We confirm our theoretical consideration of model collective learning by experimenting on new data sets and data sets commonly used in entity analysis.
relational learning
collective learning
With the increasing relevance of relational data and network data in different fields such as social network modeling, semantic web, bioinformatics and artificial intelligence, the importance of relational learning is increasing. This paper concerns the application of tensors in relational learning. Tensors and their decomposition are widely used in psychology or chemometrics.And recently, it has been applied to data mining and machine learning problems, such as modeling time effects in social networks. In relational learning, tension has just emerged to replace more common methods, such as graphical models.
From a modeling point of view, tensors provide simplicity because multiple relationships in any order can be expressed directly as higher-order tensors. In addition, it is not necessary to know or infer a priori knowledge about the structure of the problem from the data, but this is necessary for Bayesian networks or Markov Logical Networks (MLNs).. The reason for applying tensor decomposition from a learning perspective is that relational domains are usually high-dimensional and sparse, and factorization methods show excellent results.
An important characteristic of relational data is that it can generate correlations between multiple interconnection nodes.
These correlations can be captured by including attributes, relationships or classes of related entities in learning tasks.
dyadic relational data（Two relational data) indicates the direct relationship between two objects.
The three direction tensor (three-way tensor X) X (I, K, J) ==1 represents the existence of three tuples (I, J, K).