[Paper Review] Conversational Browsing

Svitlana Vakulenko, Vadim Savenkov+1 more|arXiv (Cornell University)|Dec 7, 2020

Speech and dialogue systems4 citations

TL;DR

This paper introduces conversational browsing as a novel paradigm for exploratory search that enables users to discover and navigate structured information sources—like databases or knowledge graphs—through natural, interactive dialogues without requiring explicit query formulation. By modeling user-system interaction as a dynamic, feedback-driven exploration of facets and entities, the authors develop a prototype system that outperforms traditional query-based approaches in scenarios with vague or unclear information needs, demonstrating the viability of conversational browsing as a complementary interaction model.

ABSTRACT

How can we better understand the mechanisms behind multi-turn information seeking dialogues? How can we use these insights to design a dialogue system that does not require explicit query formulation upfront as in question answering? To answer these questions, we collected observations of human participants performing a similar task to obtain inspiration for the system design. Then, we studied the structure of conversations that occurred in these settings and used the resulting insights to develop a grounded theory, design and evaluate a first system prototype. Evaluation results show that our approach is effective and can complement query-based information retrieval approaches. We contribute new insights about information-seeking behavior by analyzing and providing automated support for a type of information-seeking strategy that is effective when the clarity of the information need and familiarity with the collection content are low.

Motivation & Objective

• To understand the mechanisms behind multi-turn, conversational information-seeking dialogues in exploratory search contexts.
• To design a dialogue system that supports navigation of structured information sources without requiring users to formulate precise queries upfront.
• To develop and evaluate a conversational browsing model grounded in empirical human dialogue data, focusing on user control and cognitive load management.
• To explore how interactive storytelling principles can be applied to guide users through information sources in a dynamic, responsive manner.
• To provide a foundation for conversational exploratory search that complements existing question-answering systems.

Proposed method

• The authors collected and analyzed human-to-human dialogue transcripts from a controlled lab study where participants explored a structured information source (a database table) with an intermediary.
• They derived a grounded theory of conversational browsing based on observed interaction patterns, emphasizing user-driven topic shifts and feedback-informed navigation.
• A conversational browsing model was formalized as an interactive process where the system dynamically reveals facets and entities based on user feedback, guiding exploration without requiring explicit queries.
• The model was implemented into a prototype dialogue system that supports navigation control and adapts to user preferences and cognitive load.
• The prototype was evaluated in a user study comparing it against a traditional request-response conversational search system, measuring effectiveness in satisfying information needs under vague or unclear goals.
• The system uses semantic parsing and information retrieval techniques to map user utterances to navigational actions within the data structure, enabling seamless transitions between browsing and querying.

Experimental results

Research questions

• RQ1 How do users naturally navigate and communicate about structured information sources when their information needs are unclear or implicit?
• RQ2 What interaction patterns emerge in human-to-human conversations where one party guides another through a data source without query formulation?
• RQ3 Can a dialogue system that supports dynamic, feedback-driven navigation improve information access in exploratory search compared to traditional query-based systems?
• RQ4 How can conversational browsing be formalized as a general model applicable across different domains and data structures?
• RQ5 What role does user control and cognitive load regulation play in effective conversational browsing?

Key findings

• The conversational browsing model effectively supports information access in scenarios where users lack clarity about their information needs or are unfamiliar with the domain vocabulary.
• Evaluation results show that the conversational browsing system is effective and can complement traditional query-based information retrieval approaches, particularly in exploratory search tasks.
• Users demonstrated improved engagement and satisfaction when guided through data via interactive, feedback-driven dialogue rather than relying solely on query formulation.
• The system's ability to dynamically adjust exploration paths based on user feedback led to more efficient discovery of relevant information subsets within the data source.
• The study confirms that conversational browsing is a viable alternative to question-answering in conversational search, especially when users are not ready to articulate precise queries.
• The prototype successfully enables seamless transitions between browsing and querying, suggesting a hybrid interface as a promising future direction.