Topic-Dependent Document Ranking: Citation Network Analysis by Analogy to Memory Retrieval in the Brain

详细信息

下载全文

推荐本文 |

• 作者：Hiroshi Okamoto (12) hiroshi.okamoto@fujixerox.co.jp
• 关键词：Bibliometrics &#8211 ; Scientometrics &#8211 ; Citation network &#8211 ; PageRank &#8211 ; Memory retrieval &#8211 ; Continuous attractor &#8211 ; Graded persistent activity
• 刊名：Lecture Notes in Computer Science
• 出版时间：2011
• 出版年：2011
• 期刊代码：99_0302-9743
• 类别：cp
• 卷：6791
• 期：1
• 页码：371-378
• 数据来源：sp

摘要

We propose a method of citation analysis for evaluating the topic-dependent importance of individual scientific papers. This method assumes spreading activation in citation networks with a multi-hysteretic input/output relationship for each node (paper). The multi-hysteretic property renders the steady state of spreading activation continuously dependent on the initial state. Given a topic represented by the initial state, the importance of individual papers can be defined by the activities they have in the steady state. We have devised this method inspired by memory retrieval in the brain, where the multi-hysteretic property of single cells or neuronal networks is considered to play an essential role for cue-dependent retrieval of memory. Quantitative evaluation using a restoration problem has revealed that the performance of the proposed method is considerably higher than that of the benchmark method. We demonstrate the practical usefulness of the proposed method by applying it to a citation network of neuroscience papers.