Department of Computer Science

Permanent URI for this community

https://hdl.handle.net/1805/41295

Browse

Browsing Department of Computer Science by Author "Alfantoukh, Lina"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Using Twitter trust network for stock market analysis
    (Elsevier, 2018-04) Ruan, Yefeng; Durresi, Arjan; Alfantoukh, Lina; Computer and Information Science, School of Science
    Online social networks are now attracting a lot of attention not only from their users but also from researchers in various fields. Many researchers believe that the public mood or sentiment expressed in social media is related to financial markets. We propose to use trust among users as a filtering and amplifying mechanism for the social media to increase its correlation with financial data in the stock market. Therefore, we used the real stock market data as ground truth for our trust management system. We collected stock-related data (tweets) from Twitter, which is a very popular Micro-blogging forum, to see the correlation between the Twitter sentiment valence and abnormal stock returns for eight firms in the S&P 500. We developed a trust management framework to build a user-to-user trust network for Twitter users. Compared with existing works, in addition to analyzing and accumulating tweets’ sentiment, we take into account the source of tweets – their authors. Authors are differentiated by their power or reputation in the whole community, where power is determined by the user-to-user trust network. To validate our trust management system, we did the Pearson correlation test for an eight months period (the trading days from 01/01/2015 through 08/31/2015). Compared with treating all the authors equally important, or weighting them by their number of followers, our trust network based reputation mechanism can amplify the correlation between a specific firm’s Twitter sentiment valence and the firm’s stock abnormal returns. To further consider the possible auto-correlation property of abnormal stock returns, we constructed a linear regression model, which includes historical stock abnormal returns, to test the relation between the Twitter sentiment valence and abnormal stock returns. Again, our results showed that by using our trust network power based method to weight tweets, Twitter sentiment valence reflect abnormal stock returns better than treating all the authors equally important or weighting them by their number of followers.