Abstract | ||
---|---|---|
The distributed vector representations learned from the deep learning framework have shown its great power in capturing the semantic meaning of words, phrases and sentences, from which multiple NLP applications have benefited. As words combine to form the meaning of sentences, so do sentences combine to form the meaning of documents, the idea of representing each document with a dense distributed representation holds promise. In this paper, we propose a supervised framework (Compound RNN) for document classification based on document-level distributed representations learned from deep learning architecture. Our framework first obtains the distributed representation at sentence-level by operating on the parse tree structure from recursive neural network, and then obtains the document presentation-level by convoluting the sentence vectors from a recurrent neural network. Our framework (Compound RNN) outperforms existing document representations such as bag-of-words, LDA in multiple text classification/regression tasks. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1007/978-3-319-18038-0_17 | ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING, PART I |
Field | DocType | Volume |
Data mining,Parse tree,Computer science,Machine translation,Recurrent neural network,Artificial intelligence,Natural language processing,Deep learning,Document classification,Architecture,Support vector machine,Sentence,Machine learning | Conference | 9077 |
ISSN | Citations | PageRank |
0302-9743 | 3 | 0.37 |
References | Authors | |
39 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Rumeng Li | 1 | 3 | 0.71 |
Hiroyuki Shindo | 2 | 75 | 13.80 |