Distant supervision has attracted recent interest for training information extraction systems because it does not require any human annotation but rather isting knowledge bases to he bel a training corpus. Howe work has failed to address of false negative training exa beled due to the incompleten edge bases. To tackle this propose a simple yet novel fr combines a passage retrieval coarse features into a state-o tion extractor using multi-in ing with fine features. We formation retrieval techniqu relevance feedback to expan bases, assuming entity pairs i passages are more likely to e tion. Our proposed technique improves the quality of dis vised relation extraction, bo from 47.7% to 61.2% with a high level of precision of around 93% in the experiments.