QUIDS: Query Intent Description for Exploratory Search via Dual Space Modeling

TL;DR

The paper tackles the lack of explicit feedback in exploratory search by generating natural-language descriptions of the search system's inferred query intent in a post-retrieval setting.It introduces QUIDS, a dual-space, contrastive-learning framework comprising a representation-space with dual encoders and a disentangling space in a decoder, augmented by Intent-Driven Negative Augmentation (IDNA).The method optimizes a joint objective across $\mathcal{L}_{NLL}$, $\mathcal{L}_{Encoder}$, and $\mathcal{L}_{Decoder}$ with weights $ (\lambda_0, \lambda_1, \lambda_2) $, and demonstrates substantial gains over strong baselines on ROUGE, BERTScore, and human/LLM evaluations on the Q2ID benchmark.Empirical results show QUIDS effectively filters out distracting content and yields concise, accurate intent descriptions, providing actionable post-retrieval feedback that can guide query refinement and improve transparency in exploratory search.

Abstract

In exploratory search, users often submit vague queries to investigate unfamiliar topics, but receive limited feedback about how the search engine understood their input. This leads to a self-reinforcing cycle of mismatched results and trial-and-error reformulation. To address this, we study the task of generating user-facing natural language query intent descriptions that surface what the system likely inferred the query to mean, based on post-retrieval evidence. We propose QUIDS, a method that leverages dual-space contrastive learning to isolate intent-relevant information while suppressing irrelevant content. QUIDS combines a dual-encoder representation space with a disentangling decoder that works together to produce concise and accurate intent descriptions. Enhanced by intent-driven hard negative sampling, the model significantly outperforms state-of-the-art baselines across ROUGE, BERTScore, and human/LLM evaluations. Our qualitative analysis confirms QUIDS' effectiveness in generating accurate intent descriptions for exploratory search. Our work contributes to improving the interaction between users and search engines by providing feedback to the user in exploratory search settings. Our code is available at https://github.com/menauwy/QUIDS

Figures (7)

• Figure 1: A user-facing application of query intent generation in exploratory search. The system's inferred intent is generated by contrasting top-ranked (pseudo-relevant) and low-ranked (pseudo-irrelevant) documents. Key information contributing to the inferred intent is shown in bold.
• Figure 2: Overview of our proposed pipeline. From left to right, we show (a) Intent-Driven Negative Augmentation method, (b) Contrastive decoder structure with dual cross-attention layers, (c) and (d) Contrastive learning via dual space modeling.
• Figure 3: Distribution of human and LLM evaluation sores on four qualitative metrics.
• Figure 4: Boxplot of human and LLM evaluation scores on 4 metrics of our model on different intent types.
• Figure 5: Case study indicating the role of contrastive examples in the decoder stage. Token-level decoder cross-attention weights are shown for a generated intent token (red) are shown with (a) and without (b) an irrelevant document in the model input. Deeper color indicates a higher value.