I have to find the best 10 items from the set of items for x number of given features. I don't know what the best recommendation will be. User data is not available to validate it. After reading a number of papers I came to know about voting algorithm.
Trying to implement something like this-
- Find x number of features for static variables( that doesn't change over time), dynamic variables (that change very frequently).
- Assume some weights for the weights on the basis of requirement for dynamic variables.
- Try to come up with 3-4 regression models by combining the static and dynamic variables and find the best 10 items from each
- Take the common one out, and then calculate the error.
- Back-propagate the error to get the new weights.
- Keep on iterating till we get at-least common top 8 elements.
Is the approach right?
Any guidance for what to take as error.
My approach to error calculation was-
- Mean of common items feature values (X).
- For each model MSE where mean is X and Xi is feature values for uncommon items of that model.
I am new to it. Can anyone please guide me through it.