一种改进的slope one推荐算法研究

Research on Improved Slope One Recommendation Algorithm

随着互联网信息的不断膨胀,互联网已经进入了大数据时代。为了解决人们当前面临的信息过载问题,个性化推荐系统应运而生,系统核心是其所使用的推荐算法。slope one算法是一种简单高效的典型协同过滤推荐算法,算法通过对用户——项目评分矩阵进行线性回归,预测用户对于未评分项目的可能评分。由于算法的输入只有用户评分矩阵,而实际情况中的评分矩阵通常较为稀疏,因此数据稀疏性是影响其推荐准确率的主要问题。为了克服该问题,文章基于现有研究提出了一种改进的slope one算法。该算法根据所有用户对项目的历史评分计算其项目相似度,然后将其加入评分公式予以修正,同时针对稀疏的评分矩阵使用奇异值分解技术降低矩阵维度,生成更加稠密的相似矩阵作为slope one核心计算部分的输入。项目相似度的引入增加了算法对于项目内在联系的考虑,推荐结果更加合理。而奇异值分解则可以使稀疏的评分矩阵转换为更适用于算法计算的形式。通过项目相似性和奇异值分解两种技术的融合,文中算法实现了更好的推荐准确性和适应性。

With the continuous expansion of Internet information, the Internet has entered the big data era.The recommendation systems have come into being to solve the problem of information overload. The core of recommendation system is the recommendation algorithm. The slope one algorithm is a simple, efficient and typical collaborative filtering recommendation algorithm. This algorithm uses the linear regression of user-item score matrix to predict the scores of items. The data sparsity is the major problem affecting its accuracy because the matrix is usually very sparse. In this paper, we propose an improved slope one algorithm to solve this problem. Firstly, the similarity between items is calculated and is added to the score formula. Then, we use single value decomposition to reduce dimension of the sparse user-item value matrix and generate a similar but denser matrix to be the new input of slope one algorithm. Item similarity takes the internal relation into consideration and makes the result more reasonable. Single value decomposition converts the sparse matrix to a more appropriate form for calculation. Through the mix of the two techniques, the new algorithm has better prediction accuracy and fl exibility.