Multi-dimensional service quality evaluation method and system

Abstract

The invention discloses a multi-dimensional service quality evaluation method and system. The method comprises the following steps: collecting subjective feedback data of a client platform and objective operation data of an operation system; performing cleaning operation, duplicate removal operation and standardization processing operation on the collected subjective feedback data and objective operation data; loading a preset industry evaluation dimension template or creating a user-defined evaluation dimension; distributing an initial weight for the evaluation dimension, and dynamically adjusting the weight through a real-time iterative optimization mechanism; and calculating a comprehensive score based on the optimized weight, and generating a visual evaluation report, thereby improving the dynamic adaptability of the evaluation dimension, and ensuring that the result is real-time and accurate.

Description

Technical Field

[0001] This invention relates to the field of aviation services, and more particularly to a multi-dimensional service quality assessment method and system. Background Technology

[0002] Traditional service quality assessment systems generally suffer from structural defects when integrating multi-source heterogeneous data, especially in high-real-time demand fields such as civil aviation. These limitations manifest in three key areas: First, the heterogeneity between passenger feedback and operational system data leads to collection delays; for example, the structural differences between social media comments and flight logs make real-time parsing difficult. Second, static assessment dimension templates cannot adapt to emerging service scenarios; for instance, innovative businesses like drone freight lack effective evaluation standards. Third, cultural differences in global operations cause assessment biases; a typical example is the failure to quantify and incorporate differences in tolerance for flight delays across different regions. Existing technologies rely on manually configured rules and periodic data updates, which cannot dynamically expand the knowledge graph coverage or achieve adaptive weight adjustments across regions, ultimately causing assessment results to lag behind the evolution of service scenarios. Summary of the Invention

[0003] This invention proposes a multi-dimensional service quality assessment method, including: S1. Collect subjective feedback data from customer platforms and objective operational data from operating systems; S2. Perform cleaning, deduplication, and standardization operations on the collected subjective feedback data and objective operational data; S3. Load preset industry evaluation dimension templates or create user-defined evaluation dimensions; S4. Assign initial weights to the evaluation dimensions and dynamically adjust the weights through a real-time iterative optimization mechanism; S5. Calculate a comprehensive score based on the optimized weights and generate a visual evaluation report, thereby realizing multi-source data fusion evaluation and improving the comprehensiveness of service quality analysis.

[0004] S1 includes deploying an intelligent data source adapter generator to perform metadata learning operations to infer heterogeneous data source structures, generate lightweight edge adapters, and perform dynamic allocation of processing resources through a reinforcement learning priority scheduling mechanism, which can enhance heterogeneous data compatibility and optimize edge computing resource utilization.

[0005] Specifically, the creation of user-defined evaluation dimensions in S3 includes: performing multi-turn conversational intelligent wizards to identify user configuration intent operations; integrating voice and visual inputs to perform user intent parsing operations; performing intent deviation correction operations based on conversation history to generate custom dimension rules; simultaneously embedding a scenario simulation training interface to perform synthetic data generation operations to simulate configuration scenarios; pre-training the intent recognition model; and performing error correction enhancement operations through an error injection mechanism, thereby improving the accuracy of custom dimension creation and strengthening interactive fault tolerance.

[0006] Specifically, the knowledge graph self-update operation performed before S3 includes: performing new term frequency monitoring through streaming data processing, triggering a context-aware dynamic term base adaptive mechanism to perform real-time knowledge graph expansion; simultaneously integrating a reinforcement learning-driven hotspot detection model to perform high-dimensional sparse region identification, triggering missing data supplementation sampling through an adaptive sparse region identification mechanism, and deploying an attention-driven model to perform key dimension blind spot priority detection, thereby ensuring dynamic updates of the knowledge graph and proactively detecting data blind spots.

[0007] S3 also includes: launching the industry migration pre-training engine to perform industry rule feature extraction operations, automatically generalizing the evaluation template for emerging fields, and optimizing cross-industry dimension adaptation operations, thereby accelerating cross-industry template migration and improving the adaptability of emerging fields.

[0008] The real-time iterative optimization mechanism described in S4 includes: integrating an anomaly detection module to perform optimization parameter adjustment operations, coordinating multi-node historical data to perform federated weight smoothing operations, and calling a memory reinforcement module to store historical weight trends and perform smoothing algorithm application operations, which can suppress abnormal weight fluctuations and ensure the stability of the optimization process.

[0009] The real-time iterative optimization mechanism described in S4 also includes: adding a time series anomaly buffer module to perform continuous analysis of historical weight trends, performing anomaly fluctuation point detection operation through a sliding window algorithm, and injecting buffer delay to perform parameter smoothing operation in the optimization process, which can buffer the interference of sudden changes in time series data and smooth the parameter adjustment trajectory.

[0010] S5 includes: when generating a visual assessment report, starting a culturally adaptive report calibrator to perform multilingual semantic parsing operations, and extending the multimodal cultural context embedding interface to perform cultural symbol analysis and fusion operations to calibrate the content of the assessment report, thereby eliminating cross-cultural semantic ambiguity and enhancing the adaptability of the report to different scenarios.

[0011] The present invention also proposes a computer-readable storage medium storing a computer program that implements the above method when the program is executed by a processor, thereby providing a standardized implementation carrier and reducing deployment complexity.

[0012] This invention also proposes a multi-dimensional service quality assessment system, comprising: The data acquisition module is used to collect subjective feedback data from the customer platform and objective operational data from the operating system. The data processing module performs cleaning, deduplication, and standardization operations. The assessment dimension configuration module allows you to load preset industry assessment dimension templates or create user-defined assessment dimensions. The weight optimization module assigns initial weights to the evaluation dimensions and dynamically adjusts the weights through a real-time iterative optimization mechanism. The report generation module calculates a comprehensive score based on optimized weights and generates a visual evaluation report, thereby constructing a closed-loop evaluation system and realizing fully automated decision-making.

[0013] In summary, this invention can intelligently optimize the entire process of data collection, cleaning, weight calculation, and report generation, significantly improving the real-time performance, adaptability, and cross-domain applicability of service quality assessment, and supporting accurate decision-making. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating a multi-dimensional service quality assessment method proposed in an embodiment of the present invention. Detailed Implementation

[0015] refer to Figure 1 This invention proposes a multi-dimensional service quality assessment method, including: S1. Collect subjective feedback data from customer platforms and objective operational data from operating systems. Specifically, this includes: deploying an intelligent data source adapter generator to perform metadata learning operations to infer heterogeneous data source structures, generating lightweight edge adapters, and performing dynamic allocation of processing resources through a reinforcement learning priority scheduling mechanism.

[0016] In this exemplary embodiment, for the civil aviation service sector, subjective feedback data specifically includes satisfaction ratings or social media comments submitted by passengers through mobile apps, such as complaints about flight delays; while objective operational data covers real-time flight punctuality rates, baggage handling time records, or fuel consumption logs from the airport operation system.

[0017] Furthermore, the intelligent data source adapter generator is specifically an automated tool that analyzes the format differences of different data sources through metadata learning operations, such as comparing the structural differences between airline databases and third-party weather APIs, to generate a lightweight edge adapter; this adapter can be deployed on the airport's local server to directly process departure data streams.

[0018] As one specific implementation method, during deployment, a meta-learning model is trained based on historical flight data. The model's input is data source metadata, and its output is adaptation rules. After generating an adapter, it is integrated into the edge computing node. As another specific implementation method, a reinforcement learning priority scheduling mechanism uses the Q-learning algorithm to dynamically allocate processing resources, such as CPU or bandwidth resources. For example, during peak passenger flow periods, real-time passenger feedback data is prioritized over historical log processing to address the collection latency problem caused by data source heterogeneity, ultimately improving the integration speed and reliability of multi-source data.

[0019] Furthermore, the edge adapter specifically includes a data conversion module for standardizing the raw data into a unified JSON format; the reinforcement learning priority scheduling mechanism specifically includes a state monitor and an action selector, the former monitoring resource load and the latter adjusting the scheduling strategy according to the reward function, for example, using improved processing efficiency as the reward basis.

[0020] In summary, the implementation of the civil aviation example can ensure the timely collection of aviation service quality data.

[0021] S2. Perform cleaning, deduplication, and standardization operations on the collected subjective feedback data and objective operational data, and perform knowledge graph self-update operations. Specifically, the knowledge graph self-update operations include: monitoring the frequency of new terms through streaming data processing; triggering a context-aware dynamic terminology base adaptive mechanism to perform real-time knowledge graph expansion; integrating a reinforcement learning-driven hotspot detection model to perform high-dimensional sparse region identification; triggering missing data supplementation sampling through an adaptive sparse region identification mechanism; and deploying an attention-driven model to perform key dimension blind spot priority detection.

[0022] In this exemplary embodiment, data cleaning operations in the civil aviation service field involve removing invalid passenger comments, such as duplicate submissions or blank feedback; deduplication operations target redundant entries in flight delay records, such as merging multiple reports of the same flight; and standardization operations unify timestamp formats, such as converting departure times from different time zones to UTC format. The knowledge graph self-updating operation is based on streaming data processing, and the new terminology frequency monitoring operation scans the real-time data stream, such as new terms in passenger feedback like baggage loss rate; when the frequency exceeds a threshold, a dynamic terminology base adaptation mechanism is triggered.

[0023] As one specific implementation method, the context-aware dynamic terminology database adaptation mechanism uses natural language processing models to analyze term context, such as analyzing the semantics of delays in feedback, and adding new nodes to the knowledge graph in real time. Models used include BERT. As another specific implementation method, the reinforcement learning-driven hotspot detection model is trained based on historical graph data. The input is node distribution, and the output is sparse region identifiers. It identifies high-dimensional sparse regions, such as missing service quality dimensions for specific routes, and then activates an adaptive sparse region identification mechanism for supplementary sampling, such as automatically initiating additional passenger surveys. This addresses data quality issues and ultimately promotes the completeness and real-time performance of the knowledge graph.

[0024] Furthermore, streaming data processing specifically includes a real-time data pipeline for continuously ingesting new data; the attention-driven model is specifically a Transformer architecture that prioritizes detecting blind spots in key dimensions, such as focusing on frequently occurring boarding experience dimensions; and the high-dimensional sparse region identification operation specifically includes density clustering algorithms to partition the data space. Understandably, combining this with the data collected in S1 could improve the accuracy of civil aviation service quality assessment.

[0025] S3. Load preset industry evaluation dimension templates or create user-defined evaluation dimensions. Specifically, this includes: starting the industry migration pre-training engine to perform industry rule feature extraction, automatically generalizing emerging field evaluation templates, and optimizing cross-industry dimension adaptation. When creating user-defined evaluation dimensions: it also requires performing multi-turn conversational intelligent guidance to recognize user configuration intent; integrating voice and visual input to perform user intent parsing; performing intent deviation correction based on conversation history to generate custom dimension rules; embedding a scenario simulation training interface to perform synthetic data generation to simulate configuration scenarios; pre-training the intent recognition model and performing error correction enhancement through an error injection mechanism.

[0026] In this exemplary embodiment, the preset industry evaluation dimension templates for the civil aviation service sector are loaded from aviation industry standards. For example, the IATA service quality index template covers dimensions such as on-time performance and cabin comfort. When the industry migration pre-training engine performs feature extraction, it analyzes the template rules, taking industry documents as input and outputting feature vectors. It automatically generalizes to emerging fields such as drone cargo services, optimizing the adaptation process to ensure template compatibility. When creating user-defined evaluation dimensions, a multi-turn conversational intelligent wizard identifies user intent through a chatbot interface. For example, if an airline administrator inputs to add a baggage handling efficiency dimension, the wizard integrates voice and visual inputs to interpret the intent. Voice input includes voice commands, and visual input includes screen gestures. The wizard utilizes conversation history to correct biases, such as correcting the misinterpretation of delay as cancellation.

[0027] As one specific implementation method, the scenario simulation training interface generates synthetic data, such as passenger feedback in simulated flight delay scenarios, to train the intent recognition model. The model used includes an LSTM network. As another specific implementation method, the error injection mechanism adds noisy data, such as random intent error samples, to the training to enhance the error correction function. The output is the corrected rule, thereby solving the problem of rigid evaluation dimensions and ultimately improving the template flexibility to adapt to innovation in aviation services.

[0028] Furthermore, the industry migration pre-training engine specifically includes a transfer learning module, which extracts rule features based on pre-trained models, such as GPT; the intent bias correction operation specifically includes a context matching algorithm that compares historical dialogues. In summary, this makes the setting of civil aviation service quality dimensions more intelligent and user-friendly.

[0029] S4. Assign initial weights to the evaluation dimensions and dynamically adjust the weights through a real-time iterative optimization mechanism.

[0030] The optimization mechanisms include: integrating an anomaly detection module to perform optimization parameter adjustment operations; coordinating multi-node historical data to perform federated weight smoothing operations; calling the memory reinforcement module to store historical weight trends and perform smoothing algorithm application operations; adding a time series anomaly buffer module to perform continuous analysis of historical weight trends; performing anomaly fluctuation point detection operations through a sliding window algorithm; and injecting buffer delays to perform optimization process parameter smoothing operations.

[0031] In this exemplary embodiment, the evaluation dimensions in the civil aviation service field are dynamically adjusted through a real-time iterative optimization mechanism. For example, the on-time performance weight is initially set to 0.4, and the cabin service weight is set to 0.3. The anomaly detection module scans the weight parameters; for example, if the on-time performance suddenly drops, it triggers an optimization parameter adjustment operation and outputs new weight values. The federated weight smoothing operation coordinates historical data from multiple airport nodes, such as the weight records of Beijing and Shanghai airports, to avoid single-point bias.

[0032] As one specific implementation, the memory reinforcement module stores historical trends, such as using a time series database to record weekly weight changes and applying smoothing algorithms, such as exponential smoothing. As another specific implementation, the time series anomaly buffer module continuously analyzes trends, with historical weight sequences as input and anomaly indicators as output. A sliding window algorithm detects fluctuation points, such as identifying weight anomalies caused by peak passenger flow during holidays, injecting buffer delays to optimize smoothing operations, and outputting the stabilized weights. This solves the problem of weight instability and ultimately ensures the fairness and continuity of the aviation service quality score.

[0033] Furthermore, the federated weighted smoothing operation specifically includes a distributed computing framework, where each node processes data locally and then aggregates it; the memory reinforcement module specifically includes a weight memory and a trend analyzer; the sliding window algorithm specifically detects fluctuations within a fixed time window, such as a 24-hour time window. This can be understood as optimizing the dynamic adaptability of civil aviation assessments.

[0034] S5. Calculate a comprehensive score based on the optimized weights and generate a visual evaluation report. Specifically, this includes: launching a culturally adaptive report calibrator to perform multilingual semantic parsing during report generation, and extending the multimodal cultural context embedding interface to perform cultural symbol analysis and fusion operations to calibrate the evaluation report content.

[0035] In this exemplary embodiment, the comprehensive score for civil aviation services is calculated based on S4 optimized weights, such as a 0.35 weight for on-time performance, generating a visual report, such as a dashboard displaying route service quality rankings. The culture-adaptive report calibrator processes multilingual feedback during multilingual semantic parsing, such as translating Chinese passenger comments into English reports. A multimodal cultural context embedding interface analyzes cultural symbols, such as differences in regional tolerance for delays, and integrates them into the calibrated content of the report.

[0036] As one specific implementation method, multilingual semantic parsing uses a neural machine translation model, taking the original text as input and outputting a semantically equivalent translation. As another specific implementation method, cultural symbol analysis and fusion integrates text and image data, such as the cultural meanings of airline logos, and maps them to report templates through an embedding layer. This addresses the issue of cultural bias in reporting, ultimately promoting the fairness and comprehensibility of global aviation reports.

[0037] Furthermore, the culturally adaptive report calibrator specifically includes a cross-language parser that supports the languages ​​of major aviation markets; the multimodal cultural context embedding interface is specifically a feature fusion module that extracts symbolic features, such as the meaning of colors or icons. In summary, this makes the civil aviation assessment results more inclusive and practical.

[0038] In a further specific implementation, the present invention also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in claim 1, thereby providing a standardized implementation carrier and reducing deployment complexity.

[0039] In a further specific implementation, the present invention also proposes a multi-dimensional service quality assessment system, including: a data acquisition module, a data processing module, an assessment dimension configuration module, a weight optimization module, and a report generation module. The data acquisition module is used to collect subjective feedback data from the customer platform and objective operational data from the operating system. The data processing module is used to perform cleaning, deduplication, and standardization operations. The assessment dimension configuration module is used to load preset industry assessment dimension templates or create user-defined assessment dimensions. The weight optimization module is used to assign initial weights to the assessment dimensions and dynamically adjust the weights through a real-time iterative optimization mechanism. The report generation module is used to calculate a comprehensive score based on the optimized weights and generate a visual assessment report, thereby constructing a closed-loop assessment system and achieving fully automated decision-making throughout the entire process.

[0040] Based on the above method, the present invention proposes a computer-readable storage medium storing a computer program that implements the above method when the program is executed by a processor, thereby providing a standardized implementation carrier and reducing deployment complexity.

[0041] This invention proposes a multi-dimensional service quality assessment system, comprising: The data acquisition module is used to collect subjective feedback data from the customer platform and objective operational data from the operating system. The data processing module performs cleaning, deduplication, and standardization operations. The assessment dimension configuration module allows you to load preset industry assessment dimension templates or create user-defined assessment dimensions. The weight optimization module assigns initial weights to the evaluation dimensions and dynamically adjusts the weights through a real-time iterative optimization mechanism. The report generation module calculates a comprehensive score based on optimized weights and generates a visual evaluation report, thereby constructing a closed-loop evaluation system and realizing fully automated decision-making.

[0042] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for multi-dimensional quality of service evaluation, characterized by, Comprising: S1, collecting subjective feedback data of the customer platform and objective operation data of the operation system; S2, performing cleaning operation, de-duplication operation and standardization processing operation on the collected subjective feedback data and objective operation data; S3, loading a preset industry evaluation dimension template or creating a user-defined evaluation dimension; S4, assigning an initial weight to the evaluation dimension, and dynamically adjusting the weight through a real-time iterative optimization mechanism; S5, calculating a comprehensive score based on the optimized weight and generating a visual evaluation report.

2. The method of claim 1, wherein, S1 includes deploying an intelligent data source adapter generator to perform metadata learning operation to infer heterogeneous data source structure, generate lightweight edge adapter, and perform processing resource dynamic allocation operation through reinforcement learning priority scheduling mechanism.

3. The method of claim 1, wherein, The creation of user-defined evaluation dimensions in S3 includes: performing multi-round dialogue intelligent guide to identify user configuration intent operation, fusing voice input and visual input to perform user intent analysis operation, and performing intent bias correction operation based on conversation history to generate custom dimension rules; At the same time, embed scenario simulation training interface to perform synthetic data generation operation to simulate configuration scenarios, pre-train intent recognition model, and perform error correction function enhancement operation through error injection mechanism.

4. The method of claim 1, wherein, Before S3, perform knowledge graph self-updating operation, specifically including: through stream data processing, perform new term frequency monitoring operation, trigger context-aware dynamic term library adaptive mechanism to perform knowledge graph real-time expansion operation; At the same time, integrate reinforcement learning driven hot spot detection model to perform high-dimensional sparse area identification operation, trigger missing data supplement sampling operation through adaptive sparse area identification mechanism, and deploy attention driven model to perform key dimension blind area priority detection operation.

5. The method of claim 1, wherein, S3 also includes: starting industry migration pre-training engine to perform industry rule feature extraction operation, automatically generalizing emerging field evaluation template, and optimizing cross-industry dimension adaptation operation.

6. The method of claim 1, wherein, The real-time iterative optimization mechanism in S4 includes: integrating an anomaly detection module to perform optimization parameter adjustment operation, cooperating with multi-node historical data to perform federated weight smoothing operation, and calling a memory reinforcement module to store historical weight trend to perform smoothing algorithm application operation.

7. The method of claim 6, wherein, The real-time iterative optimization mechanism in S4 also includes: adding a time series anomaly buffer module to perform historical weight trend continuous analysis operation, detecting abnormal fluctuation points through sliding window algorithm, and injecting buffer delay to perform optimization process parameter smoothing operation.

8. The method of claim 1, wherein, S5 includes: starting cultural self-adaptive report calibrator to perform multi-language semantic analysis operation when generating a visual evaluation report, and expanding a multi-modal cultural context embedding interface to perform cultural symbol analysis fusion operation to calibrate the evaluation report content.

9. A computer-readable storage medium, characterized in that, A computer program is stored, which, when executed by a processor, implements the method of any one of claims 1-8.

10. A multi-dimensional quality of service evaluation system, characterized by, Comprising: A data acquisition module for acquiring subjective feedback data of the customer platform and objective operation data of the operation system; A data processing module for performing cleaning operation, de-duplication operation and standardization processing operation; An evaluation dimension configuration module for loading a preset industry evaluation dimension template or creating a user-defined evaluation dimension; The weight optimization module assigns initial weights to the evaluation dimensions and dynamically adjusts the weights through a real-time iterative optimization mechanism. The report generation module calculates a comprehensive score based on the optimized weights and generates a visual evaluation report.

Images