Improved Collaborative Filtering Recommendation Algorithm Adapted to User Interest Change

Abstract

Collaborative filtering algorithm is the most widely used algorithm in the recommendation system. Its core is to recommend the target according to the neighbor users or similar items. When the traditional collaborative filtering algorithm is faced with sparse data, the computational similarity accuracy is low, which indirectly leads to the quality decline of the recommendation system. To alleviate the influence of data sparsity on similarity calculation and improve recommendation quality, an improved collaborative filtering recommendation algorithm adapted to the change of user interest is proposed. Firstly, considering the user preference similarity, the traditional cosine similarity is combined with the penalty factor of popular items, while the correction factor for the user rating score difference degree and the user rating tendency similarity are introduced; then, for the user interest decay problem, the Ebbinghaus forgetting function is incorporated into the Pearson correlation coefficient and modified using the common scoring weighting function. Finally, the correlations between user preferences and user interest changes are obtained by weighted fusion, to calculate more accurate user similarity. The experimental results show that the algorithm of this paper effectively improves the prediction accuracy of the recommendation system.