Extraction and evaluation of popular online trends: A case of Pantip.com

Abstract

Popular online trends detection from crowd becomes more and more essential for both trend followers and online sellers. However, huge amount of online posts, both text and images, has prevented trends detection to be manually processed. This article, focusing on text mining, aims to automatically extract popular online trends. A case study is performed on one of the most popular discussion forum websites in Thailand - i.e., Pantip.com. The approach involves employing several unsupervised text mining techniques, namely, TF-IDF and HTML scores, and supervised learning sentiment classification, to extract key topics and assess sentiment results, respectively. Also, we conducted an experiment on the performance of sentiment classification and found that support vector machine (SVM) outperformed other learning techniques. In addition, the authors developed a web- application incorporating the proposed approach. The application collects data from Pantip.com, identifies key concepts of posts and calculates the popularity of each key concept based on statistics and sentiment results.