Cross-linguistic news topic discovery holds significant application value in natural language processing, spanning tasks such as information retrieval, international news communication, and global event influence analysis. However, existing methods face key challenges is the scarcity of parallel corpora and ignoring the propagation of associations throughout event networks.
To solve the problems, a research team led by Zhengtao YU published their new research on 15 September 2026 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed a Chinese-Vietnamese bilingual news topic discovery method based on association graph model. This method can well characterize the complex structure and relationship between events by using graph model, and can take into account the news events of multiple elements of information. Empirical results indicate that method substantially outperforms baseline models in low resource language. Additionally, the team demonstrated that association graph can weakening the impact of language differences on clustering results, can be a good way to describe the same news events in different languages together.
In their research, they use an association graph to represent the complex structure and relationships between events. This approach can weaken the effect of language differences on clustering results and enable more comprehensive and accurate extraction of news texts describing the same event.
Specifically, a Chinese-Vietnamese bilingual association graph is constructed using the features of element association, entity co-occurrence, and text content similarity between related news reports. The Chinese and Vietnamese texts are first clustered roughly using the Affinity Propagation (AP) algorithm, after which the clustering results are adjusted by dynamically updating the weights based on association strengths. News in both languages is used to supervise the clustering process, thereby weakening the effect of linguistic differences on the results. This method provides an effective way to group descriptions of the same news events across different languages. Finally, optimal local and global Chinese-Vietnamese bilingual news clusters are obtained, achieving automated clustering. The experimental results show that the proposed approach has obvious advantages in cross-language topic discovery tasks, particularly in low-resource language scenarios.
Future work can focus on on topic evolution and tracking on the basis of static topics, analyze the formation of a complete topic chain from beginning to peak and ebb, and construct a skeleton for topic development.
DOI:10.1007/s11704-025-41049-1