Doo-Sup Choi, Keyword Influence Kalman Filter for Keyword Frequency Estimation, Master's Thesis, Department of Industrial and Systems Engineering, KAIST, 2016

Abstract

News articles present a series of events which influence on our communities with great or small impacts. Each article involves representative key words which may provide an information of the corresponding phenomena. Therefore, analyzing the sequencial frequencies of the keywords in the articles will provide an insight to understand historical social events and to predict upcoming events. Additionally, a set of keywords has complex relations which give view points of causality of events. This paper used Kalman filter model, which is one of the state-of-the-art models to release sensitivity of the features white noises, to predict the word-level frequencies and catch the causal relationships between them. Specially, we applied social influence theorem to model relation between keywords. When we applied the proposed approach, our approach get improvements at prediction performance for highly related keywords with others. 

@masterthesis{Choi:2016,

author = {Doo-Sup Choi},

advisor ={Il-Chul Moon},

title = {Keyword Influence Kalman Filter for Keyword Frequency Estimation},

school = {KAIST},

year = {2016}

}