Interactive Topic Guided Thematic Analysis for Social Media Data : Thematic Analysis is a method for identifying and analyzing patterns in qualitative data.
This project was developed as a part of M.S Thesis in Information Technology.
Problem Domain
Understanding human discourse is important and requires in-depth qualitative analysis . The problem is that for social media and other large datasets, the data is too large so it's impossible for people to read everything. So an app that uses topic modeling and clustering of text or images to enable thematic analysis.
Why Thematic Analysis?
Thematic Analysis is a method for identifying and analyzing patterns in qualitative data.
Thematic Analysis can be applied within a range of theoretical frameworks, from essentialist to constructionist; thematic discourse analysis.
Thematic Analysis can be used to analyze communications such as letters, memoranda, reports, or social media texts to identify the intentions of the communicators,
to reveal the focus of the individual, communal, institutional, or societal attention, to describe trends in communication content and to examine attitudes, interests, and values.
Results from the study
To evaluate the system, we interviewed experts from social science who focus on the problem of theme generation and analysis, each who provided different feedback:
- When we are dealing with the complex set of textual data, we need to pay attention to the each of the clusters/topics as well as the intents of such producers of tweets/data.
- To better deal with this issue, our interviewees recommended allowing them to make modifications on the systems encoding.
- Our interviewees wanted to do the data cleaning process themselves. Define their own stopwords or baggerwords.
- Interviewees found our application useful as it allows them to go both top down and bottom up approach.
Conclusion
This application explores the application of computational techniques to support qualitative research. I tried to show how the complementary use of computational and qualitative techniques, and the use of topic modelling could be used to purposively sample for thematic analysis, and can yield insights into the types of discussions occurring in social media at scale, and allow human researchers/social scientists to more deeply engage with those discussions. I expect to see in depth integration of qualitative research and computing moving forward, and I think this work can serve as one model of partnership between human and machine-guided analysis techniques.