A Ranking Framework of Scientific Publications Using Temporal and Lexical Relevance and Citation Behavior

The exponential growth of scientific literature has amplified the need for ranking mechanisms that prioritize recent publications while ensuring the relevance and authenticity of cited information. The current recency-based metrics, like Price’s Index and the mean/median age of references, primarily characterize citation age but ignore textual or lexical alignment and self-citation bias. Conversely, methods prioritizing relevance highlight lexical alignment but overlook citation recency and self-citation bias. The objective of this study is to propose an interpretable multi-signal ranking framework that integrates the reference recency, lexical relevance, and self-citation behavior into a unified ranking score. Two benchmark datasets, the AMiner corpus (DBLPV13) and OpenAlex, are used for empirical evaluation of the proposed method. The proposed approach demonstrates the greater score differentiation compared with conventional recency metrics like Price’s Index and citation half-life (mean/median age). Additionally, the resulting publication rankings are further compared with the well-known methods such as AttRank, PageRank, RAM, and raw citation counts. Furthermore, the top-ranked publications are qualitatively evaluated using Computer Science Ontology (CSO) to assess topical alignment and topical coherence complemented by statistical analysis. The results show that combining temporal, lexical relevance, and self-citation signals produces publication rankings that consistently differ from traditional citation- and recency-based methods while providing interpretable approach for examining multiple aspects of scientific publications.