What is it about?
Fairness in information retrieval is a vital target. There many parameters that may issue bias towards retrieving specific set of documents more than others. In this paper , we investigate different document representations including unigrams , bigrams or both and see their effect on fairness and effectiveness as well.
Featured Image
Photo by Nicolas Jossi on Unsplash
Why is it important?
we found that Bigrams is the worst performer and fairest one as well. Also , combining unigrams & bigrams gives the best performance with a reasonable bias. This might be helpful for others to choose the document representation that mostly satisfies their requirements whether to focus on performance or fairness.
Perspectives
I ran lots of experiments while altering many parameters then I measured the results to view the trend that shows the effect of each document representation method on performance and bias. For me , it was nice , simple and long process that save time and effort for others . I hope that others try more sets to figure out the trend for example trigrams.
ABDULAZIZ ALQATAN
University of Strathclyde
Read the Original
This page is a summary of: Analyzing the Influence of Bigrams on Retrieval Bias and Effectiveness, September 2020, ACM (Association for Computing Machinery),
DOI: 10.1145/3409256.3409831.
You can read the full text:
Contributors
The following have contributed to this page
