基于Timed-HITS与协同过滤的混合推荐算法

用户间的信任关系、用户对商品的偏好兴趣及商品的时效性都会影响对商品的推荐效果.将这些因素引入到基本的HITS算法中，对HITS算法进行了改进.将用户对商品的偏好兴趣矩阵进行了改进，利用隐馈数据通过逻辑回归算法估计用户对商品的偏好兴趣，对评分为零的情况赋予了不同的偏好兴趣度，这样更符合实际.将改进的HITS算法和协同过滤算法相结合得到一个混合推荐算法，同时将用户分为活跃用户和非活跃用户分别进行推荐.将提出的算法在Movielens数据集上进行了试验，结果表明该算法在一定程度上缓解了数据稀疏和冷启动的问题，推荐效果优于基于用户的协同过滤算法.

Hybrid Recommendation Algorithm Based on Timed-HITS and Collaborative Filtering

The product recommendation effect was affected by the trust relationship among users， the preference interest for goods and the time factor. These factors were introduced to the basic HITS algorithm， and the HITS algorithm was improved. The user preference interest matrix was also improved， which uses implicit data to estimate users′ preference for goods using logistic regression algorithm. The situation with zero score gives different preference values， which is more consistent with reality. A hybrid recommendation model was proposed by combining the improved HITS algorithm with the collaborative filtering algorithm， and users were divided into active users and inactive users for recommendation. The proposed algorithm was tested using the Movielens data set， the results showed that the algorithm could generate better recommendation result in sparse data sets and cold-start situation， and it outperforms user-based collaborative filtering algorithm.