Претрага

Скуп објеката Михаило Шкорић

32 items

  • It-Sr-NER: CLARIN Compatible NER and Geoparsing Web Services for Italian and Serbian Parallel Text

    Olja Perišić, Ranka Stanković, Milica Ikonić Nešić, Mihailo Škorić (2023)

  • It-Sr-NER: Web Services for Recognizing and Linking Named Entities in Text and Displaying Them on a Web Map

    Olja Perišić, Ranka Stanković, Milica Ikonić Nešić, Mihailo Škorić (2023)
    The paper will present the results of the project `“It-Sr-NER: Web services for named entities recognition, linking and mapping,” in which teams from the University of Turin and the Society for Language Resources and Technologies JeRTeh participated, and whose goal was the development of the It-Sr-NER web service for named entity annotations in the text and displaying them on the map. Named entities in these services are names of persons, places, organizations, demonyms (ethnicities), events and works of art.
    General Engineering

  • Football terminology: compilation and transformation into OntoLex-Lemon resource

    Jelena Lazarević, Ranka Stanković, Mihailo Škorić, Biljana Rujević (2023)
    У овом раду представља се пројекат који је у развоју, креирање првог дигиталног фудбалског речника на српском језику, као и да демонстрација примене модела OntoLex и љегових модула. OntoLex-FrAC модул укључује информације о учесталости и примерима употребе екстрахованих из корпуса. У овом случају, креиран је корпус за специфичан домен под називом СрФудКо, који садржи чланке вести о фудбалу на српском језику. Вишечлани термини аутоматски су екстраховани из српског корпуса, а затим ручно евалуирани и класификовани као спортски или ...
    повезани отворени подаци, корпус, СрФудКо, OntoLex, OntoLex-FrAC

  • Transformer-Based Composite Language Models for Text Evaluation and Classification

    Mihailo Škorić, Miloš Utvić, Ranka Stanković (2023)
    Parallel natural language processing systems were previously successfully tested on the tasks of part-of-speech tagging and authorship attribution through mini-language modeling, for which they achieved significantly better results than independent methods in the cases of seven European languages. The aim of this paper is to present the advantages of using composite language models in the processing and evaluation of texts written in arbitrary highly inflective and morphology-rich natural language, particularly Serbian. A perplexity-based dataset, the main asset for the ...
    General Mathematics, Engineering (miscellaneous), Computer Science (miscellaneous)

  • An Italian-Serbian Sentence Aligned Parallel Literary Corpus

    Saša Moderc, Ranka Stanković, Aleksandra Tomašević, Mihailo Škorić (2023)
    This article presents the construction and relevance of an Italian-Serbian sentence-aligned parallel corpus, delving into the aligned sentences in order to facilitate effective translation between the two languages. The parallel corpus serves as a valuable resource for language experts, researchers, and language enthusiasts, fostering a deeper understanding of linguistic nuances and cultural expressions. By bridging the gap between Serbian and Italian, this corpus opens new avenues for cross-cultural communication and collaboration, and ultimately contributes to the improvement of language-related ...
    Aligned corpus, parallel corpus, Serbian, Italian, literature

  • Towards Semantic Interoperability: Parallel Corpora as Linked Data Incorporating Named Entity Linking

    Ranka Stanković, Milica Ikonić Nešić, Olja Perisic, Mihailo Škorić, Olivera Kitanović (2024)
    U radu se prikazuju rezultati istraživanja vezanih za pripremu paralelnih korpusa, fokusirajući se na transformaciju u RDF grafove koristeći NLP Interchange Format (NIF) za lingvističku anotaciju. Pružamo pregled paralelnog korpusa koji je korišćen u ovom studijskom slučaju, kao i proces označavanja delova govora, lematizacije i prepoznavanja imenovanih entiteta (NER). Zatim opisujemo povezivanje imenovanih entiteta (NEL), konverziju podataka u RDF, i uključivanje NIF anotacija. Proizvedene NIF datoteke su evaluirane kroz istraživanje triplestore-a korišćenjem SPARQL upita. Na kraju, razmatra se povezivanje Linked ...
    paralelni korpusi, povezivanje imenovanih entiteta, prepoznavanje imenovanih entiteta, NER, NEL, povezani podaci, NIF, Vikipodaci

  • Data from the Digital Repository of the Faculty of Mining and Geology in eScience (eNauka)

    Biljana Rujević, Mihailo Škorić (2024)
    The paper describes linking the Digital Repository of the University of Belgrade, Faculty of Mining and Geology, with the eScience system in terms of transferring metadata about the results of researchers' scientific work. The steps taken to ensure a smooth harvesting of metadata are outlined. Additionally, a presentation of additional improvements to the OAI system is provided, aiming to contribute to the automatic linking of authors with their results in the eScience system.
    репозиторијуми, еНаука, метаподаци, Рударско-геолошки факултет

  • Језички модели, шта је то?

    Ми­ха­и­ло Шко­рић (2023)
    Језички модели

  • Advancing Sentiment Analysis in Serbian Literature: A Zero and Few-Shot Learning Approach Using the Mistral Model

    Milica Ikonić Nešić, Saša Petalinkar, Mihailo Škorić, Ranka Stanković, Biljana Rujević (2024)
    Ova studija predstavlja analizu sentimenta srpskih starih romana iz perioda 1840-1920, koristeći veliki jezički model (LLM) Mistral za tehniku učenja sa zasnovani na takozvanim "zero" i "few-shot" pokušajima. Glavni pristup uvodi inovacije osmišljavanjem istraživačkih upita (promptova) uključuju tekst sa uputstvom za klasifikaciju bez primera i na osnovu nekoliko primera, omogućavajući jezičkom modelu da klasifikuje osećanja u pozitivne, negativne ili objektivne kategorije. Ova metodologija ima za cilj da pojednostavi analizu osećanja ograničavanjem odgovora, čime se povećava preciznost ...
    zero-shot, few-shot, sentiment, Serbian, Mistral model

  • New Language Models for South Slavic Languages

    Mihailo Škorić (2024)
    Izlaganje će predstaviti izazove i perspektive modelovanja južnoslovenskih jezika, sa posebnim osvrtom opšte jezičke modele građene na arhitekturi transformera (BERT, GPT), na dostupne skupove tekstova za obučavanje tih modela, te kvantitet i kvalitet tih skupova. Izlaganje će ponuditi pregled dostupnih skupova i modela, dok će posebna pažnja biti posvećena najnovijim korpusima tekstova. Prvi korpus, Kišobran, predstavlja krovni veb korpus južnoslovenskih jezika i ujedno trenutno najveći korpus tekstova na našim prostorima koji broji preko osamnaest milijardi reči i uključuje sve ...
    veliki korpusi teksta, jezički modeli, južnoslovenski jezici

  • BERT Downstream Task Analysis: Named Entity Recognition in Serbian

    Milica Ikonić Nešić, Saša Petalinkar, Mihailo Škorić, Ranka Stanković (2024)
    This paper compares different architectures and techniques for preparing named entity recognition (NER) models for the Serbian language via integrating BERT with spaCy. Models were trained to recognize seven different named entity types (persons, locations, organisations, professions, events, demonyms, and artworks), and are trained on the dataset containing Serbian novels published between 1840 and 1920, publicly available newspaper articles and sentences generated from the Wikidata knowledge base and Leximirka lexical database. We explore various configurations and several training pipelines ...

  • Semantic Textual Similarity of Courses Based on Text Embeddings

    Olivera Kitanović, Aleksandra Tomašević, Mihailo Škorić, Ranka Stanković, Ljiljana Kolonja (2024)
    This paper explores the application of textual embeddings to measure semantic similarity between educational courses’ curriculums, aiming to enhance the effectiveness of the next faculty accreditation. Leveraging state-of-the-art natural language processing techniques, we employ pre-trained embeddings to capture the semantic meaning of course descriptions. Our methodology involves transforming course curriculum texts into high-dimensional vector representations, enabling efficient and meaningful comparisons. We evaluate the proposed approach on a diverse dataset of course descriptions, employing established benchmarks for semantic textual similarity ...