Upload Part4.md

Part4.md

@@ -0,0 +1,10 @@
+## Critical Analysis
+
+1. The current model can only take first 500 words in the text due to training resources. It might cause problems when a long text is entered into the model which context is highly relevant. 
+
+2. When the input text is too short, it often judges it as fake news due to our strategy of truncating first 500 words and padding short text.
+
+   ### Further improvement
+
+   1. We may add more features such as title, author information to improve the model.      
+   2. The current pre-trained model RoBERTa is still quite slow when trainning on big datasets. Probably we can try other models and compare the runtime and performance of the models.