Information Retrieval Using Topic Modeling Techniques on User Generated Content

Urban community resilience assessment has drawn a lot of attention recently and being explored at various levels of society. For more comprehensive understanding of the urban community challenges to derive indicators to quantify resilience capabilities, social media platform can prove stronger data resources as they are widely used to disseminate peoples lived experiences. User generated content on social media platforms are goldmines of the user viewpoints and perceptions. Information Retrieval from this unstructured data lead to gather deeper insights into peoples' lives without any direct intervention. Topic modelling is text analytics technique which helps to dissect the data to generated themes with any a-priori training data. This paper presents comparison of various topic modeling algorithms and detailed analysis of Latent Dirichlet Allocation (LDA) method which used to identify topics from user generated content mined from Reddit platform as a community sharing platform. Data collected is location based sub-reddit posts for four cities in India. The paper also presents results of the topic modeling to suggest that the themes discovered through this automated process is reliable and expandable for further research.