Loading...
Please wait, while we are loading the content...
Similar Documents
InfoGather+:Semantic Matching and Annotation of Numeric and Time-Varying Attributes in Web Tables
| Content Provider | Microsoft Research |
|---|---|
| Author | Zhang, Meihui Chakrabarti, Kaushik |
| Copyright Year | 2013 |
| Abstract | Users often need to gather information about "entities" ofinterest. Recent efforts try to automate this task by lever-aging the vast corpus of HTML tables; this is referred toas "entity augmentation". The accuracy of entity augmen-tation critically depends on semantic relationships betweenweb tables as well as semantic labels of those tables. Currenttechniques work well for string-valued and static attributesbut perform poorly for numeric and time-varying attributes.In this paper, we fifirst build a semantic graph that (i) la-bels columns with unit, scale and timestamp informationand (ii) computes semantic matches between columns evenwhen the same numeric attribute is expressed in differentunits or scales. Second, we develop a novel entity augmen-tation API suited for numeric and time-varying attributesthat leverages the semantic graph. Building the graph ischallenging as such label information is often missing fromthe column headers. Our key insight is to leverage the wealthof tables on the web and infer label information from se-mantically matching columns of other web tables; this com-plements "local" extraction from column headers. However,this creates an interdependence between labels and seman-tic matches; we address this challenge by representing thetask as a probabilistic graphical model that jointly discov-ers labels and semantic matches over all columns. Our ex-periments on real-life datasets show that (i) our semanticgraph contains higher quality labels and semantic matchesand (ii) entity augmentation based on the above graph hassignificantly higher precision and recall compared with thestate-of-the-art. |
| Language | English |
| Publisher | ACM SIGMOD |
| Publisher Date | 2013-06-01 |
| Access Restriction | Open |
| Rights Holder | Microsoft Corporation |
| Subject Keyword | Search Information retrieval Knowledge management |
| Content Type | Text |
| Resource Type | Proceeding |
