School of Engineering Department of Computer Science and Engineering 146 Efficient Queries over Database Supervisor: WONG Raymond Chi Wing / CSE Student: TANG Chiu Yeung / DSCT NGUYEN Kim Hue Nam / COGBM Course: UROP 2100, Summer Our goal is to find the user’s utility function that represents the user’s preference. One common approach in the database community is to involve iterative interactions, by asking the user a series of questions. Each question displays two points and requires the user to choose a more preferred point between them. However, in real-world scenarios, the user may not be able to answer some questions. For example, the user is unsure about their preference on the given points for now, or the user does not want to provide information on a particular attribute. In that case, the user may randomly choose a point, which can lead to an undesirable output, due to the incorrect utility function learned. To address this problem, we propose a new problem of finding the most preferred tuple via interaction, where the user is able to “skip” the questions that they do not want to answer. We are currently working on an algorithm that finds the best point with asymptotically optimal round complexity. Efficient Queries over Database Supervisor: WONG Raymond Chi Wing / CSE Student: WANG Yuning / DA Course: UROP 2100, Fall This report explores the development of a novel framework for converting natural language queries into Overpass Query Language (OverpassQL) through a series of methodologies. The research begins by establishing a Speech-to-OverpassQL dataset, which pairs audio inputs generated from a Text-toOverpassQL dataset with corresponding Overpass queries. With the well-built dataset, the study investigates various approaches, including fine-tuning advanced large language models (LLMs) like GPT-4o, which has already supported audio inputs, as well as the application of a two-step approach combining automatic speech recognition and LLMs like Meta Llama. Additionally, it examines the challenges posed by the informal grammar of OverpassQL and the dynamic nature of OpenStreetMap schemas, in order to explore the feasibility of building an end-to-end approach to accomplish the Speech-to-OverpassQL task. Finally, it discusses methodologies from existing literature that can inform and enhance the conversion process, highlighting the need for a robust model capable of generating accurate and contextually relevant OverpassQL queries.
RkJQdWJsaXNoZXIy NDk5Njg=