Attribution Bias in Philosophical Knowledge Graphs: Corpus Frequency versus Temporal Sourcing

Joy Bose

arxiv: 2606.29070 · v1 · pith:TACSSX7L · submitted 2026-06-27 · cs.DL · cs.CY· cs.SI

Attribution Bias in Philosophical Knowledge Graphs: Corpus Frequency versus Temporal Sourcing

Joy Bose This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 reserved 2026-06-30 08:02 UTCgrok-4.3pith:TACSSX7L record.json open to challenge →

**Figure 1.** Figure 1: Two Attribution Pipelines for Philosophical Concepts [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] reproduced from arXiv: 2606.29070

classification cs.DL cs.CYcs.SI

keywords knowledge graphsattribution biastemporal sourcingbetweenness centralitystructural homologyIndian philosophydarshanaAdvaita Vedanta

0 comments

The pith

Corpus frequency attributes philosophical concepts to the wrong traditions by up to 2,288 years while temporal sourcing shows early pluralism and structural homologies.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that assigning concepts in knowledge graphs by which school mentions them most often mixes up how much text survives with which idea came first or mattered most. Using a graph of 28,322 links among Hindu, Buddhist, and Jain sources, it finds that seven of the top 25 concepts by betweenness centrality appear hundreds or thousands of years before the school they are usually credited to. A snapshot limited to explicitly dated texts around 300 BCE instead shows roughly 59 percent Vedic, 24 percent Jain, and 18 percent Buddhist structure. The same temporal filter reveals that later growth in the graph is dominated by texts carrying Advaita dates, pointing to survival bias rather than philosophical dominance. The dated graph then supports measuring structural similarity across traditions through ego-network vectors, recovering known pairs and identifying new ones such as Nibbana with samsara.

Core claim

Standard frequency-based attribution in philosophical knowledge graphs conflates textual survival with historical priority and significance; a temporally ordered graph of 28,322 relationships demonstrates that seven of the twenty-five highest-betweenness concepts predate their attributed schools by 288 to 2,288 years, that a 300 BCE slice using only dated sources is 59 percent Vedic, 24 percent Jain and 18 percent Buddhist, and that ego-network feature vectors recover known cross-tradition correspondences while surfacing previously unnoted structural homologies such as Nibbana-samsara and Cetana-ajiva.

What carries the argument

The darshana-graph of 28,322 relationships across Hindu, Buddhist, and Jain traditions, filtered by explicit source dates and analyzed with betweenness centrality plus ego-network feature vectors for structural similarity.

If this is right

Moksha should be reassigned to Jain sources more than 1,200 years before its usual Advaita attribution.
The apparent dominance of Advaita Vedanta between 300 CE and 800 CE is an artifact of 97.4 percent of new nodes carrying proxy dates from that school.
Ego-network similarity recovers the known pairs purusha-jiva (0.990) and prakriti-maya (0.972).
Structural homology identifies Nibbana and samsara as sharing the role of ultimate reference concept in their respective soteriologies despite doctrinal opposition.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same temporal-plus-similarity method could be applied to re-attribute concepts in other historical knowledge graphs where dated corpora exist.
High structural similarity between doctrinal opposites suggests that traditions can share functional roles in conceptual space even when their explicit teachings differ.
If betweenness centrality is accepted as a proxy for significance, then early Jain and Buddhist contributions to the shared graph were larger than frequency-based attribution currently records.

Load-bearing premise

That the set of explicitly dated sources available at 300 BCE forms an unbiased sample of the conceptual landscape at that time rather than reflecting which texts happened to survive.

What would settle it

Discovery of a large cache of explicitly dated pre-300 BCE texts whose betweenness-centrality distribution differs substantially from the reported 59-24-18 split.

Figures

Figures reproduced from arXiv: 2606.29070 by Joy Bose.

**Figure 4.** Figure 4: Network Growth Across Eras: Tier-1 vs. Tier-2 Nodes. The jump from 18 nodes at 300 CE to 1,028 nodes at 800 CE is not historical philosophical innovation. It reflects the simultaneous entry of 1,001 Advaita Vedanta Tier-2 proxy nodes dated to 788 CE. Of the 1,028 nodes at 800 CE, 97.4% are Tier-2 proxies whose dates reflect the school's founding date rather than any concept-level attestation. 4.5 The 300 B… view at source ↗

**Figure 5.** Figure 5: Network Tradition Composition at 300 BCE (17 Tier- [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗

read the original abstract

Computational knowledge graphs assign philosophical concepts to traditions based on corpus frequency: the school that mentions a concept most becomes its attributed tradition. We argue this conflates three measurements: textual power, historical priority, and philosophical significance, demonstrated using the darshana-graph, a knowledge graph of 28,322 relationships across Hindu, Buddhist, and Jain traditions. Seven of the top 25 concepts by betweenness centrality predate their attributed school by 288 to 2,288 years. Moksha, attributed to Advaita Vedanta, appears first in Jain sources over 1,200 years earlier. The most reliable snapshot, at 300 BCE using only explicitly dated sources, shows a genuinely pluralistic structure: 59% Vedic, 24% Jain, 18% Buddhist. We also quantify a critical distortion in the temporal method: between 300 CE and 800 CE the network grows from 18 to 1,028 nodes, with 97.4% carrying Advaita proxy dates, revealing that apparent dominance reflects textual survival, not philosophical history. Beyond correcting attribution bias, the temporally grounded graph enables structural homology analysis across traditions. Ego-network feature vectors applied to 48 temporally labelled concepts across eight traditions identify cross-tradition concept pairs with high structural similarity. The method recovers known correspondences including purusha-jiva (Samkhya/Jain, sim 0.990) and prakriti-maya (Samkhya/Vedic, sim 0.972), and surfaces novel homologies. Nibbana and samsara score 0.954 despite being doctrinal opposites: both function as the ultimate reference concept in their tradition's soteriology. Cetana (Buddhist intention) and ajiva (Jain non-living matter) score 0.923, a pairing absent from the literature. These are not claims of doctrinal equivalence but of measurable structural homology: different philosophical vocabularies navigating a shared conceptual space.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper shows corpus-frequency attribution distorts historical priority in philosophical KGs and offers a temporal graph plus homology method, but the quantitative claims rest on unvalidated source selection and centrality assumptions.

read the letter

The main takeaway is that frequency-based attribution in these graphs mixes up textual survival with actual historical order. The paper gives concrete numbers: seven of the top 25 betweenness concepts predate their assigned schools by 288–2,288 years, Moksha shows up in Jain sources over 1,200 years before Advaita, and the 300 BCE slice from dated sources is 59% Vedic, 24% Jain, 18% Buddhist. It also flags the big temporal skew where the graph jumps from 18 to 1,028 nodes between 300–800 CE with 97.4% Advaita proxy dates.

What works is the attempt to separate the three things the abstract names—textual power, priority, and significance—and the homology step using ego-network vectors. Recovering purusha-jiva and prakriti-maya at high similarity is a reasonable check, and the Nibbana-samsara and Cetana-ajiva pairs are at least interesting as structural observations even if they are not doctrinal claims.

The soft spots are exactly where the stress-test note points. The 300 BCE snapshot and the predating claims depend on the subset of explicitly dated sources being representative and on betweenness tracking philosophical weight rather than just denser textual linkage. No sensitivity checks on inclusion rules or alternative measures are mentioned in the abstract, and the same graph is used for both the bias demonstration and the homology scores. Without the methods section or data, it is hard to tell how much of the result is construction artifact.

This is for people working on knowledge graphs in digital humanities or philosophy-adjacent AI. It deserves a serious referee to check whether the full paper supplies the missing validation steps; the idea is worth testing if the data pipeline holds up.

Referee Report

3 major / 2 minor

Summary. The manuscript constructs the darshana-graph, a knowledge graph of 28,322 relationships across Hindu, Buddhist, and Jain traditions, to argue that corpus-frequency attribution of concepts to schools conflates textual power, historical priority, and philosophical significance. Using temporal sourcing from explicitly dated sources, it reports that seven of the top 25 concepts by betweenness centrality predate their attributed schools by 288–2,288 years (e.g., Moksha first in Jain sources >1,200 years before Advaita Vedanta), that the 300 BCE snapshot yields a 59% Vedic / 24% Jain / 18% Buddhist structure, and that network growth from 300–800 CE is 97.4% Advaita-proxy dated due to textual survival. It further applies ego-network feature vectors to 48 temporally labelled concepts to recover known homologies (purusha-jiva sim 0.990) and identify novel ones (Nibbana–samsara sim 0.954; Cetana–ajiva sim 0.923).

Significance. If the graph construction, date assignments, and centrality interpretations prove robust, the work offers a concrete, quantitative demonstration of attribution bias in philosophical KGs and a temporal corrective that could improve historical modeling in digital humanities. The homology analysis supplies a falsifiable, reproducible method for cross-tradition structural comparison that goes beyond doctrinal claims. Credit is due for grounding claims in explicit temporal snapshots and for reporting specific similarity scores that invite independent testing.

major comments (3)

[Methods (darshana-graph construction and temporal sourcing)] The 300 BCE snapshot distribution (59/24/18) and all predating claims rest on the subset of 'explicitly dated sources,' yet the manuscript provides no validation of date-assignment rules, no sensitivity analysis on inclusion criteria, and no error rates for the temporal labels; this directly undermines the representativeness assumption required for the central bias-correction result.
[Results (betweenness centrality and predating analysis)] Betweenness centrality is interpreted as tracking philosophical significance, but no comparison to alternative measures (degree, closeness, eigenvector) or correlation with independent historical significance indicators is reported; without this, the claim that the top-25 predating concepts demonstrate attribution bias rather than textual survival patterns cannot be evaluated.
[Homology analysis (ego-network feature vectors)] The ego-network feature vectors for homology detection are derived from the same 28,322-edge graph used to demonstrate bias, creating dependence between the bias-correction and homology sections; while temporal dates draw on external sources, the absence of an independent validation set for the feature vectors makes the reported similarities (e.g., Nibbana–samsara 0.954) vulnerable to graph-construction artifacts.

minor comments (2)

[Abstract and §4] The phrase 'Advaita proxy dates' appears in the abstract and results without an explicit definition or reference to the precise dating heuristic employed.
[Homology results] No table or appendix lists the 48 temporally labelled concepts used for homology, the exact feature-vector construction, or the similarity metric formula, hindering reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the positive evaluation of the manuscript's significance and for the constructive major comments. We address each point below with proposed revisions where the concerns identify gaps in the current version.

read point-by-point responses

Referee: [Methods (darshana-graph construction and temporal sourcing)] The 300 BCE snapshot distribution (59/24/18) and all predating claims rest on the subset of 'explicitly dated sources,' yet the manuscript provides no validation of date-assignment rules, no sensitivity analysis on inclusion criteria, and no error rates for the temporal labels; this directly undermines the representativeness assumption required for the central bias-correction result.

Authors: We agree that the absence of explicit validation, sensitivity analysis, and error rates for the temporal labels is a limitation that weakens the representativeness claims. The revised manuscript will add a new subsection detailing the date-assignment protocol (drawing on standard scholarly chronologies), sensitivity tests varying inclusion thresholds, and error estimates obtained via cross-referencing with secondary historical sources. These additions will directly support the 300 BCE snapshot and predating analyses. revision: yes
Referee: [Results (betweenness centrality and predating analysis)] Betweenness centrality is interpreted as tracking philosophical significance, but no comparison to alternative measures (degree, closeness, eigenvector) or correlation with independent historical significance indicators is reported; without this, the claim that the top-25 predating concepts demonstrate attribution bias rather than textual survival patterns cannot be evaluated.

Authors: Betweenness centrality was selected because it identifies concepts that bridge disparate regions of the network, aligning with the paper's focus on how attribution bias arises from uneven textual connectivity. We acknowledge that direct comparisons to other measures and external indicators are missing. The revision will report top-25 rankings under degree, closeness, and eigenvector centrality, include a correlation table, and discuss alignment with available historical metadata on textual survival, while noting data limitations for full external validation. revision: yes
Referee: [Homology analysis (ego-network feature vectors)] The ego-network feature vectors for homology detection are derived from the same 28,322-edge graph used to demonstrate bias, creating dependence between the bias-correction and homology sections; while temporal dates draw on external sources, the absence of an independent validation set for the feature vectors makes the reported similarities (e.g., Nibbana–samsara 0.954) vulnerable to graph-construction artifacts.

Authors: The temporal labels used for bias correction are drawn from independent external sources and are not derived from the graph structure itself; the feature vectors measure local topology around temporally labelled nodes. Recovery of established homologies (purusha-jiva at 0.990) supplies internal corroboration. The revision will separate the sections more clearly, add explicit discussion of this independence, test similarity robustness under edge-sampling perturbations, and note the absence of an external validation corpus as a limitation for future work. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper constructs the darshana-graph from a corpus of explicitly dated historical sources and applies standard graph algorithms (betweenness centrality, ego-network feature vectors) to derive its quantitative claims. Temporal attributions and the 300 BCE snapshot rest on external date assignments rather than being computed from or defined in terms of the graph metrics themselves. Homology scores are direct structural computations on the resulting graph without any fitted parameters, self-referential definitions, or load-bearing self-citations that reduce the central results to the inputs by construction. No equations or ansatzes in the provided text exhibit the enumerated circular patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on the accuracy of source dating and the assumption that graph metrics reflect philosophical rather than textual properties; no free parameters or invented entities are evident from the abstract.

axioms (2)

domain assumption Explicitly dated sources at 300 BCE form an unbiased sample of the conceptual landscape.
Invoked to produce the 59/24/18 percent distribution claimed as the most reliable snapshot.
domain assumption Betweenness centrality ranks concepts by philosophical significance.
Used to select the top 25 concepts whose attribution dates are then checked.

pith-pipeline@v0.9.1-grok · 5887 in / 1505 out tokens · 52749 ms · 2026-06-30T08:02:23.669060+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

2 extracted references

[1]

Bronkhorst, J. (2007). Greater Magadha: Studies in the Culture of Early India. Brill. Collins, R. (1998). The Sociology of Philosophies: A Global Theory of Intellectual Change . Harvard University Press. D'Ignazio, C., & Klein, L. F. (2020). Data Feminism. MIT Press. Drucker, J. (2011). Humanities approaches to graphical display. Digital Humanities Quarte...

2007
[2]

Jänicke, S., Franzini, G., Cheema, M

Oxford University Press. Jänicke, S., Franzini, G., Cheema, M. F., & Scheuermann, G. (2015). On close and distant reading in digital humanities: A survey and future challenges. In Proceedings of the Eurographics Conference on Visualization (EuroVis). Eurographics Association. Kalupahana, D. J. (1986). Nagarjuna: The Philosophy of the Middle Way. SUNY Pres...

2015

[1] [1]

Bronkhorst, J. (2007). Greater Magadha: Studies in the Culture of Early India. Brill. Collins, R. (1998). The Sociology of Philosophies: A Global Theory of Intellectual Change . Harvard University Press. D'Ignazio, C., & Klein, L. F. (2020). Data Feminism. MIT Press. Drucker, J. (2011). Humanities approaches to graphical display. Digital Humanities Quarte...

2007

[2] [2]

Jänicke, S., Franzini, G., Cheema, M

Oxford University Press. Jänicke, S., Franzini, G., Cheema, M. F., & Scheuermann, G. (2015). On close and distant reading in digital humanities: A survey and future challenges. In Proceedings of the Eurographics Conference on Visualization (EuroVis). Eurographics Association. Kalupahana, D. J. (1986). Nagarjuna: The Philosophy of the Middle Way. SUNY Pres...

2015