Ekspansi Kueri pada Sistem Temu Kembali Informasi Berbahasa Indonesia Menggunakan Analisis Konteks Lokal

Abstract

Users express their needs in a query to get information using a information retrieval system. However, many are not able to compose appropriate queries as the average length of their queries were too short. Another problem is word mismatch, which refers to the phenomenon that the users of information retrieval systems often use words to describe the concept in their queries which are different from words that authors use to describe the same concept in their documents. Local context analysis is an automatic query expansion which is a combination of global and local techniques. Like global techniques, local context analysis select expansion features based on their co-occurrences with the query terms. Like local techniques, it selects expansion features from the top retrieved documents for a query. Local context analysis ranks the concept by their co-occurrences with the query term in the top ranked documents and uses the highest ranked concepts for query expansion. Basically, a document consists of topics, so in this research, the top ranked documents are divided into passages which represent topics in the relevant document. The highest ranked concepts are then taken from top ranked passages. The purpose of this research is to implement query expansion with local context analysis. The performance of information retrieval system with local context analysis gave good result with around 60% average precision. The results showed that the retrieval performance using local context analysis was significantly higher based on statistical analysis using t-test. The average precision increased by 6.07% compared to retrieval without local context analysis, indicating relevant documents occur higher in the retrieval result. The results also showed that the number of top-ranked documents and passages did not significantly affect the average precision. The more influential factor was the number of query expansions added. Local context analysis is quite suitable for collection of relatively similar documents.