Measuring the Dynamic Relatedness between Chinese Entities Orienting to News Corpus

摘要

The related applications areltmited due to the static characteristics on existing relatedness calculation algorithms. We proposed a method aiming to efficiently compute the dynamic relatedness between Chinese entity-pairs, which changes over time. Our method consists of three components: using cooccurrence statistics method to mine the co-occurrence information of entities from the news texts, inducing the development law of dynamic relatedness between entity-pairs, taking the development law as basis and consulting the existing relatedness measures to design a dynamic relatedness measure algorithm. We evaluate the proposed method on the relatedness value and related entity ranking. Experimental results on a dynamic news corpus covering seven domains show a statistically significant improvement over the classical relatedness measure.