tech,0-1-N04-4028,bq baseline </term> on all three aspects . <term> Information extraction techniques </term> automatically create <term> structured
other,5-1-N04-4028,bq techniques </term> automatically create <term> structured databases </term> from <term> unstructured data sources
other,8-1-N04-4028,bq <term> structured databases </term> from <term> unstructured data sources </term> , such as the <term> Web </term> or <term>
other,15-1-N04-4028,bq unstructured data sources </term> , such as the <term> Web </term> or <term> newswire documents </term>
other,17-1-N04-4028,bq </term> , such as the <term> Web </term> or <term> newswire documents </term> . Despite the successes of these
measure(ment),7-2-N04-4028,bq Despite the successes of these systems , <term> accuracy </term> will always be imperfect . For many
other,12-3-N04-4028,bq desirable to accurately estimate the <term> confidence </term> the system has in the correctness
other,21-3-N04-4028,bq system has in the correctness of each <term> extracted field </term> . The <term> information extraction
tech,1-4-N04-4028,bq each <term> extracted field </term> . The <term> information extraction system </term> we evaluate is based on a <term> linear-chain
tech,10-4-N04-4028,bq system </term> we evaluate is based on a <term> linear-chain conditional random field ( CRF ) </term> , a <term> probabilistic model </term>
tech,19-4-N04-4028,bq conditional random field ( CRF ) </term> , a <term> probabilistic model </term> which has performed well on <term>
other,26-4-N04-4028,bq </term> which has performed well on <term> information extraction tasks </term> because of its ability to capture
other,38-4-N04-4028,bq to capture arbitrary , overlapping <term> features </term> of the input in a <term> Markov model
model,44-4-N04-4028,bq <term> features </term> of the input in a <term> Markov model </term> . We implement several techniques
other,7-5-N04-4028,bq several techniques to estimate the <term> confidence </term> of both <term> extracted fields </term>
other,10-5-N04-4028,bq the <term> confidence </term> of both <term> extracted fields </term> and entire <term> multi-field records
other,14-5-N04-4028,bq <term> extracted fields </term> and entire <term> multi-field records </term> , obtaining an <term> average precision
measure(ment),19-5-N04-4028,bq multi-field records </term> , obtaining an <term> average precision </term> of 98 % for retrieving correct <term>
other,27-5-N04-4028,bq </term> of 98 % for retrieving correct <term> fields </term> and 87 % for <term> multi-field records
other,32-5-N04-4028,bq correct <term> fields </term> and 87 % for <term> multi-field records </term> . We present a novel approach for
skrij podrobnosti