Multi-word Embeddings CNN for Identifying Informative Messages

• Simplified implementation of "Convolutional Neural Networks for Sentence Classification" paper

• We introduce a novel approach usingConvolutional NeuralNetwork augmented with multi-word embeddings. The concatenation of embeddings at the embedding layer is utilized to learn local feature for sentenceclassification. Our approach combines i) diverse version of pre-trained wordembeddings and ii) extracts features of the sentence with variable-size convolution filter. We also show the advantages and good performance whencompared with baseline methods in case of disaster data.

---

Requirement

• tensorflow
• keras
• ...

---

Running

• Configure home folder
• run with 'python train.py'
• Check output

---

References

• Kim, Yoon. "Convolutional neural networks for sentence classification." arXiv preprint arXiv:1408.5882 (2014)