RULIE, Rule Unification for Learning Information Extraction

Abstract

Today access to information is without a doubt an easy task when compared back to years ago when the web was yet to be created. The easy accessibility the World Wide Web provides is making the immense size of information resources grow out of proportion thus making its retrieval problematic. The resources are organized in such a way that makes their interpretability unmanageable for machines thus making their use ineffectual towards solving this problem. Information Extraction is an area in Artificial Intelligence dedicated to the retrieval of relevant information out of unstructured data giving machines the ability of humans to extract specific information hidden in large amounts of information resources, effectively providing a solution to the information retrieval problem. In this project, we will try to boost the effectiveness of Information Extraction by introducing RULIE an Adaptive Information Extraction algorithm that uses a hybrid technique of Rule Learning and Rule Unification to extract relative information from all types of structured data. In the first part of this project we will discuss in depth the importance and usefulness of Information Extraction such as how it will contribute to the realization of the Semantic Web and other practical day to day applications. We will then shift focus to the concepts and current techniques in Information Extraction while outlining the main inconveniences faced in various rule-based algorithms. In the second part we will review in detail the LP2 and BWI algorithms as their characteristics and mechanisms will be the basis for the realization of RULIE. These algorithms were chosen because their results best those of the other learners, and because of common attributes they share such as their ability to work on all text structures. The benefits gained from one of them should compensate the shortcomings of the other which means that the combination of these two algorithms ought to improve the results obtained. The third part of this project illustrates the methodology used for the creation of RULIE which will be shown through the design and implementation of the main functions in the algorithm. The last part will focus on the evaluation of the system which includes comparing the results obtained with the results of the best Information Extraction algorithm in existence. The outcome of these testing experiments will be analyzed so that indications on how to improve the algorithm further can be obtained for future work.