2019/04/12 by Ismini Lourentzou, Kabir Manghnani, Lourentzou, Ismini +4 · 2 citations
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Hate Speech and Cyberbullying Detection #Machine Learning (cs.LG) #Natural Language Processing Techniques #Topic Modeling
paper · pdf · doi:10.48550/arxiv.1904.06100
openalex publication_date 2019/04/12 · openalex created_date 2022/07/29 · openalex updated_date 2026/07/28
Social media offer an abundant source of valuable raw data, however informal\nwriting can quickly become a bottleneck for many natural language processing\n(NLP) tasks. Off-the-shelf tools are usually trained on formal text and cannot\nexplicitly handle noise found in short online posts. Moreover, the variety of\nfrequently occurring linguistic variations presents several challenges, even\nfor humans who might not be able to comprehend the meaning of such posts,\nespecially when they contain slang and abbreviations. Text Normalization aims\nto transform online user-generated text to a canonical form. Current text\nnormalization systems rely on string or phonetic similarity and classification\nmodels that work on a local fashion. We argue that processing contextual\ninformation is crucial for this task and introduce a social media text\nnormalization hybrid word-character attention-based encoder-decoder model that\ncan serve as a pre-processing step for NLP applications to adapt to noisy text\nin social media. Our character-based component is trained on synthetic\nadversarial examples that are designed to capture errors commonly found in\nonline user-generated text. Experiments show that our model surpasses neural\narchitectures designed for text normalization and achieves comparable\nperformance with state-of-the-art related work.\n