How to mine brand Tweets: Procedural guidelines and pretest

Abstract

This paper presents a methodological framework for using opinion mining to analyse comments on social networking sites. A series of procedural recommendations is described and compared with the content analysis method. The major steps include brand selection, determination of a classification scheme and categories, human coding, programming of the automated classification algorithm, and evaluation of the classification results. We then present the results of a pretest that examined the content of Tweets about IKEA. After human coding of 100 Tweets, the automated classification was carried out. The Precision measure achieved more than 65% for the first classification (Satisfaction, Dissatisfaction and Exclude) and 64% for the second classification (Sharing, Information, Opinion, Question, Reply and Exclude), demonstrating the efficiency of mining Tweets for emotional patterns. Combining the two classification schemes, the pretest performs a social network analysis to identify interrelationships among the Tweets. In closing, methodological implications and utility for marketing research are discussed.