Pattern Matching Based Metric for Recommending Ordered Items

Abstract

At present, we all know that the Recommendation System is essential. Our lives are continuously impacted by the Recommendation System. Its main objective is to suggest a relevant item or item list as per the user’s requirement. In many cases, it recommends the user’s desired item or item list based on rating prediction, and this prediction accuracy is considered to be the system’s actual accuracy. But can the rating prediction accuracy be considered the system’s true accuracy in ordered items prediction? Rating prediction system even after predicting a near-exact rating, there could be a difference between the actual item list and the predicted item list. We attempted to find answers to these issues by working with the College Recommendation System. We have used different machine learning-based models in our work for rating prediction. And we have measured the correlation between the actual item list and the predicted item list using the Longest Common Subsequence algorithm. Our analysis showed that the rating prediction accuracy does not always reflect the system’s actual accuracy in the scenario of ordered items prediction. The accuracy of the system should be verified by how closely the predicted item list matches the actual item list when recommending ordered items. A pattern-matching algorithm like - Longest Common Subsequence can be considered as an accuracy metric in this context.