EXPLORE PUBLICATIONS BY COUNTRIES

Biomedical Literature , Feature Selection, Hierarchical Text Classification, Ontology Alignment, Text Mining

Classifying biomedical literature is a difficult and challenging task, especially when a large number of biomedical articles should be organized into a hierarchical structure. Due to this problem, various classification methods were proposed by many researchers for classifying biomedical literature in order to help users find relevant articles on the web. In this paper, we propose a new approach to classifying a collection of biomedical text abstracts by using ontology alignment algorithm that we have developed. To accomplish our goal, we construct the OHSUMED disease hierarchy as the initial training hierarchy and the Medline abstract disease hierarchies as our testing hierarchy. For enriching our training hierarchy, we use the relevant features that extracted from selected categories in the OHSUMED dataset as feature vectors. These feature vectors then are mapped to each node or concept in the OHSUMED disease hierarchy according to their specific category. Afterward, we align and match the concepts in both hierarchies using our ontology alignment algorithm for finding probable concepts or categories. Subsequently, we compute the cosine similarity score between the feature vectors in probable concepts, in the genrichedh OHSUMED disease hierarchy and the Medline abstract disease hierarchy. Finally, we predict a category to the new Medline abstracts based on the highest cosine similarity score. The results obtained from the experiments demonstrate that our proposed approach for hierarchical classification performs slightly better than the multi-class flat classification.