Multi-task Ensemble Learning for Fake Reviews Detection and Helpfulness Prediction: A Novel Approach

Proceedings of Recent Advances in Natural Language Processing, pages 721–729 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_078 721 Multi-task Ensemble Learning for Fake Reviews Detection and Helpfulness Prediction: A Novel Approach Alimuddin Melleng Anna-Jurek Loughrey Deepak P Queen’s University Belfast, UK alimuddinmllg@gmail.com, a.jurek@qub.ac.uk, deepaksp@acm.org Abstract Research on fake reviews detection and review helpfulness prediction is prevalent, yet most studies tend to focus solely on either fake re- views detection or review helpfulness predic- tion, considering them separate research tasks. In contrast to this prevailing pattern, we address both challenges concurrently by employing a multi-task learning approach. We posit that undertaking these tasks simultaneously can en- hance the performance of each task through shared information among features. We uti- lize pre-trained RoBERTa embeddings with a document-level data representation. This is coupled with an array of deep learning and neural network models, including Bi-LSTM, LSTM, GRU, and CNN. Additionally, we em- ploy ensemble learning techniques to integrate these models, with the objective of enhancing overall prediction accuracy and mitigating the risk of overfitting. The findings of this study offer valuable insights to the fields of NLP and machine learning and present a novel perspec- tive on leveraging multi-task learning for the twin challenges of fake reviews detection and review helpfulness prediction. 1 Introduction The proliferation of online marketplaces has signif- icantly altered the way consumers purchase goods and services. As part of this transformation, user- generated reviews have become a vital factor in influencing purchasing decisions. However, the in- creased reliance on these reviews has given rise to an unsettling phenomenon: the spread of deceptive or ”fake” reviews. Fake reviews, either overly pos- itive or overly negative, can distort the perceived quality or popularity of products or services, mis- leading consumers and affecting businesses’ repu- tations. Simultaneously, the concept of review helpful- ness has emerged as another crucial aspect of user- generated reviews. Helpfulness prediction aims to rank and highlight reviews that potential consumers would find most useful. It is based on the premise that not all reviews provide the same value to con- sumers, and certain reviews are more informative and helpful than others. Accurate helpfulness pre- diction can thus enhance the shopping experience by guiding consumers towards reviews that offer the most beneficial insights. An example of helpful and unhelpful review: Helpful: ”I purchased this phone two weeks ago and have been using it ever since. The battery life is impressive, and the screen is bright and colour- ful. The camera produces high-resolution images, especially in night mode, which delivers fantastic results.” Unhelpful: ”I bought this phone as a gift for my daughter and she’s happy with it. The delivery was quick and the packaging was satisfactory.” Recently, multi-task learning, a paradigm of ma- chine learning, has been recognized as a promising approach to improve the performance of related tasks (Ruder, 2017; Xue et al., 2017; Fan et al., 2018). Multi-task learning operates on the prin- ciple that learning multiple tasks simultaneously, leveraging shared representations, can lead to im- proved generalization by …

Content truncated to 500 words