Semi-Open Information Extraction

Abstract

Open Information Extraction (OIE), the task aimed at discovering all textual facts organized in the form of (subject, predicate, object) found within a sentence, has gained much attention recently. However, in some knowledge-driven applications such as question answering, we often have a target entity and hope to obtain its structured factual knowledge for better understanding, instead of extracting all possible facts aimlessly from the corpus. In this paper, we define a new task, namely Semi-Open Information Extraction (SOIE), to address this need. The goal of SOIE is to discover domain-independent facts towards a particular entity from general and diverse web text. To facilitate research on this new task, we propose a large-scale human-annotated benchmark called SOIED, consisting of 61,984 facts for 8,013 subject entities annotated on 24,000 Chinese sentences collected from the web search engine. In addition, we propose a novel unified model called USE for this task. First, we introduce subject-guided sequence as input to a pre-trained language model and normalize the hidden representations conditioned on the subject embedding to encode the sentence in a subject-aware manner. Second, we decompose SOIE into three uncoupled subtasks: predicate extraction, object extraction, and boundary alignment. They can all be formulated as the problem of table filling by forming a two-dimensional tag table based on a task-specific tagging scheme. Third, we introduce a collaborative learning strategy that enables the interactive relations among subtasks to be better exploited by explicitly exchanging informative clues. Finally, we evaluate USE and several strong baselines on our new dataset. Experimental results demonstrate the advantages of the proposed method and reveal insight for future improvement.