FUTURE GOALS
Even though we achieved high performing models, there are many more methods we can try. We think it would’ve been really interesting to look into the lyrics, likely using the bag of words method and assigning different values to different words for both the lyrics and annotations. We could then look into what words or kind of words in an annotation are associated with more votes given the lyrics of a song. We were also proud that our feature selection lead to significant improvements so we can continue to narrow down our search using other methods such as other forms of regressions such as polynomial, kernel, or local linear regression to find non linear fits of our features on total votes. Finally, we hope to develop this as an application with an attractive front end where users can plug in an annotation and immediately find how many votes they will get.