MUMBAI, India, July 30 -- Intellectual Property India has published a patent application (202621067992 A) filed by Sayali Prakash Shinde; Vishwakarma Institute Of Technology; Prajkta Dandavate; Adwait Kavishwar; Yash Agiwal; Shripad Borbande; and Sejal Ambekar on May 30, 2026, for Bhaashashodh: Bilingual Plagiarism Detection System Using Transformer-Based Semantic Embeddings.
Inventors include Prajkta Dandavate; Adwait Kavishwar; Yash Agiwal; Shripad Borbande; Sejal Ambekar; and Sayali Shinde.
The application for the patent was published on July 24, 2026, under issue no. 30/2026.
Abstract: Abstract— This study introduces a bilingual context-aware learning approach built upon a Transformer-based SentenceTransformer biencoder architecture to produce semantically consistent sentence embeddings across languages. The proposed framework combines a pre-trained Transformer encoder with a pooling strategy and a fully connected projection layer to model rich contextual representations for both English and Marathi text. To enhance generalization and cross-lingual robustness, the model is fine-tuned using a Multi-Task Learning (MTL) paradigm leveraging the Samanantar parallel corpus alongside the MahaSTS semantic similarity dataset. Training is carried out through a customized interleaved optimization process that jointly minimizes Multiple Negatives Ranking Loss and Cosine Similarity Loss. Performance evaluation on a document-level similarity benchmark indicates that the model successfully learns monolingual meaning, cross-language semantic alignment, and contextual dependencies, attaining a cross-lingual plagiarism detection accuracy of 62.5%. Furthermore, the framework is deployed as a Flask-based web application capable of ingesting and analyzing PDF documents, demonstrating the practicality and effectiveness of the proposed method for multilingual semantic similarity, information retrieval, and cross-lingual comprehension tasks.
Disclaimer: Curated by HT Syndication.
