Social Media Mining for Assessing Brand Popularity

Abstract

Businesses seek to analyse their customer feedback to compare their brand's popularity with the popularity of competing brands. The increasing use of social media in recent years is producing large amounts of textual content, which has become rich source of data for brand popularity analysis. In this article, a novel hybrid approach of classification and lexicon based methods is proposed to assess brand popularity based on the sentiments expressed in social media posts. Two different classification models using Naïve Bayes (NB) and SVM are built based on Twitter messages for 9 different brands of 3 cosmetic products. In addition, sentiment quantification have been performed using a lexicon-based approach. Based on the overall comparison of the proposed models, the SVM classifier has the highest performance with 78.85% accuracy and 94.60% AUC, compared to 73.57% and 63.63% accuracy, 80.63% and 69.38% AUC of the NB classifier and the sentiment quantification approach respectively. Specific indices based on classification and lexicon approaches are proposed to assess the brand popularity.