Dobromir Tsolyov
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Dobromir Tsolyov has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- George Petkov
GATE Institute, Sofia University, 1164 Sofia, Bulgaria
4 shared publications - Simeon Karpuzov
GATE Institute, Sofia University, 1164 Sofia, Bulgaria
4 shared publications - Beloslava Malakova
GATE Institute, Sofia University, 1164 Sofia, Bulgaria
2 shared publications - Iglika Ivanova
Big Data for Smart Society Institute (GATE)
2 shared publications - Keith Peter Kiely
Big Data for Smart Society Institute (GATE)
2 shared publications - Milen Zamfirov
Софийски университет „Св. Климент Охридски
2 shared publications
Latest Publications
- Open Access
Risk Beyond the Variant: A Composite Metric for Autism Gene Pathogenicity
Background: Variant-based pathogenicity predictors such as REVEL evaluate missense variants in isolation, discarding the gene-length and allele-frequency context needed to compare collections of genes. Methods: We introduce a composite gene-level metric integrating Hardy–Weinberg heterozygosity, coding-sequence length, and REVEL scores. The metric returns a single value per gene expressing variant burden per unit of coding sequence within a given cohort, so that the ratio between a case and a control cohort quantifies gene-level enrichment. It was evaluated on 55 high-confidence autism genes, defined as the intersection of three large-scale ASD sequencing studies, against the 1000 Genomes reference. Results: It identifies elevated pathogenic burden in 48 of 55 genes, removes gene-length and variant-count confounds, and substantially outperforms naive gene-level aggregation of REVEL scores. Bootstrap resampling and a label-permutation control confirm the enrichment is stable and not an artefact of the scoring construction. Conclusions: The metric allows genes to be ranked within a set and aggregate burden to be compared across gene sets. We present it as a complementary gene-level layer for case–control and gene-set comparisons, with a nonlinear successor outlined as future work.
Open AccessRisk Beyond the Variant: A Composite Metric for Autism Gene Pathogenicity
Background: Variant-based pathogenicity predictors such as REVEL evaluate missense variants in isolation, discarding the gene-length and allele-frequency context needed to compare collections of genes. Methods: We introduce a composite gene-level metric integrating Hardy–Weinberg heterozygosity, coding-sequence length, and REVEL scores. The metric returns a single value per gene expressing variant burden per unit of coding sequence within a given cohort, so that the ratio between a case and a control cohort quantifies gene-level enrichment. It was evaluated on 55 high-confidence autism genes, defined as the intersection of three large-scale ASD sequencing studies, against the 1000 Genomes reference. Results: It identifies elevated pathogenic burden in 48 of 55 genes, removes gene-length and variant-count confounds, and substantially outperforms naive gene-level aggregation of REVEL scores. Bootstrap resampling and a label-permutation control confirm the enrichment is stable and not an artefact of the scoring construction. Conclusions: The metric allows genes to be ranked within a set and aggregate burden to be compared across gene sets. We present it as a complementary gene-level layer for case–control and gene-set comparisons, with a nonlinear successor outlined as future work.
- Open Access
- 16 July 2026
- Published by MDPI AG in Preprints.org
Risk Beyond the Variant: A Composite Metric for Autism Gene Pathogenicity
Variant-based pathogenicity predictors such as REVEL evaluate missense variants in isolation, discarding the gene-length and allele-frequency context needed to compare collections of genes. In this paper, we introduce a composite gene-level metric integrating Hardy-Weinberg heterozygosity, coding-sequence length, and REVEL scores, evaluated on 55 high-confidence autism genes against the 1000 Genomes reference. It identifies elevated pathogenic burden in 48 of 55 genes, removes gene-length and variant-count confounds, and outperforms REVEL-only scoring by a large margin. Bootstrap resampling and a label-permutation control confirm the enrichment is stable and not an artefact of the scoring construction. We present it as a complementary gene-level layer for case-control and gene-set comparisons, with a nonlinear successor outlined as future work.
- 16 July 2026
- Published by MDPI AG in Preprints.org
Open AccessRisk Beyond the Variant: A Composite Metric for Autism Gene Pathogenicity
Variant-based pathogenicity predictors such as REVEL evaluate missense variants in isolation, discarding the gene-length and allele-frequency context needed to compare collections of genes. In this paper, we introduce a composite gene-level metric integrating Hardy-Weinberg heterozygosity, coding-sequence length, and REVEL scores, evaluated on 55 high-confidence autism genes against the 1000 Genomes reference. It identifies elevated pathogenic burden in 48 of 55 genes, removes gene-length and variant-count confounds, and outperforms REVEL-only scoring by a large margin. Bootstrap resampling and a label-permutation control confirm the enrichment is stable and not an artefact of the scoring construction. We present it as a complementary gene-level layer for case-control and gene-set comparisons, with a nonlinear successor outlined as future work.
- Open Access
- 20 March 2026
- Published by MDPI AG in Information
- Vol. 17 (3) , 300
Exploring Optical Flow Methods for Automated Fall Detection System
Falls pose severe risks to vulnerable populations, particularly the elderly and individuals with adverse neurological conditions, necessitating reliable and non-obstructive detection systems. While previous multi-modal approaches utilizing video and audio have demonstrated strong performance, they face significant limitations regarding sensitivity to environmental noise. This paper presents a robust, video-only fall detection framework that eliminates reliance on acoustic data to enhance universality. We conduct a comprehensive comparative analysis of five optical flow (OF) algorithms—Horn–Schunck, Lucas–Kanade (LK), LK-Derivative of Gaussian, Farneback, and the spectral method SOFIA—to determine the range of applicability of each technique for capturing fall dynamics. Beyond detection accuracy, we investigate the computational efficiency of each configuration. This optimized, privacy-centric pipeline offers a scalable solution for continuous monitoring in home and clinical settings, addressing the critical need for immediate intervention following high-impact falls.
- 20 March 2026
- Published by MDPI AG in Information
- Vol. 17 (3) , 300
Open AccessExploring Optical Flow Methods for Automated Fall Detection System
Falls pose severe risks to vulnerable populations, particularly the elderly and individuals with adverse neurological conditions, necessitating reliable and non-obstructive detection systems. While previous multi-modal approaches utilizing video and audio have demonstrated strong performance, they face significant limitations regarding sensitivity to environmental noise. This paper presents a robust, video-only fall detection framework that eliminates reliance on acoustic data to enhance universality. We conduct a comprehensive comparative analysis of five optical flow (OF) algorithms—Horn–Schunck, Lucas–Kanade (LK), LK-Derivative of Gaussian, Farneback, and the spectral method SOFIA—to determine the range of applicability of each technique for capturing fall dynamics. Beyond detection accuracy, we investigate the computational efficiency of each configuration. This optimized, privacy-centric pipeline offers a scalable solution for continuous monitoring in home and clinical settings, addressing the critical need for immediate intervention following high-impact falls.
- Open Access
- 26 February 2026
- Published by MDPI AG in Preprints.org
Exploring Optical Flow Methods for Automated Fall Detection System
Falls pose severe risks to vulnerable populations, particularly the elderly and individuals with adverse neurological conditions, necessitating reliable and non-obstructive detection systems. While previous multimodal approaches utilising video and audio have demonstrated strong performance, they face significant limitations regarding sensitivity to environmental noise. This paper presents a robust, video-only fall detection framework that eliminates reliance on acoustic data to enhance universality. We conduct a comprehensive comparative analysis of five Optical Flow (OF) algorithms—Horn-Schunck, Lucas-Kanade (LK), LK-Derivative of Gaussian, Farneback, and the spectral method SOFIA—to determine the range of applicability of each technique for capturing fall dynamics. Beyond detection accuracy, we investigate the computational efficiency of each configuration. This optimised, privacy-centric pipeline offers a scalable solution for continuous monitoring in home and clinical settings, addressing the critical need for immediate intervention following high-impact falls.
- 26 February 2026
- Published by MDPI AG in Preprints.org
Open AccessExploring Optical Flow Methods for Automated Fall Detection System
Falls pose severe risks to vulnerable populations, particularly the elderly and individuals with adverse neurological conditions, necessitating reliable and non-obstructive detection systems. While previous multimodal approaches utilising video and audio have demonstrated strong performance, they face significant limitations regarding sensitivity to environmental noise. This paper presents a robust, video-only fall detection framework that eliminates reliance on acoustic data to enhance universality. We conduct a comprehensive comparative analysis of five Optical Flow (OF) algorithms—Horn-Schunck, Lucas-Kanade (LK), LK-Derivative of Gaussian, Farneback, and the spectral method SOFIA—to determine the range of applicability of each technique for capturing fall dynamics. Beyond detection accuracy, we investigate the computational efficiency of each configuration. This optimised, privacy-centric pipeline offers a scalable solution for continuous monitoring in home and clinical settings, addressing the critical need for immediate intervention following high-impact falls.
- 1 January 2026
- Published by Institute for Bulgarian Language Prof. Lyubomir Andreychin at the Bulgarian Academy of Sciences
- p. 366-375
Syntactic Patterns in Political Speech: Dependency-based Ad Hominem Detection in Bulgarian Social Media
This paper explores the automatic detection of ad hominem attacks in Bulgarian political social media discourse by leveraging dependency-based structural features. Based on a linguistic analysis of ad hominem constructions, we derive structural patterns using Universal Dependencies relations. Grounded in a corpus of Bulgarian political discourse annotated for logical fallacies, we compare three classification settings: structural features alone, lexical features based on TF–IDF representations, and a combined model integrating both. Experiments with logistic regression, decision trees, and support vector machines show that dependency-based structural features achieve competitive detection performance and provide complementary signal to lexical TF–IDF baselines, with performance estimates sensitive to evaluation protocol in this small-data setting. Overall, ad hominem discourse exhibits identifiable syntactic configurations that can complement lexical cues in its automatic detection. The study demonstrates the value of linguistically informed structural features for analyzing political rhetoric and contributes computational methods for studying argumentative strategies in a low-resource language such as Bulgarian.
- 1 January 2026
- Published by Institute for Bulgarian Language Prof. Lyubomir Andreychin at the Bulgarian Academy of Sciences
- p. 366-375
Syntactic Patterns in Political Speech: Dependency-based Ad Hominem Detection in Bulgarian Social Media
This paper explores the automatic detection of ad hominem attacks in Bulgarian political social media discourse by leveraging dependency-based structural features. Based on a linguistic analysis of ad hominem constructions, we derive structural patterns using Universal Dependencies relations. Grounded in a corpus of Bulgarian political discourse annotated for logical fallacies, we compare three classification settings: structural features alone, lexical features based on TF–IDF representations, and a combined model integrating both. Experiments with logistic regression, decision trees, and support vector machines show that dependency-based structural features achieve competitive detection performance and provide complementary signal to lexical TF–IDF baselines, with performance estimates sensitive to evaluation protocol in this small-data setting. Overall, ad hominem discourse exhibits identifiable syntactic configurations that can complement lexical cues in its automatic detection. The study demonstrates the value of linguistically informed structural features for analyzing political rhetoric and contributes computational methods for studying argumentative strategies in a low-resource language such as Bulgarian.
- 1 January 2026
- Published by Institute for Bulgarian Language Prof. Lyubomir Andreychin at the Bulgarian Academy of Sciences
- p. 309-322
Optimizing LLM-based Discourse Analysis in Low-Resource Languages: A Case Study of Bulgarian Euro Adoption Narratives
This study explores the “complexity ceiling” of LLMs in analyzing culturally specific Euro adoption narratives in Bulgarian. Navigating this high-subtext content requires pipelines that balance abstract reasoning with rigorous stability. Using the IWH and FABLE methodologies, which comprise a total of 11 dimensions scored {0, 1, 2}, we evaluated model performance on a dataset sampled for diversity via vector embeddings. We operationalize a complexity ceiling as the highest-complexity configuration within a bounded search space that satisfies a predefined stability criterion. The ceiling was identified via “descent to feasibility”: beginning from the highest complexity parameters within the predefined search space, and iteratively reducing internal and external parameters until the stability criterion was met. Stability was measured using the “80/80 rule,” requiring at least 80% of samples to maintain modal consistency in 8 out of 10 independent iterations, across all dimensions. Results indicate that pipelines capable of both stability and higher-order reasoning are achievable through calibrated trade-offs between internal and external complexity. The leading configuration met the 80/80 criteria for 10 out of 11 dimensions, suggesting that stable multi-dimensional discourse analysis is achievable under carefully calibrated complexity conditions, even in low-resource settings.
- 1 January 2026
- Published by Institute for Bulgarian Language Prof. Lyubomir Andreychin at the Bulgarian Academy of Sciences
- p. 309-322
Optimizing LLM-based Discourse Analysis in Low-Resource Languages: A Case Study of Bulgarian Euro Adoption Narratives
This study explores the “complexity ceiling” of LLMs in analyzing culturally specific Euro adoption narratives in Bulgarian. Navigating this high-subtext content requires pipelines that balance abstract reasoning with rigorous stability. Using the IWH and FABLE methodologies, which comprise a total of 11 dimensions scored {0, 1, 2}, we evaluated model performance on a dataset sampled for diversity via vector embeddings. We operationalize a complexity ceiling as the highest-complexity configuration within a bounded search space that satisfies a predefined stability criterion. The ceiling was identified via “descent to feasibility”: beginning from the highest complexity parameters within the predefined search space, and iteratively reducing internal and external parameters until the stability criterion was met. Stability was measured using the “80/80 rule,” requiring at least 80% of samples to maintain modal consistency in 8 out of 10 independent iterations, across all dimensions. Results indicate that pipelines capable of both stability and higher-order reasoning are achievable through calibrated trade-offs between internal and external complexity. The leading configuration met the 80/80 criteria for 10 out of 11 dimensions, suggesting that stable multi-dimensional discourse analysis is achievable under carefully calibrated complexity conditions, even in low-resource settings.