结合奇异值分解和时间权重的协同过滤算法

COLLABORATIVE FILTERING ALGORITHM COMBINING SINGULAR VALUE DECOMPOSITION AND TIME WEIGHT

协同过滤是现阶段最成功的推荐技术之一。提出一种结合奇异值分解和时间权重的协同过滤算法。与使用奇异值分解来降维的最近邻法不同,该算法通过梯度下降法进行奇异值分解,并直接将分解的结果用于预测评分。同时,该算法根据评分时间,为每个评分赋予不同的时间权重,考虑了用户兴趣随时间的变化。实验表明,该算法相较于传统协同过滤算法,能够获得更高的推荐精度。

Collaborative filtering is one of the most successful recommendation techniques nowadays.In this paper,a collaborative filtering combining the singular value decomposition（SVD） algorithm and the time weight is proposed.Unlike the Nearest Neighbour methods which use SVD as a way to reduce dimensions,our algorithm uses the gradient descent algorithm to calculate the SVD,and predicts the ratings directly based on the results of the SVD.Meanwhile,each rating score is assigned with a time weight based on the rating time,and thus the interest drift with time is taken into consideration.Experimental results show that this algorithm is able to acquire higher recommendation accuracy in comparison with the traditional collaborative filtering algorithms.