How People Choose in the Smart Cockpit? Behaviour Modelling with the Multimodal Data from the Intelligent Connected Vehicles

The intelligent connected vehicle (ICV) and the smart cockpit have substantially changed the driving experience and environment, where the in-cabin behaviours can be entirely different from the past. This paper focuses on revealing the choice decision mechanism of cabin users, where hundreds of ICV functions can be installed. The multimodal data of online reviews, console clicking and driving trajectories are used, where the SOTA behavioural-enhanced large language model and the WAL2vec (Word-Activity-Location sequence vector) space are employed and proposed for efficient processing of high-dimensional big data. In particular, two decision mechanisms are evaluated, namely the random utility maximization and the random regret minimization. The Metropolis-Hasting importance sampling is adopted to generate the in-cabin usage sequence as the choice set. Then, modified models for the sequence choice analysis are derived to consider the choice set generation process where the choice set size is large. Data of more than one year from a studied carmaker is used for evaluation, and simulations with the trained models are conducted for future scenarios. The results suggest that our methods are proper, practical, and precise. Practical and managerial insights are discussed, where we try to identify the critical role of the ICV system in the entangled relations among the individual behaviours, the cars, and the whole system.