Semantic Search on Flight Manual using Ensemble Approach

Abstract

Every piece of hard printed paperwork, such as articles and user manuals, has been digitized
in this new era, and everything is instantly accessible via the internet. So, if we need to find a
specific piece of content in a document, we don't have to go through the process of reading
and finding it page by page; instead, there are a number of search engines that can do it for us
in a very quick manner. When it comes to search engines, it employs a variety of NLP
techniques to do the task. NLP approaches include semantic search technique which is a very
effective method of searching that uses the context of the text rather than the keyword match
methodology. The first section of the study will look at three different Semantic search
algorithms, including Universal Sentence Encoder (USE), Latent Semantic Indexing (LSI),
and Sentence BERT (Bidirectional Encoder Representations from Transformers), and how
they work with the pilot’s flight manual. The following part focuses on ensemble these 3
semantic methods and testing it with the same flight manual. The performance of these four
semantic approaches is demonstrated in the experimental results, and the optimal way is
chosen and concluded.