NewMemoryNetwork: A Dynamic Memory Management Framework for Text Classification

He Weiyu

Keywords: Memory networks, Cognitive computing

TL;DR: Experimental results show that our proposed NewMemoryNetwork achieves 98.10% accuracy on text classification tasks, surpassing the traditional TextLSTM by a margin of 65.8 percentage points and the TextTransformer by 3.1 percentage points.

Abstract: This paper presents a novel memory network architecture that achieves intelligent information processing and long-term memory storage through dynamic memory management, novelty assessment mechanisms, and frequency tracking techniques. We conduct experimental validation on a text classification dataset containing 10 categories with 5000 samples. The experimental results demonstrate that our proposed NewMemoryNetwork achieves 98.10% accuracy on text classification tasks, only 1.7 percentage points lower than the state-of-the-art TextCNN under equivalent experimental conditions, while outperforming traditional TextLSTM by 65.8 percentage points and TextTransformer by 3.1 percentage points. The network features dynamic memory management capabilities, validated on text classification tasks, providing a new effective solution for intelligent text processing.

Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)

Submission Number: 16786