characterized in that: It also includes a deployment and adaptation solution: The client-side APP displays a breakdown of the scoring (e.g., "Safety compliance +10 points, efficiency -5 points") and records of specific deductions. The core algorithm of the model is encapsulated as an API interface, supporting pay-per-use (0.1 yuan / call); Service personnel upload photos of the cleaned areas via a WeChat mini-program, which automatically identifies the level of cleanliness.

8. A domestic service quality assessment system implementing the method of any one of claims 1-7, characterized in that, include: Dynamic indicator configuration module: Automatically adjusts weights and thresholds according to the scenario; Multi-source data acquisition engine: integrates IoT devices, voice interaction, and behavioral substitution data; Federated learning interface: Supports privacy-preserving training with cross-institutional data sharing; Anti-cheating verification module: Real-time analysis and evaluation of the compliance of data and behavioral patterns.

Images