TL;DR
This paper investigates how people collaboratively reformulate search queries when looking for travel information online. They found that individual queries and chat logs are key sources for generating new query terms, and identified five distinct operations users employ to refine their searches. This work has implications for designing better query suggestion tools in collaborative search environments.
What Problem Does This Paper Address?
Web searching is often an iterative process where users refine their queries to find better results. This is especially true for complex, exploratory tasks where users might benefit from collaborating. While individual query reformulation has been studied, collaborative query reformulation (CQ reformulation) in web search, particularly for exploratory tasks, has received less attention. This paper aims to fill this gap by examining how users collaboratively reformulate queries, what strategies they use, and what influences these reformulations in the context of travel information searching.
Methods
The researchers conducted a user study using a custom-built collaborative search interface called ColTIS (Collaborative Tourism Information Search). This interface allowed pairs of participants to search for travel information collaboratively. They analyzed the search logs, which included individual queries and collaborative queries (CQs), as well as chat logs generated during the search sessions. The analysis focused on two main areas: identifying CQ reformulation strategies and understanding the interplay between individual queries, chat logs, and CQ reformulation. They defined five specific operations for query reformulation: addition, modification, reordering, addition and modification, and addition and reordering.
Datasets
The study used data collected from 17 pairs of participants (34 individuals) who engaged in a tourism-related exploratory web search task. The data included:
- 210 valid individual queries.
- 140 valid collaborative queries (CQs).
- 486 valid chat log entries.
This data was collected through a simulated remotely-located collaboration scenario using either the ColTIS interface or Tripadvisor.com with Google Talk for communication.
Key Insights and Ideas
- Collaboration is Key for Exploratory Search: When information needs are complex and exploratory, users benefit from collaborating to explore the information space and learn together.
- Individual Efforts Fuel Collaboration: Even in a collaborative setting, individual queries and the communication (chat logs) between team members are crucial sources for generating new terms in collaborative queries.
- "Adding" is the Dominant Reformulation Strategy: The most common way users reformulate collaborative queries is by adding new terms to an existing query.
- Self-Generated Content Matters Most: Participants tended to reuse terms from their own individual queries more than from their partner's queries when reformulating. Similarly, self-generated chat logs were more influential than partner-generated chat logs for query reformulation.
- Partners' Contributions are More Balanced: Interestingly, for chat logs, the influence of self-generated logs was less pronounced compared to partner-generated logs, suggesting a more balanced contribution in communication.
Key Results
- Sources of CQ Terms: Individual queries were the most significant source of terms for collaborative query reformulation (51%), followed by chat logs (26%) and previous knowledge/experiences (23%).
- Reformulation Operations: The most frequent operations performed for CQ reformulation were "Addition" (61%) and "Modification" (20%).
- Impact of Individual vs. Chat: While users collaborated more through chat logs (mean=28.59) than individual queries (mean=12.35) during the overall collaborative search process, individual queries were not significantly different from chat logs as sources for CQ reformulation terms (p=0.206).
- Self vs. Partner Influence: Self-generated individual queries were significantly more influential for CQ reformulation than self-generated chat logs (p=0.0054). However, there was no significant difference between partners' queries and partners' chat logs for CQ reformulation (p=0.0909).
Limitations and Future Work
Limitations: * The study focused on a specific domain (travel information searching), and findings might not generalize to all types of exploratory search tasks. * The simulated remote collaboration might not fully replicate the dynamics of in-person collaboration.
Future Work: The authors suggest that their findings have implications for designing better query suggestion tools for collaborative search. They are currently working on automatically extracting query terms from individual queries and chat logs to facilitate CQ reformulation. Further research could explore the impact of different collaboration structures and task types on query reformulation behavior.
Summary updated