Utilizing Noun-Verb Extraction in Enhancing Information Retrieval

Abstract

The increasing growth of the news and a large number of users on information retrieval (IR) has resulted in making the retrieval of documents complex and more difficult. The IR process consists of pre-processing, extraction and representation, feature selection and indexing, querying, and retrieving results. The weakness of IR is concerned with the process of extraction, where most important words focused on verbs or nouns. Unsupervised Feature selection is an important task in content classification for being among the most popular and effective methods for retrieval reduction. The use of the verbs and nouns as extraction was recently introduced in the IR technique to avoid irrelevant and redundant unsupervised features. This paper aims to enhance IR using noun-verb extraction using Word-Net and Krill Herd Algorithm as Unsupervised Feature Selection (KHUFS) combine with Simulated Annealing as Unsupervised Feature Selection to find the suitable retrieval of ranking. The external of Mean Average Precision (MAP) and Mean Average Recall (MAR) internal of Mean Average Distance (MAD) as measurements were used to verify the proposed retrieval of ranking. The results demonstrate that the proposed nouns verbs method extraction outperformed other extraction methods, in which the proposed extraction was 26.15 % using MAP measure and using MAR was 45.41%. In comparison with other unsupervised feature selection algorithms such as Harmony Search, Simulated Annealing, Particle Swarm, and Genetic Algorithm, the combined combination outperformed other unsupervised feature selection algorithms with an accuracy 26.163 % MAP, and 11.653% MAR. On the other side, the effect of use proposed extraction on the proposed unsupervised feature selection was 39.563% MAP, and 8.96% MAR. The other evaluation using the number of features, the using combined Krill Herd with Simulated Annealing number of features has decreased to more than 50 %, which the total feature in the dataset was 10682 features and after used the proposed was 4723 features.