CAJ | 학술논문

针对现有知识粒度研究中大多研究信息聚类而忽略增量更新的问题，提出了一种基于相似度的信息颗粒更新方法。通过计算新文本特征权向量与信息颗粒质心的相似度，将新文本添加到相似度最高且高于阈值的信息颗粒下，然后更新信息颗粒的质心，计算颗粒间的相似度，根据结果进行泛化、剪枝、细化和拆分等操作，完成信息颗粒的更新。将其与层次聚类法进行对比实验，结果表明，信息颗粒更新方法在保证性能的同时提升了效率。
침대현유지식립도연구중대다연구신식취류이홀략증량경신적문제，제출료일충기우상사도적신식과립경신방법。통과계산신문본특정권향량여신식과립질심적상사도，장신문본첨가도상사도최고차고우역치적신식과립하，연후경신신식과립적질심，계산과립간적상사도，근거결과진행범화、전지、세화화탁분등조작，완성신식과립적경신。장기여층차취류법진행대비실험，결과표명，신식과립경신방법재보증성능적동시제승료효솔。
The existed studies of knowledge granularity mostly focus on information clustering while ignore studying incremen -tal updating .In order to solve this problem , this paper proposed an information granules updating method based on similarity . This method added a new text to an existed information granule with the highest similarity which is higher than the threshold value by comparing similarities between features ’ weight vectors of the new text and the centroids of information granules .Then the up-dating of information granules was accomplished by relocating their centroids .This step included generalization , elimination , splitting and refinement based on the similarities between the granules .The comparative experiment with level clustering demon-strates that the new method can guarantee the performance as well as improve the efficiency .