Cuoliang Li, Beng Chin Ooi, Jianhua Feng, Jianyong Wang, Lizhu Zhou Tsinghua University

1. EASE: An Effective 3-in-1 Keyword Search Method for Unstructured, Semi-structured and Structured Data Cuoliang Li, Beng Chin Ooi, Jianhua Feng, Jianyong Wang, Lizhu Zhou Tsinghua University SIGMOD 2008 2009. 03. 19. Summarized by Jaehui Park, IDS Lab., Seoul National University Presented by Jaehui Park, IDS Lab., Seoul National University

2. INTRODUCTION • Keyword search capability into text documents, XML documents, and relational databases • Graph index • Instead of traditional inverted index • Effective for unstructured data • Inadequate for complex structural information. • EASE (Efficient and Adaptive keyword Search method) • Efficient algorithmic basis for scalable top-k-style processing of large amounts of heterogeneous data • Employing and adaptive, efficient and novel index

3. Contributions Model for unstructured, semi-structured and structured data as graphs Effective graph index as opposed to the inverted index Novel ranking mechanism for both DB and IR viewpoint Extensive performance study

4. Motivation • Unstructured • Link awareness • Relevant data may be separated into different pages but linked through hyperlinks • (Semi-) Structured • LCA (Lowest common ancestors) • Connected tree with minimal cost • Ex) Steiner trees

5. r-Radius Steiner Graph Problem • Meaningful Steiner graphs with acceptable sizes • Several concepts • Centric distance • Radius • r-Radius Steiner tree • Radius of a Steiner graph cannot be larger than r

6. Example DBLP example

7. The r-Radius Seiner Graph Problem Given a graph and an input keyword query K, the r-Radius Seiner Graph Problem is to find all the r-radius Steiner graphs in , which contain all or a portion of the input keywords in K, ranked by relevancy with K.

8. EASE: An adaptive search method • Inverted indices are not effective for discovering the much richer structural relationships existing in databases with complicated structured [10]. • Index r-radius Steiner graphs for each combination • Very expensive • Proposed method • 1. Discover r-radius graphs (indexing) • 2. Extracting r-radius Steiner graphs (on the fly) • By removing non-Steiner nodes

9. EASE: An adaptive search method • Adjacency Matrix • Extracting r-radius graphs effectively

10. EASE: An adaptive search method • Determining the subgraph that are r-radius graphs • By Lemma 1. • For efficient retrieval of r-radius graphs • Graph index • r-radius graph that contain query keywords k • Extracting r-radius Steiner graphs • By Theorem 1.

11. EASE: An adaptive search method Computing the Steiner nodes 11

12. EASE: An adaptive search method • Maximal r-Radius Graph • Avoid redundancy • Keep the maximal r-radius graphs in the graph index • Overlapping graphs • Graph partitioning • Avoid the incurrence of huge storage • Only need to retrieve the corresponding relevant graph partitions • Graph similarity • Bigger overlap -> higher similarity

13. Summary 1. Obtain adjacency matrix M 2. Compute Mr 3. Extract the maximal r-radius graphs 4. Cluster the graphs by employing the existing K-means algorithm and partition the graph 5. Construct the graph index to materialize the maximal r-radius graphs

14. Others • Ranking Functions • TF-IDF based IR-ranking • Structural Compactness-based DB Ranking • Intuitively, when an r-radius Steiner graph SG is more compact, SG is more likely to be meaningful and relevant. • Indexing

15. Experimental study • Dataset: DBLife, DBLP and IMDB • Comparison • Unstructured • InfoUnit [18] • Semi-structured • SLCA [28] • Structured • DPBF [6]

16. Experimental study

17. Experimental study

18. Conclusion • Proposed an efficient and adaptive keyword search method • EASE • Keyword queries over unstructured, semi-structured and structure data • Examined the issues of indexing and ranking • By taking into account both the structural compactness • Experimental results shows that EASE achieves both high search efficiency and quality for keyword search over heterogeneous data.