I have a movie and user-ratings dataset. After implementing the content-based filtering technique, I figured, I can improvise the results even further by assigning weightage to the parameters based on intuition and practical understanding.
For e.g. Popularity of the movie plays lesser role than the Genre in recognising the similar movies. But I am not able to determine how much less? How should I know if I should set Wpopularity:Wgenre to 0.1:0.9, 0.2:0.8 or 0.3:0.7; such that it does not bias the model to an unethical extent (as this can be very critical in some critical recommender systems, like those involving human rights, gender equality, etc.)? Is there a known practice or convention for this? Is there any research paper available to refer for this matter?
Scenario Reference - refer the 50-50 weightage implemented in https://youtu.be/_hf_y-_sj5Y?t=1239
Hi, to optimize your neural network hyper-parameter first you can use different algorithms like genetic algorithm, marine predators algorithm, Gray wolf algorithm and so on to find which one is better.
the second advice is to limit hyper-parameter in small range for example for number of hidden units use [50,100] then for next Run use [100,150] so do not use wide range like [50,300] in this way your optimization algorithm will stuck in local minima and can not be efficient .
the third advice is to optimize 5 or 6 parameter not all of them, optimize 5 or 6 parameter and consider the rest as constant number for example optimize num hidden units, num layer, initial learn rate, drop factor, L2 regularization and take other parameters as suitable constant number.
don't forget if you want to optimize your neural network by yourself it will take a lot of time so it's better to use optimization algorithms, it will optimize network bias and weight.
good luck
- Badges
- Science topic
- Similar topics
- Computer Science
- Program