CS-Embed at SemEval-2020 Task 9: The effectiveness of code-switched word embeddings for sentiment analysis
Code and specs for CS-Embed's contribution to SemEval-2020 Task 9.
- tweet_ids.zip : contains the tweet-id's of the tweets used to create the code-switched embeddings
- tweet_collect.py: code used to collect tweets from twitter using Tweepy and keyword list
- cs_model.py: code used to train bilstm model
- cs_embeddings.tar.gz: word2vec code-switched embeddings with dimension 100. These are the main contribution for SemEval2020: Task 9
| Layer (type) | Output Shape | Param No. |
|---|---|---|
| embedding (Embedding) | (None, 12, 100) | 21592000 |
| bidirectional (Bidirectional) | (None, 12, 256) | 234496 |
| bidirectional_1 (Bidirectional) | (None, 256) | 394240 |
| dropout (Dropout) | (None, 256) | 0 |
| dense (Dense) | (None, 100) | 25700 |
| dropout_1 (Dropout) | (None, 100) | 0 |
| dense_1 (Dense) | (None, 100) | 10100 |
| dropout_2 (Dropout) | (None, 100) | 0 |
| dense_2 (Dense) | (None, 3) | 303 |
Total params: 22,256,839 Trainable params: 22,256,839 Non-trainable params: 0
Hyperparameters of BiLSTM Model
- Optimiser: Adamax
- Learning rate:0.0002
- EarlyStopping: min_delta=0.0001, patience=5
If any code or models are used please cite:
@InProceedings{Leon2020, author = {Frances A. Laureano De Leon and Florimond Guéniat and Harish Tayyar Madabushi}, title = {CS-Embed at SemEval-2020 Task 9: The effectiveness of code-switched word embeddings for sentiment analysis}, booktitle = {Proceedings of the 14th International Workshop on Semantic Evaluation ({S}em{E}val-2020)}, year = {2020}, address = {Barcelona, Spain}, month = {December}, publisher = {Association for Computational Linguistics}, }