Transformer induced enhanced feature engineering for contextual similarity detection in text

Availability of large data storage systems has resulted in digitization information. Question and answering communities like Quora stack overflow take advantage such to provide information users. However, as the amount stored gets larger, it becomes difficult keep track existing information, especially duplication. This work presents a similarity detection technique that can be used identify levels textual based on context which was provided. transformer contextual (TCSD), uses combination bidirectional encoder representations from transformers (BERT) metrics derive features data. The derived are train ensemble model for detection. Experiments were performed using question set. Results comparisons indicate proposed exhibits with an accuracy 92.5%, representing high efficiency.