Abstract: In order to address the issue that sentences may contain context-information with reference to the facts, this paper presents a Clause-Level Context-aware Open Information Extraction approach (ClauseContextIE). ClauseContextIE extends the scale of context-information that can be identified, and takes advantage of the dependency-parsing to extract the context-information and general clauses in a top-down way, so that it can construct a graph that expresses the hierarchical structure. Finally, ClauseContextIE assigns the corresponding context-information to each tuple extracted from general clauses in a bottom-up approach. ClauseContextIE avoids extracting context-information as a relation tuple, and assigns context-information to correct relation tuples accurately. Experiments were conducted to compare ClauseContextIE with ReVerb, OLLIE and ClausIE on three datasets, ReVerb dataset, Wiki dataset and NYT dataset. Experimental results show that ClauseContextIE achieves significantly higher accuracy and recall than the other extractors.

Key words: computer application, open information extraction, context-aware, clause-level, dependency parsing