面向海量数据的基于效用的个性化学术资源推荐系统关键技术研究

项目名称： 面向海量数据的基于效用的个性化学术资源推荐系统关键技术研究

项目编号： No.61202321

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 刘莹

作者单位： 中国科学院大学

项目金额： 24万元

中文摘要： 个性化学术资源推荐系统是学生和科研人员在学习和工作中迫切需要的工具。已有的数字图书馆系统、个性化推荐系统等都没有针对学术资源和学术用户的特点展开深入的研究。为此，本项目首先要转变只从用户感兴趣的内容中提取用户兴趣的传统思路，提出将内容个性化与行为个性化相结合的用户兴趣模型。为了跟踪用户在内容兴趣方面的变化，提出动态发现用户兴趣的思路，主动地收集用户显式的、隐式的反馈，通过数据挖掘、信息检索等技术，挖掘用户最新的兴趣并适时更新用户兴趣模型。为了对行为偏好不同的用户提供个性化的推荐，提出基于用户历史行为的效用模型，并通过对模型的训练，获得效用函数，进而实现基于效用的个性化推荐。面对海量的学术资源和用户，为了解决系统的可扩展性问题和性能问题，将研究分布式的并行推荐方法。系统原型将在两个有代表性的实际应用中得到检验。本项目的研究将有利于学生和科研人员提高科研水平，丰富推荐系统的理论和技术。

中文关键词： 推荐系统；数据挖掘；并行计算；最优化；效用

英文摘要： Personalized academic resource recommendation system is in real demand by students and researchers. Existing systems, such as digital library systems, personalized recommendation systems, etc. cannot meet the requirements of academic users. Therefore, in this proposal, first of all, we propose a novel user profile model based on both content and behavior, which overcomes the weakness of content-based profile model that only takes into consideration the text content the user looked through. Next, in order to follow the user's interersts in content, we propose various methods to discover the patterns hidden in Web logs, comments provided by the user, and user click streams obtained at runtime when the user is browsing around in the recommendation system.Then, the user interest profile will be updated dynamically when necessary. In order to recommend different academic resources for users who share common interest in content but with different preference, we propose a utility-based model, which models a user's preference by a classification model. The utility function will be obtained through the training of the classification model. Since the number of academic resources is huge, so as the number of users, the performance is a big challenge. We propose to implement the recommendation in data parallelism and task p

英文关键词： recommendation system；data mining；parallel computing；optimization；utility