Detecting negative deceptive opinion from tweets

Abstract

Nowadays, a huge amount of opinions about specific brands of a company are shared on the Web. Such opinions are an important source of information for customers and companies. Unfortunately, there is an increasing number of deceptive opinions in order to deceive consumers by promoting a low quality product (positive deceptive) or by criticizing a potentially better quality product (negative deceptive). This paper focuses on the detection of negative deceptive opinions from tweets on specific brands of a company. We developed a classifier that detects negative deceptive opinions by combining lexical features of a tweet and personal profile and behavioural features of the writer. One of the challenges to develop this system is the lack of labeled dataset for training and testing. To resolve this issue, we collect our own dataset and label each tweet by multiple experts. Our experimental results show that the proposed system is a promising approach for detecting negative deceptive opinions. Our approach can help to identify defamers by analyzing personal profiles and writing style of each writer.