Retrieval with Semantic Query Expansion Using Context-Based Graph Approaches

Semantic Query Expansion (Semantic QE) is a method used to better understand user queries that have ambiguity such as synonyms and polysemy, and to overcome vocabulary mismatch problems. Semantic QE can provide other topics on candidate terms relevant to the query, however, the provided topics can lead to the retrieval of irrelevant documents. In this paper, we proposed semantic QE using graph Word2Vec on Wikipedia knowledge for document searching. Article information in Wikipedia was used for candidate terms. All words in articles were tokenized and used as the corpus, and semantic similarity will be calculated using Word2Vec. Graph representation using each word in the article as nodes and the similarity of all the words results from Word2Vec as edges. Nodes with a minimum similarity threshold will be selected and used as an additional extended query together with the original query in the final document search. The experiment was conducted with two types of data: monolingual data with 950 questions and answers pair and Cross-Lingual Information Retrieval (CLIR) with 50 questions and 200 answers pair. The results of the two data are: monolingual problem with the proposed method has the highest result compared to others (traditional method, Word2Vec, and WordNet) with the recall of 78.6% and the result of CLIR problem with second-highest results on success rate $\mathbf{1 0}$ documents (SR10K) of $\mathbf{6 7 \%}$. It can be concluded that the proposed method effectively improves the relevance of results in document searching.