1. XSEarch: A Semantic Search Engine for XML Sara Cohen Jonathan Mamou Yaron Kanza Yehoshua Sagiv Presented at VLDB 2003, Germany

2. XSEarch an XML Search Engine Our Goal: Find the “relevant” XML fragments, given tag names and keywords Our Goal: Find the “relevant” XML fragments, given tag names and keywords

3. Excerpt from the XML Version of DBLP Moshe Y. Vardi Querying Logical Databases Victor Vianu A Web Odyssey: From Codd to XML

4. A Search Example Find papers by Vianu on the topic of “logical databases” How can we find such papers?

5. Attempt 1: Standard Search Engine Each document in the corpus is treated as an integral unit. A document containing some of the three query terms is considered as a result.

6. The document is not relevant to the query. This does not work!!! The document is returned BUT it does not contain any paper on “logical databases” by Vianu This fragment does not represent a paper by Vianu This fragment does not represent a paper about logical databases Moshe Y. Vardi Querying Logical Databases Victor Vianu A Web Odyssey: From Codd to XML The document contains the three query terms. Hence, it is returned by a standard search engine. BUT

7. Attempt 2: XML Query Language FOR $i IN document(“bib.xml”)//inproceedings WHERE $i/author contains ‘Vianu’ AND $i/title contains ‘Logical’ AND $i/title contains ‘Databases’ RETURN $i/author $i/title FOR $i IN document(“bib.xml”)//inproceedings WHERE $i/author contains ‘Vianu’ AND $i/title contains ‘Logical’ AND $i/title contains ‘Databases’ RETURN $i/author $i/title Complicated syntax Extensive knowledge of the document structure required to write the query No mechanism for ranking results This does work, BUT

8. Our Requirements from the Search Tool A simple syntax that can be used by naive users Search results should include XML fragments and not necessarily full documents The XML fragments in an answer, should be semantically related –For example, a paper and an author should be in an answer only if the paper was written by this author Search results should be ranked Search results should be returned in “reasonable” time

9. Overall Architecture User Interface Query Processor Ranker XML Files 1 2 3 4 L1L1 L2L2 L3L3 L4L4 Indexer Indices Index Repository

10. Query Syntax and Semantics User Interface Query Processor Ranker XML Files 1 2 3 4 L1L1 L2L2 L3L3 L4L4 Indexer Indices Index Repository User Interface Query Processor

11. XSEarch Query Syntax A query is a list of query terms A query term can be a –Keyword, e.g., database –Tag, e.g., inproceedings: –Tag-keyword combination, e.g., author:Vianu Optionally preceded by a ‘+’

12. Appearance of logical in the fragment increases the rank of this fragment Note that the different document fragments matching these query terms must be “semantically related” Example Find papers by Vianu on the topic of “logical databases” logical +database inproceedings: author:Vianu Appearance of the tag inproceedings, in the fragment, increases the rank of this fragment Appearance of Vianu under the tag author, in the fragment, increases the rank of this fragment The keyword database must appear in the fragment

13. Semantic Relation: The Intuition Semantic Relation: The Intuition

14. XSEarch: Moshe Y. Vardi Querying Logical Databases Victor Vianu A Web Odyssey: From Codd to XML A Web Odyssey: From Codd to XML Victor Vianu Good Result! title and author elements ARE semantically related author:Vianu title:

15. Moshe Y. Vardi Querying Logical Databases Victor Vianu A Web Odyssey: From Codd to XML Querying Logical Databases Victor Vianu Bad Result! title and author elements ARE NOT semantically related XSEarch: author:Vianu title:

16. Semantic Relation: Formalization Semantic Relation: Formalization

17. Data Model: Document Tree Tags are colored in green Data is colored in red proceedings Moshe Y. Vardi inproceedings author title Querying Logical Databases author title Victor Vianu A Web Odyssey: From Codd to XML inproceedings GOAL: Find pairs of semantically related titles and authors.

18. Relationship Trees Relationship tree of n 1, n 2, …, n k n1n1 n2n2 nknk … Lowest common ancestor of n 1, n 2, …, n k

19. Our “Semantic Relation”: Interconnection n 1,..., n k are strongly interconnected if the relationship tree of n 1,..., n k does not contain two nodes with the same label n 1,..., n k are interconnected if either –they are strongly interconnected, or –the only nodes with the same label in the relationship tree of n 1,..., n k, are among n 1,..., n k

20. proceedings Moshe Y. Vardi inproceedings author title Querying Logical Databases author title Victor Vianu A Web Odyssey: From Codd to XML inproceedings Circled nodes belong to different inproceedings entities. They ARE NOT strongly interconnected nor interconnected! Relationship tree Lowest common ancestor of circled nodes Example (1)

21. proceedings Moshe Y. Vardi inproceedings author title Querying Logical Databases author title Victor Vianu A Web Odyssey: From Codd to XML inproceedings Circled nodes belong to the same inproceedings entity. They ARE strongly interconnected, thus, interconnected! Relationship tree Lowest common ancestor of circled nodes Example (2)

22. proceedings Moshe Y. Vardi inproceedings author title Querying Logical Databases author title Victor Vianu Queries and Computation on the Web inproceedings Circled nodes belong to the same inproceedings entity, but are labeled with the same tag. They ARE interconnected, BUT NOT strongly interconnected! Relationship tree Lowest common ancestor of circled nodes Example (3) author Serge Abiteboul We can see the advantage of using interconnection rather than strong interconnection. These two author nodes ARE semantically related.

23. Interconnection Based on theoretical results of “Generating relations from XML documents”, S. Cohen, Y.Kanza, Y. Sagiv, ICDT 2003. –Three types of interconnection We have implemented two types of interconnection XSEarch can easily accommodate different types of interconnection, or other semantic relations between nodes

24. Checking Whether Two Nodes Are Interconnected Given a document T, it is possible to check whether nodes n and n’ are interconnected in O(|T|) time Too expensive to do it during query processing! During query processing, we need to check whether pairs of nodes are interconnected

25. Interconnection Index Is built offline Allows for checking interconnection between two nodes, during query processing, in O(1) time We have two implementations –as a hash table –as a symmetric matrix The Indexer is responsible for building the Interconnection Index

26. Indexer User Query Processor Ranker XML Files 1 2 3 4 L1L1 L2L2 L3L3 L4L4 Indexer Indices Index Repository

27. For each pair of nodes, check whether this pair is interconnected –There are O(|T| 2 ) pairs –Checking interconnection is in O(|T|) time As a result, checking for interconnection of all pairs of nodes in T is in O(|T| 3 ) time  Too expensive also if it is done offline!!! Building the Interconnection Index: Naïve Approach

28. Idea: Checking whether two nodes are interconnected can be done by checking interconnection between their parents/children There are two characterizations of nodes interconnection –For child-ancestor nodes –For non child-ancestor nodes Building the Interconnection Index: Dynamic Programming Approach

29. Interconnection Characterization: n is an ancestor of n’ n and n’ are interconnected if and only if: – the parent of n’ is strongly interconnected with n – the child of n on the path to n’ is strongly interconnected with n’ parent of n’ child of n … n n’ – the parent of n’ is strongly interconnected with n parent of n’ n child of n n’ – the child of n on the path to n’ is strongly interconnected with n’

30. Interconnection Characterization: n is not an ancestor of n’ n and n’ are interconnected if and only if: – the parent of n’ is strongly interconnected with n – the parent of n is strongly interconnected with n’ n’n parent of n’ parent of n … lca … – the parent of n is strongly interconnected with n’ n’ parent of n – the parent of n’ is strongly interconnected with n n parent of n’

31. Building the Interconnection Index Using Dynamic Programming Theorem: Let T be a document. Then it is possible to determine interconnection of all pairs of nodes in T in O(|T| 2 ) time Proof hint: –Derive nodes numbers in T by a depth-first traversal of T –Compute the index using dynamic programming, based on the characterizations

32. Query Processing Document fragments are extracted using the interconnection index and other indices Extracted fragments are returned ranked by the estimated relevance

33. Ranker User Query Processor Ranker XML Files Indexer Indices Index Repository 1 2 3 4 L1L1 L2L2 L3L3 L4L4

34. Ranking Factors Several factors increase the rank of a result Similarity between query and result Weight of labels appearing in the result Characteristics of result tree

35. Query and Result Similarity TFILF –Extension of TFIDF, classical in IR –Term Frequency: number of occurrences of a query term in a fragment –Inverse Leaf Frequency: number of leaves containing a query term divided by number of leaves in the corpus

36. TFILF Term frequency of keyword k in a leaf node n l Inverse leaf frequency TFILF is the product between tf and ilf

37. Weight of Labels Some labels are considered more important than others –Text under an element labeled with title is more “important” than text under element labeled with section Label weights can be –system generated –user defined

38. Relationship between Nodes Size of the relationship tree: small fragment indicates that its nodes are closer, and thus, probably, “more related” article: title:XML 2 nodes 3 nodes This fragment will obtain an higher rank article title XML article section title XML

39. Relationship between Nodes Ancestor-descendant relationships between a pair of nodes in a fragment, indicates “strong relation” between these nodes section: title:XML article title XML section section node is an ancestor of title node This fragment will obtain an higher rank section title article XML

40. Experimental Results

41. Hardware and Software Used Language: Java Processor: 1.6 GHZ Pentium 4 RAM: 2 GB (limited to 1.46 GB by JVM) OS: Windows XP

42. Choosing the Implementation for the Interconnection Index We have experimented the two implementations of the interconnection index 1. IIH: the index is an hash table 2. IIM: the index is a symmetric matrix We compare the two implementations –Cost of building the index –Cost of query processing, i.e., using the index

43. IIM is better than IIH, because of the additional overhead of hashing Time For Building Indices IIH time (ms) IIM time (ms) Number of nodes Size (KB) XML corpus 36293,360146Dream 1851146,635281Hamlet 1,7291,55221,246704Sigmod 7,8376,23149,4221,198Mondial Both implementations are reasonable

44. On the Fly Indexing (OFI) Fully building the indices as a preprocess of querying is expensive in memory for “huge” corpuses! –Also expensive in time because of the additional overhead of using virtual memory Instead, compute interconnection index incrementally on-the-fly during query processing for each pair that must be checked –By how much will query processing be slowed down?

45. Time For Building Indices: Comparing IIH, IIM, OFI IIM time (ms) IIH time (ms) OFI time (ms) Number of nodes Size (KB) XML corpus 29360.63,360146Dream 1141851.16,635281Hamlet 1,5521,7292.221,246704Sigmod 6,2317,83710.049,4221,198Mondial For these corpuses, OFI time is less than 10 ms. Actually it is the time to build all the indices other than the interconnection index.

46. Query Execution Time We generated 1000 random queries for the Sigmod Record corpus Each query had: –At most 3 optional search terms –At most 3 required search terms We checked time with IIH, IIM and OFI

47. IIH/IIM: Query Processing Time Note: Logarithmic scale Both approaches lead to similar results Average run time for queries: 35 ms

48. After processing the 1000 queries, 0.75% of all pairs of nodes were checked for interconnection. Space saved in main memory Slowdown in response time not too large! Locality property: queries tend to be similar in the parts of the document that they may access More than 50% of the queries processed in under 10 ms OFI: Query Processing Time

49. How Good are the Results? We measured recall & precision for the query: –Find papers written by Buneman that contain the keyword database in the title We tried two different queries that reflect different amounts of user knowledge –Kw: +Buneman +database (classical search engine query) –Tag-kw: +author:Buneman +title:database Corpus: Sigmod, DBLP

50. We computed the "correct answers" using XQuery Recall  Perfect recall, i.e., XSEarch returns all the correct answers Precision at n Precision and Recall correct returned answers correct answers correct answers in the first n returned answers n

51. Precision at 5, 10 and 20 Sigmod: Perfect precision DBLP: 0.8/0.9 for query containing only keywords Combining tags and keywords leads to perfect precision

52. Conclusions Paradigm for querying XML combining IR and database techniques Returns semantically related fragments, ranked by estimated relevance Combining tags and keywords in the query leads to good results

53. Conclusions Efficient index structures –IIM/IIH for “small” documents –OFI for “big” documents Efficient evaluation algorithms –Dynamic algorithm for computing interconnection Extensible implementation –The system can easily accommodate different types of semantic relations between nodes, other than interconnection

54. Thank You. Questions?