协同过滤与基于内容的混合推荐算法研究

发布时间：2018-10-24 14:22

【摘要】：网络技术的飞速发展影响并改变着人类的生活。网络技术的先进性已经不再是网络能给让人们干什么,而是体现在人们如何在网络上更好的体验生活。这是技术由简单到繁琐,由粗放到精细的转变。信息过载问题俨然已经影响了人们网络生活的舒适度,而推荐系统的诞生为满足人们的个性化需求带来福音。协同过滤和基于内容是两种主要的推荐算法,虽然在不同领域都已经被应用,但是仍存在自适应能力不强,个性化推荐能力不足等问题,另外,两种算法虽然优缺点比较互补,但是由于受到非文本项目的特征提取困难的束缚,基于内容推荐算法一般只用于文本类项目的推荐系统之中,因此,当前融合两种推荐算法的混合推荐的普适度不高。为了提高推荐算法的推荐质量,发挥协同过滤与基于内容两种推荐算法各自的优势,提出了一种混合推荐算法。该算法的主算法是协同过滤算法,在主算法寻找可信邻居时融入了基于内容推荐的算法思想,最后利用可信邻居协同推荐。该策略在构建中的创新点主要包括:第一,引入项目热度优化皮尔逊相关系数。第二,以项目标签作为非文本项目的属性特征,并给出了为用户建立二维兴趣模型,衡量兴趣模型相似度的方法。第三,结合兴趣模型相似度公式的结构特征,提出了利用方差来求解相似度权重系数的方法。最终的实验证明,混合推荐方法提高了推荐质量,为有效方法,与两个现存的混合策略相比具有推荐优势。而且在混合算法模型中的每一步优化以及计算方法的有效性都通过实验得以验证。由于算法以项目标签来提取项目特征,所以对于非文本类项目的混合推荐具有一定的普适性。
[Abstract]:The rapid development of network technology influences and changes the human life. The advancement of network technology is no longer what the network can give people, but how people experience life better on the network. This is the technology from simple to cumbersome, from extensive to fine transformation. The problem of information overload has affected the comfort of people's online life, and the birth of recommendation system has brought good news to meet the personal needs of people. Collaborative filtering and content-based recommendation algorithms are two major recommendation algorithms. Although they have been applied in different fields, there are still some problems such as weak adaptive ability and insufficient personalized recommendation ability. In addition, the two algorithms complement each other in their advantages and disadvantages. However, due to the difficulty of feature extraction of non-text items, the content-based recommendation algorithm is generally only used in the recommendation system of text items. In order to improve the recommendation quality of the recommendation algorithm, a hybrid recommendation algorithm is proposed to give full play to the advantages of both collaborative filtering and content-based recommendation algorithms. The main algorithm of this algorithm is the collaborative filtering algorithm. When the main algorithm finds trusted neighbors, it integrates the idea of content-based recommendation. Finally, the trusted neighbor collaborative recommendation is used. The innovation of the strategy includes: first, introducing project heat to optimize Pearson correlation coefficient. Secondly, the item label is taken as the attribute feature of non-text items, and the method of building two-dimensional interest model for users to measure the similarity of interest model is given. Thirdly, based on the structural features of the similarity formula of interest model, a method is proposed to solve the similarity weight coefficient by using variance. The final experiments show that the hybrid recommendation method improves the quality of the recommendation. It is an effective method and has the advantages of recommendation compared with the two existing hybrid strategies. Moreover, the effectiveness of each step optimization and calculation method in the hybrid algorithm model is verified by experiments. Because the project feature is extracted by the item label, the mixed recommendation of non-text item has a certain universality.
【学位授予单位】：天津财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】