In this repository, we are working on studying propagation of bias towards author groups in a book recommendation setting.
We are applying our research methods to the Book-Crossing dataset, as processed in our repository EnrichBookCrossing.
Our overall goal is to evaluate hidden bias towards author characteristics that comes as a direct result of popularity bias. In this work, we are looking into author country of citizenship. We find that popularity is not independent to author country of citizenship. Specifically, American-authored books tend to be more popular in the data. We also find that certain algorithms on average recommend more American-authored books compared to the users' profile. Finally, we find that these are the same algorithms that propagate popularity bias according to previous work by Naghiaei et al.
- Book data analysis
- Book Recommendation (on Colab)
- Book results analysis
Certain code segments taken from Fairbook .