Mining User’s Opinions and Emojis For Reputation Generation Using Deep Learning

摘要

Millions of people express their opinions and viewpoints about products and services directly online. Twitter is one of the popular platforms used by internet users. This massive amount of people’s opinions is very relevant for automatically computing and drawing a reputation score. Therefore, having a system capable of generating a reputation value from reviews shared by the users will be a beneficial tool for companies to assess the reputation of their outputs and make favorable changes. However, a limited number of studies have inquired mining opinions expressed in natural language for reputation generation. Therefore, our approach consists of a system that generates a reputation by applying the deep learning Bidirectional Encoder Representations from Transformers (BERT) as an embedding layer and using a multi-layer Gated recurrent units (GRU) which learns from the representations produced by the transformer. Also, processing the emojis included in the opinions with a simple technique for a reliable reputation value. Experimental results conducted on two twitter datasets about different products show that our system provides the nearest reputation value to the ground truth, which is weighted average vote for each of our products provided by (IMDB and Yelp). This implies that the proposed approach can be applied in real world applications.