Park, Sungrae. Associated Topic Model with Social Measurement, Master's Thesis, Department of Industrial and Systems Engineering, KAIST, 2014

Abstract : 

Topic modeling is useful tools to analyze a huge size of text data. Some applications used topic modeling to understand relationship between text data and a time series social measurement, i.e. stock price index, approval rate, etc. However, previous works has regarded topic models as preprocessing models and proposed no joint probabilistic models with the social measurement. This paper suggests two novel probabilistic topic models: Associated Topic Models (ATM) and Indirect ATM. The proposed models automatically find topics, identified in the crawled corpus, associated with the social measurement. This process is ultimately discovering the relationship between text data topics and measurement trends. When we applied the proposed models to financial news articles (Bloomberg) and stock price index (DJIA), ATMs show a higher capability, such as R2, in explaining the relation between the two sources of data compared to the integration of separate models for the two sources, i.e. dynamic topic modeling with regression.

@masterthesis{Park:2014,

author = {Sungrae Park},

advisor ={Il-Chul Moon},

title = {Associated Topic Model with Social Measurement},

school = {KAIST},

year = {2014}

}