A Computational Framework for Modelling Organisation-Level Semantic Identity from Longitudinal Textual Data
Authors: Brinda Murali Krishna, Oktay Karakuş, Can Eyupoglu
Organizations: School of Computer Science and Informatics, Cardiff University, UK · Department of Computer Engineering, Turkish Air Force Academy, National Defence University, Istanbul, Türkiye
Organisations continuously generate large volumes of textual data that capture how they communicate, evolve and differentiate themselves over time. Although recent advances in natural language processing have substantially improved organisation-level text analytics, existing approaches primarily represent organisations as latent embeddings or predictive feature vectors for similarity estimation, classification or retrieval. Consequently, there is currently no general computational framework for modelling organisation-level semantic identity as an interpretable and evolving semantic construct derived from longitudinal textual evidence. This paper introduces a computational framework that integrates semantic representation learning, graph-based semantic modelling, organisation-level semantic fingerprints, temporal semantic evolution and evidence-driven validation within a unified analytical methodology. Organisations are characterised through complementary semantic dimensions describing diversity, concentration, connectivity, novelty and semantic community composition, which are analysed longitudinally to infer evidence-supported semantic identities. The framework is demonstrated using a longitudinal corpus of K-pop lyrics from artists affiliated with the four major South Korean entertainment companies. The empirical analyses reveal distinguishable multidimensional semantic identities, diverse temporal evolutionary trajectories and coherent integrated identity profiles. Comprehensive validation demonstrates that the inferred identities are statistically supported, robust under alternative analytical assumptions, reproducible and operationally informative. Beyond the case study, the proposed framework establishes organisation-level semantic identity and provides a transferable methodology for modelling organisational behaviour from longitudinal textual data.
Figures & tables
Figure 1. Relationship between the validated semantic foundation established in ( Karakuş, 2026 ) and the organisational semantic identity framework proposed in this paper. The study in ( Karakuş, 2026 ) constructs a multilingual semantic landscape through representation learning, graph construction, community discovery and semantic validation. Building upon this frozen analytical foundation, the present work introduces an organisation-level framework that models entertainment companies through hierarchical aggregation, multidimensional semantic fingerprints, temporal semantic evolution and comprehensive validation. The figure highlights that the construction of the semantic landscape and the modelling of organisational identity constitute two distinct methodological contributions.
Company
Songs
Artists
First Year
Last Year
Communities
HYBE
170
15
2003
2023
14
JYP
178
19
2001
2023
15
SM
528
59
1998
2023
15
YG
413
33
1997
2023
15
Total
1,289
126
1997
2023
15
Table 1. Summary of the analysed longitudinal K-pop lyric dataset after application of the eligibility criteria.
Figure 2. End-to-end computational framework for organisational semantic identity modelling. Longitudinal lyrical observations and associated artist, company and temporal metadata are mapped into a validated semantic representation, from which song-level evidence is hierarchically aggregated into artist and company profiles. Multidimensional company semantic fingerprints capture the implemented organisational characteristics, while temporal analysis models their stability, drift and evolution. Fingerprint evidence, temporal evidence and qualified interpretation are subsequently integrated to infer organisational semantic identities. Statistical, robustness, predictive, reproducibility, traceability and validity analyses form a supporting validation foundation across the framework rather than a single terminal processing stage.
Fingerprint Component
Implemented Metrics
Interpretation
Community composition
Dominant semantic community, dominant community proportion, top-three community proportion
Describes how the company’s lyrical catalogue is distributed across the validated semantic communities and identifies its principal semantic themes.
Semantic diversity
Normalised Shannon entropy, effective number of communities, Simpson diversity
Quantifies the breadth and balance of the company’s semantic coverage. Higher values indicate a broader and more evenly distributed semantic portfolio.
Semantic concentration
Herfindahl–Hirschman Index (HHI)
Measures the extent to which lyrical output is concentrated within a limited number of semantic communities rather than being broadly distributed.
Graph integration
Mean participation coefficient, mean neighbour entropy
Characterises how strongly the company’s songs connect multiple semantic regions within the semantic similarity graph, indicating boundary-spanning behaviour and structural integration.
Semantic novelty
Mean centroid novelty, mean local novelty
Measures the originality and distinctiveness of the company’s semantic profile relative to the overall semantic landscape and its local neighbourhood.
Organisational scale
Number of eligible songs, number of artists, number of occupied semantic communities
Provides descriptive characteristics of the analysed catalogue and contextualises the remaining fingerprint dimensions without contributing directly to identity interpretation.
Table 2. Canonical components of the organisational semantic fingerprint. Rather than representing an entertainment company by a single scalar score, the proposed framework models each organisation through multiple complementary semantic properties derived from the validated semantic-community structure. Each component is operationalised using one or more implemented quantitative metrics and collectively forms the multidimensional company semantic fingerprint.
Figure 3. Multi-layer validation architecture supporting the proposed computational framework. The organisational semantic fingerprints produced by the proposed framework are evaluated through six complementary validation perspectives: (1) statistical validation, (2) robustness and sensitivity analysis, (3) predictive validation, (4) reproducibility auditing, (5) evidence traceability, and (6) validity and scope assessment. Rather than relying on a single evaluation criterion, confidence in the inferred organisational semantic identities is established through the convergence of complementary evidence across these validation layers.
Figure 4. Organisation-level semantic fingerprints of the four entertainment companies. The heatmap compares HYBE, JYP, SM and YG across the ten canonical semantic dimensions. Values are scaled independently within each dimension to emphasise relative company-level differences; consequently, colour intensity should be interpreted within rows rather than compared across different metrics. The figure shows that the four companies exhibit distinct multidimensional semantic profiles spanning diversity, concentration, connectivity and novelty.
Category
Semantic Dimension
HYBE
JYP
SM
YG
Diversity
Semantic Diversity
0.821
0.942
0.935
0.903
Effective Communities
9.234
12.820
12.584
11.534
Simpson Diversity
0.855
0.912
0.907
0.889
Concentration
Semantic Concentration
0.145
0.088
0.093
0.111
Dominant Community Share (%)
28.2
14.0
17.6
24.2
Top-3 Community Share (%)
58.2
37.6
42.6
44.6
Table 3. Organisation-level semantic fingerprint summary of the four entertainment companies. Values correspond to the canonical semantic dimensions visualised in Figure 4 . Boldface indicates the highest observed value for each semantic dimension.
Figure 5. Semantic community prevalence and company enrichment. (A) Number of songs in each semantic community, ordered from largest to smallest and coloured by the company contributing the largest share of songs. (B) Company-by-community enrichment expressed as log 2 enrichment ratios. Stars denote associations that remain significant after Holm correction. Community labels show the first two evidence terms; complete three-term descriptors are reported in Table 4 .
Community
Representative Vocabulary
Songs
Largest Contributor
Notable Association
C7
The / You / Like
266
YG (37.6%)
None
C8
You / Love / Baby
163
SM (44.2%)
None
C9
You / Know / Don
108
YG (40.7%)
SM under-represented
C6
You / Baby / For
103
YG (38.8%)
None
C13
You / Baby / The
102
SM (58.8%)
SM over-represented
C12
You / Baby / Let
83
SM (47.0%)
None
Table 4. Reference summary of the semantic communities underlying the organisation-level semantic fingerprints. Communities are ordered by size, consistent with Figure 5 . Representative vocabulary comprises the three highest-ranked evidence terms for each community.
Figure 6. Statistical differentiation of artist-level semantic metrics across the Big Four entertainment companies. (A) Global Kruskal–Wallis tests summarising differences across all companies. Horizontal bars represent the corresponding epsilon-squared ( ε2 ) effect sizes, with annotations reporting the Kruskal–Wallis statistic ( H ) and Holm-adjusted p -value. Metrics remaining statistically significant after Holm correction are indicated by a star ( ⋆ ). (B) Holm-corrected pairwise Mann–Whitney comparisons between companies. Cell colour represents −log10(pHolm) , while stars denote statistically significant comparisons ( α=0.05 ). (C) Rank-biserial correlation coefficients for the corresponding pairwise comparisons, illustrating both the magnitude and direction of the observed effects. Positive values indicate larger metric values for the first company in each comparison, whereas negative values indicate larger values for the second company.
Metric
Global
Significant Pair(s)
Largest ∣r∣
Interpretation
Participation Coefficient
No
–
Small
Cross-community participation is statistically comparable across companies.
Community Entropy
No
–
Small
Semantic diversity within artist neighbourhoods does not differ significantly.
External Strength Fraction
No
–
Small
External semantic connectivity remains broadly consistent across companies.
External Communities Reached
No
–
Small
Artists exhibit comparable breadth of semantic exploration across organisations.
Centroid Novelty
Yes
JYP–SM
0.43
Companies differ in the global distinctiveness of artist semantic profiles.
Local Novelty
Yes
HYBE–SM
0.49
Local semantic uniqueness differs significantly between companies.
Table 5. Summary of inferential statistical comparisons across artist-level semantic profile metrics. Statistically significant differences after Holm correction are concentrated in novelty-related measures, whereas semantic diversity and cross-community connectivity remain comparable across companies.
Figure 7. Temporal evolution of organisational semantic identity derived from overlapping five-year rolling windows. (A) Mean consecutive Jensen–Shannon semantic drift across the early, middle, and late phases of each company’s temporal history. (B) Typical magnitude of consecutive semantic drift summarised by the median and interquartile range. (C) Long-term temporal trend estimated using the Theil–Sen slope and its confidence interval, where negative values indicate progressively smaller semantic changes over time.
Figure 8. Temporal stability of organisational semantic identity. (A) Composite stability index derived from normalised measures of mean semantic drift, drift variability, maximum semantic drift, and change-point frequency. Higher values indicate greater long-term semantic stability. (B) Min–max-normalised values of the principal drift-based instability components: mean drift, drift variability, and maximum drift. (C) Frequency of semantic change points relative to the number of consecutive temporal transitions. Labels indicate the absolute number of detected change points and the maximum Jensen–Shannon divergence associated with those events.
Figure 9. Integrated organisational semantic identity profiles of the four entertainment companies. Each panel summarises the complete evidence supporting the inferred organisation-level semantic identity. The horizontal bars represent the relative structural fingerprint across the four principal semantic dimensions (exploration, integration, breadth and adaptation). The accompanying evidence panel reports the organisation’s relative semantic distinctiveness, dominant fingerprint dimensions, leading semantic communities, temporal qualification derived from the rolling five-year analysis, and comparative semantic relationships within the four-company landscape. Together, these complementary evidence sources provide an interpretable, traceable synthesis of organisation-level semantic identity inferred from the analysed lyrical corpus.
Figure 10. Robustness and sensitivity analysis of the proposed organisational semantic identity framework. (A) Summary of robustness conclusions across the six sensitivity analyses, distinguishing continuous semantic structure from exact ordinal rankings. (B) Sensitivity of temporal semantic drift estimates to alternative temporal aggregation windows. (C) Quantitative robustness evidence derived from similarity-metric sensitivity, bootstrap resampling, leave-one-artist-out analysis, temporal aggregation and community-detection sensitivity.
Figure 11. Predictive validation of the organisational semantic identity framework. (A) Company prediction performance under song-level and artist-grouped validation protocols. (B) Minimum catalogue fraction required to recover each organisation with predefined identification accuracies. (C) Temporal persistence of organisational semantic communities across alternative rolling-window definitions measured using community cosine similarity.
Validation dimension
Evidence
Principal finding
Outcome
Statistical validation
Artist-level non-parametric inference
Company differences are statistically supported for all three novelty-related measures, while diversity and cross-community connectivity remain comparatively similar after Holm correction.
Partially supported
Robustness and sensitivity
Six predefined perturbation analyses
Continuous company geometry remains stable under alternative metrics, resampling, artist exclusion, balanced sampling, temporal windows and community partitions; exact rankings are less stable.
Robust structure
Predictive validation
Classification, catalogue recovery and temporal persistence
Company prediction is above chance under song-level and artist-grouped protocols; partial catalogues become highly identifiable and community composition remains temporally persistent.
Informative
Computational reproducibility
Audit of 27 retained analytical scripts
All audited scripts executed successfully with controlled stochastic components and no reported execution failures.
Passed
Evidence traceability
Fingerprint, temporal and validation evidence records
Every integrated organisational interpretation is linked to explicit quantitative evidence generated by the analytical workflow.
Complete
Validity and scope assessment
Statistical, computational, data and construct validity
The evidence supports comparative, corpus-dependent semantic interpretations; causal and universal organisational claims remain outside the validated scope.
Supported within scope
Table 6. Integrated validation synthesis of the proposed organisational semantic identity framework. The table follows the six validation perspectives introduced in Section 5 and concludes with their integrated framework-level assessment.
High-dimensional dense text embeddings and large language models face real obstacles in financial-disclosure analysis: context-window limits, hallucination risk, high computational cost, and the arbitrary rotation of vector spaces across independently trained models. We present a training-free, alignment-free framework for corporate intelligence built on deterministic sparse seed vectors. Hashing word strings into a fixed high-dimensional basis places all documents and all temporal epochs in a common coordinate system by construction, removing any need for training or alignment. Accumulating these seed vectors across sentence contexts yields corpus-specific semantic signatures that compose linearly, supporting sub-second document comparison, issuer fingerprinting, tracking of how an issuer's vocabulary shifts between filings, and thematic sentence extraction, all on ordinary CPU hardware. Demonstrating the approach on a multi-year corpus of SEC filings (10-K, 10-Q, 8-K), we show how material corporate events, among them Boeing's 737 MAX crisis, Intel's supply-chain disruptions, and Bunge's acquisition of Viterra, emerge as distinct, interpretable semantic profiles, each traceable to the exact source sentences that produced it, with no domain-specific training and no LLM inference.
Calculating semantic textual similarity is a foundational task in natural language processing. Current large language models (LLMs) based methods typically rely on extracting last-layer hidden states with fixed dimensions to compute similarity for every text pairs. We argue that this paradigm is suffer from two limitations: (i) The last hidden layer encodes more general knowledge rather than just semantic knowledge, making it suboptimal for semantic similarity computation; (ii) The hidden layer dimensions of LLMs are generally very large, which introduces some redundancy and noise for representing semantics. In this work, we propose DySem, a novel training-free framework that investigates more semantic-related internal components of LLMs via multilingual consensus, and shifts away from static representation spaces in favor of dynamic, sample-specific semantic dimensions by constructing text-dependent joint semantic set and computes similarity over this shared dimensional subset. Extensive experiments across various LLMs show that our method consistently outperforms recent baselines while maintaining lower dimensions for similarity calculation. The code is released at https://github.com/szu-tera/DySem.
Kaijie Zheng, Weiqin Wang, Yile Wang +1
College of Computer Science and Software Engineering, Shenzhen University
Collective semantic change in populations of language model agents is a measurable dynamical phenomenon. We present a passive longitudinal instrument called Kopterix that observes the semantic state of an agent population as a sequence of bounded observations under a protocol defined before the observations begin. Each observation divides the sampled feed by post age into surface, mid-stream, and residue layers, which makes semantic differences across content age measurable alongside run-to-run change. We validate the instrument on Moltbook, an agent-native social platform, over a two-month window of scheduled observations, with the periodicity check extended across approximately four months. At the lexical level, rarefied entropy resolves an April-May difference in the evenness of the stored top 200 unigram distributions, and adjacent states are lexically closer than states paired after timestamp shuffling. At the geometric level, grand mean centering exposes the scale of a common embedding direction, and scheduled shuffle checks support a recurring excess in the mid-stream to residue separation relative to the shuffled reference. At the temporal level, detrended scalar quantities and centered layer centroids lose much of their similarity over several hours, and a weaker positive component declines across longer separations with no strong weekly recurrence. Several attractive apparent structures failed their controls, and each reading is limited to the level its controls support. The design applies wherever a population of agents produces a timestamped language environment that can be observed repeatedly and divided by content age.
Elena Kopteva
The Grainger College of Engineering, Department of Physics, University of Illinois Urbana-Champaign, Urbana, Illinois 61801, USA