The Web provides a platform for people to share their data, leading to an abundance of accessible information. In recent years, significant research effort has been directed especially at tables on the Web, which form a rich resource for factual and relational data. Applications such as fact search and knowledge base construction benefit from this data, as it is often less ambiguous than unstructured text. However, many traditional information extraction and retrieval techniques are not well suited for Web tables, as they generally do not consider the role of the table structure in reflecting the semantics of the content. Tables provide a compact representation of similarly structured data. Yet, on the Web, tables are very heterogeneous, often with ambiguous semantics and inconsistencies in the quality of the data. Consequently, recognizing the structure and inferring the semantics of these tables is a challenging task that requires a designated table recovery and understanding process. In the literature, many important contributions have been made to implement such a table understanding process that specifically targets Web tables, addressing tasks such as table detection or header recovery. However, the precision and coverage of the data extracted from Web tables is often still quite limited. Due to the complexity of Web table understanding, many techniques developed so far make simplifying assumptions about the table layout or content to limit the amount of contributing factors that must be considered. Thanks to these assumptions, many sub-tasks become manageable. However, the resulting algorithms and techniques often have a limited scope, leading to imprecise or inaccurate results when applied to tables that do not conform to these assumptions. In this thesis, our objective is to extend the Web table understanding process with techniques that enable some of these assumptions to be relaxed, thus improving the scope and accuracy. We have conducted a comprehensive analysis of tables available on the Web to examine the characteristic features of these tables, but also identify unique challenges that arise from these characteristics in the table understanding process. To extend the scope of the table understanding process, we introduce extensions to the sub-tasks of table classification and conceptualization. First, we review various table layouts and evaluate alternative approaches to incorporate layout classification into the process. Instead of assuming a single, uniform layout across all tables, recognizing different table layouts enables a wide range of tables to be analyzed in a more accurate and systematic fashion. In addition to the layout, we also consider the conceptual level. To relax the single concept assumption, which expects all attributes in a table to describe the same semantic concept, we propose a semantic normalization approach. By decomposing multi-concept tables into several single-concept tables, we further extend the range of Web tables that can be processed correctly, enabling existing techniques to be applied without significant changes. Furthermore, we address the quality of data extracted from Web tables, by studying the role of context information. Supplementary information from the context is often required to correctly understand the table content, however, the verbosity of the surrounding text can also mislead any table relevance decisions. We first propose a selection algorithm to evaluate the relevance of context information with respect to the table content in order to reduce the noise. Then, we introduce a set of extraction techniques to recover attribute-specific information from the relevant context in order to provide a richer description of the table content. With the extensions proposed in this thesis, we increase the scope and accuracy of Web table understanding, leading to a better utilization of the information contained in tables on the Web.

The Web provides a platform for people to share their data, leading to an abundance of accessible information. In recent years, significant research effort has been directed especially at tables on the Web, which form a rich resource for factual and relational data. Applications such as fact search and knowledge base construction benefit from this data, as it is often less ambiguous than unstructured text. However, many traditional information extraction and retrieval techniques are not well suited for Web tables, as they generally do not consider the role of the table structure in reflecting the semantics of the content. Tables provide a compact representation of similarly structured data. Yet, on the Web, tables are very heterogeneous, often with ambiguous semantics and inconsistencies in the quality of the data. Consequently, recognizing the structure and inferring the semantics of these tables is a challenging task that requires a designated table recovery and understanding process. In the literature, many important contributions have been made to implement such a table understanding process that specifically targets Web tables, addressing tasks such as table detection or header recovery. However, the precision and coverage of the data extracted from Web tables is often still quite limited. Due to the complexity of Web table understanding, many techniques developed so far make simplifying assumptions about the table layout or content to limit the amount of contributing factors that must be considered. Thanks to these assumptions, many sub-tasks become manageable. However, the resulting algorithms and techniques often have a limited scope, leading to imprecise or inaccurate results when applied to tables that do not conform to these assumptions. In this thesis, our objective is to extend the Web table understanding process with techniques that enable some of these assumptions to be relaxed, thus improving the scope and accuracy. We have conducted a comprehensive analysis of tables available on the Web to examine the characteristic features of these tables, but also identify unique challenges that arise from these characteristics in the table understanding process. To extend the scope of the table understanding process, we introduce extensions to the sub-tasks of table classification and conceptualization. First, we review various table layouts and evaluate alternative approaches to incorporate layout classification into the process. Instead of assuming a single, uniform layout across all tables, recognizing different table layouts enables a wide range of tables to be analyzed in a more accurate and systematic fashion. In addition to the layout, we also consider the conceptual level. To relax the single concept assumption, which expects all attributes in a table to describe the same semantic concept, we propose a semantic normalization approach. By decomposing multi-concept tables into several single-concept tables, we further extend the range of Web tables that can be processed correctly, enabling existing techniques to be applied without significant changes. Furthermore, we address the quality of data extracted from Web tables, by studying the role of context information. Supplementary information from the context is often required to correctly understand the table content, however, the verbosity of the surrounding text can also mislead any table relevance decisions. We first propose a selection algorithm to evaluate the relevance of context information with respect to the table content in order to reduce the noise. Then, we introduce a set of extraction techniques to recover attribute-specific information from the relevant context in order to provide a richer description of the table content. With the extensions proposed in this thesis, we increase the scope and accuracy of Web table understanding, leading to a better utilization of the information contained in tables on the Web.

Spreadsheets compose a valuable and notably large dataset of documents within many enterprise organizations and on the Web. Although spreadsheets are intuitive to use and equipped with powerful functionalities, extraction and transformation of the data remain a cumbersome and mostly manual task. The great flexibility they provide to the user results in data that is arbitrarily structured and hard to process for other applications. In this paper, we propose a novel architecture that combines supervised layout inference and multimodal candidate classification to allow knowledge base augmentation from arbitrary spreadsheets. In our design, we consider the need for repairing misclassifications and allow for verification and ranking of ambiguous candidates. We evaluate the performance of our system on two datasets, one with single-table spreadsheets, another with spreadsheets of arbitrary format. The evaluation result shows that the proposed system achieves similar performance on single-table spreadsheets compared to state-of-the-art rule-based solutions. Additionally, the flexibility of the system allows us to process arbitrary spreadsheet formats, including horizontally and vertically aligned tables, multiple worksheets, and contextualizing metadata. This was not possible with existing purely text-based or table-based solutions. The experiments demonstrate that it can achieve high effectiveness with an F1 score of 95.71 on arbitrary spreadsheets that require the interpretation of surrounding metadata. The precision of the system can be further increased by applying candidate schema-matching based on semantic similarity of column headers. / Kalkylblad består av ett värdefullt och särskilt stort datasätt av dokument inom många företagsorganisationer och på webben. Även om kalkylblad är intuitivt att använda och är utrustad med kraftfulla funktioner, utvinning och transformation av data är fortfarande en besvärlig och manuell uppgift. Den stora flexibiliteten som de ger användaren resulterar i data som är godtyckligt strukturerade och svåra att bearbeta för andra applikationer. I det här förslaget föreslår vi en ny arkitektur som kombinerar övervakad layoutinferens och multimodal kandidatklassificering för att tillåta kunskapsbasförstärkning från godtyckliga kalkylblad. I vår design överväger vi behovet av att reparera felklassificeringar och möjliggöra verifiering och rangordning av tvetydiga kandidater. Vi utvärderar systemets utförande på två datasätt, en med singeltabellkalkylblad, en annan med kalkylblad av godtyckligt format. Utvärderingsresultatet visar att det föreslagna systemet uppnår liknande prestanda på singel-tabellkalkylblad jämfört med state-of-the-art regelbaserade lösningar. Dessutom tillåter systemets flexibilitet oss att bearbeta godtyckliga kalkylark format, inklusive horisontella och vertikala inriktade tabeller, flera kalkylblad och sammanhangsförande metadata. Detta var inte möjligt med existerande rent textbaserade eller tabellbaserade lösningar. Experimenten visar att det kan uppnå hög effektivitet med en F1-poäng på 95.71 på godtyckliga kalkylblad som kräver tolkning av omgivande metadata. Systemets precision kan ökas ytterligare genom att applicera schema-matchning av kandidater baserat på semantisk likhet mellan kolumnrubriker.