A Novel Recommender System Using Interest Extracting Agents and User Feedback
Bilal Hawashin, Ayman Mansour, Farshad A. Fotouhi, Shadi Mahmoud Faleh AlZu’bi, Tarek Kanan · 2021
Recommender Systems have been used widely to provide suggestions to users based on their interests. Although many works have proposed recommender systems in the literature, the majority of these works did not consider the hidden user interests. Even those that considered the user latent interests did not resolve the conflict in interests that could occur. In our previous works, we proposed the use of intelligent agents to extract the interests of user categories such as the interests of a certain gender or the interests of those who work in a certain job. However, there could be some conflicting interests as the user would belong to many categories simultaneously. In this work, we rank the agents according to their ability to extract interests that best represent the users. For this sake, this work takes the user feedback about the most representative agent. Later, it uses classification to predict the most representative agent for new users given their user information. Experimental work proved that our method is efficient in terms of accuracy and training time.