mosh98 / Swedish_Sentiment_BERTIL

Sentiment analysis using KB-BERT

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Swedish_Sentiment_BERTIL

I named it BERT_IL becuase its Swedish ;)

I wrote a small paper on how to fine tune KB BERT model for sentiemnt analysis using around ~15000 examples. This was an exploratory investigation for current swedish BERT model(released 2020).

The results show KB-BERT performs with a 60% accuracy when it comes to sentiment analysis. Which is very poor however there were a few papers that depicts that if bert is trained with dataset containing around ~20k examples, it could perform very well.

Bert performing results from this paper graph

F1 score: 0.5711702412679611

Precision Score: 0.6927536231884058

Recall Score: 0.6867816091954023

You can find the paper here. mm

About

Sentiment analysis using KB-BERT


Languages

Language:Jupyter Notebook 100.0%