GroupLens: Scalable Collaborative Filtering for the Internet

Grouplens研究项目探索协同过滤方法，以解决互联网信息系统用户的信息过载问题。协同过滤是一种利用文档某些读者的意见来预测其他潜在读者对该文档的兴趣的技术。本项目建立在成功的Grouplens Usenet新闻协同过滤系统的基础上，更详细地探索协同过滤的人为因素和算法问题。该项目解决了协同过滤中的三个基本问题：如何以最小的用户努力有效地测量用户兴趣，如何使用这些兴趣度量来产生其他用户兴趣的准确预测，以及如何集成非协作和协作过滤方法来创建健壮的过滤环境。研究方法依赖于三种实验方法。通过一组对照实验，收集完整的评分数据集，直接衡量协同过滤算法的有效性。一个小规模的公开试验被用来收集关于GroupLens系统实际使用情况的数据。调查研究是用来获取用户的意见和反馈，从两个试验。

The Grouplens research project explores collaborative filtering approaches to address information overload for users of Internet information systems. Collaborative filtering is a technology that uses the opinions of some readers of a document to predict the interest of other potential readers in that document. This project builds upon the successful Grouplens collaborative filtering system for Usenet news to explore in more detail the human factors and algorithmic issues of collaborative filtering. This project addresses three fundamental problems in collaborative filtering: how to effectively measure user interest with minimal effort from users, how to use these measures of interest to produce accurate predictions of other user's interest, and how to integrate non-collaborative and collaborative filtering approaches to create robust filtering environments. The research approach relies upon three experimental methods. A set of controlled experimental trials is used to gather complete ratings datasets to directly measure the effectiveness of collaborative filtering algorithms. A small- scale public trial is used to collect data on the actual usage of the GroupLens system. Survey research is used to obtain user opinions and feedback from both trials.