Effective interpretation, integration and querying of web tables

EFFECTIVE INTERPRETATION, INTEGRATION AND QUERYING OF WEB TABLES LU MEIYU Bachelor of Engineering Harbin Institute of Technology, China A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY DEPARTMENT OF COMPUTER SCIENCE NATIONAL UNIVERSITY OF SINGAPORE 2013 DECLARATION I hereby declare that this thesis is my original work and it has been written by me in its entirety. I have duly acknowledged all the sources of information which have been used in the thesis. This thesis has also not been submitted for any degree in any university previously. Lu Meiyu 29 July 2013 ACKNOWLEDGEMENT This thesis would not have been finished without the help and guidance of many people who gave me valuable assistance during my Ph.D period. It is now my great pleasure to express my thanks to them. First and foremost, my sincere gratitude goes to my supervisors, Professor Anthony K. H. Tung and Professor Beng Chin Ooi. They taught me research skills with patience and knowledge, shared with me their experiences in life, provided me financial support, and gave me internship opportunities. Their continuous encouragement helped me keep motivated during my Ph.D pursuit. I would not have completed this thesis without their guidance and support. It is my great honor to be their student. I would like to thank Dr. Divesh Srivastava, Dr. Graham Cormode, Dr. Marios Hadjieleftheriou and Dr. Srinivas Bangalore for their valuable insights and advice during my internship in AT&T Labs Research. I had two great summers with them. I learnt a lot from their rich experiences in solving real problems and building systems. I am grateful to my thesis committee, Professor Kian-Lee Tan, Professor Wynne Hsu, Professor Chee Yong Chan and the external examiner, for their insightful comments and suggestions to this thesis. Their comments helped me improve the presentation of this thesis in many aspects. I would like to express my thanks to the collaborators during my Ph.D study, especially Professor Divyakant Agrawal, Professor Wang-Chiew Tan, Dr. Bing Tian Dai, and Dr. Ju Fan, for the helpful discussion and suggestions to my research work. i My fellow group members have made my graduate school experience better in many ways. I would like to thank them for their help to my course work and research. I will not forget all the days I was embraced by your friendship, help and support. I am particularly grateful to my best friend, Meihui Zhang, for her company and help during the last nine years. The friendship with you is one of the most valuable assets I own. I owe my deepest thanks to my parents for their unconditional love, understanding, and their faith in me. All the support they have provided me over the years is the greatest gift I ever received. They are the wonderful parents. Last but not the least, I would like to thank my dear husband, Feng Guo. Thank you for your unwavering love in our long-distance relationship in the past eight years. Your support and encouragement helped me through the hard times in finishing this work. ii CONTENTS Acknowledgement i Abstract vii Introduction 1.1 Web Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Regular Tables . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Complicated Tables . . . . . . . . . . . . . . . . . . . . . Web Table Exploration . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Schema Extraction . . . . . . . . . . . . . . . . . . . . . 1.2.2 Schema Matching Discovery . . . . . . . . . . . . . . . . 1.2.3 Querying Facilitation . . . . . . . . . . . . . . . . . . . . 12 1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.4 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.2 Literature Review 2.1 2.2 2.3 18 Web Table Extraction and Interpretation . . . . . . . . . . . . . 18 2.1.1 High-quality Table Discovery . . . . . . . . . . . . . . . 19 2.1.2 Table Header Identification . . . . . . . . . . . . . . . . 20 Schema Matching . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.2.1 Traditional Schema Matching Techniques . . . . . . . . . 21 2.2.2 Web Table Annotation . . . . . . . . . . . . . . . . . . . 22 Data Integration . . . . . . . . . . . . . . . . . . . . . . . . . . 23 iii CONTENTS 2.4 2.3.1 Traditional Data Integration Systems . . . . . . . . . . . 23 2.3.2 Probabilistic Data Integration . . . . . . . . . . . . . . . 24 2.3.3 User Feedback based Data Integration . . . . . . . . . . 25 Data/Query Relaxation and Probabilistic Databases . . . . . . . 25 2.4.1 Database Relaxation . . . . . . . . . . . . . . . . . . . . 25 2.4.2 Query Relaxation . . . . . . . . . . . . . . . . . . . . . . 27 2.4.3 Probabilistic Databases . . . . . . . . . . . . . . . . . . . 28 Schema Extraction for Web Tables 29 3.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.2 The Web Table Corpus . . . . . . . . . . . . . . . . . . . . . . . 31 3.2.1 Table Extraction . . . . . . . . . . . . . . . . . . . . . . 31 3.2.2 Table Header Heterogeneity . . . . . . . . . . . . . . . . 33 Feature Engineering . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.3.1 Cell Features . . . . . . . . . . . . . . . . . . . . . . . . 38 3.3.2 Row/Column Features . . . . . . . . . . . . . . . . . . . 41 3.4 Single and Separate Classification . . . . . . . . . . . . . . . . . 43 3.5 Holistic and Two-phase Classification . . . . . . . . . . . . . . . 44 3.5.1 Holistic Header Identification . . . . . . . . . . . . . . . 46 3.5.2 Two-phase Header Identification . . . . . . . . . . . . . . 46 Training Data and Classifiers . . . . . . . . . . . . . . . . . . . 48 3.6.1 Training Data Collection . . . . . . . . . . . . . . . . . . 48 3.6.2 Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Schema Construction . . . . . . . . . . . . . . . . . . . . . . . . 52 3.7.1 Rows/Columns are Headers . . . . . . . . . . . . . . . . 52 3.7.2 Both Rows and Columns are Headers . . . . . . . . . . . 53 Experimental Evaluation . . . . . . . . . . . . . . . . . . . . . . 53 3.8.1 Baselines . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.8.2 Effect of Features . . . . . . . . . . . . . . . . . . . . . . 56 3.8.3 Effectiveness of Post-Processing . . . . . . . . . . . . . . 57 3.8.4 Single vs. Separate Classification . . . . . . . . . . . . . 59 3.8.5 Holistic vs. Two-phase Classification . . . . . . . . . . . 62 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.3 3.6 3.7 3.8 3.9 iv CONTENTS A Machine-Crowdsourcing Hybrid Approach to Matching Web Tables 65 4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.2 Hybrid Machine-Crowdsourcing Framework . . . . . . . . . . . . 67 4.2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.2.2 System Architecture . . . . . . . . . . . . . . . . . . . . 69 Column Utility . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.3.1 Candidate Concept Generation . . . . . . . . . . . . . . 72 4.3.2 Modeling the Difficulty of a Column . . . . . . . . . . . 73 4.3.3 Modeling the Column Influence . . . . . . . . . . . . . . 75 Utility-Based Column Selection . . . . . . . . . . . . . . . . . . 79 4.4.1 Expected Utility of Columns . . . . . . . . . . . . . . . . 79 4.4.2 Algorithm for Column Selection . . . . . . . . . . . . . . 81 4.5 Concept Determination . . . . . . . . . . . . . . . . . . . . . . . 86 4.6 Experimental Evaluation . . . . . . . . . . . . . . . . . . . . . . 86 4.6.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . 87 4.6.2 Value Incompleteness and Freebase Coverage . . . . . 89 4.6.3 Hybrid Machine-Crowdsourcing Method . . . . . . . . . 90 4.6.4 Comparison with Table Annotation Techniques . . . . . 95 4.6.5 Evaluation on Table Matching . . . . . . . . . . . . . . . 96 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.3 4.4 4.7 Probabilistic Tagging and Querying of Web Tables 100 5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 5.2 System Overview and Usage Scenario . . . . . . . . . . . . . . . 102 5.3 5.4 5.5 5.2.1 System Overview . . . . . . . . . . . . . . . . . . . . . . 102 5.2.2 Usage Scenario . . . . . . . . . . . . . . . . . . . . . . . 102 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . 106 5.3.1 Probabilistic Tagging . . . . . . . . . . . . . . . . . . . . 106 5.3.2 Query Semantics . . . . . . . . . . . . . . . . . . . . . . 107 Tag Inference . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 5.4.1 Probabilistic Matches Generation . . . . . . . . . . . . . 111 5.4.2 Probabilistic Tag Inference . . . . . . . . . . . . . . . . . 113 Top-k Query Processing . . . . . . . . . . . . . . . . . . . . . . 114 5.5.1 Data Organization . . . . . . . . . . . . . . . . . . . . . 114 v CONTENTS 5.6 5.7 5.5.2 Dynamic Instantiation . . . . . . . . . . 5.5.3 Top-k Query Answering . . . . . . . . . Experimental Evaluation . . . . . . . . . . . . . 5.6.1 Experimental Setup . . . . . . . . . . . . 5.6.2 Effectiveness of Top-k Query Processing 5.6.3 Comparison with OpenII . . . . . . . . . 5.6.4 Efficiency of Top-k Query Processing . . Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 115 119 119 120 122 123 125 Conclusion 126 6.1 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 6.2 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . 128 Bibliography 130 vi CHAPTER 5. PROBABILISTIC TAGGING AND QUERYING OF WEB TABLES η ŽĨ dĂŐƐ с Ϯ η ŽĨ dĂŐƐ с ϯ η ŽĨ dĂŐƐ с Ϯ η ŽĨ dĂŐƐ с ϰ η ŽĨ dĂŐƐ с ϰ ϭ ϭ Ϭ͘ϴ Ϭ͘ϴ Ϭ͘ϲ Ϭ͘ϲ ZĞĐĂůů WƌĞĐŝƐŝŽŶ η ŽĨ dĂŐƐ с ϯ Ϭ͘ϰ Ϭ͘ϰ Ϭ͘Ϯ Ϭ͘Ϯ Ϭ Ϭ ϮϬ ϰϬ ϲϬ dŽƉͲŬй ϴϬ ϮϬ ϭϬϬ (a) Precision. ϰϬ ϲϬ dŽƉͲŬй ϴϬ ϭϬϬ (b) Recall. Figure 5.9: Precision and recall in DIR domain by varying k%. of tags than car dataset, we vary its distinct number of tags from to 4. We observe similar behavior as over car dataset. Since the number of tags in the directory dataset is less, and most are well-formatted, its recall is much better than the car dataset. 5.6.3 Comparison with OpenII We next compare our probabilistic tagging and querying approach with Open II (Open Information Integration) [68] on query processing. Open II is an open source data integration toolkit that has implemented several schema matching methods3 . In this experiment, for Open II, we choose all the matching methods it implemented to produce its matching result. These matchers include “Name Similarity Matcher”, “Documentation Matcher”, “Mapping Matcher”, “Exact Matcher”, “Quick Matcher”, and “WordNet Matcher”. The results of precision and recall are shown in Figure 5.10(a) and Figure 5.10(b), respectively. In particular, we only list the results for T case over car dataset. The reason is that Open II makes many wrong inferences, as such, precision in most cases is really low. Here we choose its best case, i.e., T 3, for comparison. It is indicated that our approach provides better precision and recall than Open II. Specifically, among the queries in T 3, Open II produces 0% precision for two of them because they contain wrong matched tags, and 100% precision for the rest three. In average, Open II has 60% precision across all k% cases. With less correct records returned, Open II has lower recall than Note that Open II is a schema level mapper while our approach is an instance-based approach [64]. 122 CHAPTER 5. PROBABILISTIC TAGGING AND QUERYING OF WEB TABLES ϭ ϭ Ϭ͘ϴ Ϭ͘ϴ Ϭ͘ϲ Ϭ͘ϲ KƉĞŶ// ZĞĐĂůů WƌĞĐŝƐŝŽŶ WƚĂŐŐŝŶŐ Ϭ͘ϰ Ϭ͘Ϯ KƉĞŶ// Ϭ͘ϰ WƚĂŐŐŝŶŐ Ϭ͘Ϯ Ϭ Ϭ ϮϬ ϰϬ ϲϬ dŽƉͲŬй ϴϬ ϭϬϬ ϮϬ ϰϬ ϲϬ dŽƉͲŬй ϴϬ ϭϬϬ (a) Probabilistic tagging vs. Open II on (b) Probabilistic tagging vs. Open II on precision. recall. Figure 5.10: Probabilistic tagging vs. Open II on precision and recall by varying k%. ZƵŶŶŝŶŐ dŝŵĞ ;ƐĞĐͿ Ϭ͘Ϯϱ ^Y> ĂŐĞƌ ZĞƚƌŝĞǀĂů >ĂnjLJ ZĞƚƌŝĞǀĂů Ϭ͘Ϯ Ϭ͘ϭϱ Ϭ͘ϭ Ϭ͘Ϭϱ Ϭ ϮϬ ϰϬ ϲϬ dŽƉͲŬй ϴϬ ϭϬϬ Figure 5.11: Running time by varying k%. ours. 5.6.4 Efficiency of Top-k Query Processing We validate the efficiency of our top-k query processing over T on car dataset. For the directory dataset and the other queries, the overall trend is similar. We first compare the running time of our approaches with the translated SQL queries, as shown in Figure 5.11. Note that when k% is set to 100%, our approaches retrieve almost the same number of tuples as the translated SQLs. As illustrated, the running time of the translated SQL queries is quite stable with respect to different top-k%, because they always retrieve the gold standard (i.e., all the true answers), no matter how k% changes. In contrast, the running 123 CHAPTER 5. PROBABILISTIC TAGGING AND QUERYING OF WEB TABLES ĂŐĞƌ ZĞƚƌŝĞǀĂů >ĂnjLJ ZĞƚƌŝĞǀĂů Ϯϱ ϮϬ ϭϱ ϭϬ ϱ Ϭ ϮϬ ϰϬ ϲϬ ϴϬ ϭϬϬ ϭϮϬ η ŽĨ ^ĞƋƵĞŶƚŝĂů ĐĐĞƐƐĞƐ ;yϭϬϬͿ η ŽĨ ZĂŶĚŽŵ ĐĐĞƐƐĞƐ ;yϭϬϬͿ >ĂnjLJ ZĞƚƌŝĞǀĂů ϯϬ ĂŐĞƌ ZĞƚƌŝĞǀĂů ϯϬ Ϯϱ ϮϬ ϭϱ ϭϬ dŽƉͲŬ ϱ Ϭ ϮϬ ϰϬ ϲϬ ϴϬ ϭϬϬ ϭϮϬ dŽƉͲŬ (a) Lazy Retrieval vs. Eager Retrieval on ran- (b) Lazy Retrieval vs. Eager Retrieval on sedom I/Os. quential I/Os. Figure 5.12: Lazy Retrieval vs. Eager Retrieval on random I/O and sequential I/O. time of our approaches, i.e., lazy retrieval and eager retrieval, increases linearly when top-k% increases. However, eager retrieval takes longer time than lazy retrieval by almost 1.5 times since eager retrieval maintains and probes the sorted lists for QTags, but most tuples in them are invalid. The running time that lazy retrieval takes is almost k% of translated SQLs, which clearly shows the efficiency of our lazy retrieval approach. We next look at how the number of random and sequential accesses vary as we vary k. Figure 5.12(a) shows the number of random accesses for eager and lazy retrieval when varying k from 20 to 120. The corresponding graphs for sequential I/Os are in Figure 5.12(b). Note that in both figures, we adopt top-k rather than top-k%, as the number of tuples returned by top-k% varies proportionally to query’s answer set. Consistent with earlier observation that eager retrieval takes almost 1.5 times longer than lazy retrieval, the same conclusion holds for both the number of sequential accesses and random accesses. From the figures, we observe that when k equals to 20, we need about 1,500 sequential accesses and random accesses, but we only need approximately 80 more accesses for each 20 more answers. The large 1,500 I/Os is caused by the tables which only contain some of user queried tags. Tuples from such tables may rank higher than the correct answers in sorted lists, so probing these tuples introduces more I/Os. But in the later process, we only need about 4*k more accesses if we want to retrieve k more answers. In a word, our top-k query processor is quite efficient. 124 CHAPTER 5. PROBABILISTIC TAGGING AND QUERYING OF WEB TABLES 5.7 Summary In this chapter, we proposed an effective approach to improving the usability of integrated Web tables for ordinary users. We introduced the idea of probabilistic tagging, where each value in the database is associated with multiple semantic-relevant tags in a probabilistic way. With the enriched tags, users are allowed to issue structured queries using any tag they like, rather than over a predefined mediated schema. We have designed an efficient and effective dynamic instantiation scheme to process user-issued queries, where the semantics of queried tags are determined on-the-fly. We believe our approach is able to grant more privileges to non-expert database users and to better maintain the database quality. The work in this chapter has been published as a full research paper in the 2011 SIGMOD Conference [55]. 125 CHAPTER Conclusion In this work we proposed and implemented a holistic Web table processing framework to explore the rich knowledge embedded in Web tables. Our framework consists of three main components: a machine learning based approach to extract table schema, a machine-crowdsourcing hybrid approach to integrate the tables, and a probabilistic tagging and querying scheme to improve the usability of integrated tables. In the following, we will first summarize our main contributions with respect to the three components in Section 6.1, and discuss some interesting directions for future work in Section 6.2. 6.1 Contributions We proposed a series of machine learning based approaches to identify the schemata for the whole range of Web tables, rather than regular tables only as prior works did. We extracted a rich set of features and built several classifier variants to identify the header cells. A post-processing procedure was specially designed for cell-level classifiers, to guarantee the consistency among the header cells. We conducted an extensive experimental study over millions of tables extracted from Wikipedia. It was found that the visual features coupled with a suitable classifier are essential for improving the accuracy of identified headers. Also, post-processing is a necessary phase for achieving consistent headers. Compared with existing heuristic based and simple machine learning based 126 CHAPTER 6. CONCLUSION methods, our approach is more general (applicable to the whole range of Web tables), and at the same time can improve both precision and recall significantly. The schemata identified by our approach are important for understanding the table semantics, and could be further exploited by other Web table applications. Another work in this study was that we presented a machine-crowdsourcing hybrid approach to discover the schema matches between Web table columns. Here crowdsourcing was adopted to resolve the semantic heterogeneity in Web tables. To reduce the crowdsourcing cost and improve the accuracy at the same time, we defined an effective utility function to select the most valuable candidate matches for crowdsourcing. Matches that are difficult for machine algorithms and have greater influence on other matches are preferred. We conducted experiments on two real-world Web table datasets. The results show that crowdsourcing is quite effective to improve the matching quality. Compared with pure machine algorithms, our approach is able to provide a much better accuracy while with a low crowdsourcing cost. To the best of our knowledge, we are the first to leverage the power of crowdsourcing to facilitate the Web-scale schema matching problem. Although we focus on schema matching only, our work has laid the foundation for more sophisticated integration procedures, such as data fusion [28] and table relationship discovery. We introduced a probabilistic tagging and querying scheme to improve the usability of integrated Web tables for ordinary users. We treated each matched attribute as a tag and associated a probability with it. In this way, each underlying data value is associated with multiple tags which can express its semantics. With the enriched tags, users are allowed to issue queries in a SQL-like fashion by using any tag they like, rather than over a large mediated schema. A dynamic instantiation was proposed to process the posed queries. The experiments on two real datasets show that probabilistic tagging is effective in relaxing the query complexity. It was also found that dynamic instantiation is efficient in query processing. In summary, the integrated data produced by our system would contribute to a better understanding of the Web table semantics, and may have potentials in many applications, for instance knowledge base augmentation, query answering systems and OLAP analysis. It might also be useful to enhance the search engine performance, e.g., supporting structured results. 127 CHAPTER 6. CONCLUSION 6.2 Future Directions In Web table processing, many directions deserve further research in future work. These are summarized below. • Relational modeling. In interpreting Web tables, our work identifies one schema for each Web table. In practice, we also observe that some Web tables are not normalized. An ideal case is to first normalize the original Web table, such as dividing it into several sub-tables and assign one schema to each sub-table, and then we can easily transform them into relational databases. We did not take this problem into account in this work due to its complexity. To solve this problem, a reasonable approach to analyze the content in We tables, and model the relationship among different cells is needed. • Data fusion. In integrating Web tables, we mainly focused on schema matching discovery. The matching problem in value level, such as data fusion [28], was not considered in this work, as Web-scale data fusion is a challenging problem itself. However, in the real world, some Web tables may provide incorrect information, resulting in inconsistency for the same entity after integration. In view of this, data fusion is required to estimate the accuracy of each Web table, and discover the true values. Current algorithms [53] for data fusion are usually NP-hard, and how to implement them in Web scale is important and worthy of future research. • Relationship discovery. On querying Web tables, our probabilistic tagging and querying scheme executes the query on each Web table in isolation. Nevertheless, tables might be related to each other via some relationships such as primary-primary key and primary-foreign key. More specifically, there is a primary-primary key relationship between two tables if they contain entities from the same/similar domain, and there is a primary-foreign key relationship if one property column in one table has the same domain with the entity in another table. Some relationship, such as primary-primary key relationship, is specific to Web tables only. Identifying the possible relationships between Web tables and discovering them are important and necessary. Such relationships would be valuable for more advanced table operations such as join/union [67], and 128 CHAPTER 6. CONCLUSION more flexible querying schemes as well. For example, with the help of primary-primary key relationship, we can easily union the Web tables which contain entities from the same domain, while with primary-foreign key relationship, we can join the Web tables in a more meaningful way. 129 Bibliography [1] Flickr - Photo Sharing. http://www.flickr.com. [2] YouTube - Broadcast Yourself. http://www.youtube.com. [3] B. Aditya, Gaurav Bhalotia, Soumen Chakrabarti, Arvind Hulgeri, Charuta Nakhe, Parag, and S. Sudarshan. Banks: Browsing and keyword searching in relational databases. In VLDB, pages 1083–1086, 2002. [4] Sanjay Agrawal, Surajit Chaudhuri, and Gautam Das. Dbxplorer: A system for keyword-based search over relational databases. In ICDE, pages 5–16, 2002. [5] Zohra Bellahsene, Angela Bonifati, and Erhard Rahm, editors. Schema Matching and Mapping. Springer, 2011. [6] Sonia Bergamaschi, Silvana Castano, and Maurizio Vincini. Semantic integration of semistructured and structured data sources. SIGMOD Record, 28(1):54–59, 1999. [7] Jacob Berlin and Amihai Motro. Database schema matching using machine learning with feature selection. In CAiSE, pages 452–466, 2002. [8] Philip A. Bernstein, Jayant Madhavan, and Erhard Rahm. Generic schema matching, ten years later. PVLDB, 4(11):695–701, 2011. [9] Bloomberg. Market data. http://www.bloomberg.com/markets. 130 BIBLIOGRAPHY [10] Kurt D. Bollacker, Colin Evans, Praveen Paritosh, Tim Sturge, and Jamie Taylor. Freebase: a collaboratively created graph database for structuring human knowledge. In SIGMOD Conference, pages 1247–1250, 2008. [11] Michael J. Cafarella, Alon Y. Halevy, and Nodira Khoussainova. Data integration for the relational web. PVLDB, 2(1):1090–1101, 2009. [12] Michael J. Cafarella, Alon Y. Halevy, Daisy Zhe Wang, Eugene Wu, and Yang Zhang. Webtables: exploring the power of tables on the web. PVLDB, 1(1):538–549, 2008. [13] Michael J. Cafarella, Alon Y. Halevy, Yang Zhang, Daisy Zhe Wang, and Eugene Wu. Uncovering the relational web. In WebDB, 2008. [14] Silvana Castano, Valeria De Antonellis, and Sabrina De Capitani di Vimercati. Global viewing of heterogeneous data sources. IEEE Trans. Knowl. Data Eng., 13(2):277–297, 2001. [15] Central Intelligence Agency. The world factbook. https://www.cia. gov/library/publications/the-world-factbook. [16] C.-C. Chang and C.-J. Lin. LIBSVM: A library for support vector machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1– 27:27, 2011. [17] Fay Chang, Jeffrey Dean, Sanjay Ghemawat, Wilson C. Hsieh, Deborah A. Wallach, Michael Burrows, Tushar Chandra, Andrew Fikes, and Robert Gruber. Bigtable: A distributed storage system for structured data. In OSDI, pages 205–218, 2006. [18] Hsin-Hsi Chen, Shih-Chung Tsai, and Jin-He Tsai. Mining tables from large scale html texts. In Proceedings of the 18th conference on Computational linguistics - Volume 1, COLING ’00, pages 166–172, Stroudsburg, PA, USA, 2000. Association for Computational Linguistics. [19] Eric Chu, Jennifer L. Beckmann, and Jeffrey F. Naughton. The case for a wide-table approach to manage sparse relational data sets. In SIGMOD Conference, pages 821–832, 2007. 131 BIBLIOGRAPHY [20] Michael Cole and Jacek Gwizdka. Tagging semantics: investigations with wordnet. In JCDL, page 446, 2008. [21] Decide. Decide.com. http://www.decide.com/markets. [22] AnHai Doan, Pedro Domingos, and Alon Y. Halevy. Reconciling schemas of disparate data sources: A machine-learning approach. In SIGMOD Conference, pages 509–520, 2001. [23] AnHai Doan, Pedro Domingos, and Alon Y. Levy. Learning source description for data integration. In WebDB (Informal Proceedings), pages 81–86, 2000. [24] AnHai Doan and Alon Y. Halevy. Semantic integration research in the database community: A brief survey. AI Magazine, 26(1):83–94, 2005. [25] Anhai Doan, Raghu Ramakrishnan, and Alon Y. Halevy. Crowdsourcing systems on the world-wide web. Commun. ACM, 54(4):86–96, April 2011. [26] Xin Dong and Alon Y. Halevy. Malleable schemas: A preliminary report. In WebDB, pages 139–144, 2005. [27] Xin Luna Dong, Alon Y. Halevy, and Cong Yu. Data integration with uncertainty. In VLDB, pages 687–698, 2007. [28] Xin Luna Dong and Felix Naumann. Data fusion: resolving data conflicts for integration. Proc. VLDB Endow., 2(2):1654–1655, August 2009. [29] Ahmed K. Elmagarmid, Panagiotis G. Ipeirotis, and Vassilios S. Verykios. Duplicate record detection: A survey. IEEE Trans. Knowl. Data Eng., 19(1):1–16, 2007. [30] Hazem Elmeleegy, Jayant Madhavan, and Alon Y. Halevy. Harvesting relational tables from lists on the web. VLDB J., 20(2):209–226, 2011. [31] J´erˆome Euzenat and Pavel Shvaiko. Ontology matching. Springer, 2007. [32] Ronald Fagin, Amnon Lotem, and Moni Naor. Optimal aggregation algorithms for middleware. In PODS, 2001. 132 BIBLIOGRAPHY [33] Ju Fan, Meiyu Lu, Beng Chin Ooi, Wang-Chiew Tan, and Meihui Zhang. A hybrid machine-crowdsourcing system for matching web tables. In ICDE, 2014. [34] Wolfgang Gatterbauer and Paul Bohunsky. Table extraction using spatial reasoning on the css2 visual box model, 2006. [35] Tingjian Ge, Stanley B. Zdonik, and Samuel Madden. Top-k queries on uncertain data: on score distribution and typical answers. In SIGMOD Conference, pages 375–388, 2009. [36] Scott A. Golder and Bernardo A. Huberman. The structure of collaborative tagging systems. CoRR, 2005. [37] Roy Goldman and Jennifer Widom. Dataguides: Enabling query formulation and optimization in semistructured databases. In VLDB, pages 436–445, 1997. [38] Google. Freebase Data Dumps. https://developers.google.com/ freebase/data. [39] Rahul Gupta and Sunita Sarawagi. Answering table augmentation queries from unstructured lists on the web. PVLDB, 2(1):289–300, 2009. [40] Laura M. Haas. Beauty and the beast: The theory and practice of information integration. In ICDT, pages 28–43, 2007. [41] Alon Y. Halevy, Anand Rajaraman, and Joann J. Ordille. Data integration: The teenage years. In VLDB, pages 9–16, 2006. [42] Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Peter Reutemann, and Ian H. Witten. The weka data mining software: an update. SIGKDD Explor. Newsl., 11(1):10–18, November 2009. [43] Vagelis Hristidis, Luis Gravano, and Yannis Papakonstantinou. Efficient ir-style keyword search over relational databases. In VLDB, pages 850–861, 2003. [44] Ekaterini Ioannou, Wolfgang Nejdl, Claudia Nieder´ee, and Yannis Velegrakis. On-the-fly entity-aware query processing in the presence of linkage. PVLDB, 3(1):429–438, 2010. 133 BIBLIOGRAPHY [45] Shawn R. Jeffery, Michael J. Franklin, and Alon Y. Halevy. Pay-as-yougo user feedback for dataspace systems. In SIGMOD Conference, pages 847–860, 2008. [46] Manas Joglekar, Hector Garcia-Molina, and Aditya Parameswaran. Evaluating the crowd with confidence. 2012. [47] Eugene L. Lawler. Combinatorial Optimization: Networks and Matroids. Dover Publications, 2001. [48] Maurizio Lenzerini. Data integration: A theoretical perspective. In PODS, pages 233–246, 2002. [49] M. Ley. DBLP database. http://dblp.uni-trier.de/xml. [50] Guoliang Li, Beng Chin Ooi, Jianhua Feng, Jianyong Wang, and Lizhu Zhou. Ease: an effective 3-in-1 keyword search method for unstructured, semi-structured and structured data. In SIGMOD Conference, pages 903– 914, 2008. [51] Wen-Syan Li and Chris Clifton. Semantic integration in heterogeneous databases using neural networks. In VLDB, pages 1–12, 1994. [52] Girija Limaye, Sunita Sarawagi, and Soumen Chakrabarti. Annotating and searching web tables using entities, types and relationships. PVLDB, 3(1):1338–1347, 2010. [53] Xuan Liu, Xin Luna Dong, Beng Chin Ooi, and Divesh Srivastava. Online data fusion. PVLDB, 4(11):932–943, 2011. [54] Xuan Liu, Meiyu Lu, Beng Chin Ooi, Yanyan Shen, Sai Wu, and Meihui Zhang. Cdas: A crowdsourcing data analytics system. PVLDB, 5(10):1040–1051, 2012. [55] Meiyu Lu, Divyakant Agrawal, Bing Tian Dai, and Anthony K. H. Tung. Schema-as-you-go: on probabilistic tagging and querying of wide tables. In SIGMOD Conference, pages 181–192, 2011. [56] Benjamin Markines, Ciro Cattuto, Filippo Menczer, Dominik Benz, Andreas Hotho, and Gerd Stumme. Evaluating similarity measures for emergent semantics of social tagging. In WWW, pages 641–650, 2009. 134 BIBLIOGRAPHY [57] Metaweb Technologies. Freebase wikipedia extraction (wex). http:// download.freebase.com/wex. [58] G L Nemhauser, L A Wolsey, and M L Fisher. An analysis of approximations for maximizing submodular set functionsi. Mathematical Programming, 14(1):265–294, 1978. [59] Thomas Neumann and Gerhard Weikum. Rdf-3x: a risc-style engine for rdf. PVLDB, 1(1):647–659, 2008. [60] Luigi Palopoli, Domenico Sacc`a, and Domenico Ursino. Semi-automatic semantic discovery of properties from database schemas. In IDEAS, pages 244–253, 1998. [61] Gerald Penn, Jianying Hu, Hengbin Luo, and Ryan Mcdonald. Flexible web document analysis for delivery to narrow-bandwidth devices. In Proceedings of the 6th International Conference on Document Analysis and Recognition (ICDAR), pages 1074–1078, 2001. [62] Lu Qin, Jeffrey Xu Yu, and Lijun Chang. Keyword search in databases: the power of rdbms. In SIGMOD Conference, pages 681–694, 2009. [63] Sriram Raghavan and Hector Garcia-Molina. Crawling the hidden web. In VLDB, pages 129–138, 2001. [64] Erhard Rahm and Philip A. Bernstein. A survey of approaches to automatic schema matching. VLDB J., 10(4):334–350, 2001. [65] Christopher R´e and Dan Suciu. Approximate lineage for probabilistic databases. PVLDB, 1(1):797–808, 2008. [66] Anish Das Sarma, Xin Dong, and Alon Y. Halevy. Bootstrapping pay-asyou-go data integration systems. In SIGMOD Conference, pages 861–874, 2008. [67] Anish Das Sarma, Lujun Fang, Nitin Gupta, Alon Y. Halevy, Hongrae Lee, Fei Wu, Reynold Xin, and Cong Yu. Finding related tables. In SIGMOD Conference, pages 817–828, 2012. 135 BIBLIOGRAPHY [68] Len Seligman, Peter Mork, Alon Halevy, Ken Smith, Michael J. Carey, Kuang Chen, Chris Wolf, Jayant Madhavan, Akshay Kannan, and Doug Burdick. Openii: an open source information integration toolkit. In SIGMOD ’10, pages 1057–1060, 2010. [69] Mohamed A. Soliman, Ihab F. Ilyas, and Kevin Chen-Chuan Chang. Top-k query processing in uncertain databases. In ICDE, pages 896–905, 2007. [70] Fabian M. Suchanek, Gjergji Kasneci, and Gerhard Weikum. Yago: a core of semantic knowledge. In WWW, pages 697–706, 2007. [71] Jeffrey D. Ullman. Information integration using logical views. Theor. Comput. Sci., 239(2):189–210, 2000. [72] United States Goverment. Data.gov: Empowering people. http://www. data.gov. [73] Petros Venetis, Alon Y. Halevy, Jayant Madhavan, Marius Pasca, Warren Shen, Fei Wu, Gengxin Miao, and Chung Wu. Recovering semantics of tables on the web. PVLDB, 4(9):528–538, 2011. [74] W3Schools. Html table. http://www.w3schools.com/html/html_ tables.asp. [75] Jiannan Wang, Tim Kraska, Michael J. Franklin, and Jianhua Feng. Crowder: Crowdsourcing entity resolution. PVLDB, 5(11):1483–1494, 2012. [76] Qiu Yue Wang, Jeffrey Xu Yu, and Kam-Fai Wong. Approximate graph schema extraction for semi-structured data. In EDBT, pages 302–316, 2000. [77] Yalin Wang and Jianying Hu. A machine learning based approach for table detection on the web. In WWW, pages 242–250, 2002. [78] Steven Euijong Whang, Peter Lofgren, and Hector Garcia-Molina. Question selection for crowd entity resolution. In PVLDB. Stanford InfoLab, August 2013. [79] Jennifer Widom. Trio: A system for integrated management of data, accuracy, and lineage. In CIDR, pages 262–276, 2005. 136 BIBLIOGRAPHY [80] I.H. Witten and E. Frank. Data Mining: Practical machine learning tools and techniques. Morgan Kaufmann, 2005. [81] Wensheng Wu, AnHai Doan, and Clement T. Yu. Webiq: Learning from the web to match deep-web query interfaces. In ICDE, page 44, 2006. [82] Mohamed Yakout, Kris Ganjam, Kaushik Chakrabarti, and Surajit Chaudhuri. Infogather: entity augmentation and attribute discovery by holistic matching with web tables. In SIGMOD Conference, pages 97–108, 2012. [83] Bei Yu, Guoliang Li, Karen R. Sollins, and Anthony K. H. Tung. Effective keyword-based selection of relational databases. In SIGMOD Conference, pages 139–150, 2007. [84] Jeffrey Xu Yu, Lu Qin, and Lijun Chang. Keyword search in relational databases: A survey. IEEE Data Eng. Bull., 33(1):67–78, 2010. [85] Xuan Zhou, Julien Gaugaz, Wolf-Tilo Balke, and Wolfgang Nejdl. Query relaxation using malleable schemas. In SIGMOD Conference, pages 545– 556, 2007. 137 [...]... present and share structured data on the Web The amount of Web tables on the whole Internet is huge and continuously increasing over the years For examples, Cafarella et al reported 154M Web tables from a snapshot of Googles crawl in 2008 [12]; and Yakout et al extracted 573M Web tables from a crawl of Microsoft Bing search engine in 2012 [82] The Wikipedia1 alone contains around one million Web tables Web. .. consists of three main components: Web table interpretation, integration and querying Our first work is to present a generic solution to extract the schema (i.e., attribute names and data types) of Web tables The main challenge arises from the diversity in Web tables, especially in those with complex structure For instance, the ways to organize tables and present table headers may vary widely across tables. .. relational form and subsequently use existing annotation/labeling techniques to derive attribute labels [67, 73, 52] Given the large diversity of Web tables and the lack of standardization, the problem of extracting the metadata relating to these tables and constructing table schemata becomes very challenging Previous works on Web tables handle regular tables only [13, 12, 52, 73, 82] (the definition of regular... matching problem, Web tables are inherently incomplete and heterogeneous, making existing schema matching solutions inadequate for matching columns of Web tables Value Incompleteness The incompleteness in Web tables arises from the fact that Web tables typically contain only a limited amount of information, since a Web table is usually extracted from a single Web page Hence, given two Web tables, even if...ABSTRACT The World Wide Web contains a vast amount of structured and semi-structured information in the form of HTML tables (a.k.a., Web tables) The rich information embedded in those Web tables provides us an opportunity to build a valuable knowledge base and make it usable and queryable for ordinary users In this work, we aim to propose and implement a holistic Web table processing framework... matching discovery, and querying facilitation More specifically, schema extraction is to interpret the semantics of each Web table, schema matching discovery is to find the semantic relationships between Web tables and perform the integration, and querying facilitation is to provide a querying mechanism where even ordinary users are able to conveniently perform searches over Web tables 5 CHAPTER 1 INTRODUCTION... user-issued queries, where the semantics of queried tags are determined on-the-fly We validate our proposed approaches via extensive experiments on real-world Web table datasets viii LIST OF TABLES 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 Statistics of header types in Web tables Top 10 popular HTML tags and attributes in Web tables Categorization of tables in the English language Wikipedia... 3 An example of non-regular Web table, with row span and column span 4 An example of crosstab, where both rows and columns are table headers 7 Two Web table examples Table T1 is about movies and T2 is about books 11 Distribution of Web tables w.r.t the number of rows, columns, and cells ... schema for Web tables is almost infeasible This is mainly attributed to the large scale of Web tables: millions of tables that 12 CHAPTER 1 INTRODUCTION cover various domains/topics Collected on regular tables only, we found a total number of 132,062 distinct attribute names This is actually a very conservative estimation, as only a fraction (∼55%) of the whole Web table corpus are regular tables Manually... approach to extract the schemata for the whole range of Web tables (including both regular Web tables and non-regular Web tables) We first extract a variety of features which are essential for table schema discovery, and then apply a series of classification variants to distinguish the header cells from the non-header cells In addition, we also propose a set of other techniques to improve the accuracy, including . EFFECTIVE INTERPRETATION, INTEGRATION AND QUERYING OF WEB TABLES LU MEIYU Bachelor of Engineering Harbin Institute of Technology, China A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY DEPARTMENT. 130 vi ABSTRACT The World Wide Web contains a vast amount of structured and semi-structured information in the form of HTML tables (a.k.a., Web tables) . The rich infor- mation embedded in those Web tables provides. consists of three main components: Web table interpretation, integration and querying. Our first work is to present a generic solution to extract the schema (i.e., attribute names and data types) of Web