Analysis of Airline Service Quality Through Online Reviews: A Multi-Agent LLM Approach Based on SERVPERF
Author
Ashrafi, Jaleh
Term
4. term
Education
Publication year
2025
Abstract
This thesis examines how AI-based methods can turn large-scale online customer reviews into measurable insights on airline service quality. Using SERVPERF (tangibles, reliability, assurance, responsiveness, empathy) within the Smart Tourism Ecosystem, it analyzes 10,115 verified Skytrax reviews from 11 European airlines. The three-stage method includes: (1) a qualitative content analysis of 50 reviews to identify service themes mapped to SERVPERF, (2) aspect-based sentiment analysis with a multi-agent setup of large language models (Gemini 1.5 Flash and GPT-3.5 Turbo) that assigns positive, neutral, or negative polarity to each SERVPERF dimension and is validated via manual review and micro F1, and (3) OLS regression on sentiment-labeled aspects to assess which dimensions more strongly shape customer satisfaction. The study shows that LLMs can extract structured insights from unstructured data and complement traditional surveys with scalable, near–real-time evaluation across service dimensions. The findings offer practical implications for managers and service designers by indicating which dimensions matter most to passengers, enabling resource prioritization, continuous feedback monitoring, and data-driven decision-making. Overall, the thesis demonstrates that classic models like SERVPERF can be extended with AI tools to foster service innovation, adapt to evolving customer expectations, and support a smarter tourism system.
Denne afhandling undersøger, hvordan AI-baserede metoder kan omsætte store mængder online kundeanmeldelser til målbar viden om flyselskabers servicekvalitet. Med SERVPERF som ramme (tangibles, reliability, assurance, responsiveness, empathy) og inden for Smart Tourism Ecosystem analyseres 10.115 verificerede Skytrax-anmeldelser fra 11 europæiske flyselskaber. Metoden er tretrins: (1) en kvalitativ indholdsanalys af 50 anmeldelser til at identificere service-temaer, som mappes til SERVPERF, (2) aspektbaseret sentimentsanalyse med et fleragent-setup af store sprogmodeller (Gemini 1.5 Flash og GPT-3.5 Turbo), der tildeler positiv, neutral eller negativ polarisering pr. SERVPERF-dimension og valideres via manuel gennemgang og micro F1, samt (3) OLS-regression på de sentimentmærkede aspekter for at vurdere, hvilke dimensioner der i højere grad former kundetilfredshed. Studiet dokumenterer, at LLM’er kan udtrække struktureret indsigt fra ustrukturerede data og supplere traditionelle spørgeskemer med skalerbar, tæt-på-realtids evaluering på tværs af service-dimensioner. Resultaterne giver praktiske implikationer for ledere og servicedesignere ved at pege på hvilke dimensioner, der betyder mest for passagerer, så ressourcer kan prioriteres, feedback kan overvåges løbende, og beslutninger kan træffes datadrevet. Samlet viser afhandlingen, at klassiske modeller som SERVPERF kan udvides med AI-værktøjer til at fremme serviceinnovation, tilpasse sig dynamiske kunde-forventninger og understøtte en smartere turismeinfrastruktur.
[This apstract has been generated with the help of AI directly from the project full text]