<![CDATA[Identifying fake reviews in the current digital era has become an interesting study, especially in cross-domain context. This process is based on the limitations of the situation, where not all review domains have labels, and the labeling process takes a long time. In the context of machine learning, cross-domain learning involves learning from data and prediction processes, using data from different domains. Identifying fake reviews, several previous studies have conducted cross-domain, and the results of these studies indicate that there are still several issues in detecting fake reviews. The main problem with cross-domain methods is the difficulty of the model in understanding the differences in characteristics between domains, such as differences in language style, word structure in reviews, and the context present in the reviews. The main problem with the cross-domain method is the difficulty of the model in understanding the differences in characteristics between domains, such as differences in language style, word structure in reviews, and the context present in reviews. Based on these issues, this research adopts an approach to identify fake reviews using the Convolutional Neural Network and Bidirectional Long Short Term Memory models, utilizing the Multi Feature Rolling Collaborative Training (MRCT) algorithm with data from Yelp dan Amazon. The experimental results show that by conducting two scenarios, Scenario-1 provides better performance with an accuracy of 98.59%, while scenario-2 is only capable of providing an accuracy performance of 79.64%. Additionally, by using multi-features, the model experienced a 24.96% improvement in detecting fake reviews across domains. Based on these results, it can be seen that the use of multi-features and rolling collaborative training with the CNN-BiLSTM model works effectively in identifying fake reviews across domains.]]>
