Bootstrapping estimates of regression analysis for a non-random sample and its application in the research on anti-proliferation effects of triptolide

Objective: To solve the problem of parameter estimate in the regression analysis of non-random sample. Methods: Calculating residuals according to the regression function based on original data. Modifying residuals and correcting them with mean. Adding mean-corrected residuals on original response and bootstrapping them to get 1000 samples. Fitting regression functions of 1000 resampling samples and calculating the 2.5th percentile and 97.5th percentile of corresponding coefficient. Results: The interval estimates deriving from bootstrap method had more statistical significance than that from usual method. Conclusion: Bootstrapping a regression with residuals is a valid method for estimating parameter in regression analysis.

协同过滤系统中基于种子集评分的新用户冷启动推荐研究

认为建立种子集引导用户评分是解决协同过滤推荐系统新用户冷启动问题的方法之一。尝试将关联度引入种子集的构建策略,提出基于多属性综合评价的种子集策略,并利用公开数据集MovieLens设计实验,模拟推荐系统的新用户环境,对比不同种子集策略的预测准确度和成功率。实验结果表明,在更符合实际推荐系统需求的少量种子集情况下,考虑种子之间的关联性可以改善推荐效果。