A node-similarity based algorithm for tree generation and evolution
Wenjun Zhang · DOAJ (DOAJ: Directory of Open Access Journals) · 2016
In present study we proposed a node-similarity based algorithm for tree generation and evolution. In this algorithm, we assume that each isolated node is a node set at the beginning, two node sets with the greatest similarity tend to connect into a new node set firstly. Repeat this procedure, until all isolated nodes are combined into a tree. Pearson correlation measure, cosine measure, and (negative) Euclidean distance measure (the three measures are for interval attributes), contingency correlation measure (for nominal attributes), or Jaccard coefficient measure (for binary attributes) were used as the between-node similarity. In this way, all connections are sequentially generated and it thus forms the evolution process of a spanning tree of maximum likelihood. The similarity value of a connection can be considered as the weight of the connection. Matlab codes of the algorithm are provided.