Convolutional Neural Networks for Sentence Classification

Сверточные нейронные сети для классификации предложений
Yoon Kim
2014-01-01

Convolutional Neural Networksfine-tuningpre-trained word vectorssentence classificationsentiment analysis
We report on a series of experiments with convolutional neural networks (CNN) trained on top of pre-trained word vectors for sentence-level classification tasks.We show that a simple CNN with little hyperparameter tuning and static vectors achieves excellent results on multiple benchmarks.Learning task-specific vectors through fine-tuning offers further gains in performance.We additionally propose a simple modification to the architecture to allow for the use of both task-specific and static vectors.The CNN models discussed herein improve upon the state of the art on 4 out of 7 tasks, which include sentiment analysis and question classification.
1
A proposed simple architecture modification enables simultaneous use of both task-specific (fine-tuned) and static vectors.
2
A simple CNN trained on pre-trained word vectors with little hyperparameter tuning achieves excellent sentence-level classification results across multiple benchmarks.
3
Fine-tuning word vectors for specific tasks provides additional performance gains over static vectors.
4
The described CNN models improve state-of-the-art results on 4 out of 7 evaluated tasks, including sentiment analysis and question classification.
5
Using static (non-fine-tuned) pre-trained word vectors already yields strong performance for sentence classification tasks.

Convolutional neural network models for sentence-level classification trained on top of pre-trained word vectors

Effect of CNN architectures and word-vector strategies (static vs. fine-tuned task-specific, and combined) on sentence-level classification performance across benchmarks

Publication Details
Publication Date
2014-01-01
Journal
Publisher
ISSN
Cited by
14831
Access Type
Author Information
Authors
Yoon Kim
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%