Abstract: To solve the problem that collaborative filtering algorithm only relies on scoring matrix to generate prediction, which leads to its low recommendation accuracy, a hybrid recommendation algorithm based on time weighted and LDA clustering is proposed. Firstly, the time column model is constructed, time weighted similarity is generated by user rating and scoring time, and the prediction score of time weighted is generated with weighted proportional means. Then the project type is generated by LDA clustering to produce the topic cluster, the prediction score of the LDA clustering is generated by probability transition. Finally, the adjustment factor determines the weight coefficient of two grades, and the final score is generated by linear weighting. Experimental results show that the new algorithm can give reasonable recommendation based on the number of adjacent neighbors and improve the recommendation accuracy.

Key words: collaborative filtering, LDA clustering, time column, probability weighting, topic model

摘要： 针对协同过滤算法仅依赖评分矩阵产生预测,推荐准确度不高的问题，提出一种结合时间加权和LDA聚类的混合推荐算法。先构造时间柱模型，根据用户评分及时刻生成时间加权相似度，采用加权平均偏差法生成时间加权的预测评分；再对项目类型进行LDA聚类生成主题项目簇，经过概率转移生成LDA聚类的预测评分；最后通过调节因子确定两种评分的权重系数，从而线性加权生成最终评分。实验结果表明，新算法能够根据具体的近邻数目给出合理的推荐，提高推荐的准确度。

关键词: 协同过滤, LDA聚类, 时间柱, 概率加权, 主题模型