Neural joint learning for classifying wikipedia articles into fine-grained named entity types

This paper addresses the task of assigning finegrained NE type labels to Wikipedia articles. To address the data sparseness problem, which is salient particularly in fine-grained type classification, we introduce a multi-task learning framework where type classifiers are all jointly learned by a neural network with a hidden layer. In addition, we also propose to learn article vectors (i.e. entity embeddings) from Wikipedia's hypertext structure using a Skipgram model and incorporate them into the input feature set. To conduct large-scale practical experiments, we created a new dataset containing over 22,000 manually labeled instances. The dataset is available. The results of our experiments show that both ideas gained their own statistically significant improvement separately in classification accuracy.