A Comprehensive Survey of Cutting-Edge Fine-Grained Analysis Methods in Text Classification Tasks

With the arrival of the big data era, the volume of text data on the internet is experiencing explosive growth. As a key technology in natural language processing, text classification is widely used in many scenarios. Text classification tasks are divided into two major categories, multi-classification and binary classification, based on practical needs. Binary classification focuses on binary judgment of texts, meaning dividing them into two categories to determine if a text is “positive” or “negative”. This is usually represented by 0 and 1, such as distinguishing between “good reviews” and “bad reviews” in internet comment sections. Binary classification offers the advantages of simple models and efficient computation, providing quick binary guidance for decision-making. On the other hand, multi-classification is designed for complex situations, involving dividing texts into more than two categories with independent and non-overlapping labels. For example, news articles can be categorized into “sports events”, “economic insights” and “technological frontiers”. While multi-classification enhances the management of information, the diversity of categories also poses a rigorous challenge to the model's comprehensive judgment ability. This paper comprehensively and deeply reviews the fine grained analysis methods for multi - classification and binary classification in text classification tasks. By carefully organizing and categorizing research methods for different granularity classification results, it presents the cutting - edge achievements and innovative breakthroughs in both classifications. Crucially, it delves into the existing problems of current methods in handling complex text semantics, addressing data imbalance, and weak generalization abilities. Based on these pain points, it looks to the future and discusses how to advance text classification technology through new approaches like deep learning architecture optimization, cross - domain knowledge fusion, and reinforcement learning expansion. This aims to revitalize the natural language processing field and broaden its application.