# What similarity measures would you recommend for product vectors given only 1, 0, and -1?

I am currently building a product recommendation engine using item based filtering techniques.

Each user on our site can leave the following feedback: Like (represented by 1), Not seen (v = 0) and Dislike (v = -1)

We are trying to calculate similarities between two items by measuring the cosine distance between two product vectors, where item_i 's product vector is defined to be { user1, user2, user3, .. } 's score for that particular item.

Currently, we only have "likes" available at this moment to seed our data. This means that each vector is going to end up having only 1s and 0s, i.e. {1, 0, 0, 1, ... }.

We are using cosine similarity atm but we found out that the similarity score between our products are either 1 or 0 given the nature of seed data.

I am interested to look for a similarity measure that can fit my case. Can anyone recommend me something useful for now?

-
Would correlation be interesting? – Tunococ Jan 22 '13 at 11:53
@Tunococ not really. – vng Jan 23 '13 at 0:38
What is the "nature of seed data"? Why do you get either $1$ or $0$ and not any bigger integers? – Tunococ Jan 23 '13 at 9:52
This book contains a table of similarities/dissimilarities: amzn.to/VxG1tr – Memming Jan 31 '13 at 15:55