基于共同购买和用户行为的矩阵分解推荐算法

发布时间：2018-06-02 06:48

本文选题：推荐算法 + 矩阵分解　；参考：《浙江大学》2017年硕士论文

【摘要】：在推荐算法中,基于矩阵分解的协同过滤算法是使用最为广泛的推荐技术之一。本文将对传统的矩阵分解算法在共同购买的模式上进行扩展,基于word2Vec中点际关系的概念构建物品、用户之间共同购买的关联矩阵,并结合用户行为建模和概率矩阵分解,研究共同购买关系矩阵对推荐系统的影响。首先,本文根据物品共同购买的关系,把物品视作节点,构建点际关系矩阵,同样的得到用户点际关系矩阵。接着本文综合历史评分和共同购买的因素,把这两个矩阵和用户-物品评分矩阵进行分解,得到用户向量和物品向量。然后本文使用主题模型对用户的物品集合和评论文本进行建模得到主题向量,使用这两个主题向量来线性拟合用户向量。最后基于概率矩阵分解算法对本文模型考虑的因素进行建模,并利用最终的用户和物品向量来预测用户对物品的购买情况。本文在公开的movielens和amazon数据集上进行实验,结果表明,基于共同购买和用户行为的矩阵分解算法在推荐质量上对比PMF、CTR等优秀的推荐算法有一定的提升,并且在推荐物品共同购买集上有明显的提高。
[Abstract]:Among the recommendation algorithms, the collaborative filtering algorithm based on matrix decomposition is one of the most widely used recommendation techniques. In this paper, the traditional matrix factorization algorithm is extended to construct the items based on the concept of point relation in word2Vec, and the association matrix between users to buy together, combined with user behavior modeling and probability matrix decomposition. This paper studies the influence of the common purchase relation matrix on the recommendation system. First of all, according to the relationship of joint purchase of items, this paper regards the items as nodes, constructs the matrix of point relationships, and obtains the same matrix of user points. Then this paper synthesizes the historical score and the common purchase factor, and decomposes the two matrices and the user-item score matrix to obtain the user vector and the item vector. Then, we use the topic model to model the user's item set and comment text to get the topic vector, and use these two theme vectors to fit the user vector linearly. Finally, the factors considered in the model are modeled based on the probabilistic matrix decomposition algorithm, and the final user and item vectors are used to predict the purchase of the items. In this paper, experiments are carried out on the open data sets of movielens and amazon. The results show that the matrix decomposition algorithm based on common purchase and user behavior can improve the quality of recommendation. And in the recommended items on the common purchase set has a significant improvement.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】