Using Inductive Transfer Learning to Improve Hotel Review Spam Detection

The buying decisions of consumers across the world are increasingly influenced by online reviews. With this in mind, there exists increased opportunity for unscrupulous business owners or spammers to create false reviews to promote their brand or smear those of their competitors. As a result, there have been many studies of ways to identify review spam. However, a common limitation in these studies is the lack of large amounts of labeled training data. This is problematic because Deep Learning is susceptible to overfitting when trained on small datasets. In this paper, we empirically explore the use of inductive transfer learning to help with the problem by using a language model trained on Wikipedia and fine-tuning it for the task of hotel review spam detection. We show that it is possible to use inductive transfer learning to create a deep learning model that outperforms traditional bag of words based approaches on datasets with limited labeled data.