Graph-based Model for Negative e-WOM Influence in Social Media

Nowadays, several companies use social media marketing to increase profit and control the market. The customer's feedback has a powerful influence on company reputation by conveying their experience in social media. Customers exchange their feedback about the services using electronic Word-of-Mouth (e-WOM). Negative feedback could help companies improve their service to increase profit. In this work, we propose an approach to determine the effect of negative e-WOM relating to a company's products or services. Firstly, we apply a machine-learning algorithm called random forest to classify e-WOM on three classes based on polarity: Positive, negative, or neutral. Secondly, we group negative e-WOM into different clusters based on their topics using a similarity method named cosine similarity. Thirdly, we generate an influence graph of negative e-WOM based on time precedence and social ties. Finally, we analyze the resulted graph to identify risk patterns and convey useful information. The provided method is implemented using Python and is tested with collected data.