基于加权融合的E2LSH用户相似度计算

摘要

针对传统协同过滤算法遇到现实场景中海量、高维时难以直接进行计算用户相似度和实时性较差的问题,使用基于p-stable分布的局部敏感哈希E2LSH(Exact Euclidean Locality Sensitive Hashing)算法对海量高维数据降维后计算用户之间的相似度,进一步针对该相似度计算的精度问题使用模型融合技术对多个E2LSH算法计算的相似用户进行加权融合得到最相似用户.同时对于具有较高相似度的用户使用加权平均方法对目标用户进行未交互商品的评分预测并对商品进行排序推荐,从而提高推荐实时性和准确率.实验结果表明,算法在用户相似度计算和推荐准确率方面都有较大提高.%When the traditional collaborative filtering algorithm encounters the massive, high-dimensionality of the real scene,it is difficult to directly calculate the user similarity and the real-time performance is poor.This paper used the local sensitive hash(E2LSH)algorithm based on p-stable distribution to calculate the similarity between users after dimension reduction of massive high-dimensional data.Based on the accuracy of the similarity calculation,we used the model fusion technology to weight the similar users calculated by multiple E 2LSH algorithms to get the most similar users.At the same time, for users with higher similarity, we used a weighted average method to target users to score predictions for uninteractive products and to sort products for recommendation, thereby improving the real-time and accuracy of recommendations.The experimental results show that the proposed algorithm had greatly improved the similarity calculation and recommendation accuracy of users.