Masters Final Oral: Rihui Li

Event
Tuesday, May 10, 2016 - 10:00am
Event Type: 

Title:  Tweet classification for political policy topics
Date/Time: May 10th, 2016 @ 10:00 AM
Place: 223 Atanasoff Hall
Faculty Advisor: Professor Johnny Wong

Abstract:

Twitter is a popular online social software for people's social networking: sharing their current status, mind, etc. It is also widely used in politics, such as running e-campaigns, mining or affecting public opinions etc. We study the problem of automatically classifying the tweets posted from state's senate accounts or representative accounts in the United States to 21 topics which are defined by policy agenda project. To get a even distribution for tweets number in different topics in training set, we grouped several topics together into a combined group by using the technique of sliding windows and sorting the tweets by date. For the testing set, as we don't know which topic the tweet belongs to, we tried several approach for the grouping in testing set, such as similarity based method, center based method, Convolutional Neural Networks based method. After grouping training set and testing set respectively, the overall accuracy of tweet topic classification could be around 75%.

Rihui Li.pdf

Category: 
Tags: