I investigated whether there was any bias towards reviews written as part of the Vine Program from Amazon, INC.
I used Google Colaboratory and Pyspark for ETL. I also loaded the data into a relational database using PGAdmin. Lastly, I created filtered dataframes from the original dataset after processing to discover any bias with vine reviews.
The following output from the notebook shows the results from the filtering:
-
There were 26 total reviews for paid vine programs.
-
There were 29,264 total reviews for unpaid ones.
-
There were 5 five-star rated reviews for the paid vine programs
-
There were 15,784 five star reviews for those that were unpaid
-
I calculated the precentage of each type of vine and got these results:
- 34.62% were five star ratings in the paid dataframe...
- 53.94% were five stars in the unpaid dataframe.
-
There seems to be a strong positivity bias in favor of reviews which are unpaid.
- This could be because there are a lot more of them as compared to the ones that are from paid vines.
-
I could perform a sample T-test for further analysis.
- A sample T-test would show exactly how skewed and biased the results were based on the p-value generated from T-Tests on both paid and unpaid dataframes.