Unsupervised Learning of Relational Patterns.
Abstract
Learning relational patterns without supervision is a challenging and open problem for machine learning, yet a very desirable feature for real world applications. In this paper, we present a new algorithm for unsupervised learning of relational patterns. Given a relational database, this algorithm can find function-free Horn clauses without requiring users to label the data as positive or negative examples, and specify what target concepts to learn. The algorithm is a heuristic search through the relational pattern space. It starts with general patterns derived from a given database schema, and then iteratively generates new promising patterns using the knowledge dynamically collected in the learning process, such as previously learned patterns. To demonstrate the capability of the algorithm, we show some abstracted database examples, and we also show that some of the well-known examples in the relational concept learning literature can be learned without supervision.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1996
- Accession Number
- ADA308923
Entities
People
- Bing Leng
- Wei-min Shen
Organizations
- University of Southern California