https://github.com/zhaozhengChen/RegionEmbedding

Mxnet implementation of an ICLR 2018 paper: A new method of region embedding for text classification.
https://github.com/zhaozhengChen/RegionEmbedding

Last synced: 2 months ago
JSON representation

Mxnet implementation of an ICLR 2018 paper: A new method of region embedding for text classification.

• Host: GitHub
• URL: https://github.com/zhaozhengChen/RegionEmbedding
• Owner: zhaozhengChen
• Created: 2018-08-31T10:18:03.000Z (over 6 years ago)
• Default Branch: master
• Last Pushed: 2018-10-14T08:23:22.000Z (about 6 years ago)
• Last Synced: 2024-08-01T22:41:48.794Z (5 months ago)
• Language: Python
• Homepage:
• Size: 54.7 KB
• Stars: 10
• Watchers: 1
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-MXNet - **RegionEmbedding**

README

        
# RegionEmbedding

MXNet implementation of ICLR 2018 paper: [A new method of region embedding for text classification](https://openreview.net/forum?id=BkSDMA36Z).

Official implementation in [TensorFlow](https://github.com/text-representation/local-context-unit).

## 0.Notes

- I implemented both Word-Context and Context-Word Region Embedding in the paper.

- Please see the original papar about the datasets and pre-pocessing.

- All the hyper-parameters I used are copied from the official implementation.

## 1.Requirements

- Python2 or Python3

- Mxnet 1.2.1

## 2.Results

|Datasets| Accuracy(%)
WordContext|Best Epoch
WordContext|Accuracy(%)
ContextWord|Best Epoch
ContextWord|Running Time
Per Epoch(mins)

| :-- | :--: | :--: | :--: | :--: | :--: |

|Yahoo Answer|73.07(73.7)|2|73.42(73.4)|3|110|

|Amazon Polarity|95.27(95.1)|2|95.36(95.3)|3|247|

|Amazon Full|61.58(60.9)|2|61.59(60.8)|2|183|

|Ag news| 92.96(92.8)|6|92.89(92.8)|8|2|

|DBPedia|98.91(98.9)|4|98.88(98.9)|3|23|

|Yelp Full| 64.98(64.9)|3|64.94(64.5)|2|25|

Note: 

- The accuracy in brackets are results reported in the original paper.

- The running speed is much faster than the origin implementation in Tensorflow. 

  
The running time was tested on the model of context-word region embedding, which run roughly the same as the word-context region embedding. 

- The code run on a Titan Xp GPU.