A New Approach for Better Document Retrieval and Classification Performance Using Supervised WSD and Concept Graph

Word Sense Disambiguation (WSD) is main task in the area of natural language processing (NLP). Supervised WSD methods are shown to be more effective than other WSD methods with the limitation of the size of manual annotated learning set. On the other hand, Concept graph is a weighted graph with each of its edges representing the relationships between concepts (relevancy of each pair of concepts). In this paper, we propose a method to improve the retrieval and classification performance of documents from different sources by means of concept graph. In our method, some features are initially selected from a training set by applying a well-known feature selection algorithm. Then, by injecting suggested relevant words for each class from the concept graph, a more enriched feature set is produced to apply to the test set. Our experimental results exhibit an improvement of 14.6% and 18.4% (few and more term injection evaluations, respectfully) in classification and also some improvements in retrieval performance.