Semantic Searching and Ranking of Documents using Hybrid Learning System and WordNet

Semantic searching seeks to improve search accuracy of the search engine by understanding searcher’s intent and the contextual meaning of the terms present in the query to retrieve more relevant results. To find out the semantic similarity between the query terms, WordNet is used as the underlying reference database. Various approaches of Learning to Rank are compared. A new hybrid learning system is introduced which combines learning using Neural Network and Support Vector Machine. As the size of the training set highly affects the performance of the Neural Network, we have used Support Vector Machine to reduce the size of the data set by extracting support vectors that are critical for the learning. The data set containing support vectors is then used for learning a ranking function using Neural Network. The proposed system is compared with RankNet. The experimental results demonstrated very promising performance improvements. For experiments, we have used English-Hindi parallel corpus, Gyannidhi from CDAC. F-measure and Average Interpolated Precision are used for evaluation.