Abstract | ||
---|---|---|
The Neural Bag-of-Words (NBOW) modelperforms classification with an average ofthe input word vectors and achieves an impressiveperformance. While the NBOWmodel learns word vectors targeted forthe classification task it does not explicitlymodel which words are important forgiven task. In this paper we propose animproved NBOW model with this abilityto learn task specific word importanceweights. The word importance weightsare learned by introducing a new weightedsum composition of the word vectors.With experiments on standard topic andsentiment classification tasks, we showthat (a) our proposed model learns meaningfulword importance for a given task (b)our model gives best accuracies among theBOW approaches. We also show that thelearned word importance weights are comparableto tf-idf based word weights whenused as features in a BOWSVM classifier. |
Year | Venue | Field |
---|---|---|
2016 | Rep4NLP@ACL | Bag-of-words model,Computer science,Speech recognition,Natural language processing,Artificial intelligence,Classifier (linguistics),Machine learning |
DocType | Citations | PageRank |
Conference | 0 | 0.34 |
References | Authors | |
0 | 4 |
Name | Order | Citations | PageRank |
---|---|---|---|
imran sheikh | 1 | 15 | 4.01 |
irina illina | 2 | 85 | 20.05 |
dominique fohr | 3 | 239 | 49.61 |
georges linar es | 4 | 136 | 29.55 |