基于改进MPLSH的协同过滤推荐算法

Collaborative Filtering Recommendation Algorithm Based on Improved MPLSH

针对基于项目的协同过滤推荐算法(Item-CF)在处理高维项目评分数据时出现计算效率急剧下降的不足,提出一种将改进的多探寻局部敏感哈希算法(MPLSH)和Item-CF相结合的推荐算法。改进的MPLSH通过将待搜索哈希桶的探寻方式由原始的哈希值差异导向替换为由距离远近导向,从而减少MPLSH需要探寻哈希桶的个数,缩小了Item-CF中相似项目集合的查找范围。并利用MPLSH本身具有的高效数据降维特性,提高Item-CF在高维项目评分数据中寻找相似项目集合的速度,从而有效改善Item-CF在处理高维项目评分数据时计算效率下降的问题。通过在MovieLens电影评分数据集上进行实验和算法比较,验证了该算法的有效性。

Aimed at the insufficient of computation efficiency drops dramatically when item-based collaborative filtering recommendation algorithm（Item-CF）processing high dimension item scoring data,we proposed a new recommendation algorithm which combine an improved multi probing local sensitive hashing algorithm（MPLSH）with Item-CF.This improved MPLSH decrease the number of hash buckets that need to be explored by replace the search way of hash buckets from hash value difference orientation to distance guide,decrease the search range of the similar item set in Item-CF.And make use of the efficient data dimension reduction characteristic of MPLSH,increase the speed of finding similar item sets in high dimension item scoring data,So as to effective improve the problem of computation efficiency reduce of Item-CF when it handle high dimension item scoring data.By conduct experiment on film scoring data set of MovieLens and compare it with other algorithms,the effectiveness of this algorithm is verified.