Coin_flipper at eHealth-KD Challenge 2019: Voting LSTMs for Key Phrases\n and Semantic Relation Identification Applied to Spanish eHealth Texts

This paper describes our approach presented for the eHealth-KD 2019\nchallenge. Our participation was aimed at testing how far we could go using\ngeneric tools for Text-Processing but, at the same time, using common\noptimization techniques in the field of Data Mining. The architecture proposed\nfor both tasks of the challenge is a standard stacked 2-layer bi-LSTM. The main\nparticularities of our approach are: (a) The use of a surrogate function of F1\nas loss function to close the gap between the minimization function and the\nevaluation metric, and (b) The generation of an ensemble of models for\ngenerating predictions by majority vote. Our system ranked second with an F1\nscore of 62.18% in the main task by a narrow margin with the winner that scored\n63.94%.\n