An automata based approach to biomedical named entity recognition

ing an automata learning algorithm: Causal-State Splitting Reconstruction [1]. This algorithm has previously been applied to Named Entity Recognition [2] obtaining good results given the simplicity of the approach. The same approach has been applied to Biomedical NE identi?cation, using GENIA corpus 3.0, with 10-fold cross-validation. Our system attained F1 = 73.14%. These results can be compared directly to [3] and [4], which used the same data. First system obtains F1 = 57.4% using ME Models, and the second one reports F1 = 79.2% using SVMs. Both improve their results using post-processing techniques, reaching F1 = 76.9% and F1 = 79.9% respectively. Our system does not use any post-processing techniques, and takes into acount few features, so the results are considered very promising. In future work some post-processing will be developed to improve the results.