Case Study on Veracity in Twitter Data Using Oil Company Related Tweets

Abstract

Twitter is a powerful real-time micro-blogging service and it is a platform where users provide and obtain information, called tweets, at a rapid pace. Because of the volume, velocity, and unstructured nature of tweets, Twitter data can be viewed as big data. In this thesis we study the veracity of tweets using oil industry related tweets. Previous research has shown that most of the tweets posted on twitter are truthful. But the same platform (Twitter) is also used often to spread misinformation intentionally or unintentionally. There is no definitive measures to determine the veracity of tweets based on the tweets themselves. So there is a need for better mechanisms to measure levels of accuracy from tweets.In this thesis, we propose three measures to estimate the veracity/accuracy of topics based on analysis of tweets. They are topic diffusion, geographic dispersion, and spam rate. We collect tweets associated to topics. Using the tweets we compute the measures and estimate the veracity of topics. Reliable geographic dispersion data was not available in our data set and hence it is not used in validation process. To validate measures, we verity the tweeted information using official data. For this study we streamed oil industry data. Several topics were identified for our analysis. In the case of each topic, tweets unrelated to the topic are considered noise. After noise elimination, tweets are classified according to company names, then the proposed measures are computed. The results are compared against the verification results. In majority of cases, the estimates of veracity of topics by the proposed measures are confirmed by the verification results.