50 likes | 177 Vues
This presentation discusses the Stanford Events Crawler, a system designed to extract and classify event data efficiently. It features advanced techniques such as decision tree classifiers, hand-written rules for field extraction, and normalization processes for date, time, and building names. The system enables extraction at a rate of 10 events per second, utilizing LR wrappers and segmentation for emails. Key fields extracted include title, location, category, contact information, and more. The talk also covers free text search capabilities with Lucene for enhanced data retrieval.
E N D
Stanford Events Crawler Zoe Chu Michael Tung
Architectural Overview Crawler Event? classifier http web nntp newsgroup pop Extractor mailing list backend frontend Event tuple DBMS Presentation layer User Applications Notification
Crawling/Classification • Event pages • Index and detail pages • 10 events/sec
Extraction/Normalization • LR Wrappers • Segmentation for email – decision tree classifier • Hand written rules for field extraction • Date & Time Normalization • Building Normalization • Edit distance against a lexicon of Stanford building names • Free text search - Lucene
Which fields? • Title • Date • Time • Location • Category • Sponsor • Contact info • Admission/fees • Speaker • Food • Description • Building • X,Y physical coordinates • Picture of building • Map • Nearby buildings