An Exploration of Conversation Enhanced Browsing with Hypergraph of Text

Despite many finding great utility in LLM chatbots such as ChatGPT for their information needs, current systems are not well suited for all tasks. Particularly in knowledge work, hallucinations undermine trustworthiness and effortless answers may undermine long-term learning. We argue that browsing is a promising paradigm to complement traditional technologies for tasks where trust and engagement are highly important. However, unlike search or question answering, which benefit from the general technologies of search engines and RAG, there isn’t yet a general technology to support making any collection browsable. Previous and ongoing work has focused on articulating Hypergraph of Text (HoT) as a framework to support semantic organization and browsing for arbitrary text collections. We have also proposed the Text Information Navigation Kit (TINK) which acts as a toolkit for turning the HoT data structure into a browsing system. Given this, planned and future research focuses on how to evaluate such systems, and how to overcome information overload issues we have seen in preliminary works. Towards this end, we propose a user focused study on both TINK and a conversationally enhanced TINK system. We believe that building such a system and evaluating it with user studies is a key step towards better supporting a wide variety of information needs, especially when trust and learning are of high importance.