Introduction
In contemporary academia, bibliometric indices such as the h-index significantly influence hiring, funding, and career progression. However, these metrics are often criticized for being easily manipulated. A more recent and sophisticated metric, the Consolidation-Disruption (CD) index, has gained traction for its ability to distinguish between research that incrementally improves existing knowledge (consolidative) and research that shifts a field in a new direction (disruptive).
While the CD index is influential—notably used in high-profile studies to argue that small teams are more disruptive than large ones—its robustness against strategic manipulation has remained largely unexplored. This paper investigates the computational complexity of "gaming" the CD index through various strategies, providing a theoretical foundation for its reliability as a metric.
Understanding the CD Index
The CD index measures the nature of a paper's impact by analyzing its citation context. Specifically, it looks at how subsequent papers cite a target paper $p$ in relation to its own references.
As shown in Figure 1, the index categorizes papers into three sets:
- $N_F^D(p)$ (Forward): Papers that cite $p$ but do not cite any of $p$'s references. These represent disruptive impact.
- $N_B^D(p)$ (Both): Papers that cite both $p$ and at least one of its references. These represent consolidative impact.
- $N_R^D(p)$ (References): Papers that cite $p$'s references but do not cite $p$ itself.
The CD index is calculated based on the sizes of these sets, ranging from -1 (purely consolidative) to 1 (purely disruptive).
Methods: Modeling Manipulation as Graph Problems
The author frames the manipulation of the CD index as a series of graph-theoretic problems. A citation network is modeled as a directed acyclic graph (DAG) where nodes represent papers and edges represent citations. The study identifies three primary manipulation tactics:
- CD-MANI-MERGE: Merging multiple papers into a single entry (e.g., on Google Scholar) to consolidate citations.
- CD-MANI-ADD: Adding new "dummy" papers to the network to cite the target paper disruptive-style.
- CD-MANI-DEL: Deleting existing papers from the network to remove consolidative citations.
To determine the difficulty of these tasks, the author employs parameterized complexity. This involves mathematical reductions from known hard problems (like CLIQUE) to see if an optimal manipulation strategy can be found in polynomial time or if it belongs to harder classes like NP-hard or W[1]-hard.
Key Insights and Results
The core finding of the paper is that most strategies to artificially inflate a paper's CD index are computationally "hard." This suggests that the metric is inherently resistant to simple, automated manipulation.
The Hardness of Merging and Adding
The study proves that CD-MANI-MERGE is W[1]-hard, even in highly restricted scenarios where the citation graph is simple. This means that as the number of papers to be merged increases, the time required to find the optimal combination grows exponentially, making it impractical for a human or a standard computer to solve quickly.
Similarly, CD-MANI-ADD is shown to be W[2]-hard with respect to the number of papers added ($k$). This indicates that strategically adding dummy papers to reach a specific disruption score is a significantly difficult optimization problem.
The Complexity Dichotomy in Deletion
An interesting "complexity dichotomy" was discovered regarding the deletion of papers. If a scholar aims for a "perfectly disruptive" score of 1, the problem (CD-MANI-DEL) is Fixed-Parameter Tractable (FPT), meaning it is relatively easy if only a few deletions are needed. However, if the target is any fractional value less than 1, the problem immediately becomes W[1]-hard.
Conclusion
This theoretical analysis suggests that the CD index is more robust than simpler metrics like citation counts. Because the "optimal" way to cheat involves solving complex mathematical problems, the metric is naturally protected against strategic manipulation. While these results represent worst-case scenarios, they provide a strong argument for the continued use of the CD index in high-stakes academic evaluations. Future research may focus on whether heuristic or "greedy" algorithms—which are easier to compute—can still provide effective, if sub-optimal, ways to boost the index in real-world networks.
Summary updated