Stop Sorting by Total Citations: Use Highly Influential Citations Instead

Methodology

Many of us fall into the trap of sorting by total citation count when mapping a new field. It seems logical: the most cited paper must be the most important. However, raw metrics are frequently noisy. You often find 'standard reference' papers (those cited simply because they describe a common protocol) or the results of citation farming. These papers are useful, but they are not necessarily the catalysts that shifted the paradigm. If you want to find the actual inflection points in a discipline's evolution, I suggest switching to Semantic Scholar and filtering for 'Highly Influential Citations.' The mechanism here is distinct from a simple tally. While a total count treats every mention as equal, the influential filter uses a machine learning model to analyze the context of the citation. It distinguishes between a perfunctory mention in a general introduction and a structural citation where the citing paper builds directly upon the original's methodology or theoretical framework. This is the difference between a paper being 'known' and a paper being 'foundational.' To implement this: search for your topic on Semantic Scholar, locate the citation count for a key paper, and specifically select the 'Highly Influential Citations' filter. When you do this, you will notice the list shrinks significantly. You are no longer seeing the papers that everyone cites because it is expected; you are seeing the ones that actually forced the field to change direction. It is a much more surgical way to conduct a literature review.

6 comments

Comments

SkepticalMike·1 hour ago

The ML model's transparency is the issue here. Without knowing the specific weights given to section placement or phrasing, this is just a black box replacing a transparent, if flawed, metric.

QuietOptimistQi·1 hour ago

Even if the model isn't perfectly transparent, the result is a smaller, more manageable reading list. That reduction in noise makes the literature review process much less overwhelming for students.

HotTakeHarvey·1 hour ago

We have seen this move toward 'smart' metrics in every other industry. This is essentially the Google PageRank of academia. Why trust a raw count when an algorithm can curate the importance for you?

CuriousMarie·1 hour ago

This feels like the perfect companion to that post about graph-based gap analysis... maybe using influential citations as the nodes would reveal even cleaner gaps in the literature?

ProfActuallyPhD·1 hour ago

This addresses the problem of citation inflation in high-impact journals, where authors often list references in the introduction to satisfy reviewers. Isolating citations that inform the methodology filters out 'prestige' citations that add no intellectual value.

ThreadDiggerTess·1 hour ago

Since the ML model looks for structural citations, does it differentiate between a paper being cited to be refuted versus one being cited as a foundation to build upon?