Choosing Better Seeds for Entity Set Expansion by Leveraging Wikipedia Semantic Knowledge

Entity Set Expansion, which refers to expanding a human-input seed set to a more complete set which belongs to the same semantic category, is an important task for open information extraction. Because human-input seeds may be ambiguous, sparse etc., the quality of seeds has a great influence on expansion performance, which has been proved by many previous researches. To improve seeds quality, this paper proposes a novel method which can choose better seeds from original input ones. In our method, we leverage Wikipedia semantic knowledge to measure semantic relatedness and ambiguity of each seed. Moreover, to avoid the sparseness of the seed, we use web corpus to measure its population. Lastly, we use a linear model to combine these factors to determine the final selection. Experimental results show that new seed sets chosen by our method can improve expansion performance by up to average 13.4% over random selected seed sets.