amal572 / Multi-task-Training-NER-and-Sentiment-Analysis

MULTI-TASK TRAINING ModeLs WITH HUGGING FACE Arabic TRANSFORMERS (Arabert, Marbert, and Qarib)

Multi-task-Training-NER-and-Sentiment-Analysis

Problem statement

MULTI-TASK TRAINING MODELS WITH HUGGING FACE Arabic TRANSFORMERS:

1- we work on Real Arabic Hotel Reviews to :

• classify them into good, natural, and bad reviews

• at the same time, we get the NER for each word.

2- Explore data then Clean it and Analysis the Result.

3- Using Huggingface Arabic Transformers such as Arabert, Marbert, and Qarib.

4- Build and Train a Multi-Task Model then we use it to Predict the test set for every single Task.

5- Evaluate each multitask model with both classification and NER tasks:

result

model sentimenet_valid_acc sentiment_test_acc ner_test_acc

arabert 0.847211 0.849907 0.732296

marbert 0.845594 0.838789 0.735117

qarib 0.854487 0.843731 0.724035