Users' Rating Predictions Using Collaborating Filtering Based on Users and Items Similarity Measures.

Area of Science:

• Computer Science
• Information Science
• Artificial Intelligence

Background:

• The proliferation of online information necessitates effective filtering mechanisms.
• Recommender systems (RSs) are crucial for navigating vast digital content and personalizing user experiences.
• Predicting user ratings is a key challenge in collaborative filtering (CF) for RSs.

Purpose of the Study:

• To compare the effectiveness of various similarity measures and collaborative filtering approaches for user rating prediction.
• To evaluate user-user and item-item based CF algorithms against baseline and matrix factorization methods.
• To identify the optimal CF approach for enhancing recommender system performance.

Main Methods:

• Explored user-user and item-item similarity measures including cosine similarity and Pearson correlation.
• Applied K-Nearest Neighbour (K-NN) algorithms for both user-based and item-based CF.
• Compared K-NN approaches with baseline methods (slope one, random, global average) and matrix factorization techniques (MF, biased MF, factor wise MF).
• Utilized three real-world datasets: MovieLens 1M, CiaoDVD, and MovieLens 100k.

Main Results:

• Item-item K-NN utilizing Pearson correlation demonstrated superior performance compared to all other evaluated approaches.
• The study provides an empirical comparison of diverse CF algorithms on standard datasets.
• Performance was evaluated using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE).

Conclusions:

• Item-item K-NN with Pearson correlation is a highly effective method for user rating prediction in recommender systems.
• The findings offer valuable insights for developing more accurate and efficient recommender systems.
• This research contributes to improving online business and e-commerce through enhanced information filtering.