CFE 2015: Start Registration
View Submission - CFE
A1112
Title: In quest of significance: Identifying types of Twitter sentiment spikes that predict events in sales Authors:  Olga Kolchyna - University College London (United Kingdom) [presenting]
Tharsis Souza - University College London (United Kingdom)
Tomaso Aste - University College London (United Kingdom)
Abstract: We introduce a sentiment classification approach that combines the traditional lexicon-based approach with support vector machines algorithm. Using this model we analyse over 150 million tweets related to 75 companies from the retail sector and study the power of Twitter sentiment to predict sales events for the selected companies. To perform the analysis we developed a robust method for identifying and clustering bursts in sales and Twitter series based on their shape. The result of events clustering suggests that Twitter time series can be separated into six clusters with unique signatures. We demonstrate that some types of Twitter events have predictive power that is more significant than the predictive power of the aggregated Twitter sentiment signal. The methodology can be easily extended to the field of finance.