Cha Cha A system for Organizing Intranet Search Results Contextualized Hierarchical information Access by Chen, Hearst

1
Cha Cha A system for Organizing Intranet
Search ResultsContextualized Hierarchical
information Access by Chen, Hearst Associates
2
Problems

• Most search Engines present search results as a
  ranked list
• No information about the context in which pages
  exist and their relationship with each other
• Difficult to tell about the relationships between
  the search results
• Query is vague, only a few words heterogeneous
  result set

3
(No Transcript)
4
Prior work

• Super Book
• AMIT
• WebTOC
• WebGlimse
• Connectivity Server
• WebCutter (Mapuccino)

5
(No Transcript)
6
Off-Line Processing
Phase 1 Web crawler
Phase 2 Meta File Generator
WWW
Web Pages
Meta Files
Text
Phase 3 Inverted Index Generator
Inverted Index
7
On-Line Processing
USER
Cha Cha Front end
Query Processor
Graph Generator
Abstract Generator
HTML Generator
Inverted Index
Inverted Index
8
Goals

• Group as many pages together within a sub
  hierarchy as possible avoid branches that
  terminate into only one leaf
• Remove as many internal nodes as possible while
  retaining at least one path to every leaf

9
Algorithm

• Form a graph
• Top down First pass to set the number of children
  and depth of each node
• Bottom up pass. For each level from the bottom to
  the root
• Find all the non leaf node at this level
• Sort them according to the active number of
  children
• For each node in this level
• i.      See if it has at least one active
  children which does not have another parent.
•        ii.    If no eliminate this node
• Sort the leaf nodes in ascending order according
  to their rank in the search result.
• Start with an empty tree and first hit, a path is
  found from that hit through the active nodes at
  each level above it to the root. Whenever there
  is a choice of parent choose the parent with most
  number of nodes.
• Repeat this for all other leaf nodes according to
  their rank. Whenever there is a choice of parent
  choose the parent that is already in the tree.
• Convert the tree to HTML hierarchy by traversing
  it in a depth first manner. The choice of which
  sibling to traverse next is determined by the
  order in which the sibling were entered into the
  tree.

10
Graph Merging Example
Before pruning
After pruning
1
3
1
2
2
3
4
1
3
2
3
4
3
2
7
6
5
8
7
6
5
8
11
Results

• 5 hours to build Inverted Index for a Intranet
  with 200,000 webpage
• Query time 3.02 seconds, of which 1 sec by Java
  front end ( Sparc Ultra II)
• 2.4 secs on Sun Enterprise 450

12
Questions