Add tf & tf-idf cosine based similarity comparisons

[SailSlick]

connects to #373, #374

Add new similarity function based on methods described in Content-based Recommendation in Social Tagging Systems link

• Add 2 new similarity metrics
• Clean all output cells of jupyter notebook (had no idea you could do this before)
• Clean tag input a bit

Cosine similarity

Averages from 3 runs:

n1   n2   n3   AUC
1933 	102  	3110 	0.426623
1923 	107  	3115 	0.436296
1958 	99   	3088 	0.431699

Cosine TF-IDF frequency

Averages from 3 runs:

n1   n2   n3   AUC
2047 	149  	2949 	0.455078
2023 	142  	2981 	0.449217
2032 	131  	2982 	0.451105

issues:

• Most movie tags are not repeated often/ not that many tags.
  • Could be solved by using the larger movielens dataset (3,600 tag applications applied to 9,000 movies) vs (1,100,000 tag applications applied to 58,000 movies)

[iandioch]

The AUC metric shouldn't give results below 0.5. I'm not sure why this metric guarantees this (it doesn't seem apparent in the basic AUC formula we're using), or why exactly here we're getting low results, but your work here seems mostly fine.