Disha Purohit
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Disha Purohit has a total of 56 co-authors. Here are the top co-authors by number of shared publications:
- Maria-Esther Vidal
TIB Leibniz Information Centre for Science and Technology
14 shared publications - Yashrajsinh Chudasama
TIB-Leibniz Information Centre for Science and Technology
13 shared publications - Enrique Iglesias
L3S Research Center
3 shared publications - Maria Torrente
Medical Oncology Department, Hospital Universitario Puerta de Hierro Majadahonda, Madrid, Spain
2 shared publications - Philipp D. Rohde
TIB Leibniz Information Centre for Science and Technology, Hannover, Germany
2 shared publications Aaron García
Universidad Politécnica de Madrid, Spain
1 shared publication
Latest Publications
- 26 September 2026
- Published by Springer Nature
- p. 3-12
From Normal Forms to Knowledge Graph Ecosystems: Foundations for Trustworthy AI
- 26 September 2026
- Published by Springer Nature
- p. 3-12
From Normal Forms to Knowledge Graph Ecosystems: Foundations for Trustworthy AI
- Open Access
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
PRISM: Path-guided Reasoning and Interpretation for Semantic Models
Knowledge Graph Completion (KGC) relies on Link Prediction (LP) models to infer missing facts in incomplete Knowledge Graphs (KGs). Although LP models have achieved remarkable predictive performance, identifying the facts that explain their predictions remains a fundamental challenge. Existing explainability methods either rely on computationally expensive combinatorial exploration of candidate facts or exploit only symbolic or sub-symbolic representations, limiting their scalability, interpretability, or explanatory power. Consequently, developing methodologies capable of generating faithful, interpretable, and computationally efficient explanations remains an open research problem. This paper presents a neuro-symbolic methodology for generating explanations of LP decisions through the identification of explanatory facts. The proposed methodology associates each predicted relationship with two complementary forms of explanation: necessary explanations, which identify the facts essential for a prediction, and sufficient explanations, which identify additional facts supporting the predicted relationship. The methodology is realized through PRISM, a neuro-symbolic framework that integrates symbolic reasoning with Knowledge Graph Embeddings (KGEs) to exploit both explicit semantic knowledge and latent relational representations. By guiding the explanation process toward semantically relevant candidate facts, PRISM substantially reduces the explanation search space while preserving explanation quality. Experimental results on real-world KGs demonstrate that the proposed methodology consistently generates faithful and interpretable explanations while requiring significantly fewer candidate explorations than state-of-the-art symbolic and sub-symbolic explainability methods. These results establish the effectiveness and scalability of the proposedapproach for explainable knowledge-based AI systems.
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessPRISM: Path-guided Reasoning and Interpretation for Semantic Models
Knowledge Graph Completion (KGC) relies on Link Prediction (LP) models to infer missing facts in incomplete Knowledge Graphs (KGs). Although LP models have achieved remarkable predictive performance, identifying the facts that explain their predictions remains a fundamental challenge. Existing explainability methods either rely on computationally expensive combinatorial exploration of candidate facts or exploit only symbolic or sub-symbolic representations, limiting their scalability, interpretability, or explanatory power. Consequently, developing methodologies capable of generating faithful, interpretable, and computationally efficient explanations remains an open research problem. This paper presents a neuro-symbolic methodology for generating explanations of LP decisions through the identification of explanatory facts. The proposed methodology associates each predicted relationship with two complementary forms of explanation: necessary explanations, which identify the facts essential for a prediction, and sufficient explanations, which identify additional facts supporting the predicted relationship. The methodology is realized through PRISM, a neuro-symbolic framework that integrates symbolic reasoning with Knowledge Graph Embeddings (KGEs) to exploit both explicit semantic knowledge and latent relational representations. By guiding the explanation process toward semantically relevant candidate facts, PRISM substantially reduces the explanation search space while preserving explanation quality. Experimental results on real-world KGs demonstrate that the proposed methodology consistently generates faithful and interpretable explanations while requiring significantly fewer candidate explorations than state-of-the-art symbolic and sub-symbolic explainability methods. These results establish the effectiveness and scalability of the proposedapproach for explainable knowledge-based AI systems.
- Open Access
- 1 July 2026
- Published by Elsevier in SSRN Electronic Journal
Generating Symbolic Chain-of-Thoughts from Knowledge Graphs to Fine-Tune Language Models for Dataset Metadata Extraction in Scientific Publications
Scientific progress increasingly depends on the ability to discover, understand, and reuse datasets described in scientific publications. However, dataset metadata is often embedded in unstructured text, making dataset discovery, benchmarking, and reuse labor-intensive. Although large language models (LLMs) have shown promise for metadata extraction, they primarily rely on textual evidence and frequently produce incomplete or inconsistent descriptions that are difficult to align with semantic standards and knowledge-driven discovery systems. This paper addresses the problem of extracting dataset metadata from scientific publications in a semantically grounded and interpretable manner. Given a corpus of research papers and a metadata vocabulary, the objective is to automatically identify datasets, determine the scientific tasks for which they are used, and generate structured metadata describing task–dataset relationships as standards-compliant RDF aligned with the DCAT vocabulary. We present METAMINE, a neuro-symbolic framework that integrates knowledge extraction, knowledge graph (KG) construction, semantic reasoning, and LLM adaptation. Relations among scientific tasks, datasets, methods, and publications are extracted from research papers to construct a KG. Semantic traversal algorithms identify high-confidence multi-hop paths connecting tasks and datasets, which are then transformed into symbolic Chain-of-Thoughts (CoTs) that provide explicit reasoning traces. These KG-derived CoTs are used to fine-tune an LLM, aligning neural reasoning with the relational structure encoded in the KG. By combining symbolic reasoning with neural language models, METAMINE generates semantically grounded metadata together with transparent explanations that justify dataset–task associations. The resulting metadata supports interoperability, semantic querying, FAIR dataset descriptions, and more effective dataset discovery, benchmarking, and scientific knowledge management.
- 1 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessGenerating Symbolic Chain-of-Thoughts from Knowledge Graphs to Fine-Tune Language Models for Dataset Metadata Extraction in Scientific Publications
Scientific progress increasingly depends on the ability to discover, understand, and reuse datasets described in scientific publications. However, dataset metadata is often embedded in unstructured text, making dataset discovery, benchmarking, and reuse labor-intensive. Although large language models (LLMs) have shown promise for metadata extraction, they primarily rely on textual evidence and frequently produce incomplete or inconsistent descriptions that are difficult to align with semantic standards and knowledge-driven discovery systems. This paper addresses the problem of extracting dataset metadata from scientific publications in a semantically grounded and interpretable manner. Given a corpus of research papers and a metadata vocabulary, the objective is to automatically identify datasets, determine the scientific tasks for which they are used, and generate structured metadata describing task–dataset relationships as standards-compliant RDF aligned with the DCAT vocabulary. We present METAMINE, a neuro-symbolic framework that integrates knowledge extraction, knowledge graph (KG) construction, semantic reasoning, and LLM adaptation. Relations among scientific tasks, datasets, methods, and publications are extracted from research papers to construct a KG. Semantic traversal algorithms identify high-confidence multi-hop paths connecting tasks and datasets, which are then transformed into symbolic Chain-of-Thoughts (CoTs) that provide explicit reasoning traces. These KG-derived CoTs are used to fine-tune an LLM, aligning neural reasoning with the relational structure encoded in the KG. By combining symbolic reasoning with neural language models, METAMINE generates semantically grounded metadata together with transparent explanations that justify dataset–task associations. The resulting metadata supports interoperability, semantic querying, FAIR dataset descriptions, and more effective dataset discovery, benchmarking, and scientific knowledge management.
- Open Access
- 10 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 111-118
Capturing Symbolic Knowledge of Constraints and Incompleteness to Guide Inductive Learning in Neuro-Symbolic Knowledge Graph Completion
- 10 December 2025
- Published by Association for Computing Machinery (ACM)
- p. 111-118
Open AccessCapturing Symbolic Knowledge of Constraints and Incompleteness to Guide Inductive Learning in Neuro-Symbolic Knowledge Graph Completion
- Open Access
- 20 June 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 325
VANILLA: Validated knowledge graph completion—A Normalization-based framework for Integrity, Link prediction, and Logical Accuracy
- 20 June 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 325
Open AccessVANILLA: Validated knowledge graph completion—A Normalization-based framework for Integrity, Link prediction, and Logical Accuracy
- Open Access
- 24 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 74705-74723
SimE: A Knowledge Graph Embedding Model to Encode Self-Similar Structures Through Algebraic and Geometric Transformations
Knowledge Graphs (KGs), with their intricate hierarchies and semantic relationships, present unique challenges for graph representation learning, necessitating tailored approaches to effectively capture and encode their complex structures into useful numerical representations. The fractal-like nature of these graphs, where patterns repeat at various scales and complexities, requires specialized algorithms that can adapt and learn from the multi-level structures inherent in the data. This similarity to fractals requires methods that preserve the recursive detail of knowledge graphs while facilitating efficient learning and extraction of relational patterns. In this study, we explore the integration of similarity group with attention mechanisms to represent knowledge graphs in complex spaces. In our approach, SimE, we make use of the algebraic (bijection) and geometric (similarity) properties of the similarity transformations to enhance the representation of self-similar fractals in KGs. We empirically validate the capability of providing representations of bijections and similarities in benchmark KGs. We also conducted controlled experiments that captured one-to-one, one-to-many, and many-to-many relational patterns and studied the behavior of state-of-the-art models including the proposed SimE model. Because of the lack of benchmark fractal-like KG datasets, we created a set of fractal-like testbeds to assess the subgraph similarity learning ability of models. The observed results suggest that SimE captures the complex geometric structures of KGs whose statements satisfy these algebraic and geometric properties. In particular, SimE is competitive with state-of-the-art KG embedding models and is able to achieve high values of Hits@1. As a result, SimE is capable of effectively predicting correct links and ranking them with the highest ranks. SimE is publicly available on GitHub https://github.com/NIMI-research/SimE.
- 24 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 74705-74723
Open AccessSimE: A Knowledge Graph Embedding Model to Encode Self-Similar Structures Through Algebraic and Geometric Transformations
Knowledge Graphs (KGs), with their intricate hierarchies and semantic relationships, present unique challenges for graph representation learning, necessitating tailored approaches to effectively capture and encode their complex structures into useful numerical representations. The fractal-like nature of these graphs, where patterns repeat at various scales and complexities, requires specialized algorithms that can adapt and learn from the multi-level structures inherent in the data. This similarity to fractals requires methods that preserve the recursive detail of knowledge graphs while facilitating efficient learning and extraction of relational patterns. In this study, we explore the integration of similarity group with attention mechanisms to represent knowledge graphs in complex spaces. In our approach, SimE, we make use of the algebraic (bijection) and geometric (similarity) properties of the similarity transformations to enhance the representation of self-similar fractals in KGs. We empirically validate the capability of providing representations of bijections and similarities in benchmark KGs. We also conducted controlled experiments that captured one-to-one, one-to-many, and many-to-many relational patterns and studied the behavior of state-of-the-art models including the proposed SimE model. Because of the lack of benchmark fractal-like KG datasets, we created a set of fractal-like testbeds to assess the subgraph similarity learning ability of models. The observed results suggest that SimE captures the complex geometric structures of KGs whose statements satisfy these algebraic and geometric properties. In particular, SimE is competitive with state-of-the-art KG embedding models and is able to achieve high values of Hits@1. As a result, SimE is capable of effectively predicting correct links and ranking them with the highest ranks. SimE is publicly available on GitHub https://github.com/NIMI-research/SimE.
- Open Access
- 10 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 1108-1109
Enhancing Medical Knowledge Discovery: A Neuro-symbolic System for Inductive Learning over Medical KGs
Medical knowledge graphs (KGs) excel at integrating heterogeneous healthcare data with domain knowledge, but face challenges due to incompleteness. While Knowledge Graph Embedding (KGE) models show promise in link prediction, they often fail to incorporate crucial semantic constraints from medical ontologies and clinical guidelines. We propose a neuro-symbolic system that enhances medical knowledge discovery by combining symbolic learning from medical ontologies, inductive learning through KGE, and semantic constraint validation. Applied to lung cancer care, our system demonstrates enhanced performance in predicting novel medical relationships while maintaining semantic consistency with medical knowledge. Experimental results show our approach enhances the KGE model's performance while ensuring clinical validity and the implementation is publicly accessible on GitHub https://github.com/SDM-TIB/KOSMOS.
- 10 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 1108-1109
Open AccessEnhancing Medical Knowledge Discovery: A Neuro-symbolic System for Inductive Learning over Medical KGs
Medical knowledge graphs (KGs) excel at integrating heterogeneous healthcare data with domain knowledge, but face challenges due to incompleteness. While Knowledge Graph Embedding (KGE) models show promise in link prediction, they often fail to incorporate crucial semantic constraints from medical ontologies and clinical guidelines. We propose a neuro-symbolic system that enhances medical knowledge discovery by combining symbolic learning from medical ontologies, inductive learning through KGE, and semantic constraint validation. Applied to lung cancer care, our system demonstrates enhanced performance in predicting novel medical relationships while maintaining semantic consistency with medical knowledge. Experimental results show our approach enhances the KGE model's performance while ensuring clinical validity and the implementation is publicly accessible on GitHub https://github.com/SDM-TIB/KOSMOS.
- Open Access
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 39489-39509
Toward Interpretable Hybrid AI: Integrating Knowledge Graphs and Symbolic Reasoning in Medicine
Knowledge Graphs (KGs) are data structures that enable the integration of heterogeneous data sources and supporting both knowledge representation and formal reasoning. In this paper, we introduce TrustKG, a KG-based framework designed to enhance the interpretability and reliability of hybrid AI systems in healthcare. Positioned within the context of lung cancer, TrustKG supports link prediction, which uncovers hidden relationships within medical data, and counterfactual prediction, which explores alternative scenarios to understand causal factors. These tasks are addressed through two specialized hybrid AI systems, VISE and HealthCareAI, which combine symbolic reasoning with inductive learning over KGs to provide interpretable AI solutions for clinical decision-making. Leveraging KGs to represent biomedical properties and relationships, and augmenting them with learned patterns through symbolic reasoning, our hybrid approach produces models that are both accurate and transparent. This interpretability is particularly important in medical applications, where trust and reliability in AI-driven predictions are paramount. Our empirical analysis demonstrates the effectiveness of VISE and HealthCareAI in improving the predictive accuracy and clarity of model outputs. By addressing challenges in link prediction—such as discovering previously unknown connections between medical entities—and in counterfactual prediction, TrustKG, with VISE and HealthCareAI, underscores the potential of integrating KGs with symbolic AI to create trustworthy, interpretable AI systems in healthcare. This paper contributes to the advancement of semantic AI, offering a pathway for robust and reliable AI solutions in clinical settings.
- 13 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 39489-39509
Open AccessToward Interpretable Hybrid AI: Integrating Knowledge Graphs and Symbolic Reasoning in Medicine
Knowledge Graphs (KGs) are data structures that enable the integration of heterogeneous data sources and supporting both knowledge representation and formal reasoning. In this paper, we introduce TrustKG, a KG-based framework designed to enhance the interpretability and reliability of hybrid AI systems in healthcare. Positioned within the context of lung cancer, TrustKG supports link prediction, which uncovers hidden relationships within medical data, and counterfactual prediction, which explores alternative scenarios to understand causal factors. These tasks are addressed through two specialized hybrid AI systems, VISE and HealthCareAI, which combine symbolic reasoning with inductive learning over KGs to provide interpretable AI solutions for clinical decision-making. Leveraging KGs to represent biomedical properties and relationships, and augmenting them with learned patterns through symbolic reasoning, our hybrid approach produces models that are both accurate and transparent. This interpretability is particularly important in medical applications, where trust and reliability in AI-driven predictions are paramount. Our empirical analysis demonstrates the effectiveness of VISE and HealthCareAI in improving the predictive accuracy and clarity of model outputs. By addressing challenges in link prediction—such as discovering previously unknown connections between medical entities—and in counterfactual prediction, TrustKG, with VISE and HealthCareAI, underscores the potential of integrating KGs with symbolic AI to create trustworthy, interpretable AI systems in healthcare. This paper contributes to the advancement of semantic AI, offering a pathway for robust and reliable AI solutions in clinical settings.
- Open Access
- 27 December 2024
- Published by Elsevier in Journal of Web Semantics
- Vol. 84
Integrating Knowledge Graphs with Symbolic AI: The Path to Interpretable Hybrid AI Systems in Medicine
- 27 December 2024
- Published by Elsevier in Journal of Web Semantics
- Vol. 84
Open AccessIntegrating Knowledge Graphs with Symbolic AI: The Path to Interpretable Hybrid AI Systems in Medicine
- Open Access
- 23 November 2024
- Published by Elsevier in Data in Brief
- Vol. 57, 111167
A lung cancer diagnosis and treatment dataset with geno- and phenotypical characteristics of the patient
- 23 November 2024
- Published by Elsevier in Data in Brief
- Vol. 57, 111167
Open AccessA lung cancer diagnosis and treatment dataset with geno- and phenotypical characteristics of the patient