IDE: A Differential Evolution-Based Algorithm for Summarizing Multiple Vietnamese Comments

Abstract

Summarizing multiple comments is essential. However, comments are often short, subjective, grammatically inconsistent, and emotionally charged. Therefore, summarizing them is challenging and complex, especially for resource-poor languages like Vietnamese. Based on Differential Evolution and Word2Vec/PhoBERT, this paper proposes an extractive summarization method for mutiple Vietnamese comments, called Improved Differential Evolution (IDE) algorithm. This algorithm optimizes an objective function that balances content coverage and deduplication by modeling semantic relationships between sentences and documents. Combining Word2Vec/PhoBERT enables the algorithm to capture both lexical and contextual nuances effectively. We compare this proposal with state-of-the-art models ChatGPT and Grok3. Experimental results indicate that the IDE algorithm, especially with PhoBERT embedding, outperforms these two models in terms of summarization metrics. This result demonstrates that IDE algorithm can generate good summaries for low-resource languages.