1. Keyword Searching and Browsing in Databases using BANKS
 Gaurav Bhalotia  Arvind Hulgeri  Charuta Nakhe
 Soumen Chakrabarti  S. Sudarshan
18th International Conference on Data Engineering (ICDE'02), 2002
Kushal Bansal
Today, we are going to see the paper “Keyword Searching and Browsing in Databases using BANKS”
This paper was presented by Gaurav, Arvind, Charuta, Soumen and Sudarshan in the 18th ICDE in the year 2002.

2. Outline Introduction Database and Query Model
Searching for the Best Answers
Interface and Templates of BANK System
Experiment and Performance
Conclusion
My outline of the talk will be on the following topics
Introduction – In this we will see the need for the BANK system
Different types of Database and Query Models
How to search for the best answers
Various features of BANKS
A small experiment
Conclusion

3. Introduction Web Search engines make use of unstructured queries
Users have to type in keywords and follow hyperlinks
Relational databases use structured query languages like SQL
Users need to know the schema of the database
Difficult for naïve users
For data stored in databases, keyword based techniques is not much useful
Data often splits across the tables due to normalization
Read the first bullet
What do we mean by unstructured query? In case of search engines like Google, there is no particular format in which we have to enter the search terms. Such style of querying is called unstructured querying. After entering the search terms when we hit enter, we get a list of links ordered based on the PageRank algorithm.
In relational databases, information (needed to answer a keyword query) / is often split across the tables/tuples

4. Introduction BANKS (Browsing And Keyword Searching)
It is a system which provides search engine type interface to search and browse relational databases.
Allows interaction with controls on the displayed results.
No query language or programming required.
The full form for BANKS is Browsing and Keyword Searching
It is a system which provides us search engine type interface to search and browse relational databases.
A user connects to the Bank system using HTTP protocol and then interacts with the database with the help of the JDBC.

5. Outline Introduction Database and Query Model
Informal Model
Formal Model
Query and Answer Model
Searching for the Best Answers
Interface and Templates of BANK System
Experiment and Performance
Conclusion

6. Database and Query Model
Informal Model
Each database is modeled as a directed graph
Each tuple in the database is modeled as a node in the graph.
Every Primary – Foreign key relation is modeled as a directed edge.
In Informal model, the entire database is represented as a directed graph.
Each tuple in the database is represented as a node in the graph and
Every foreign key – primary key is represented as a directed edge.

7. Database and Query Model
Informal Model
4. An answer to a query is a subgraph connecting nodes matching the keywords.
5. The importance of a link depends upon the relations it connects and on its semantics

8. Database and Query Model
The Schema

9. Database and Query Model
Fragment of the Database

10. Database and Query Model
Formal Database Model
Node Weight
Each node u in the graph is assigned a weight N(u)
Node weight is also known as the node prestige
N (u) = Indegree of the node
Node score N = Root node weight + Sum of leaf node weights

11. Database and Query Model
Formal Database Model
Edge Weights
Weight of the directed edge (u,v) given by
(u,v) exists but (v,u) does not = s (R(u), R(v))
(v,u) exists but (u,v) does not = IN(u) s (R(v),R(u))
If both exists = min [ s(R(u),R(v)), IN(u) s (R(v),R(u)) ]

12. Database and Query Model
Formal Database Model
Edge Weights
Escore(e) of an edge = w(e)/w min
Escore overall = 1/ (1 + ∑ Escore(e))
Escore overall is in the range [0,1]

13. Database and Query Model
Formal Database Model
Overall relevance score
= Node weights + Edge Weight
Using weighting factor 
Additive: (1- ) E + N
multiplicative: E * N 

14. Database and Query Model
Query and Answer Model
Query
Query consists of search terms t1 ,t2, ……tn
For each term ti we find set of nodes Si that are relevant to ti
S = {S1,S2,…Sn}
Answer Model
An answer to a query is a rooted directed tree connecting keyword nodes
Relevance score of an answer tree
Relevance scores of it nodes and its edge weight

15. Database and Query Model
Result of query “soumen and sunita”

16. Outline Introduction Database and Query Model
Searching for the best answers
Backward expanding search algorithm
Interface and Templates of BANKS
Experiment and Performance
Conclusion

17. Searching for the Best Answer
Backward expanding search algorithm
Assumes that the graph of the database fits in memory
Starts at leaf nodes each containing a query keyword
Run concurrent single source shortest path algorithm from each such node
Traverses the graph edges in reverse direction
Common vertex along the backward paths identify answer tree roots
Tree formed is a connection tree and root of tree is information node.

18. Outline Introduction Database and Query Model
Searching for the best answers
Interface and Templates of BANKS
Experiment and Performance
Conclusion

19. Interface BANKS system provides
A rich interface to browse data stored in a relational database
Schema browsing and data browsing
Hyperlink to the referenced tuple
Columns can be projected away (dropped)
Selections can be imposed on any column
Tuples can be sorted by a specified column

20. Templates BANKS system provides several predefined templates
Cross – tabs
Group by
Folder Views
Graphical Interface for display in bar, line or pie chart

21. Outline Introduction Database and Query Model
Searching for the best answers
Interface and Templates of BANKS
Experiment and Performance
Conclusion

22. Experiment and Performance
Computed absolute value of the rank difference of the ideal answer and answer for each parameter setting.
Sum of the rank differences gives the raw Error score
Setting  = 0.2 with log scaling of edge weights did best, with an error score of 0.0

23. Error scores vs. parameter choices

24. Outline Introduction Database and Query Model
Searching for the best answers
Interface and Templates of BANKS
Experiment and Performance
Conclusion

25. Conclusion BANKS system
Provides an integrated browsing and keyword querying system for relational databases
Allows users with no knowledge of database systems or schema to query and browse relational database with ease
Reduces the effort involved in publishing relational data on the web and makes it searchable.