融合用户社交信息和动态兴趣的推荐算法研究

Research on Recommendation Algorithm Integrating Users’ Social Information and Dynamic Interests

摘要: 个性化推荐系统能够有效缓解信息过载问题，被广泛运用于电子商务、内容资讯推广等应用领域，但仍存在冷启动和数据稀疏等问题。本文基于传统的协同过滤算法，通过融合社交关系和兴趣偏好特征建立用户相似性矩阵，并基于形如指数函数的时间衰减函数对用户兴趣偏好进行动态加权，最后通过改进的矩阵分解算法得到预测评分。经过实验表明，本算法在推荐准确率和精度等方面优于协同过滤等传统推荐算法。

Abstract: Personalized recommendation systems can effectively alleviate the problem of information overload and are widely used in application fields such as e-commerce and content information promotion. However, there are still problems such as cold start and data sparsity. This paper is based on the traditional collaborative filtering algorithm. By integrating social relationships and interest preference features, a user similarity matrix is established. The user interest preferences are dynamically weighted based on the time attenuation function in the form of an exponential function. Finally, the predicted scores are obtained through the improved matrix factorization algorithm. Experiments show that this algorithm is superior to traditional recommendation algorithms such as collaborative filtering in terms of recommendation accuracy and precision.