Improving Slot Filling Performance with Attentive Neural Networks on Dependency Structures

摘要

Slot Filling (SF) aims to extract the values of certain types of attributes(or slots, such as person:cities\_of\_residence) for a given entity from alarge collection of source documents. In this paper we propose an effective DNNarchitecture for SF with the following new strategies: (1). Take a regularizeddependency graph instead of a raw sentence as input to DNN, to compress thewide contexts between query and candidate filler; (2). Incorporate twoattention mechanisms: local attention learned from query and candidate filler,and global attention learned from external knowledge bases, to guide the modelto better select indicative contexts to determine slot type. Experiments showthat this framework outperforms state-of-the-art on both relation extraction(16\% absolute F-score gain) and slot filling validation for each individualsystem (up to 8.5\% absolute F-score gain).