Improving Query Expansion Performances with Pseudo Relevance Feedback and Wu-Palmer Similarity on Cross Language Information Retrieval

• Published in: 2022 9th International Conference on Advanced Informatics: Concepts, Theory and Applications (ICAICTA) pp. 1 - 6
• Main Authors: Pratama, Muhammad Akmal, Mandala, Rila
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.09.2022
• Subjects: Informatics; Information retrieval; mean average precision; query expansion; relevance feedback; Semantics; Wu-Palmer similarity
• DOI: 10.1109/ICAICTA56449.2022.9932984

With more information available in multiple languages, the need to search for relevant information is no longer fixated only on one language. Cross language information retrieval, a system that search for relevant information in different language, experienced a decrease in performance due to the loss of meanings during translation process or the initial query that was not descriptive of the information to be sought. One method to improve the performance of the information retrieval system relevance feedback query expansion. Another method utilizes external resources such as WordNet as the basis for generating query expansion term. This study utilizes the Wu-Palmer semantic similarity measurement in WordNet to improve the performance of pseudo relevance feedback query expansion. The terms contained in the feedback document considered as candidate expansion terms. Each candidate is given weight based on IDF score, Wu-Palmer similarity score, and document score using Okapi ranking function. Several candidate terms with the highest weight are used in query expansion. Experiment is conducted by comparing mean average precision value of query expansion based Rocchio feedback with query expansion based Wu-Palmer semantic similarity using NPL collection. The result show that the query expansion method of pseudo relevance feedback with WuPalmer semantic similarity can have better performance than the Rocchio pseudo relevance feedback for document feedback less than five. The highest mean average precision is obtained when two document feedback and ten expansion term is used with 0.5605 compared to Rocchio pseudo relevance feedback with only 0.4499.