Never Ending Language metaLearning: model management for CMU's ReadTheWeb project

Abstract

The following dissertation has the purpose of characterising the work done last year. It starts with a bilbiographical search to learn about the state-of-the-art approaches in the areas related to the project. Then, the problem was defined, the data collected, experimentation occured and results and conclusion were drawn. In specific this aims to make an introduction to Never Ending Language Learner (better known as NELL), its main goals, as well as its main procedures. During this explanation, some will be exposed. These are the issues that motivate this work, which will investigate better procedures to improve NELL. The second main subject is the possibility to improve the previous results. The subject is Metalearning a technique consisting on extracting information about a particular dataset (described by a set of variables) and, with that, verifying which algorithm is recommended to process new data with similar characteristics. After some theory added for the final semester, information was collected, with information about its format; how data was edited to be better suited for work; which procedures were adopted to get relevant results; the actual results; and the conclusions drew from what was produced during the work.