relation_extractor is a program for training relation extraction models, written in Python and with models implemented using TensorFlow. The model architecture is a bi-directional LSTM with attention, as outlined in the paper by Zhou et al. [1].
The implementation here is in part an object-oriented reimagining of this version from SeoSangwoo, with some additional tweaks.
Models trained using the current implementation reach 77.28 F1 (macro-averaged official score for SemEval2010 Task #8) - a little shy of the 84.00 reported in the paper, so still some improvements to be made.
[1] Zhou, P., Shi, W., Tian, J., Qi, Z., Li, B., Hao, H. and Xu, B. (2016) "Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification". Proceedings of the 54th Annual Meeting of the Association for Computational Linguitsics (ACL '16) [pdf]