CAJ | 학술논문

由于在某一次合并后不能改变对象所属类，致使分层聚类算法聚类质量差。为了提高聚类质量，提出一种新的分层聚类机制---基于启发式的分层聚类。与现有的分层聚类不同，基于启发式的分层聚类首先利用现有分层聚类算法构建初始聚类树，然后通过一定的策略改变当前聚类树以使得某个目标函数最小。借鉴计算分子生物学中的最小进化原理，将目标函数定义为聚类树的树长。最后，通过实验验证了新算法能有效地提高现有分层聚类算法。
유우재모일차합병후불능개변대상소속류，치사분층취류산법취류질량차。위료제고취류질량，제출일충신적분층취류궤제---기우계발식적분층취류。여현유적분층취류불동，기우계발식적분층취류수선이용현유분층취류산법구건초시취류수，연후통과일정적책략개변당전취류수이사득모개목표함수최소。차감계산분자생물학중적최소진화원리，장목표함수정의위취류수적수장。최후，통과실험험증료신산법능유효지제고현유분층취류산법。
Since objects are unable to switch its belonging class after one merge,the hierarchical clustering algorithm behaves poor on clustering quality.In order to improve clustering quality,the paper proposes a new hierarchical clustering mechanism-heuristic hierarchical clustering.Unlike existing hierarchical clustering,heuristic hierarchical clustering first of all makes use of existing hierarchical clustering algo-rithm to construct an initial clustering tree and then change the current clustering tree by some strategy to minimize the objective function.Mo-tivated by the minimum evolution principle in molecular biology,the objective function is defined as the length of clustering tree.Finally,ex-periments verify that the algorithm can efficiently improve existing hierarchical clustering algorithm.