Lexicalized Dependency Paths Based Supervised Learning for Relation Extraction

Log-linear models and more recently neural network used for supervised relation extraction requires substantial amounts of training data time, limiting the portability to new relations domains. To this end, we propose a representation based on dependency paths between entities in tree which call lexicalized (LDPs). We show that is fast, efficient transparent. further representations utilizing entity types its subtypes refine our model alleviate sparsity problem. apply learning using ACE corpus it can achieve similar performance level other state-of-the-art methods even surpass them several categories.