基于改进的协同过滤相似性度量算法研究

A New Similarity Measurement Method in Collaborative Filtering

随着工程测量和工业控制的发展,在多样的工程测量环境和工业控制环境中选择合适的测量和控制理论、方法和技术也将成为难题,推荐技术的引入可以提升工程测量的的自动化程度和工业控制的实时性;但是推荐系统中经典的相似性度量方法在数据稀疏的情况下处理能力较弱,影响了推荐的准确性;针对这一问题,将杰卡徳相似系数加以修正,并利用杰卡德相似系数能够衡量两个集合的相似度的特点,将修正后的杰卡德相似系数作为权重系数,对经典的相似性度量方法加以修正,得到新的相似性度量方法;选取5个测评指标,分别在基于用户和基于项目的协同过滤推荐算法中,对经典的相似性度量方法和改进的相似性度量方法进行测试;对比实验表明,改进的相似性度量方法表现优于传统的相似性度量方法,提升比例约为20%。

With the development of engineering survey and industrial control, selecting the appropriate theories, methods, and techniquese in various engineering survey and industrial control environment will become a problem, the introduction of recommendation technology can improve the degree of automation of engineering survey and real- time performance of industrial control. At the same time, the classi- cal similarity measurement method in recommendation system is weak in the case of sparse data, which affects the accuracy of recommendation. To solve this problem, Jaccard similarity coefficient is improved, because Jaccard similarity coefficient is able to measure the similarity coefficient of two sets, the improved Jaccard similarity coefficient is is used as the weight coefficient of the classical similarity measurement result, and a new similarity measurement method is obtained. MovieLens datasets and 5 evaluation indicators are selected, and the classical similarity measurement methods and the improved similarity measurement methods are tested respectively on the user-based and item- based collaborative filtering recommendation algorithm. The contrast experiment shows that the improved similarity measure is superior to the traditional similarity measure, and the promotion ratio is about 20%.