JunKeon Park, Dynamic Topic Modeling with Neural Variational Inference, Master's Thesis, Department of Industrial and Systems Engineering, KAIST, 2018

Abstract

The vast amount of time-series documentation that occurs over time requires efficient and effective analysis. In this paper, we propose a topic model for time series documents by applying neural variational inference to analyze dynamic topics. The proposed model capture the topic change over time using LSTM, which is a kind of Recurrent Neural Network. The model introduces attention network to increase the prediction accuracy so that the past information can be used positively. As a result, our proposed model shows better performance than the existing topic models for three time series documents.

@masterthesis{Park:2018,

author = {JunKeon Park},

advisor ={Il-Chul Moon},

title = {Dynamic Topic Modeling with Neural Variational Inference},

school = {KAIST},

year = {2018}

}