基于社会选择理论的群组推荐关键技术研究

With the rapid development of Internet Technology, the problem of information overload has been more and more serious than ever. How to realize effective and efficient service recommendation according to users’ requirements among massive information services has been one of the hot topics in both academy and industry of Computer Science. Traditional service recommendation mainly concerns about how to recommend a service to an individual user. However, users commonly exist in groups in real applications. Group is a basic element in social networks and therefore research on group recommendation is of great importance. This project aims at effective group recommendation for massive data oriented environments, including the research on how to extract user preferences individually and present a formal expression, how to cluster users and discover groups, how to integrate users’ preferences with each other to get a group preference, and how to avoid explosion of users’ privacy and guarantee the well execution of group recommendation. According to the characteristics of group recommendation, the project is to apply social choice theory, machine learning, collaborative filtering techniques and other related theory and technologies to make a comprehensive study on key technologies in group recommendation. Based on the study, the project is also to design a prototype as well as some application group recommendation systems to validate the practicality and effectiveness of theoretical results. It is supposed that research results from this project will be of great theoretical significance and valuable reference for further popularization of applications of group recommendation in massive data oriented environments.

随着信息技术的飞速发展，“信息过载”问题日益突显。如何在海量信息中实现高效的服务推荐，已成为计算机科学学术界与产业界的一个热点研究问题。传统的服务推荐主要研究如何向个体用户进行推荐，然而，在现实生活的许多应用场景中，用户以群组形式出现，群组是社交网络的基本要素，群组推荐研究具有重要意义。本项目致力于寻求面向海量数据的群组推荐方法，研究如何准确提取个体用户偏好并进行形式化表达、如何将个体用户进行聚类以发现群组、如何将个体用户偏好有机融合以生成群组偏好、以及如何避免用户隐私泄露，确保群组推荐服务的正常运行四个方面。项目将根据群组推荐的特征，利用社会选择理论，机器学习理论，协同过滤技术等对群组推荐的关键技术展开研究。在此基础上，实现基于社会选择理论的群组推荐应用示范，以测试和验证理论研究成果的可行性与有效性。本项目的研究成果将对面向海量数据的群组推荐应用技术的推广具有重要的理论意义和参考价值。

本项目致力于通过群组推荐来缓解海量数据环境中的信息过载问题。与传统的个性化推荐不同，群组推荐在精准提取群组中个体用户偏好并进行形式化表达的基础上，利用用户偏好信息进行群组发现，融合群组中不同用户偏好信息形成群组偏好，同时研究此过程中的用户隐私保护问题，最终实现精准化群组推荐服务，为群组推荐在更多新型应用领域的发展和普及提供理论基础和技术支撑。本项目在以下五个方面取得了具有重要意义的研究成果：（1）提出了基于协同知识图谱的用户偏好提取及表达方法，利用知识图谱挖掘用户的显式和隐式偏好。（2）提出了用户聚类与基于结构化交互信息的群组发现方法，利用图神经网络和聚类算法进行群组划分。（3）提出了基于社会选择理论的群组偏好融合方法，融合了图论、矩阵分解，社交网络以及深度学习的方法挖掘群组用户间的潜在关系，并根据潜在关系融合群组偏好。（4）提出了群组推荐中隐私保护理论与方法，采用了区块链、密码学方法以及签名策略对群组推荐中用户的隐私进行保护。（5）基于以上理论研究成果，构建了两套群组推荐应用示范系统，并取得软件著作权。