Elaborative Subtopic Query Reformulation for Broad and Indirect Queries in Travel Destination Recommendation

TL;DR

EQR is introduced, a large language model-based QR method that combines both breadth and depth by generating potential query subtopics with information-rich elaborations and achieves significant improvements in recall and precision over existing state-of-the-art QR methods.

Abstract

In Query-driven Travel Recommender Systems (RSs), it is crucial to understand the user intent behind challenging natural language(NL) destination queries such as the broadly worded "youth-friendly activities" or the indirect description "a high school graduation trip". Such queries are challenging due to the wide scope and subtlety of potential user intents that confound the ability of retrieval methods to infer relevant destinations from available textual descriptions such as WikiVoyage. While query reformulation (QR) has proven effective in enhancing retrieval by addressing user intent, existing QR methods tend to focus only on expanding the range of potentially matching query subtopics (breadth) or elaborating on the potential meaning of a query (depth), but not both. In this paper, we introduce Elaborative Subtopic Query Reformulation (EQR), a large language model-based QR method that combines both breadth and depth by generating potential query subtopics with information-rich elaborations. We also release TravelDest, a novel dataset for query-driven travel destination RSs. Experiments on TravelDest show that EQR achieves significant improvements in recall and precision over existing state-of-the-art QR methods.

Figures (4)

• Figure 1: Examples of broad (Cities for youth-friendly activities, right) and indirect (Cities for a high school graduation trip, left) queries. The comparisons are made among different QR methods discussed in Section \ref{['qr']}: Q2E q2dq2e (top), Query2Doc query2doc (middle), and EQR (bottom). The top 5 recommendations are shown, with irrelevant results marked in red with an $X$. One can clearly identify that Q2E covers subtopic breadth, but lacks depth of description, while Query2Doc provides in-depth concrete suggestions without covering general subtopics. EQR covers both general subtopic breadth with deep subtopic elaboration.
• Figure 2: LLM prompts for various QR methods discussed in Section \ref{['qr']}, with LLM output shown in Figure \ref{['fig:examples']} using two broad and indirect query examples. Q2E q2dq2e (top-left), Query2Doc query2doc (top-right), GenQR gqrgqe (bottom-left), and EQR (bottom-right).
• Figure 3: Effect of top-n paragraphs across different retrievers: Dense - TAS-B (top), Dense - MiniLM (middle), Sparse - BM25 (bottom)
• Figure 4: Effect of the number of subtopics in EQR across different retrievers: Dense - TAS-B (top) and Dense - MiniLM (bottom)