HeroCon: Citation Network Diagnostic
- HeroCon is a bibliometric tool that decomposes citations based on social network proximity within scholarly collaborations.
- It assigns graduated weights from self-citations to external citations to reveal in-group citation dynamics and diagnostic gaps.
- The methodology integrates co-author graphs, temporal affiliation matching, and venue governance for an auditable citation audit.
Searching arXiv for the referenced papers and term variants to ground the article. In the provided arXiv literature, HeroCon most directly corresponds to HEROCON, the Holistic Equilibrated Research Outreach CONstellation score introduced with Citation-Constellation, a free, open-source, no-code, and auditable tool for citation network decomposition. HEROCON is an author-level bibliometric score that assigns graduated weights to citations according to the network proximity between citing and cited authors, thereby decomposing a citation profile by social and institutional structure rather than treating all citations as equivalent. It is presented together with BARON as a structural diagnostic, not a quality indicator, and its default weights are explicitly described as experimental and in need of empirical calibration (Alam, 25 Mar 2026).
1. Conceptual definition and motivation
HEROCON is defined against a specific critique of standard bibliometrics: standard citation metrics treat all citations as equal, obscuring the social and structural pathways through which scholarly influence propagates. In response, Citation-Constellation asks where citations come from within the social graph surrounding a focal researcher, rather than only how many citations exist in aggregate (Alam, 25 Mar 2026).
The paired metric BARON is a strict binary metric that counts only citations from outside the detected collaborative and institutional network. HEROCON, by contrast, is a graduated, weighted metric that gives partial credit to in-group citations according to relationship proximity. In the terminology of the paper, HEROCON is “holistic” because it represents the full pattern of citation origins across self, co-author, institutional, venue-governance, and external layers, and “equilibrated” because it down-weights structurally likely in-group citations rather than discarding them entirely (Alam, 25 Mar 2026).
This design places HEROCON between raw citation totals and strict external-only counting. A self-citation receives no credit; a citation from a direct co-author receives limited credit; a citation from a more distant relation receives more; and a citation with no detected relationship receives full credit. The resulting score is intended to describe where in the social graph citations originate, not whether the underlying research is of high or low quality (Alam, 25 Mar 2026).
2. Network layers, weighting scheme, and score construction
HEROCON operationalizes citation origin through concentric layers of proximity between the citing author(s) and the focal researcher. The paper distinguishes the following classes: Layer 0 – Self, Layer 1 – Direct co-author, Layer 2 – Transitive co-author, Layer 3 – Institutional colleagues (temporal), Layer 4 – Venue governance relationships, and Layer 5 – External (Alam, 25 Mar 2026).
BARON is binary: all in-group citations receive weight 0, and only EXTERNAL citations receive weight 1. HEROCON instead assigns the following default weights.
| Layer / class | Relationship | HEROCON weight |
|---|---|---|
| 0 | Self-citation | 0.0 |
| 1 | Direct co-author | 0.2 |
| 2 | Transitive co-author | 0.5 |
| 3a | Same department | 0.1 |
| 3b | Same institution, different department | 0.4 |
| 3c | Same parent organization | 0.7 |
| 4a | Venue self-governance | 0.05 |
| 4b | Venue editor is co-author | 0.15 |
| 4c | Venue editor at same institution | 0.3 |
| 4d | Committee member in network | 0.4 |
| 5 | External | 1.0 |
The score is computed over classifiable citations, excluding UNKNOWN cases. In prose, BARON is the percentage of classifiable citations that are external, whereas HEROCON is the percentage obtained by summing citation weights and dividing by classifiable citations. The paper also defines a diagnostic gap as:
This gap is interpreted as the contribution of in-group citations under the current weighting scheme. The weights are configuration-dependent, can be overridden through a JSON configuration, and are explicitly described as a testable hypothesis rather than a finalized empirical standard (Alam, 25 Mar 2026).
3. Graph-theoretic and temporal classification procedure
The classification pipeline begins from a co-authorship graph. For every pair of authors who co-authored a paper, the tool assigns an edge strength
with the exponential factor yielding a half-life of about 7 years. The paper states, however, that this recency weighting currently modulates metadata in the audit trail rather than altering BARON/HEROCON decisions, which are based on graph distance (Alam, 25 Mar 2026).
A Breadth-First Search (BFS) from the focal researcher to configurable depth determines co-author classes. With default depth 2, distance 0 maps to SELF, distance 1 to DIRECT_COAUTHOR, and distance 2 to TRANSITIVE_COAUTHOR. Authors beyond depth are not counted as part of the co-author network and can subsequently be classified through institutional or venue-governance relations instead (Alam, 25 Mar 2026).
Institutional proximity is handled temporally. The tool constructs a time-stamped affiliation timeline for the focal researcher and compares citing authors’ affiliations at the year of the citing work using ROR IDs and parent hierarchies. This yields SAME_DEPT, SAME_INSTITUTION, or SAME_PARENT_ORG where appropriate. If neither co-authorship nor institutional overlap is detected, the system proceeds to venue-governance checks; if no relation is found there either, the citation is classified as EXTERNAL (Alam, 25 Mar 2026).
The decision order is explicit: self-citation, then co-author radius, then temporal institutional overlap, then venue governance, then external. This procedural ordering matters because HEROCON is a per-citation decomposition: each citation is assigned a specific structural rationale rather than only contributing to an aggregate score (Alam, 25 Mar 2026).
4. Data sources, identity resolution, and UNKNOWN handling
Citation-Constellation relies on OpenAlex as the main source of works, authors, citation links, affiliations, and venues; ORCID as a trust anchor for focal-author identity and work validation; and ROR for institution matching and parent-organization hierarchy (Alam, 25 Mar 2026).
Identity resolution is treated as a central methodological constraint. When the focal author has high ORCID coverage—stated as ≥ ~70% of their works—the tool trusts only works appearing in both ORCID and OpenAlex for network construction and scoring. When ORCID coverage is low, OpenAlex works are provisionally retained, but the system performs affiliation anomaly detection against the known institutional timeline and flags potential mismatches for manual review. Manual confirmation decisions are logged in the audit trail (Alam, 25 Mar 2026).
The category UNKNOWN addresses incomplete metadata. Typical cases include missing focal-author affiliations at the relevant time or missing affiliations for all citing authors on a citation. The paper emphasizes that UNKNOWN citations are excluded from the denominator of both BARON and HEROCON. This avoids silently treating unknowns as external and artificially inflating external-reach scores, but the paper also notes that this introduces a known selection bias and identifies sensitivity analyses on UNKNOWN handling as a priority for future work (Alam, 25 Mar 2026).
A plausible implication is that HEROCON is not merely a weighting scheme but a metadata-quality-sensitive audit instrument. The score depends on author disambiguation, temporal affiliation coverage, and venue metadata; accordingly, the tool reports data quality statistics, including the percentage of classifiable citations and reliability labels such as HIGH/MODERATE/LOW (Alam, 25 Mar 2026).
5. Phased architecture, software implementation, and auditability
The software implements a four-phase pipeline. Phase 1 performs self-citation analysis and yields a rudimentary based on NON_SELF citations. Phase 2 constructs the co-author graph, introduces HEROCON, and distinguishes SELF, DIRECT_COAUTHOR, TRANSITIVE_COAUTHOR, and EXTERNAL. Phase 3 performs temporal institutional affiliation matching via ROR and introduces SAME_DEPT, SAME_INSTITUTION, SAME_PARENT_ORG, and UNKNOWN. Phase 4, still under active development, adds AI-agent-driven venue governance extraction (Alam, 25 Mar 2026).
The abstract states that Phases 1–3 are fully operational; Phase 4 is under development. For Phase 4, the paper describes a governance-database construction step that uses a local LLM—specified as Qwen 3.5 8B, quantized, served via llama.cpp—to parse venue HTML and extract editorial or committee membership, followed by citation reclassification into governance-linked categories such as VENUE_SELF_GOVERNANCE, VENUE_EDITOR_COAUTHOR, VENUE_EDITOR_AFFIL, and VENUE_COMMITTEE (Alam, 25 Mar 2026).
The tool is available both as a no-code web interface and as a Python CLI. The web workflow supports ORCID or OpenAlex author IDs, configurable co-author graph depth, optional manual validation of flagged works, and custom HEROCON weight files in JSON. Typical execution time is stated as 1–4 minutes for 50–100 papers. Each run produces a detailed audit JSON containing included and excluded works, per-citation classifications, the assigning phase, confidence levels, human-readable rationales, network and affiliation snapshots, summary counts, and BARON/HEROCON outputs (Alam, 25 Mar 2026).
This audit orientation is central to the concept. The paper characterizes Citation-Constellation as auditable, and every classification decision is intended to be inspectable evidence rather than opaque score production. In methodological terms, HEROCON is therefore coupled to a provenance system rather than presented as a standalone scalar (Alam, 25 Mar 2026).
6. Interpretation, intended use, and relation to other “HERO” usages
HEROCON is explicitly framed as a structural diagnostic. The paper proposes qualitative reading of the HEROCON − BARON gap: small gap (e.g., < 3%), moderate gap (3–10%), and large gap (> 10%). These ranges are presented as interpretive guidance about inner-circle dependence, while the text simultaneously emphasizes that such patterns are not inherently good or bad and may reflect field size, collaboration structure, or institutional context (Alam, 25 Mar 2026).
The intended uses are self-auditing / self-reflection, science-of-science and policy research, and contextualization in evaluation only with strong caveats. The paper is unambiguous that BARON and HEROCON are not quality metrics and must not be used as stand-alone criteria in hiring, promotion, tenure, or funding decisions. They cannot distinguish healthy collaboration from gaming, small-field structure from insularity, or merit-based citation from network-mediated citation; they describe only where citations originate in the social graph (Alam, 25 Mar 2026).
Within bibliometrics, HEROCON differs from the h-index, g-index, total citations, and field-normalized indicators because those aggregate counts do not decompose citations by social/network context. It also differs from general network centrality measures and from tools such as VOSviewer, CiteSpace, Bibliometrix, and Scite.ai, since Citation-Constellation aims at an individual-level decomposition of incoming citations by network proximity plus an auditable two-score summary (Alam, 25 Mar 2026).
The term HeroCon is potentially ambiguous because two unrelated arXiv papers use the acronym HERO rather than HEROCON. One introduces “HERO: Hardware-Efficient RL-based Optimization Framework for NeRF Quantization”, where the paper explicitly states that there is no explicit “HeroCon” in the paper; the other introduces “HERO: Hindsight-Enhanced Reflection from Environment Observations for Agentic Self-Distillation”, where “HeroCon” is presented only as a cue for discussing agentic systems rather than as a defined technical term (Zhang et al., 10 Oct 2025, Liu et al., 10 Jun 2026). In the literature at hand, therefore, HEROCON is the formally defined concept associated with the name HeroCon, whereas the two HERO systems concern, respectively, hardware-aware NeRF quantization and hindsight-enhanced self-distillation for multi-turn agents, not citation-network diagnostics (Zhang et al., 10 Oct 2025, Liu et al., 10 Jun 2026).