IMPROVING HOTEL REVIEW ANALYSIS WITH DEEP LEARNING: A FOCUS ON SUPERVISED AND SEMI SUPERVISED MODELS

Abstract

Online reviews have great impact on today’s business and commerce. Decision making for purchase of online products mostly depends on reviews given by the users. Hence, opportunistic individuals or groups try to manipulate product reviews for their own interests. In recent years, online reviews have become the most important resource of customers’ opinions. These reviews are used increasingly by individuals and organizations to make purchase and business decisions. Unfortunately, driven by the desire for profit or publicity, fraudsters have produced deceptive (spam) reviews. The fraudsters’ activities mislead potential customers and organizations reshaping their businesses and prevent opinion-mining techniques from reaching accurate conclusions. Fake reviews can be created in two main ways. First, in a (a) human-generated way by paying human content creators to write authentic-appearing but not real reviews of products — in this case, the review author never saw said products but still writes about them. Second, in a (b) computer-generated way by using text-generation algorithms to automate the fake review creation. Traditionally, human-generated fake reviews have been traded like commodities in a “market of fakes” – one can simply order reviews online in each quantity, and human writers would carry out the work. However, the technological progress in text generation – natural language processing (NLP) and machine learning (ML) to be more specific – has incentivized the automation of fake reviews, as with generative language models, fake reviews could be generated at scale and a fraction of the cost compared to human-generated fake reviews. This work introduces some semi-supervised and supervised text mining models to detect fake online reviews as well as compares the efficiency of both techniques on dataset containing hotel reviews.