Abstract: Set expansion refers to expanding a partial set of "seed" objects into a more complete set. A widely employed approach to set expansion is based on iterative bootstrapping, which can be applied with only small amounts of supervision and which scales bad to very large corpus. A well-known problem with iterative bootstrapping is a phenomenon known as semantic drift: as bootstrapping proceeds it is likely that unreliable patterns will lead to false extractions. To address this issue, a hybrid method for entity set expansion was proposed based on LDA and label propagation. The whole entities in an entity list were considered to prevent words ambiguity; and the LDA used model to mine semantic information in contexts between entity lists to resolve the semantic drift phenomenon. Experiments were conducted with some datasets, and the evaluation demonstrates the effectiveness, efficiency, and scalability of the proposed solution.

Key words: topic model, seed, LDA, label propagation, entity set expansion