Computer science Department at University of Toronto

1
Computer science Department at University of
Toronto

• Devi Alagappan
• 3/24/06

2
Research activities Database groups

• Cilio A tool for generating mappings between
  SQL and XML schemas. It can also generate queries
  that convert data that conforms to one schema to
  the data that conforms to another.
• Conquer converts the given SQL query into one
  that retrieves only the results that does not
  violate any integrity constraints
• Data Exchange The algorithmic and foundational
  issues to deal with the problems of taking a data
  in a source schema and produce a data in target
  schema

3
Research activities Database groups contd..

• ToXgene This generates consistent collection of
  XML documents that are complex enough and conform
  to the benchmark integrity constraints.
• Limbo Hierarchical categorical clustering
  algorithm that uses Information Bottleneck to
  provide a distance measure and preserve the
  relevant data when clustering.

4
Adaptive Processing of Top-k Queries in XML

• Goal is to compute Top-k matches for queries in
  XML.
• Main reason for interests in Top-k matches for
  queries is the increasing volume of XML
  repository.
• And user might not be interested in viewing less
  relevant matches
• Similar to what is done in search engine

5
Query evaluation

• Compute exact and approximate matches to XPath
  query
• Approximate matches are obtained by relaxing the
  XPath query
• Use answer scores to prune irrelevant data
  earlier when obtaining intermediate results
• Choose a different execution plan for different
  partial matches

6
Query relaxation
7
Query Relaxation
8
Query Relaxation

• 3 Types of relaxation
• - edge generalization replacing a pc edge
  with ad-(2b)
• - leaf deletion making a leaf node
  optional-(2d)
• - subtree promotion moving a subtree from
  its parent node to its grand parent node (2c)
• Notion is that these Exact matches to the relaxed
  queries are the approximate matches to the
  original
• query

9
Scoring Functions

• Extension of the traditional tfidf function is
  used
• Query is decomposed into component predicates.
  Ex /a./b and ./c.//d is decomposed into
  aparentdoc-root, a./b, a./c, a.//d.
• XML idf of a component predicate p(q0,qi) and a
  XML database D the extent to which q0 nodes in
  D additionally satisfies p(q0,qi)
• XML tf of a component predicate p(q0,qi) and a
  node n that belongs to D the number of distinct
  ways in which n satisifies predicate p.

10
Whirlpool architecture
11
Whirlpool Architecture

• One server queue for each node in XPath tree.
• Root node doesnt have a queue as it initializes
  set of partial matches.
• Every other server maintains a priority queue of
  partial matches. For each of the partial match,
  it computes a set of matches that extends the
  partial match with that node
• Then it computes scores for each of the matches.
• Checks and updates the top-k set

12
Top-k set

• Determines if the new match does one of the
  following
• Updates the score of an existing match in the set
• Replaces an existing match in the set
• The match is pruned.

13
Router and Router Queue

• Maintains matches based on the maximum possible
  final scores
• Determines the next server to process the partial
  match and sends that partial match to the queue
  of the respective servers.

14
Conclusion

• Adaptive query processing ensures that different
  partial matches might have gone through different
  sets of server operations.
• The Whirlpool Architecture incrementally computes
  the scores for the partial matches during query
  evaluation
• It also prunes the intermediate matches similar
  to the Query optimization in RDBMS.