From Chat Logs to Collective Insights: Aggregative Question Answering
This addresses the challenge of analyzing societal trends from conversational data for researchers and developers, but it is incremental as it builds on existing LLM and QA frameworks.
The paper tackles the problem of extracting collective insights from large-scale chatbot conversations by introducing Aggregative Question Answering, a novel task that requires reasoning over thousands of interactions, and constructs a benchmark with 6,027 questions from 182,330 real-world conversations, showing that existing methods struggle with effective reasoning or high computational costs.
Conversational agents powered by large language models (LLMs) are rapidly becoming integral to our daily interactions, generating unprecedented amounts of conversational data. Such datasets offer a powerful lens into societal interests, trending topics, and collective concerns. Yet, existing approaches typically treat these interactions as independent and miss critical insights that could emerge from aggregating and reasoning across large-scale conversation logs. In this paper, we introduce Aggregative Question Answering, a novel task requiring models to reason explicitly over thousands of user-chatbot interactions to answer aggregative queries, such as identifying emerging concerns among specific demographics. To enable research in this direction, we construct a benchmark, WildChat-AQA, comprising 6,027 aggregative questions derived from 182,330 real-world chatbot conversations. Experiments show that existing methods either struggle to reason effectively or incur prohibitive computational costs, underscoring the need for new approaches capable of extracting collective insights from large-scale conversational data.