【摘要】 随着科研人员之间的合作关系逐渐固化，合作群体中的信息茧房问题将抑制学科交叉融合及科研合作创新水平提升。本文提出一套面向科研合作群体的信息茧房识别及破茧主题推荐方法。一方面，构建时间序列的作者合著-关键词语义相似度双层网络，利用增量社区发现算法获取各年累积合著网络中的社区结构，基于双层网络的作者和关键词的对应关系计算每个作者对应的研究主题向量，综合考虑研究主题同质性和新颖性指标，对科研合作群体中的信息茧房进行识别；另一方面，构建信息传播影响力度量模型测度作者节点的破茧潜力，生成考虑破茧潜力排序的作者合著-关键词语义相似度双层网络，并提出基于重启随机游走的作者研究主题推荐算法帮助科研人员破茧。最后，选取计算机科学领域开展实证研究，对本文方法进行有效性验证。更多还原

【Abstract】 As collaborative relationships among researchers become increasingly entrenched, the emergence of information cocoons within scientific collaboration groups may hinder interdisciplinary integration and limit advancements in collaborative scientific innovation. This paper proposes a method for identifying information cocoons and recommending breakthrough research topics in scientific collaboration groups. First, a time-series, two-layer network comprising co-authorship and keyword semantic similarity was constructed. An incremental community detection algorithm is applied to extract the evolving community structure in the cumulative co-authorship network over time. Each author’s research topic vector was calculated based on the correspondence between the authors and keywords across a two-layer network. Information cocoons are identified by jointly considering topic homogeneity and novelty metrics. Second, an information dissemination influence measurement model is constructed to measure the potential of author nodes to break out of a cocoon. Then, a co-authorship-keyword semantic two-layer network considering the ranking of potential to break out of the cocoon is generated, and an author topic recommendation algorithm based on restarted random walk(ATR＿RWR) is proposed to help researchers break out of the cocoon. An empirical analysis was conducted in the field of computer science to validate the effectiveness of the proposed method.更多还原