Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ruhillo Alayev has a total of 18 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Consistency-Constrained Joint Learning of Dependency, Constituency, and Grammatical Roles in Low-Resource Uzbek

    by Botir Boltayevich Elov, Mukhabbat Matyakubovna Kurbanova, Ruhillo Habibovich Alaev, Oqila Xolmo'Minovna Abdullayeva, Shakhnoza Burievna Nashirova, Sayyora Abdinaviyevna Nurmatova

    Complementary syntactic annotations may support learning in low-resource languages, but compatibility penalties require evaluation beyond parameter sharing. We study dependency parsing, constituency parsing, grammatical-role labeling and chunking in Uzbek using a shared XLM-R encoder, structured predictors and empirical implication penalties during training. UzSyn contains 1,123 sentences and 4,955 tokens, partitioned into 898 training, 112 development and 113 test sentences. Comparisons used five seeds, independent models and unconstrained multi-task learning. With all training sentences, the constrained model reached 85.53% labeled attachment score and 76.55% constituency F1. Differences from unconstrained multi-task learning were +0.52 and +0.06 percentage points; paired 97.5% confidence intervals of [−1.71, 2.93] and [−2.08, 2.52] included zero. At the exploratory 50% final-fitting fraction, constituency and chunk F1 increased by 4.04 and 2.97 points, with descriptive 95% intervals excluding zero. Fitting budgets excluded prior annotation access during rule-template design. Coherence changes varied by rule family. Constrained fitting required 3.66 times the mean elapsed fitting time of unconstrained multi-task learning. The findings identify intermediate-budget gains for replication while leaving full-data superiority unresolved. Independent linguistic validation and external evaluation are needed before broader accuracy or annotation-saving claims.

  • Open Access

    An annotated Uzbek social media corpus for emotion and sentiment analysis

    by Botir Elov, Alayev Ruhillo, Zilola Xusainova

    UzEDSA is an annotated corpus for emotion detection and sentiment analysis in Uzbek social media comments. It contains 304,364 unique comments after batch merging, invalid-record removal, and duplicate filtering. Each comment is labelled with a multi-attribute schema covering primary emotion, multi-emotion annotation, sentiment polarity, emotional intensity, sarcasm, and domain. The emotion inventory includes JOY, SAD, ANG, FEA, SUR, TRU, DIS, ANT, MIX, and NEU. The corpus preserves social media features such as spelling variation, Uzbek Latin-Cyrillic script mixing, Uzbek-Russian code-switching, emojis, slang, repeated characters, markup artifacts, short reactions, and long comments. It covers general, personal, news, technology, medical, education, transport, service, and marketing domains. This Data Descriptor documents corpus construction, preprocessing, annotation, deduplication, validation, repository records, and usage notes. UzEDSA supports reusable research on low-resource Uzbek affective NLP, including emotion classification, polarity classification, intensity prediction, sarcasm-aware modelling, domain adaptation, and corpus-based analysis.

  • Open Access

    Multimodal Symptom-to-ICD-11 Group Classification with External Validation Across Emergency Departments

    by Botir Elov, Ruhillo Alayev

    Background: Emergency department (ED) decisions are made under time pressure and uncertainty, yet externally validated symptom-based decision-support systems remain limited, especially for low-resource clinical languages.Objective: To develop and externally validate a multimodal model for classifying ED encounters into ICD-11 disease groups using early-available Uzbek complaint text and structured triage information.Methods: We conducted a retrospective multicenter prediction study of 3,360 ED encounters (2,348 development; 1,012 external validation). Uzbek complaint narratives were processed using a UZ-EDBench-derived symptom extraction and normalization pipeline and fused with structured symptom descriptors, demographics, and intake vital signs. Performance was compared with multinomial logistic regression, gradient boosting, and a text-only transformer. The primary endpoint was macro-F1; secondary endpoints included top-3 accuracy, macro-AUROC, Brier score, expected calibration error (ECE), and subgroup robustness.Results: The Uzbek symptom extraction layer achieved token micro-F1 0.944 and entity exact-match F1 0.861. The multimodal model outperformed all baselines on both internal temporal testing and external validation, reaching macro-F1 0.691 and 0.654, top-3 accuracy 0.901 and 0.868, and macro-AUROC 0.914 and 0.889, respectively. Compared with the best structured baseline, macro-F1 improved by 0.043 internally and 0.056 externally. Temperature scaling reduced ECE from 0.049 to 0.024 internally and from 0.056 to 0.031 externally.Conclusions: Integrating Uzbek complaint text with structured triage variables supported clinically meaningful ICD-11 disease-group stratification and retained performance under cross-site validation. The findings support calibration-aware, language-aware decision support for early stratification and differential prioritization rather than autonomous diagnosis.

  • Open Access

    Aspect-Based Sentiment Analysis of Hospital Service Reviews Using BERTbek and XLM-R

    by Botir Elov, Ruhillo Alayev

    Aspect-based sentiment analysis (ABSA) provides a more fine-grained view of patient experience than document-level sentiment classification, yet low-resource medical settings remain underexplored. This study investigates Uzbek medical and hospital-service feedback using UzMedSentiment as the main benchmark and UZMedNER as an auxiliary medical entity resource. We compare sparse lexical baselines and transformer-based encoders, including mBERT, XLM-R, and BERTbek, under a leakage-safe group-stratified 5-fold cross-validation protocol. In addition to standard pair-based fine-tuning, we propose an Entity-Aware Aspect Fusion framework with two entity-informed variants: sequential medical NER transfer and joint multi-task learning. On the full medical-domain benchmark, the best model, BERTbek-AF-MTL, achieved 82.74 Macro-F1, 85.19 Weighted-F1, and 85.65 Accuracy, outperforming the strongest sparse baseline by 6.79 Macro-F1 points and the strongest plain transformer baseline by 1.83 points. On the hospital-service subset, the same model reached 84.91 Macro-F1, 86.83 Weighted-F1, and 87.19 Accuracy. Ablation results showed cumulative gains from explicit aspect conditioning, aspect-aware fusion, and auxiliary entity supervision. The largest aspect-wise improvements were observed on the rare but managerially important categories waiting time (+7.77) and price (+6.44). Robustness analysis further showed consistent gains on clinically and pragmatically difficult slices, including ADR-positive (+6.45), negation-positive (+6.35), speculation-positive (+4.39), and sarcasm-positive (+5.67) samples. Statistical testing confirmed that the best proposed system significantly outperformed both the strongest sparse baseline and the strongest plain transformer baseline. Overall, the findings show that Uzbek-specific pretraining, when combined with aspect-aware fusion and entity-aware supervision, provides the most effective and robust solution for aspect-based sentiment analysis of Uzbek medical and hospital-service reviews.

  • Open Access

    UZ-EDBench: A Leakage-Aware Benchmark for Uzbek Symptom-to-Specialist Triage from Raw and Normalized Patient Narratives

    by Botir Elov, Ruhillo Alayev

    Objective. To develop and evaluate a leakage-aware benchmark for Uzbek symptom-to-specialist triage from raw and normalized patient-authored health narratives. Materials and methods. UZ-EDBench was constructed from paired raw and cleaned complaint texts with structured triage annotations. After repeated-header removal, malformed-row filtering, ambiguous-label exclusion, semantic deduplication, and canonicalization, the strict benchmark contained 2,480 instances, 24 specialist-routing classes, and 4 urgency levels. Baselines included majority, linear lexical, character-level, and text-plus-safe-metadata models. Routing was evaluated under both a realistic Full24 regime and a more stable Core19 regime. Results. The best Core19 routing model achieved 0.7805 Macro-F1. In the more realistic Full24 setting, the strongest model achieved 0.7210 ± 0.0246 Macro-F1. For urgency prediction, the best leakage-safe model achieved 0.6949 Macro-F1. Leakage analysis showed that target-proximal fields substantially inflated apparent performance, with suspected_system increasing routing Macro-F1 to 0.8666 and severity increasing urgency Macro-F1 to 0.9139. Conclusion. UZ-EDBench is a practical low-resource benchmark for Uzbek clinical NLP and triage modeling. Its main methodological contribution is an explicit separation between official evaluation and leakage-prone upper-bound analysis.

  • Open Access

    Multimodal Symptom-to-ICD-11 Group Classification with External Validation Across Emergency Departments (Preprint)

    by Botir Elov, Ruhillo Alaev

    BACKGROUND Emergency department (ED) decisions are often made under severe time pressure and clinical uncertainty, yet symptom-based decision-support systems with external validation remain limited, particularly for low-resource clinical languages. OBJECTIVE To develop and externally validate a multimodal model for classifying ED encounters into ICD-11 disease groups using early-available Uzbek-language complaint text and structured triage information. METHODS We conducted a retrospective multicenter prediction study of 3,360 ED encounters, including 2,348 encounters in a development cohort and 1,012 in an independent external validation cohort. Uzbek complaint narratives were processed through a UZ-EDBench-derived symptom extraction and normalization pipeline and combined with structured symptom descriptors, demographics, and intake vital signs in a multimodal fusion model. Performance was compared with multinomial logistic regression, gradient boosting, and a text-only transformer. The primary endpoint was macro-F1; secondary endpoints included top-3 accuracy, macro-AUROC, Brier score, expected calibration error (ECE), and subgroup robustness. RESULTS The Uzbek symptom extraction layer achieved a token micro-F1 of 0.944 and an entity exact-match F1 of 0.861. The multimodal model outperformed all baselines on both internal temporal testing and external validation, achieving macro-F1 values of 0.691 and 0.654, top-3 accuracies of 0.901 and 0.868, and macro-AUROC values of 0.914 and 0.889, respectively. Compared with the best structured baseline, macro-F1 improved by 0.043 internally and 0.056 externally. Temperature scaling reduced ECE from 0.049 to 0.024 internally and from 0.056 to 0.031 externally. CONCLUSIONS Integrating Uzbek complaint text with structured triage variables enabled clinically meaningful ICD-11 disease-group stratification and retained performance under cross-site validation. These findings support calibration-aware, language-aware ED decision support for early stratification and differential prioritization rather than autonomous diagnosis.

  • Stages of Creating an Uzbek-English Parallel Corpus and Principles of Selecting a Linguistic Base

    by Elov Botir Boltayevich, Ruhillo Alaev Habibovich, Marufjon Amirkulov Alikulovich, Sabohat Kenjayeva, Jamshid Elov Bekmurodovich

    This paper is a conceptual study that explores the fundamental stages of creating an Uzbek-English parallel corpus, with special emphasis on the linguistic and methodological principles of selecting the base texts. The study identifies and reviews criteria for the inclusion of texts, such as genre diversity, representativeness, alignment accuracy, and linguistic relevance. Particular attention is given to balancing modern and classical texts, as well as to the role of technological tools in achieving consistent sentence-level alignment. The pipeline and recommendations presented in this paper are based on a synthesis of existing research and are proposed as a guideline for corpus developers aiming to construct a reliable and research-oriented bilingual resource.


  • Morphotactic Models and Algorithms of the Uzbek Language

    by Khamroeva Shahlo Mirdjonovna, Elov Botir Boltayevich, Alaev Ruhillo Habibovich, Saydullayeva Dilbar Farhod Qizi

    This article presents a comprehensive study of morphotactic models and algorithms developed for the Uzbek language. Given the agglutinative nature of Uzbek, morphotactics—rules governing the internal structure and sequence of morphemes within a word—play a crucial role in accurate morphological analysis and generation. The paper outlines rule-based and data-driven approaches to modeling Uzbek morphotactics, emphasizing the integration of grammatical constraints and affix ordering principles. It further discusses the implementation of finite-state transducers (FSTs) and machine learning techniques for automatic morphological tagging and generation. The proposed models aim to enhance the performance of natural language processing (NLP) applications such as part-of-speech tagging, lemmatization, and syntactic parsing for Uzbek. Evaluation results demonstrate that the developed algorithms achieve high accuracy in capturing the complex morphological patterns of the language. This research contributes to the advancement of computational resources for low-resourced Turkic languages.


  • Processing of Uzbek Language Texts Using the Skip-Gram Method

    by Elov Botir, Ergasheva Guli, Alaev Ruhillo, Eshmuminov Askar, Hamroyeva Shahlo, Abdullayeva Oqila, Khusainova Zilola