Related Works:
	"Amazon.com Recommendations" surveys many recommendation systems, including collaborative filtering, cluster models, and item-to-item collaborative filtering. The study defines a clear set of evaluation metrics, click through and conversion rates. In this study, the proxy for success will be a user's clicking into a data visualization suggestion; there is no concept of conversion, or rating that the visualization was helpful. Collaborative filtering finds a neighborhood of users with the smallest distance from a new user. The items bought by the neighborhood are then rank-ordered for the new user. Cluster model first clusters the user base and then calculates distance with a new user.

	"Analysis of Recommendation Algorithms for E-Commerce" elaborates on traditional collaborative filtering. The framework involves three stages: representation of feature vectors, neighborhood formation, and recommendation generation. Representation involves deciding how to store user feature vectors. Neighborhood formation is how to choose a set of users similar to a given user; KNN and aggregation can be used to form neighborhoods. Finally recommendation generation includes either most-frequent item recommendation and association rule-based recommendation to select items given a neighborhood of users.

	"Automatic Selection of Partitioning Variables for Small Multiple Displays" is a machine learning application in data visualization. The study focuses on finding data partitions that score highest in a pre-defined "goodness criteria," which includes visually rich, information, well-supported, and parsimonious. The goal of this study is to not assert a single "goodness critera," but to vary the behavior of the recommendation system based on information about a user. 





Introduction - An explanation of the problem and the motivation for solving it.
Related Work - A description of previous papers related to your project.
	Amazon Recommendations: 
		click through and conversion rates as important measures of recommendation effectiveness.
			In our case, time spent looking at visualization is proxy for success. Downside is time could be correlated with not understanding a visualization.
		Traditional collaborative filtering:
			Find customers whose distance is small from new customer. 
			Rank order all visualizations from those customers.
		Cluster Model:
			Cluster previous customers into segments and find cluster that new customer is most similar to.
	Analysis of Recommendation Algorithms for E-Commerce: http://files.grouplens.org/papers/ec00.pdf
		Association rules to rank products

		Collaborative Filtering
			Representation
			Neighborhood Formation
			Recommendation Generation
	K means clustering: https://sites.google.com/site/dataclusteringalgorithms/k-means-clustering-algorithm

	Small Multiple system: https://research.tableau.com/paper/automatic-selection-partitioning-variables-small-multiple-displays
		Define a "goodness criteria" for small multiple displays. In our system we learn this criteria for specific users.
			Perhaps an aggregate goodness criteria would include what others thing as well.
			Visually rich, informative, well-supported, parsimonious






Methods - A detailed explanation of the techniques and algorithms you used to solve the problem.
Results - The visualizations your system produces and data to help evaluate your approach. For example you may include running times, or the time users typically spend generating a visualization using your system.
Discussion - What has the audience learned about visualization from your work?
Future Work - A description of how your system could be extended.