Portal de Periódicos da CAPES

Twitter Sentiment Tracking for Predicting Marketing Trends

2015; Springer International Publishing; Linguagem: Inglês

10.1007/978-3-319-14178-7_2

2191-6594

Cagdas Esiyok, Şahin Albayrak,

Web Data Mining and Analysis

We present a web-based Twitter sentiment tracking tool for brands. The tweets about four companies, namely, Facebook, Twitter, Apple, and Microsoft are collected by this system. The collection is implemented in an hourly basis in 17 Anglophone cities from which these tweets are sent. After collecting the tweets, the system classifies them as positive or negative by using the Naïve Bayes and Maximum Entropy classification methods. Later on, the system determines the winner brand of each city according to the percentage of positive tweets sent by users located in the aforementioned cities. Lastly, the winner brands of the day can be monitored on a web page using Google Maps. To increase the performance of classification methods, the tweet texts are preprocessed, such as through converting all the letters to lower case, both for training hand-classified dataset and for the collected tweets. Furthermore, statistical tracking charts can be viewed via web page of the system. A dataset is produced by collecting 362,529 tweets in 9 days via Twitter API for the research, which is automatically classified by the system. Performance of the Naïve Bayes and Maximum Entropy classification methods is also evaluated with the hand-classified dataset.