[论文解读] On Good and Bad Intentions behind Anomalous Citation Patterns among Journals in Computer Sciences

Joyita Chakraborty, Dinesh K. Pradhan+3 more|arXiv (Cornell University)|Jul 27, 2018

Complex Network Analysis Techniques被引用 3

一句话总结

本文利用加权有向图模型，对1990至2012年间2,500余种计算机科学期刊中的异常引文模式（如相互引文、引文链、三角形、卡特尔等）进行了研究。该研究区分了这些模式背后的‘良性’（合法）与‘恶性’（操纵性）意图，揭示了突然的期刊影响因子膨胀和引文激增往往与引文堆叠、自引环等不道德行为密切相关，从而对当前文献计量指标的可靠性提出质疑。

ABSTRACT

Scientific journals are an important choice of publication venue for most authors. Publishing in prestigious journal plays a decisive role for authors in hiring and promotions. In last decade, citation pressure has become intact for all scientific entities more than ever before. Unethical publication practices has started to manipulate widely used performance metric such as "impact factor" for journals and citation based indices for authors. This threatens the integrity of scientific quality and takes away deserved credit of legitimate authors and their authentic publications. In this paper we extract all possible anomalous citation patterns between journals from a Computer Science bibliographic dataset which contains more than 2,500 journals. Apart from excessive self-citations, we mostly focus on finding several patterns between two or more journals such as bi-directional mutual citations, chains, triangles, mesh, cartel relationships. On a macroscopic scale, the motivation is to understand the nature of these patterns by modeling how journals mutually interact through citations. On microscopic level, we differentiate between possible intentions (good or bad) behind such patterns. We see whether such patterns prevail for long period or during any specific time duration. For abnormal citation behavior, we study the nature of sudden inflation in impact factor of journals on a time basis which may occur due to addition of irrelevant and superfluous citations in such closed pattern interaction. We also study possible influences such as abrupt increase in paper count due to the presence of self-referential papers or duplicate manuscripts, author self-citation, author co-authorship network, author-editor network, publication houses etc. The entire study is done to question the reliability of existing bibliometrics, and hence, it is an urgent need to curtail their usage or redefine them.

研究动机与目标

• 识别并分类计算机科学期刊中的异常引文模式，如相互引文、引文链、三角形、网格结构及卡特尔。
• 区分这些引文模式背后的合法（良性）与操纵性（恶性）意图。
• 研究此类模式是否具有时间特定性，或在长期内持续存在。
• 分析引文异常对期刊影响因子和论文数量的影响，特别是因冗余引文导致的突然膨胀。
• 质疑现有文献计量指标（如影响因子）的可靠性，并倡导对其重新定义或限制使用。

提出的方法

• 基于微软学术搜索数据集（包含2,500余种计算机科学期刊及490,249篇论文，时间范围为1990–2012年）构建加权有向图。
• 通过图分析识别异常模式：相互引文、引文链、三角形、网格结构、自环以及引文卡特尔。
• 追踪影响因子膨胀与引文激增的时间动态，以检测突然出现的异常。
• 分析自引、重复稿件、作者合作者网络、作者-编辑网络及出版商影响等贡献因素。
• 利用引文网络拓扑结构检测闭合回路引文行为，以识别引文卡特尔或引文堆叠的迹象。
• 结合宏观（网络层面）与微观（意图导向）分析，评估引文模式的合法性。

实验结果

研究问题

• RQ1 能否通过引文图的网络层面分析，区分合法与操纵性引文模式？
• RQ2 如引文卡特尔或相互引文环等异常引文模式是否具有时间持续性，或仅局限于特定时期？
• RQ3 影响因子的突然增长在多大程度上与冗余或自引用相关？
• RQ4 自引、合作者网络及出版商影响等因素在多大程度上促成引文异常？
• RQ5 在存在此类操纵行为的情况下，现有文献计量指标（如影响因子）在多大程度上仍具可靠性？

主要发现

• 从1990年到2012年，出版率增长60%，引文率上升74%，加剧了引文操纵的风险。
• 引文卡特尔、相互引文及引文链等异常模式普遍存在，尤其在影响因子突然膨胀的期刊中更为显著。
• 影响因子的突然飙升与少数期刊之间存在过多单向或相互引文的现象密切相关。
• 自引和自引用论文是论文数量膨胀与引文激增的重要驱动因素，尤其在外部引文多样性较低的期刊中更为明显。
• 作者合作者网络与作者-编辑网络被发现有助于协调引文行为，暗示存在机构或协作性共谋。
• 本研究结论认为，当前的影响因子等文献计量指标易受操纵，亟需对这些指标进行重新评估或重新设计。