Author self-citation is double-edged, fostering knowledge continuity yet potentially skewing evaluations. Current models, fixated on frequency or first-author status, ignore qualitative nuances like content and contribution. We propose a multidimensional model (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">${\text{V}}_{\text{s}\text{c}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>V</mtext> <mtext>sc</mtext> </msub> </math></EquationSource> </InlineEquation>) integrating author contribution, citation depth (semantic similarity), intensity (frequency), and location (section weight). Citation depth is assessed via semantic similarity between citing sentences and cited documents’ titles, keywords, and abstracts using the bert-base-chinese model. Validated on 3,046 self-citations from 30 highly-cited authors across three disciplines, it offers a nuanced assessment beyond conventional metrics. The <InlineEquation ID="IEq2"> <EquationSource Format="TEX">${V}_{\text{s}\text{c}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mtext>sc</mtext> </msub> </math></EquationSource> </InlineEquation> index demonstrated superior performance compared to two traditional models (full-author and first-author-only). It effectively balanced the overcounting bias of the full-author model and the under-recognition of collaborative contributions in the first-author-only model. For example, in Agricultural Sciences, the full-author model yielded a value of 2874 for one author, the first-author model 66, and the <InlineEquation ID="IEq3"> <EquationSource Format="TEX">${V}_{\text{s}\text{c}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mtext>sc</mtext> </msub> </math></EquationSource> </InlineEquation> index 1968.71. The model also showed strong disciplinary adaptability: it emphasized technological innovation in Engineering (e.g., citation depth), methodological continuity in Agricultural Sciences (e.g., citation location), and theoretical contribution in Education (e.g., author role). Spearman correlation analysis confirmed that all four dimensions positively correlated with <InlineEquation ID="IEq4"> <EquationSource Format="TEX">${\text{V}}_{\text{s}\text{c}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mtext>V</mtext> <mtext>sc</mtext> </msub> </math></EquationSource> </InlineEquation> values (p < 0.05), with author contribution being the most influential dimension across all disciplines (r = 0.65–0.79). The <InlineEquation ID="IEq5"> <EquationSource Format="TEX">${V}_{\text{s}\text{c}}$</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>V</mi> <mtext>sc</mtext> </msub> </math></EquationSource> </InlineEquation> index provides a more nuanced, equitable, and discipline-sensitive tool for evaluating author self-citation. It supports ethical citation practices, enhances the fairness of academic evaluation, and offers practical utility for journals, institutions, and policy-makers. Future work should extend validation to more disciplines and incorporate finer-grained contribution metrics.

A multidimensional evaluation model for author self-citation: integrating contribution, depth, intensity, and location

期刊论文
发表日期：

2026年6月2日

查看全文

Yunmei Liu,
Jiaqian Yang,
Guoqi Liu,
Lei Huang

摘要

Author self-citation is double-edged, fostering knowledge continuity yet potentially skewing evaluations. Current models, fixated on frequency or first-author status, ignore qualitative nuances like content and contribution. We propose a multidimensional model ( ${\text{V}}_{\text{s}\text{c}}$ $V_{sc}$ ) integrating author contribution, citation depth (semantic similarity), intensity (frequency), and location (section weight). Citation depth is assessed via semantic similarity between citing sentences and cited documents’ titles, keywords, and abstracts using the bert-base-chinese model. Validated on 3,046 self-citations from 30 highly-cited authors across three disciplines, it offers a nuanced assessment beyond conventional metrics. The ${V}_{\text{s}\text{c}}$ $V_{sc}$ index demonstrated superior performance compared to two traditional models (full-author and first-author-only). It effectively balanced the overcounting bias of the full-author model and the under-recognition of collaborative contributions in the first-author-only model. For example, in Agricultural Sciences, the full-author model yielded a value of 2874 for one author, the first-author model 66, and the ${V}_{\text{s}\text{c}}$ $V_{sc}$ index 1968.71. The model also showed strong disciplinary adaptability: it emphasized technological innovation in Engineering (e.g., citation depth), methodological continuity in Agricultural Sciences (e.g., citation location), and theoretical contribution in Education (e.g., author role). Spearman correlation analysis confirmed that all four dimensions positively correlated with ${\text{V}}_{\text{s}\text{c}}$ $V_{sc}$ values (p < 0.05), with author contribution being the most influential dimension across all disciplines (r = 0.65–0.79). The ${V}_{\text{s}\text{c}}$ $V_{sc}$ index provides a more nuanced, equitable, and discipline-sensitive tool for evaluating author self-citation. It supports ethical citation practices, enhances the fairness of academic evaluation, and offers practical utility for journals, institutions, and policy-makers. Future work should extend validation to more disciplines and incorporate finer-grained contribution metrics.