anilguven/bert_tr_turkish_tweet

Model Info

This model was developed/finetuned for tweet emotion detection task for the Turkish Language. This model was finetuned via tweet dataset. This dataset contains 5 classes: angry, happy, sad, surprised and afraid.

• LABEL_0: angry
• LABEL_1: afraid
• LABEL_2: happy
• LABEL_3: surprised
• LABEL_4: sad

Model Sources

• Dataset: https://huggingface.co/datasets/anilguven/turkish_tweet_emotion_dataset
• Paper: https://ieeexplore.ieee.org/document/9559014
• Demo-Coding [optional]: https://github.com/anil1055/Turkish_tweet_emotion_analysis_with_language_models
• Finetuned from model [optional]: https://huggingface.co/dbmdz/bert-base-turkish-uncased

Preprocessing

You must apply removing stopwords, stemming, or lemmatization process for Turkish.

Results

• eval_loss = 0.06813859832385788
• mcc = 0.9843707754295762
• Accuracy: %98.75

Citation

BibTeX:

@INPROCEEDINGS{9559014, author={Guven, Zekeriya Anil}, booktitle={2021 6th International Conference on Computer Science and Engineering (UBMK)}, title={Comparison of BERT Models and Machine Learning Methods for Sentiment Analysis on Turkish Tweets}, year={2021}, volume={}, number={}, pages={98-101}, keywords={Computer science;Sentiment analysis;Analytical models;Social networking (online);Computational modeling;Bit error rate;Random forests;Sentiment Analysis;BERT;Machine Learning;Text Classification;Tweet Analysis.}, doi={10.1109/UBMK52708.2021.9559014}}

APA:

Guven, Z. A. (2021, September). Comparison of BERT models and machine learning methods for sentiment analysis on Turkish tweets. In 2021 6th International Conference on Computer Science and Engineering (UBMK) (pp. 98-101). IEEE.