Silhouette in Citespace clusters

[Glib Voloskyi]

One of the main property of a cluster in Citespace is a silhouette, which is defined as the homogeneity of a cluster. As far as I understand, the higher the silhouette score, the more consistent the cluster members are. But in what terms this 'consistency' is measured? I would much appreciate the clarification.

[Chaomei Chen]

The consistency is essentially measured as how members in the same cluster are more similar to members outside the cluster.

[Stephan De Spiegeleire]

Thanks Chaomei . But so what does that similarity REALLY mean in, for instance, a co-citation network? Or maybe easier yet - a co-term network? Explained in understandable terms that preferably do not include cosine distance :)...
And this is a more general comment too. I think more and more people in various disciplines would be willing to start using this broader epistemic exploration of the field they work in using tools like CiteSpace. I also suspect that, like myself, they' be willing to trust bibliometric experts on the math. But at least a better understanding of the underlying intuition behind things like this would, I expect, help the uptake of these tools.

[Chaomei Chen]

I am adding a glossary to explain concepts such as this without using too much of jargon.
https://citespace.podia.com/glossary-cocitations

[Nia]

尊敬的陈教授：
您好！在使用CiteSpace的聚类功能时，我遇到如下问题，因经过大量查阅您的论文和答疑帖还是未能解决，所以发此询问，希望能得到您的回复，十分感谢！

1. 聚类完成的网络图色块填充后（如图1），有两个聚类没有内部连接线、没有轮廓也没有填充颜色（#10和#14），但是Silhouette分别为0.968和0.994，请问这样的聚类可以保留吗？
   

2. 得到的聚类结果，有两个类Silhouette=1（如图2，对应聚类图见图1），我看您和李杰老师的讲解都是要接近1最好，没有具体说等于1的情况，且我看的很多论文得到的值也都没有等于1。请问这样的聚类结果是否可靠呢？Silhouette等于1的聚类结果可以保留吗？
   

3. 关于使用Keyword和Term，进行研究热点、研究趋势、知识结构分析时，我看了您的很多答疑，还是有些疑惑：因为Keyword和Term的结果会有些类似，所以进行热点分析时，是比较这两种的结果，再做出最优选择吗？还是可以同时进行分析？如果需要对这两个的结果进行选择，那哪一个更可靠呢，是要结合实际情况分析吗？

最后，再次感谢您抽出的宝贵时间！盼能尽快收到您的回复！

祝您身体健康，万事顺遂！

[Chaomei Chen]

用Cluster Explorer查看相应的聚类里的Citing Articles是否只有少数几个或只有一个。如果是的话，应避免过多解读。