From trip metrics to corporate disclosures: predicting ride-hailing prices for longitudinal data using an integrated Cumulative Link Mixed Model and generative AI-based methodology
研究了Lyft和Uber在2019年2月至2024年12月期间109,435次行程的价格分类预测，发现财务属性、语义变量（从年报中提取）以及时间因素对价格有显著影响，并提出了平台特定建议。
The ride-hailing platforms often attract severe criticisms for setting arbitrary ride prices, emphasising the importance of accurate fare prediction and key influencing factor identification. In this context, operational, financial, and temporal factors, along with semantic variables captured by Generative AI, can play an instrumental role in pricing. Motivated by this issue, we focus on the prediction of price categories for ride-hailing platforms Lyft and Uber between February 2019 and December 2024 with a dataset of 109,435 trips. We investigate the effect of operational, financial, and temporal attributes on the price classification. Further, we introduce two novel semantic variables, namely Price, Demand, and Tech Cooccurrence and Sentiment, captured from company annual reports using a leading GenAI tool, ChatGPT 4.0, and explore the impact of these variables on the price categories. We incorporate a Cumulative Link Mixed Model (CLMM)-based regression methodology to handle ordinal and non-continuous longitudinal data. We find that the financial attributes, sales tax and driver payment, and semantic variables always remain significant predictors for both, whereas the temporal attribute week of the year emerges as significant for Uber. Additionally, we explore how weather conditions, locations, and the COVID-19 pandemic influence the pricing. Finally, we propose several platform-specific recommendations.