Association link network based semantic coherence measurement for short texts of web events

Rinton Press
Publication Type:
Journal Article
Journal of Web Engineering, 2017, 16, (1-2), pp. 39-62
Issue Date:
Filename Description Size
RP_Journal_1540-9589_1612.pdf839.44 kB
Adobe PDF
Full metadata record
As novel web social Media emerges on the web, large-scale short texts are springing up. Although these massive short texts contain rich information, their disorder nature makes users difficult to obtain the desired knowledge from them, especially the semantic coherent knowledge. Different orders of these short texts often express different semantic coherence states. Therefore, how to automatically measure semantic coherence of short texts is a fundamental and significant problem for web knowledge services. Existing related works on the semantic coherence measurement of different orders of short texts/sentences seldom focus on graph structure of semantic link network for reecting coherence change, measuring coherence by these graph-based features and discovering some interesting coherence patterns. In this paper, we propose an association link network based semantic coherence measurement for short texts of web events. Our method firstly construct an association link network from which some graph-based features are then extracted to measure semantic coherence of different orders and lastly some coherence patterns are discovered for guiding automatically text ordering/generation. To validate correctness of our method, we conduct a series of experiments including sentence order permutation, sentence removal and adding/replacing sentence and compare with other two methods. The results show that our method can measure semantic coherence with higher accuracy and outperforms other methods in some experiments. Such method can be widely applied in web text automatic generation, web short text organization and web event summarization etc.
Please use this identifier to cite or link to this item: