Abstract: In view of the problem of low accuracy in the traditional recommendation algorithm. The neighborhood factor decomposition recommendation algorithm based on time deviation information（NFDRA） is presented. The main part of it is factorization algorithm, stochastic gradient descent optimization is complementary, and the integration of user ratings neighborhood information and time deviation. The experiments show, the neighborhood factor decomposition recommendation algorithm based on time deviation information, can get more accurate results and have significant difference.

Key words: time deviation, factorization, accuracy of recommendation, gradient descent

摘要： 针对传统推荐算法在进行评分预测时推荐精度低这一问题，提出了融合时间偏差信息的邻域型因子分解推荐算法（简称NFDRA）。它以因子分解算法为主，随机梯度下降寻优为辅，并融合了用户评分的邻域信息以及三种时间偏差信息。实验表明，融合时间偏差的邻域型因子分解推荐算法，相比传统的因子分解推荐可以产生更高精度的推荐结果并具有显著性差异。

关键词: 时间偏差, 因子分解, 推荐精度, 梯度下降