Topic Sentiment Analysis

A paper submitted to AERA2018

Project Description

This paper applied the Latent Dirichlet Allocation (LDA) methods to select 30 topics for 14817 documents in a school blogging system at Teachers College, Columbia University. Sentiment Analysis was conducted to the subsets of documents with different topics by AFINN. Multidimensional (MDS) unfolding method was performed to overlapping of topics, affective and cognitive processes, to identify the associations between emotions and topics. The results indicated that the topics “think and learn”, “music education”, “child development”, “work and think”, “know and think”, “government”, and “build and know” were associated with negative emotions such as anxiety, anger, and sadness. Probabilistic graphical model of 30 LDA topics suggested that there were hidden topics not detected by the word frequencies in LDA algorithms. The directed model of affective and cognitive processes indicated that positive emotions interacted with insights, anxiety led to tentative thoughts and discrepancy. In summary, emotions enhanced thinking processes.

Keywords—blog; online learning community; natural language processing; topic modeling; Latent Dirichlet Allocation; sentiment analysis; unfolding; MDS

People communicate and share thoughts through social engagement and discussion of topics (Vygotsky, 1978). It is important to figure out others’ emotions and opinions during the processing of information gathering (Pang &Lee, 2008). People share their thoughts on Facebook, Twitter, or blogs. The massive user-generated content in social media have become available for social science research. How can we utilize the unstructured contents for quantitative research? This paper approached this question from the text mining perspective, converting the text information into numeric vectors through natural language processing (NLP) techniques, and clustering the posts and comments in a school blogging system with Latent Dirichlet Allocation (LDA) topic modeling.
People express diverse opinions and emotions toward the same issues in social media. Individuals’ behavioral patterns and background information were contained in the self-publishing history at social media. Social scientists were able to interpret the reasoning of diverse decision making and emotions toward the same social issues through the social media data. In this paper, sentiment analysis was applied to the text data in different blogging topics. Considering the topics as nodes, the probabilistic graphical model from the Bayesian network perspective was utilized to map the network of topics and the emotion/cognitive processes.
The goal of this paper was to explore the analytics methods to investigate students’ interested discussion topics and emotions from the text mining perspectives.

[contact-form-7 404 "Not Found"]

Related Projects


We're not around right now. But you can send us an email and we'll get back to you, asap.


©2018 Xiaoting Kuang

Log in with your credentials

Forgot your details?