A transparent way to find the papers at the center of a field, and the connections waiting at its edges.
Rankings are good at showing what closely matches a query. Exploratory research also needs the useful surprise: a relevant paper that uses different language, sits at a disciplinary edge, or reveals an indirect connection.
Pennant retrieval produced the highest relevance and DCG values across the evaluated cut-offs.
LDA placed more topically diverse papers near the top, contributing breadth across subject categories.
The combined ranking delivered the strongest NDCG pattern by preserving relevance while broadening conceptual coverage.
Performance fell by 10.9% for the integrated method, compared with 14.0% for pennant retrieval and 26.0% for LDA.
The method is grounded in the idea that relevance depends on the cognitive effect a paper creates and the effort required to recognize the connection.
Co-citation with the seed papers signals conceptual proximity and potential intellectual impact.
Total citation visibility helps estimate the effort needed to recognize and interpret the relationship.
The strongest discovery signal comes from the balance, not from word overlap alone.
Latent Dirichlet Allocation finds thematic structure in titles and abstracts. It contributes breadth and surfaces papers across related topics.
Pennant retrieval uses co-citation and citation visibility signals. It reveals conceptually related work, including papers that may span disciplinary boundaries.
LDA ranks papers from the words in their titles and abstracts.
The top five LDA results become transparent starting points for citation exploration.
Pennant retrieval finds papers repeatedly cited alongside each seed.
Topic and citation scores are placed on a common scale and combined.
A user-controlled weight adjusts the balance between depth and scope.
Core papers explain the established center of a topic. Some boundary-spanning papers do something different: they connect that center to another method, vocabulary, or discipline. They may have fewer obvious links, yet they can be especially valuable when the goal is to discover a relationship no familiar keyword would reveal.
The best next paper may be the one you did not know how to search for.
Incremental refinement makes room for that paper.