Navigating in Streams of Opinions

The task of qualitative text analysis, in particular, identification of opinions, topics, and arguments recurring in long documents or text collections, requires a lot of effort from the analyst. Computational extraction of main topics in a document or a corpus has been shown to be an effective first step for such analyses. However, the typical output of topic modeling algorithms at the detailed level is also overwhelming. The fields of information visualization and visual analytics provide approaches for representing and interacting with textual data and results of various text analyses (including topic modeling, sentiment and stance classification, etc.) to solve this problem.

This project centers on an interactive computer-assisted argument extraction approach for textual data called Topics2Themes. The data processing pipeline of this tool includes preprocessing, optional classification or tagging of text documents (or individual utterances), and topic modeling with one of the supported algorithms. The resulting data is then presented with an interactive visual interface, which allows the user to investigate the topics and prominent keywords and to define groups of recurring topics, i.e., themes. The application scenarios for this approach include investigation of heated, opinionated discussions on controversial issues such as vaccination hesistancy or political views in online forums and social media.

This project was performed from 2017 until the end of 2019 in cooperation with University of Potsdam, Hokkaido University, and Institute for Language and Folklore (Sweden).