Abstract:

Information Extraction (IE) proposes techniques capable of extracting, from unstructured text, relevant segments in a given domain and represent them in a structured format. Most of the scientific proposals in IE so far aim at increasing the accuracy of the extraction results. However, the existing IE techniques still have
efficiency problems when processing large data volumes. IE optimization aims at executing IE processes as fast as possible with minimal or no impact in the accuracy results.

In this talk, we will first describe the state of the art in IE optimization. Then, we will present a novel approach for IE optimization. The key idea is to make IE programs faster by using sub-optimal extraction algorithms, which are are typically fast but
may produce some erroneous results or not produce some of the results of traditional algorithms (thus, leading to a negative impact on the recall and precision values). We propose a cost model that is able to evaluate not only the expected execution time of a given IE execution plan but also the quality of the results produced, in terms of the expected number of good and bad tuples. Using this cost model, our solution is able to choose a fast execution that is able to fulfill a set of objectives imposed by a user (e.g., minimum number of good tuples desired, minimum precision desired). Finally, we will report the preliminary experimental results obtained with two data sets and three IE programs, that show the gains brought by our approach with respect to the state-of-the-art solutions.

Date: 2011-Apr-15     Time: 16:00:00     Room: N7.1

For more information:

• bruno.g.martins@ist.utl.pt