A Novel Context-Aware Deep Learning Algorithm for Enhanced Movie Recommendation Systems

Abstract

Recommendation systems serve as a pivotal solution to address the increasing issue of information overload. While traditional recommendation algorithms have been grounded primarily on user-item interactions, the significance of a user's contextual information influencing decision-making has often been overlooked. Such neglect becomes more evident in the realm of systems integrating contextual mechanisms, which encounter pronounced data sparsity challenges. Existing studies in contextual recommendation systems tend to treat all contextual features uniformly as influencers of user decisions. Yet, a prevalent dilemma is the frequent absence of contextual data, leading to potential misallocations of contextual features. To mitigate these challenges, a novel deep learning-based recommendation system, termed the CAW-NeuMF Model, has been designed. Accompanying this model, a Context-aware Weighted high-order Tensor Factorization algorithm (CAWTF) has been introduced. This algorithm facilitates the calculation of correlations between user ratings in varied contexts, relying on the said context. Additionally, it ascertains the weight of context features grounded on the user ratings correlation. Such a process aids in isolating the most influential contextual features, thereby amplifying the efficiency of personalized recommendations. Empirical evaluations using the LDOS CoMoDa dataset revealed that the proposed model substantially enhances prediction score accuracy. Comparative analyses against alternative recommendation models further affirmed the superior efficacy of the introduced approach.