Mongolian Topic Extraction Using Pre-trained Models
Ailiya Wang, Qintu Si, Siriguleng Wang · Advances in Computer and Communication · 2024
The paper introduces a Mongolian topic extraction method that utilizes a pre-trained language model to improve the quality of Mongolian intelligent question answering. Initially, the Mongolian text data undergoes preprocessing steps, including text correction, data cleaning, and word segmentation, to ensure accurate and readable data. Stop words are then removed to reduce noise while filtering high- and low-frequency words to emphasize key terms for constructing a Mongolian thesaurus. After preprocessing, the pre-trained model is used to represent Mongolian word vectors that capture semantic meanings in the language. Based on these representations, an unsupervised topic extraction method employing a topic model identifies and clusters similar topics within the text, providing a structured representation of the data. Experimental results demonstrate that this proposed method outperforms traditional topic extraction methods such as latent Dirichlet allocation and embedded topic models with improvements of 0.3406 and 0.0675 in terms of topic extraction quality respectively, showcasing its effectiveness and efficiency in extracting relevant topics from Mongolian text. This approach enhances text comprehensibility.