Abstract:
Natural language processing is a rapidly evolving technology. Conventionalrule-based methods are insufficient to meet today's complex challenges. One significant task in natural language processing is recognizing textual entailment, which enables systems to understand relationships between texts. It is crucial for applications like text summarization, sentiment analysis, information verification, question answering, text classification, and machine translation. One challenge is reducing input vector size while maintaining good prediction results and F1 scores. To address this, this research presents a novel encoding technique by reducing the encoding size for sentences in the com-positional knowledge dataset, which consists of English sentence pairs classified into entailment, neutral, and contradiction. This research proposes the KIRINLog technique, which consists of six methods: preprocessing, removing duplicate words, max-length padding and truncation for sentence alignment, adding a flag value, word embedding by Word2Vec, and dimensionality reduction with PCA. These steps create feature vectors for recognizing textual entailment. This study employs an attention-based bidirectional LSTM as a classification model. The model combines bidirectional LSTM with attention to analyze sequences in both directions and focus on important sentence information. The proposed technique achieves 92.496 accuracy, with 88.8% precision, 88.696 recall, and an F1-score of 88.696 on the 2014 compositional knowledge dataset.