基于事件的社会网络中群组推荐关键技术

Recent years have witnessed increased development and popularity of event-based social networks (EBSN), such as Meetup, Plancast and Douban Event. EBSN is a new kind of social network, which is characterized as event-driven, group participation, varying dynamically, rich context information, complex relation etc. In EBSN, a group is usually the basic organization unit for users to attend event, that is, people tend to participate the event with their friends or families. So the investigation on the event recommendation for groups becomes necessary. In order to improve the participation experiences of groups and strengthen the attraction of EBSN platforms, the project plan to study in depth several key technologies on the EBSN event recommendation for groups, including the general group recommendation framework, the technology of discovering users’ preference using implicit feedback, the technology of automatically group identification in heterogeneous networks, the recommendation strategy based on deeply preference merging, the social recommendation strategy, and the context-aware recommendation strategy.The research can help solve well the problems such as the implicit information extraction, the preferences affecting and merging with the consideration of the uses’ interaction, and the cold start. And the work is not only significant for enriching the theories in the fields of social networks and recommendation systems, but also will contribute to extending the application scopes of EBSN.

近年来，以Meetup、Plancast和豆瓣同城为代表的基于事件的社会网络(event-based social networks，EBSN)得到快速发展和流行。EBSN是一种新型社会网络，具有事件驱动性、群组参与性、动态变化性、上下文多样性、关系复杂性等特征。在EBSN中，用户通常是以群组(朋友或家人)为单位去参加事件或活动，因此研究面向群组的推荐策略十分必要。为改善群组的参与体验和增强平台的吸引力，本课题拟深入研究EBSN群组推荐关键技术，包括总体推荐框架、基于隐式反馈的用户偏好发现技术、基于异构网络的隐式群组自动识别技术、群组偏好深层融合推荐策略、社会化推荐策略和上下文感知推荐策略等。研究成果可较好地解决隐式信息提取、考虑用户交互行为的偏好影响与偏好融合以及冷启动问题。开展本课题研究不仅对丰富社会网络、推荐系统等领域的理论具有重要意义，而且对扩大EBSN应用范围也具有重要价值。

事件社交网络（Event-based Social Network，ESBN）每天发布的事件数量越来越多，“信息过载”现象愈发严重，用户从中发现感兴趣事件越来越困难。在EBSN 实际活动中，用户常常以“群组”为单位去参加事件，推荐系统应尽量满足群组内每个用户的偏好进行推荐。该类型推荐系统称之为群组推荐系统。如何融合群组成员偏好以缓解成员的偏好冲突，使得推荐结果尽可能满足所有成员需求是群组推荐系统需要解决的关键问题。本项目着重围绕EBSN环境中的群组推荐关键技术展开研究，研究内容包括：基于事件社会网络推荐系统框架、用户隐式偏好发现策略、隐式群组识别技术、偏好深度融合群组事件推荐策略、事件社交网络下一个事件推荐策略、上下文感知群组推荐策略、基于知识图谱和协同注意力机制的群组事件推荐策略、群组事件安排策略等，取得了较多研究成果。本项目的主要特色表现为：围绕EBSN环境下向群组推荐事件的任务，提出了一系列群组事件安排和推荐策略，并通过实验验证了所提出方法和策略的可行性、正确性和有效性。因此，开展本项目研究既大大丰富了社交网络、深度学习及推荐理论，又为群组推荐系统的应用推广提供有效技术支撑。