Research Article Matching Methods using Attention, Hybrid, RNN, and SIF

In the digital era, search engines such as Google Scholar make it easier to access scientific information, but there are challenges in determining the relevance and accuracy of articles. One technique used to select scientific articles relevant to a particular topic is entity matching. This study aims to analyze the performance of entity matching using four models owned by DeepMatcher. The models include Attention, Hybrid, RNN, and SIF, which are applied to a dataset of articles with keyword fairness in AI. The collected dataset consisted of 50 articles retrieved from Publish or Perish 8. The dataset was preprocessed up to pairing, resulting in a final dataset of 1225 rows and eight columns. The columns follow the column types that must be present in the dataset that will be processed by DeepMatcher, namely the "left," "right," "label," and "ID" attributes. Based on the results of the analysis, it was found that the most suitable DeepMatcher model, namely the RNN model, had the highest average F1 score of 54.08%.