facebookresearch / DrQA

Reading Wikipedia to Answer Open-Domain Questions
Other
4.48k stars 898 forks source link

Bias In Retriever towards Long Texts #254

Open kuldeep7688 opened 4 years ago

kuldeep7688 commented 4 years ago

I tried the retriever module for getting documents related to the question but unfortunately almost everytime long documents were suggested as the best matched.

I tried to find out whether the tf vector is normalized in the compressed sparse matrix creation but couldn't.

Can someone help me whether I am right or wrong ? And has anyone noticed this ?