Our website uses cookies to provide you with a better online experience. Please select “Accept”.

Embrace Users’ Contexts with LSCAR Recommender System

Researchers with mutual interest in recommender systems known as the Recommender Community are spread across continents, and this includes Dr. Tipajin Thaipisutikul, an instructor at Mahidol University’s Faculty of Information and Communication Technology. Since she realizes its potential to drive economic growth and increase revenue for e-commerce and service industries, especially entertainment services, she mindfully crafted the research article “Exploiting Long-and Short-Term Preferences for Deep Context-Aware Recommendations”.

“In this original research, I applied the model with movie recommendations as the entertainment service income is directly affected by the effectiveness of the recommendation system. Another reason is that the research team wanted to test the generalization of the model with other objects of study because we first developed the model to recommend travel destinations for seniors in Taiwan. Then, we wanted to adjust it to be used for movie recommendations by adding more users’ information. As a result, the analyzed dataset includes movie titles, locations, and user’s context like day of the week, time of the day, month, and year,” stated Dr. Thaipisutikul.

LSCAR or Long-and Short-Term Preferences for Deep Context-Aware Recommendations differs from existing models applying Recurrent Neural Networks (RNN) and Collaborative Filtering (CF) due to the attention of users’ contexts, interests, and evolving preferences.

“We began with brainstorming to screen for significant factors affecting users’ decisions on movie choices as we need factors with variations. For example, some users preferred that the day of the week and movie genre matter while some care for genre, actors, and directors at the same time. Even some of them might have dynamic long-term and short-term interests. The deep learning model, LSCAR, could explain how the decision was processed. It analyzes the comparison score in each factor such as how the user prefers genres and directors. The model would also tell which factor influences more and the most,” added Dr. Thaipisutikul.

However, conducting research on the recommendation system encountered a challenge in assuring the novelty and contribution to the Recommender Community. The LSCAR must be empirically experimented on in order to test for preciseness and embrace the context of users.

“The tough part was that we didn’t know if it was going to work. So we reviewed what had been done before as much as possible to make sure that it wouldn’t repeat other models. Once we found the gap, we made assumptions and looked for the theoretical models as well as the most recent technologies to represent each assumption. It was quite a challenge to detail everything spontaneously. Finally, we were satisfied with the accuracy of the results when experimenting. Anyway, it’s not up-to-date anymore as of now (December 2022),” Dr. Thaipisutikul added.

It is safe to say that the quality of the recommender system attributes accuracy as a principal, but to combine it with personalization would make it standout. “Don’t miss novelty by being surprised with our offerings,” Dr. Thaipisutikul noted.