Hybrid ensemble and soft computing approaches for review spam detection on different spam datasets

Abstract

Online customer reviews play an important influence in a client's choice to purchase an item and are seen as a significant source of knowledge that can be used to foresee public opinion when making purchase or commodity from an online store. These opinions are frequently manipulated by people who haven't been registered on the websites or who are paid by other organisations to create negative or misleading reviews about an item or brand. Spam reviews are generally fraudulent reviews, and they can have a significant impact on the digital marketplace's behaviour. With the growing need to convert such facts into usable evidence and knowledge, information retrieval and data mining have gotten a lot of attention. The modelling of documents into a clear and accurate set of characteristic single and numerous word phrases, which has an impact on overall actions and achievements, is predicted to be a key problem in natural language processing (NLP) known as feature extraction process. Quite a few kinds of features could be used to perform different tasks through various feature collection methods such as: Text categorization (Bag of words), Linguistic characteristics (LIWC), Genre Identification (POS tagging), Sentimental feature extraction, Term frequencies (TFIDF) and Word2Vec. For traditional classification tasks with smaller datasets, machine learning algorithms have produced the greatest results. Soft computing techniques, on the other hand, when optimised with larger datasets, provide a better solution for real-world scenario problems. This work is organised into several sections, each of which examines a distinct feature selection method, machine ensemble learning, and soft computing approaches in depth. Finally, the hybrid approach for detecting review spam is introduced.