2021/04/14 by Rohan Kumar Yadav, Yadav, Rohan Kumar, Lei Jiao +5 · 2 citations
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Multimodal Machine Learning Applications #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2104.06901
openalex publication_date 2021/04/14 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Tsetlin Machine (TM) is an interpretable pattern recognition algorithm based\non propositional logic, which has demonstrated competitive performance in many\nNatural Language Processing (NLP) tasks, including sentiment analysis, text\nclassification, and Word Sense Disambiguation. To obtain human-level\ninterpretability, legacy TM employs Boolean input features such as bag-of-words\n(BOW). However, the BOW representation makes it difficult to use any\npre-trained information, for instance, word2vec and GloVe word representations.\nThis restriction has constrained the performance of TM compared to deep neural\nnetworks (DNNs) in NLP. To reduce the performance gap, in this paper, we\npropose a novel way of using pre-trained word representations for TM. The\napproach significantly enhances the performance and interpretability of TM. We\nachieve this by extracting semantically related words from pre-trained word\nrepresentations as input features to the TM. Our experiments show that the\naccuracy of the proposed approach is significantly higher than the previous\nBOW-based TM, reaching the level of DNN-based models.\n