Recommendation systems have become pervasive in every aspects in our lives. We think it would be interesting to understand, implement, and test out different recommendation methods and compare their performance on the MovieLens dataset.
What it does
Our project is the Stochastic Gradient Descent method to file out the blank ratings so that we can recommend the movies with the movies with top predicted ratings to our users. To compare the performance of our algorithm, we buily two base lines using item-item Collaborative Filtering and user-user Collaborative Filtering as our benchmarks.
How we built it
Challenges we ran into
The biggest challenges is that the RMSEs (Root Mean Square Errors) for the two base line methods were way to large. Thus, we converted the problem into an optimization problem to minimize the RMSE, which will trigger the overfitting of course.
Accomplishments that we're proud of
We were able to figure out a way to solve the overfitting problem by using Stochastic gradient descent
What we learned
We learned more machine learning methods and got more practice with Python and debugging
What's next for Movie Recommendation Analysis
We still think that there are a lot of rooms for improvement. Especially, due to the process capacity and time limits, we were not able to implement more robust method. But we would love to try to account the similarity of other features of movies and users into our similarity computation; Moreover, we want to add the influence factors to the time of the ratings, so that we can value the more recent ratings more than the older ones.