A Deep Ranking Weighted Multihashing Recommender System for Item Recommendation.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Collaborative filtering (CF) is widely used in recommender systems but faces challenges with data sparsity and the cold start problem (CSP).
• Existing CF methods often lack interpretability, failing to explain recommendation reasoning.
• These limitations hinder the effectiveness of personalized recommendations in e-commerce and social platforms.

Purpose of the Study:

• To propose a novel Deep Ranking Weighted Multihash Recommender (DRWMR) system.
• To address and mitigate the sparsity and cold start problems inherent in traditional CF techniques.
• To improve the accuracy and interpretability of recommendations generated by recommender systems.

Main Methods:

• Utilized a deep convolutional neural network (CNN) for feature extraction from input data.
• Incorporated an additional CNN layer to generate hash codes by minimizing pairwise ranking and classification loss.
• Developed a weighted multihash approach, assigning weights to hash tables and bits for enhanced recommendations.
• Calculated user similarity using weighted hammering distance to form user neighborhoods.
• Generated item ratings via a weighted average of neighborhood ratings.

Main Results:

• The DRWMR system demonstrated improved performance on the MovieLens 100K dataset.
• Achieved a precision of 0.16, recall of 0.08, and F1-score of 0.101.
• Reported a Root Mean Squared Error (RMSE) of 0.73 and Mean Absolute Error (MAE) of 0.57.
• Outperformed existing methods in terms of recommendation accuracy and effectiveness.

Conclusions:

• The proposed DRWMR system effectively suppresses sparsity and cold start problems in recommender systems.
• The deep ranking weighted multihash approach enhances recommendation quality and provides better interpretability.
• The system's performance validates its potential for practical applications in e-commerce and social media.