Name Nationality Identification with

LSTM + Character Level Embedding + Attention Mechanism

This is the project for Stanford CS230 at spring quarter of 2020. The project is developed based on Jinhyuk Lee et al's work of [Name Nationaltiy Classification with Recurrent Neural Networks]. Our model adds an additional quadgram embedding and the attention mechanism on LSTM to extract more features in order to improve the performance. The dataset is from https://github.com/d4em0n/nationality-classify which contains (name,nationality) pairs from 18 countries. Due to sudden time-cutting for the project, we don't have enough time to fine-tune the model. But the good news is that the two figures below give an encouraging experiment results. We can see that for an attention length of 5 and a batch size of 1024, the learning rate within 0.001 and 0.003 shows very exciting and promising accuracy for test set.

Requirements

• Python 3
• Tensorflow 1.0.1 (GPU enabled)
• Gensim

New (added) directories

(Please check https://github.com/63coldnoodle/ethnicity-tensorflow for origin directories)

• main_ad.py : Running our advanced model with adjustable hyperparameters
• model_ad.py : Model structure for the advanced model
• ops_ad.py : Tensorflow ops used in model_ad.py
• dataset_ad.py : Dataset reading and experiment workflow
• preprocess_own.py : Preprocessing 18-countries dataset for the advanced model
• preprocess_ad.py : Preprocessing crawled dataset in Jinhyuk Lee et al's work for advanced model
• preprocess_test.py : Preprocessing 18-countries dataset to fit in Jinhyuk Lee et al's model
• data/own/ : Preprocessed data from preprocess_own.py
• data/name_nationality/ : 18-countries dataset
• data/ad/ : Preprocessed data from preprocess_ad.py
• data/test/ : Preprocessed data from preprocess_test.py
• data/test with $/ : Preprocessed data from preprocess_test.py (cleaned verison as mentioned in Jinhyuk Lee et al's work)
• Own dataset/ : Experiment results of preprocessed data from preprocess_own.py and preprocess_test.py
• RNN dataset/ : Experiment results of preprocessed data from preprocess_ad.py

Run the code

$ python main_ad.py

Or you can try the ipynb file in 'Own dataset' or 'RNN dataset' for convenience.

Contact

• Wei Ren ([email protected])
• ZhenHuan Hong ([email protected])
• Jiate Li ([email protected])

Acknowledgement

We thank for Jinhyuk Lee's response to help the development of the project. We also thank the teaching team of CS230 for your support and patience.