摘要
针对传统电影推荐算法中数据维度高,缺乏考虑用户和电影长短期状态的问题,提出一个基于GRU网络和矩阵分解的混合推荐算法的研究。该算法利用自编码器对时序数据进行降维处理,然后使用GRU网络处理降维后的时序数据以捕获用户和电影的短期动态状态,利用矩阵分解算法处理原始评分矩阵得到用户和电影的长期固定状态,然后利用线性回归模型将长期状态的内积和短期状态的内积的混合加权评分作为最终预测评分,以提高推荐质量。
In view of the high data dimension in traditional movie recommendation algorithms and the lack of consideration of the long-term andshort-term status of users and movies, proposes a research on hybrid recommendation algorithm based on GRU network and matrix decomposition. The algorithm uses self-encoder to reduce the dimensional data, and then uses GRU network to process the reduced dimension data to capture the short-term dynamic state of the user and the movie. The matrix decomposition algorithm is used to process the originalscore matrix to obtain the long-term fixed statate of users and movies. Uses the linear regression model to the mixed weighted score of theinner product of the long-term state and the inner product of the short-term state as the final predictive score to improve the recommendation quality.
作者
徐彬源
高茂庭
XU Bin-yuan;GAO Mao-ting(Collage of Information Engineering,Shanghai Maritime University,Shanghai 201306)
基金
国家自然科学基金(No.61202022)
关键词
推荐算法
门控循环单元
数据降维
长期状态
短期状态
Recommender Algorithm
Gated Recurrent Unit
Data Reduction
Long-Term State
Short-Term State
