Clustering based causal topic mining

Events in the world generate an enormous amount of textual data like tweets and news articles. These events also manifest in the form of changes to time-series numeric data. This thesis deals with the problem of extracting these events from the timestamped document collection in the form of topics that cause a change in a time-series. We develop a conceptual framework for that can be used to analyze different causal topic mining algorithms. We also propose two novel clustering based algorithms - cCTM-CF and cCTM-CoF to generate causal topics. We evaluate these algorithms both qualitatively, and quantitatively by comparing their coherence and correlation scores to that of the baseline generative causal topic model - gCTM. We found that cCTM-CoF performs 35% and 62.5% better according to these metrics as compared to the baseline.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2017

Author dc:creator

• Mohan, Vishaal

Contributors dc:contributor

• Zhai, ChengXiang

• Text mining
• Topic models
• Time series

Statement dc:rights

• Copyright 2017 Vishaal Mohan

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/97626

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/97626