Transcript UIST 2006
1 of 20 Enabling web browsers to augment* web sites’ filter*ing and sort*ing functionalities David Huynh · Rob Miller · David Karger MIT Computer Science & Artificial Intelligence Laboratory UIST 2006 · Montreux, Switzerland 1 of 20 2 of 20 WHERE TO GO NEXT? • Letizia · Lieberman, 1995 • WebWatcher · Joachims, et. al., 1996 • WBI · Barrett, et. al. 1997 2 of 20 3 of 20 WHAT’S THE BIG PICTURE? • WebBook · Card, et. al., 1996 • WebNet · Cockburn, Jones, 2000 3 of 20 4 of 20 DATA TIDBITS • Selection Recognition Agent · Pandit, Kalbag, 1997 • Apple Data Detectors · Nardi, et. al., 1998 • Microsoft Smart Tags • Google AutoLink • Creo · Faaborg, Lieberman, 2006 4 of 20 5 of 20 Internet Scrapbook · Sugiura, Koseki, UIST 1998 5 of 20 Hunter Gatherer · schraefel, et. al., WWW 2002 6 of 20 6 of 20 7 of 20 Piggy-Bank · Huynh, et. al., Int’l Semantic Web Conf 2005 7 of 20 8 of 20 Dontcheva, et. al., UIST 2006 8 of 20 9 of 20 OUTLINE • Motivations • Contributions • interaction design • data extraction algorithm • Evaluations • Future Work 9 of 20 10 of 20 MOTIVATIONS • Internet scrapbook, Hunter Gatherer, Piggy Bank, Mira’s work — all address user needs to “save for later use” • Immediate use • Intervention by the web browser • Unique user needs; web sites not effective • Private user information • Sub-goal of “save for later use” 10 of 20 11 of 20 UNIQUE NEEDS 11 of 20 12 of 20 MOTIVATIONS • Immediate use in-place interaction • Keep context • Leverage web site • (Ultimately, seamless integration of features provided by browser and by web site) 12 of 20 13 of 20 CONTRIBUTIONS • UI design for • Extracting, then • Sorting/filtering structured data in sequences of web pages • Data extraction algorithm • For sequences of web pages • Yielding field values Firefox extension “Sifter”, open source 13 of 20 14 of 20 UI DESIGN – EXTRACTION INTERFACE • Not a user’s goal • Poorly understood • Browser and web server are “Mr. Computer” • It doesn’t let me do this… • WYSI-all-YG • Semantic/structure recovery from HTML • Seemingly unnecessary • Lossy recovery • WYSI-what-the-computer-sees (understands) • Lengthy, error-prone 14 of 20 15 of 20 UI DESIGN – EXTRACTION INTERFACE • Preview of what Sifter will do • Give user better sense of control • Offer a chance to make correction early • Make lengthy wait more acceptable • While keeping the Confirmation steps short 15 of 20 16 of 20 UI DESIGN – AUGMENTATION INTERFACE • Consistency with rest of page • Paging controls, Browsing controls, Sorting controls, Indicators 16 of 20 17 of 20 UI DESIGN – EXTRACTION INTERFACE 17 of 20 18 of 20 UI DESIGN – AUGMENTATION INTERFACE • Possible solutions? • Remove them • Leave them as-is • Take control of them • Dim them out • Focus interaction on items • Suggest disconnection with rest of page • Still support interaction 18 of 20 19 of 20 UI DESIGN – AUGMENTATION INTERFACE 19 of 20 20 of 20 UI DESIGN – AUGMENTATION INTERFACE 20 of 20 21 of 20 DATA EXTRACTION • Item Detection • • • • Items can be precisely addressed by an XPath Each item contains a link On search results pages, items occupy most of the page area No handling of items made up of sibling/cousin nodes • Subsequent Page Detection • Link label heuristic (e.g., “Pages: [1] 2 3 4 … 23”) • URL parameter heuristic • http://.../...?...&idx=10&... • http://.../...?...&idx=20&... • http://.../...?...&idx=30&... • Field Detection • Greedy tree alignment • Yields typed field values 21 of 20 22 of 20 EVALUATIONS • Data Extraction • Collection-level accuracy, not item-level accuracy • 30 collections • 19 perfectly extracted (63% collection-level accuracy) • With some user intervention • (Link to test data and detailed results is in paper) • UI Design • Web augmentation is a novel concept • Mira needed a tutorial in her informal evaluation • Formative evaluation: usable? useful? • Assuming that extraction is perfect 22 of 20 23 of 20 EVALUATIONS – UI DESIGN • Similar to Flamenco study, Yee, et. al., CHI 2003 • Task 1: Structured • Perform specified complex filter/sort operations • 3 cheapest paperbacks (if bought used) by John Grisham in 2005 • Using Sifter • Using only the web site (Amazon) in 5 minutes • High-level instruction build-up • No specific help • Task 1: Unstructured • Judge whether a sale is good • Using Sifter • Using only the web site (Ashford.com) 23 of 20 24 of 20 EVALUATIONS – UI DESIGN • Task 1 • 8/8 successful using Sifter • 5/8 successful using Amazon • Only 1 knew Advanced Search • Unified browsing UI useful • Task 2 • 7/8 successful using Sifter • 6/7 looked at distribution of percent saving 24 of 20 25 of 20 SUMMARY • In-place interaction • Structured data in web pages • Advanced browsing features • Over sequences of pages • Correction UI • More in-place interaction • Interactors 25 of 20