A relevant document search system model using word2vec approaches

Abstract

USU Repository is an institutional digital information system provided by Universitas Sumatera Utara (USU) that preserves and distributes academic papers, such as thesis and dissertation, from all departments in USU. A search box is provided to help search relevant topics from this repository. However, sometimes the search results returned were irrelevant and did not satisfy the user’s expectations. One of the causes for this situation is that the search engine could not perform optimally, particularly if the query were long and complicated. One approach that can be used to solve the problem is using semantic search. Semantic search is an information retrieval process from a sentence that involves understanding the results returned by natural language processing. In light of this approach, this paper aimed to propose a semantic search to seek the relevance between the user’s input query and academic papers returned as search results. This study implemented word2vec method in converting sentences into vectors. This study indicated average precision scores for small datasets as 46% and 73% for larger datasets.