Data set for recommendation system

5
votes

I want to create own simple recommendation system, about books. But there are some problems - it's impossible (at least, very hard) for one person to organize training data set for algorithms.

So, are there any free data sets or quizzes with information about people's votes, which books and how much they like?

Second question is about book's parameters. For some item-based predictions is really must have to use book's ratings (e.g. language, average length of words, average number of words in paragraph, i have counted about 30 parameters like those) and their weights (for example, book's language is rated in 1 point, and average length of words in 0.314). So, are there any prepared information about that?

In fact, if i got an answer for first question, i could find solution for second question, but i am sure, that needed information exists.

Also, i am reading Recommender Systems Handbook, it gives full information (with references), but it is hard to read. Can you advise some extra books in this case?

2
data-miningrecommendation-engine

2 Answers

9
votes

Could you check Books.txt.gz at; https://snap.stanford.edu/data/web-Amazon.html which consists book ratings from Amazon. it also has product title, price, review summary etc.

Also bookcrossing dataset might be useful http://grouplens.org/datasets/book-crossing/

I guess your second question is a feature selection problem and weights will be different for each dataset.

this course at coursera gives brief information for recommendation systems and it also has a reading part. unfortunately quizzes are no longer available

course:https://www.coursera.org/course/recsys

readings: http://recsys.cs.umn.edu/readings.html

Edit: Yet another dataset for books.

Goodbooks:

http://fastml.com/goodbooks-10k-a-new-dataset-for-book-recommendations/

0
votes

This dataset is about movies rather than books, but you might find the Netflix Prize dataset useful as a way of testing recommendation algorithms. The underlying issues are the same with both datasets : needing out-of-band features, having to combine features with different weights, etc.

As for extra books to read, I recommend "Programming Collective Intelligence." I found it to be clearly written and very helpful. It also includes code for all of the example algorithms.