I would like to develop a weighted hybrid recommendation system from multiple data sources.

Given are:
1. Explicit feedback: on different products in the range of 0 to 10
(0 means no feedback exists here)

Implicite feedback:
2. Exact purchases data coded binary (0 means no purchase by userX on itemY)
-> very sparse 1% of users
3. Click data coded as integer from 0 to XX (means how often a user has clicked on a particular product page) -> sparse 10% of users

The first question is: How could I transform all the different feedback ranges to one comparable Rating-Matrix?

And the second question is: How I could design the recommander?
My first idea was to calculate 3 different recommender systems based on every single matrix and than combine those in a hybrid system. But I don't know if this is a very useful way in terms of data source sparsity of matrix 2 and 3.

Is there any other idea?


1 Answer 1


Take a look at Hybrid Recommender Systems: Survey and Experiments, specifically table 3 has a list of approaches for combining different kinds of data sources.

Regarding your first question, you can scale the different metrics to lie in the same range (for eg. between 0 and 1). A popular scaling approach to compute the scaled rating $R_{scaled}$ between some $a$ and $b$ is: $R_{scaled} = \frac{(b - a)(R - min)}{max - min}$, where $min$ and $max$ are the minimum and maximum of the original rating scale.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.