Biomedical semantic indexing by deep neural network with multi-task learning
BMC Bioinformatics, pp. 5022018.
Our model uses the word2vec word embedding to represent the words in the citations, and the Bidirectional Gated Recurrent Units is used to create the representation of the data
Rather than parallel in an ordinary multi-task structure, the tasks in our model are serial and tightly coupled. It can achieve satisfied performance without any handcrafted feature.
PPT (Upload PPT)