| Nowadays, the amount of biomedical literatures is growing at an explosive speed. Researchers struggle to maintain expertise and knowledge of developments in their fields. Dealing with the huge amount of information has led to a fragmentation of scientific literature, which promoting poor communication between specialties. Swanson initiated hidden knowledge discovery in biomedical literature and formed several hypothesis.Many other researchers have successfully replicated Swanson's discoveries, and literature based discovery has become an popular topic in text mining.The popular methods based on co-occurrence produce too many target concepts which will lead to the decline of really relevant target concepts in ranking. This paper presents a new method for selecting linking concepts. This method uses the statistical and textual features to represent each linking concept and then classifies them as relevant or irrelevant to the starting concepts. The relevant linking concepts are used to discover target concepts. In this way, the amount of target concepts is greatly reduced and the really relevant target concepts can gain higher rankings, which helps the biomedical experts to discover potential target concepts efficiently. We also employ this method in the investigation of H1N1, which achieves better precision and F score. At last, we make a prediction of the substances which may affect H1N1.Many researchers utilize UMLS's semantic resource in literature based discovery. Event similarity calculated by semantic similarity between concepts show better result than statistical methods such as tf*idf. But events with high semantic similarity may lead to unreasonable hypotheses due to lacking of semantic relevancy. This paper uses UMLS's semantic network to calculate semantic relevancy between concepts, and apply F score to trade-off semantic similarity and semantic relevancy. The experimental results show Fish oils and Magnesium obtains better rankings. |
PDF Full Text Request