Realtime Semantic Similarity Analysis of Bulk Outlook Emails Using BERT

摘要

Semantic similarity (SS) analysis is a technique for finding similarities between words/sentences/documents based on their meaning. In natural language processing (NLP), SS is an important element to find a suitable mail from a bulk inbox. As the number of mails and mail content increases, it becomes difficult to get the matches with keywords and nearly impossible for many cases. This paper presents a method to find SS between query statements and mail content using BERT (Bidirectional Encoder Representations from Transformers). BERT is a pre-trained unsupervised NLP model developed by Google. The results are presented and compared with the existing keyword-based search to prove the efficiency of the proposed approach.