Session: 16-01-01: Government Agency Student Poster Competition

Paper Number: 150054

150054 - Characterizing Rural Resident Acceptance of Drone Delivery: A Large Language Model Empowered Approach 

Introduction: With the increasing demand for rural logistics services and the notable disparities in service provision between urban and rural areas, there arises a compelling need to explore innovative drone-based delivery solutions. This study aims to address the challenges hindering the adoption of drone-based delivery, due to technological and physical barriers, which consequently affect service quality for rural residents. Such disparities amplify concerns regarding delivery equity and residents' acceptance of potential drone delivery services.

Contribution: Our research presents an inaugural investigation into residents' direct willingness and sentiment toward drone delivery services in rural areas using a Large Language Model (LLM)-empowered machine learning framework. Leveraging the LLM-driven Light Gradient-Boosting Machine (LightGBM) method, our prediction model mitigates cognitive bias and significantly enhances the predictive accuracy of residents' acceptance categories compared to traditional ordinal logistic regression models. Our research advances the understanding of rural residents' acceptance to adopt drone delivery services, addressing pertinent challenges within the rural logistic landscape and the evolution of the drone delivery market. Moreover, it reveals the gap between the supply of rural drone delivery services and the demand from the rural consumer base, exploring the intricate interplay between socioeconomic factors and delivery preferences. This approach fosters a comprehensive drone-based delivery ecosystem that inclusively benefits all rural residents, irrespective of their geographical location.

Methodology: The methodology employed in this study includes surveying rural residents in the U.S. to understand their views on drone delivery, the efficiency of current Same-Day Delivery services, and other socioeconomic aspects. We utilized an integrated LLM-driven LightGBM model to analyze the survey data. This model explores residents' sentiments and accurately identifies factors influencing rural residents' acceptance of drone delivery services. The analytical power of Large Language Models (LLM) combined with the precision of machine learning techniques enabled a thorough investigation into the impacts of four primary factors: (1) premium pricing of drone delivery, (2) attitude affected by the pandemic, (3) equitable same-day delivery demand, and (4) disability in the household.

Conclusions: Preliminary results indicate that the first three factors have a significantly positive relationship with the acceptance of drone delivery. This shows that rural residents who are willing to pay premium prices, exhibit significant attitude adjustments toward drone delivery after the pandemic, and desire equitable same-day delivery are more likely to utilize drone delivery services. However, the last factor is not significant, indicating that disability in the household does not necessarily impact acceptance. Additionally, a significant positive correlation between positive sentiment and acceptance highlights the importance of using LLMs to uncover latent sentiments beyond traditional survey analysis. Although some factors displayed no significant correlation with acceptance, urban accessibility and various socioeconomic features, including household income, age, gender, and occupation, demonstrated varied and significant connections. These findings effectively define the current rural customer groups for drone delivery services and identify the crucial factors affecting their acceptance. Furthermore, this study advances beyond the limitations of traditional discrete choice models by developing an LLM-driven machine learning approach to analyze both explicit and latent willingness towards advanced drone delivery services. In conclusion, this study provides logistics policymakers and delivery companies with a practical methodology to identify target rural customer groups. Comprehensive delivery services for rural areas, including ground delivery and multiple drone-delivery options, can be further designed to meet diverse delivery demands. By examining trends in LLM and machine learning, this study overcomes the shortcomings of previous research that focused on traditional regression models and homogeneous numerical datasets. The proposed LLM-driven LightGBM model effectively handles heterogeneous survey datasets, including numerical and textual data, and maintains better accuracy than traditional models. Drone companies and their delivery partners can utilize similar methods to explore and expand the drone-based delivery market in rural areas more effectively.

Presenting Author: Henan Zhu Rensselaer Polytechnic Institute

Presenting Author Biography: Henan Zhu is currently a PhD student in Transportation Engineering at Rensselaer Polytechnic Institute (RPI). She earned her master’s degree in Transportation Engineering from Beijing Jiaotong University. Her research focuses on multimodal transportation planning and transportation equity analysis, with a particular interest in the application of emerging technologies, including AI and Large Language Models (LLM). She has presented her work at prestigious and influential conferences, including the Transportation Research Board Annual Meeting and the ITS-NY Annual Meeting. As the president of RPI’s Women in Transportation Seminar chapter, Henan is committed to promoting female students’ involvement in the transportation industry.

Authors:

Henan Zhu Rensselaer Polytechnic Institute
Xiaozheng He Rensselaer Polytechnic Institute