Measuring influence by including Latent Semantic Analysis in Twitter conversations

Abstract

As the mount of information is growing in social media, online influence estimation is becoming significant and an time consuming element in social media analytic. In the last few years therefore there have been several algorithmic approaches to automate the estimations. Examples of such algorithms are ExpertiseRank and Klout Score. In this thesis, we propose an online influence estimation algorithm. We name it XRank. XRank is a novel approach to include content analysis into the traditional influence estimation domain. In traditional estimation techniques they mainly use metadata like followers or friends. In our proposed solution, Latent Semantic Analysis(LSA) enables XRank algorithm to have capability of estimating online influence based on given topic. By designing and implementing an algorithm prototype and testing with different dataset sizes, vocabulary size and vocabularies with different topics, we measure how these parameters affect XRank result. We also compare the XRank estimation result with another online influence estimation algorithm called Klout Score.The testing results suggests that XRank shows satisfactory performance based on given topic. We believe that the result will provide a new point of view to online influence estimation.