Enriquecimiento del conocimiento previo en ILP
Inductive Logic Programming (ILP) induces concepts from a set of negative examples, a set of positive examples, and background knowledge. ILP has been applied on tasks in areas such as natural language processing, finite element mesh design, network mining, robotics, drug discovery, and more. These...
| Author: | |
|---|---|
| Format: | Doctoral thesis |
| Status: | Published version |
| Publication date: | 2023 |
| Country: | México |
| Institution: | UAM |
| Repository: | Repositorio Institucional de la UAM Iztapalapa |
| Language: | Spanish |
| OAI Identifier: | oai:bindani.izt.uam.mx:zg64tm61p |
| Online Access: | https://doi.org/10.24275/uami.zg64tm61p |
| Access Level: | Open access |
| Keyword: | info:eu-repo/classification/LEM/Discretization (Mathematics) info:eu-repo/classification/LEM/Programación lógica info:eu-repo/classification/LEM/Discretización (Matemáticas) info:eu-repo/classification/LEM/Logic programming info:eu-repo/classification/cti/7 |
| Summary: | Inductive Logic Programming (ILP) induces concepts from a set of negative examples, a set of positive examples, and background knowledge. ILP has been applied on tasks in areas such as natural language processing, finite element mesh design, network mining, robotics, drug discovery, and more. These datasets typically contain both numerical and categorical attributes; however, few relational learning systems efficiently handle such data. This thesis introduces an evolutionary method called ”Grouping and Discretization for Enriching the Background Knowledge (GDEBaK),”which enables the handling of numerical and categorical attributes. This method employs evolutionary operators to create and test different split points (for numerical attributes) and subsets of values (for categorical attributes) based on a fitness function. Subsequently, the best split points and category subsets are added to the background knowledge before the learning process, to be used during the induction of the final theory. We implemented GDEBaK embedded in the Aleph system and compared it with Aleph’s lazy discretization and the discretization performed by the Top-down Induction of Logical Decision Trees (TILDE) system [6]. Aleph is one of the most widely used ILP systems for learning concepts that require high representational power [55]. It is crucial in ILP, incorporating functionalities from other systems such as Progol, FOIL, FORS, or TILDE. The obtained results indicate that the presented method improves the accuracy of final theories and reduces the number of rules in the majority of cases. |
|---|