DEEP LEARNING FOR HOTEL REVIEW SENTIMENT: A COMPARATIVE STUDY OF SUPERVISED AND SEMI-SUPERVISED MODELS
Abstract
Modern business and trade are greatly impacted by online reviews. The majority of the decision-making process when buying things online is influenced by user reviews. Because of this, shady people or organizations attempt to influence product reviews in order to forward their own agendas. Online reviews have emerged as the most significant source of consumer feedback in recent years. More and more people and businesses are using these reviews to guide their business and buying decisions. Unfortunately, scammers have created false (spam) evaluations because they are motivated by the need for money or attention. Because of the fraudsters' actions, enterprises that are restructuring their businesses and prospective clients are misled, and opinion-mining tools are unable to draw reliable findings. There are two basic methods for producing fake reviews. First, via the use of human content writers who are paid to create evaluations of things that look legitimate but are really written by someone who has never seen the products in question. Secondly, in a (b) computer-generated manner by automating the construction of the fictitious review utilizing text-generation algorithms. In the past, human-generated fake reviews were exchanged on an internet platform like commodities in a "market of fakes"; buyers could order reviews in any amount, and human authors would do the task. But as text generation technology has advanced, particularly in natural language processing (NLP) and machine learning (ML), automated fake reviews have become more and more attractive. With generative language models, for example, fake reviews can now be produced at a much lower cost than those created by humans, and at a larger scale. In order to identify fraudulent online reviews, this study presents a few semi-supervised and supervised text mining algorithms and evaluates the effectiveness of each method using a dataset of hotel reviews.
