Investigating Lstm with k-Max Pooling for Text Classification
Bo Shu, Fuji Ren, Yanwei Bao · 2018
For most deep learning practitioners, text classification is synonymous with recurrent networks and is tackled at word level. The presented study is demonstrated that word-level inputs are better than character-level inputs. We describe the proposed model that utilizes Spatial-Dropout1D, LSTM and k-max pool. Because the word of these representations is measured in a word similarity task, so SpatialDropout can reduce information redundancy; LSTM keep track of information over longer periods of time. k-max pooling keep more useful information. The proposed model is better than the strong baseline methods on six benchmark datasets for sentiment classification and topic categorization. The code will be open.