Hybrid Index Structures for Location-based Web Search

1
Hybrid Index Structures for Location-based Web
Search

• Yinghua Zhou, Xing Xie, Chuang Wang, Yuchang
  Gong, Wei-Ying Ma
• CIKM 2005

2
Outline

• Location representation
• A framework of Location-based Web Search Engine
• Hybrid Index
• Experiment results
• Conclusion

3
Location representation

• Textual keywords
• two-dimensional spatial objects

4
Textual keywords

• Textual postal codes, telephone numbers, place
  name, etc.
• Place name easy to extract and detect in web
  content, but it can not describe the detail shape
  of a place

5
Two-dimensional spatial objects

• vector model vs. raster model
• Raster model
• heavily depends on the size of grid cells
• Difficult to balance the storage requirement and
  precision
• Vector model
• Points and polygons or MBR
• MBR only two diagonal points are needed to
  represent location infomation

6
MBR

• Minimum bounding rectangle

7
Location index

• Place names can be organized as flat list or
  hierarchy tree
• Spatial index
• R-tree family, quad-tree, grid files

8
R-tree

• Similar to R-tree, but reinsert entries upon
  overflow, rather than splitting

9
Index combination scheme
10
A framework of Location-based Web Search Engine
11
Ranker

• Contain
• grank(Q,R)R/Q
• Inside
• grank(Q,R)Q/R
• Overlap
• grank(Q,R)(QnR)/(QR- QnR)
• Nearby
• Transform to Overlap

12
Hybrid index

• Inverted file and R-tree double index
• First inverted file then R-tree
• First R-tree then inverted file

13
Inverted file and R-tree double index
14
First inverted file then R-tree
15
First R-tree then inverted file
16
Experiment results
17
Experiment results (cont.)
18
Conclusion