ميلاد محمديان

Named Entity Linking (NEL) is a fundamental task in Natural Language Processing (NLP) that aims to identify named entities within a text and link them to their corresponding entities in a knowledge graph such as Wikidata. This task is particularly challenging in Persian due to issues such as ambiguous, incomplete, and unlinkable entities, as well as the linguistic complexities of the language. Additionally, implementing an efficient NEL system for real-time inference in production environments remains a significant challenge. This research presents a comprehensive and efficient framework for addressing NEL in Persian, consisting of three main stages: named entity recognition, candidate entity retrieva‎l from the knowledge graph, and disambiguation. To eva‎luate the model, a dataset was constructed and annotated from diverse Persian sources, including news articles, literary texts, and conversational data. This dataset includes precise, ambiguous, incomplete, and unlinkable entities, covering various linguistic challenges in Persian. Experimental results demonstrated that the proposed framework achieved an F1 score of 89.06%. Moreover, its low inference time makes it suitable for deployment in production environments. A comparative analysis with existing methods showed that the proposed framework outperforms others in handling Persian-specific challenges, particularly ambiguous and incomplete entities. By designing a comprehensive framework and developing a standardized dataset, this research contributes significantly to advancing Persian NLP tools. The findings of this study can be applied to various applications, such as question-answering systems and Persian text analysis.

پيونددهي موجوديت , شناسايي موجوديت‌هاي نام‌دار , ابهام‌زدايي موجوديت , گراف دانش , ويكي‌ديتا

Entity Linking , Named Entity Recognition , Entity Disambiguation , Knowledge Graph , Wikidata

Milad Mohammadian

Behrouz Minaei Bidgoli