One of the most popular methods for link prediction problem is similarity-based approaches. Do these similarity-based approaches fall into the category of supervised or unsupervised approaches? In some papers, similarity-based approaches are referred to as supervised methods while in others as unsupervised. could you help me to solve this challenge?
https://www.nature.com/articles/s41598-017-17157-9
https://arxiv.org/pdf/1304.6257.pdf;
Similarity based-methods for link prediction consist in defining a similarity function to measure the probability of link existence between any pair of nodes based on numerous information such as structural and content information. This technique is well suited to both supervised and unsupervised classification.
In supervised similarity based-methods a training set of "labelled" edges is given, while in unsupervised similarity based-methods no such training set is available.
After study some papers, I concluded that similarity measures like the common neighbor, path and community structure are just scoring functions not link prediction approaches or algorithms.
When we consider the link prediction problem as a classification problem, we are using a supervised strategy, otherwise, we are using unsupervised one.
- Badges
- Science topic