Solving arithmetic word problems by scoring equations with recursive neural networks

Solving arithmetic word problems is a cornerstone task in assessing language understanding and reasoning capabilities NLP systems. Recent works use automatic extraction ranking of candidate solution equations providing the answer to problems. In this work, we explore novel approaches score such using tree-structured recursive neural network (Tree-RNN) configurations. The advantage Tree-RNN approach over more established sequential representations, that it can naturally capture structure equations. Our proposed method consists transforming mathematical expression equation into an tree. Further, encode tree by different Tree-LSTM architectures. Experimental results show our (i) improves overall performance with than 3% accuracy points compared previous state-of-the-art, 15% on subset require complex reasoning, (ii) outperforms LSTMs 4%