Mono-Lingual Search Engine: Combining Keywords with Context for Semantic Search Engine

The success of pre-trained models relies on their capacity to understand human language and the success of word embeddings and transformers that allow access to large pre-trained NLP models. However, finding suitable architectures and techniques for monolingual semantic search is a challenge as this research focuses deeply on English. Our approach focuses on combining keyword-based matching and semantic search using different search methods to increase precision and recall. In this paper, we present an analysis of the different approaches adopted by our semantic search engine and the comparison between them. Our experiment on datasets from different domains has shown the efficiency and effectiveness of the adopted method. Indeed, our method achieved an accuracy of 0.89 which improves by up to 4% when combining keywords with semantics in search engines.