A Visual Interface for Analyzing Text Conversations.

This paper presents a visual, intelligent interface intended to help user analyze possibly long and complex conversations. So far, the interface can only deal with synchronous conversations, but most of its components could be also applied to asynchronous conversations such as blogs and emails. In a Business Intelligent scenario, our interface could support the analysis of real-time conversation occurring for instance in blogs, discussion fora, or in sites intended to collect customer feedback. The design of our interface aims to effectively combine the power of human perceptual and cognitive skills with the ability of Natural Language Processing (NLP) techniques to mine and summarize text. Since we are targeting a large user population, with no, or minimal expertise in data analysis, we selected interface elements based on simple and common visual metaphors. Furthermore, to accommodate for user differences in such a large population, we provided users with the ability to satisfy the same information needs in different ways. We have tested our interface in a formative user study, in which participants used the interface to analyze four long and complex conversations, in order to answer questions addressing specific information needs in a business scenario. This evaluation revealed that the interface is intuitive, easy to use and provides the tools necessary for the task. Participants also found all the interface components quite useful, with the main problems coming from inaccuracies in the information extraction process, as well as from deficiencies in the generated summaries. Finally, it seems that the choice of offering redundant functionalities was beneficial. The logged interaction behaviors reveal that different users actually selected very different strategies, even independently from their performance. And this was consistent with the high variability in the user preferences for the different interface components.