xiii, 149 p. : ill. (some col.) / Information extraction (IE) aims to recognize and retrieve certain types of information from natural language text. For instance, an information extraction system may extract key geopolitical indicators about countries from a set of web pages while ignoring other types of information. IE has existed as a research field for a few decades, and ontology-based information extraction (OBIE) has recently emerged as one of its subfields. Here, the general idea is to use ontologies--which provide formal and explicit specifications of shared conceptualizations--to guide the information extraction process. This dissertation presents two novel directions for ontology-based information extraction in which ontologies are used to improve the information extraction process. First, I describe how a component-based approach for information extraction can be designed through the use of ontologies in information extraction. A key idea in this approach is identifying components of information extraction systems which make extractions with respect to specific ontological concepts. These components are termed "information extractors". The component-based approach explores how information extractors as well as other types of components can be used in developing information extraction systems. This approach has the potential to make a significant contribution towards the widespread usage and commercialization of information extraction. Second, I describe how an ontology-based information extraction system can make use of multiple ontologies. Almost all previous systems use a single ontology, although multiple ontologies are available for most domains. Using multiple ontologies in information extraction has the potential to extract more information from text and thus leads to an improvement in performance measures. The concept of information extractor, conceived in the component-based approach for information extraction, is used in designing the principles for accommodating multiple ontologies in an ontology-based information extraction system. / Committee in charge: Dr. Dejing Dou, Chair; Dr. Arthur Farley, Member; Dr. Michal Young, Member; Dr. Monte Westerfield, Outside Member

abstract: A story is defined as "an actor(s) taking action(s) that culminates in a resolution(s)''. I present novel sets of features to facilitate story detection among text via supervised classification and further reveal different forms within stories via unsupervised clustering. First, I investigate the utility of a new set of semantic features compared to standard keyword features combined with statistical features, such as density of part-of-speech (POS) tags and named entities, to develop a story classifier. The proposed semantic features are based on <Subject, Verb, Object> triplets that can be extracted using a shallow parser. Experimental results show that a model of memory-based semantic linguistic features alongside statistical features achieves better accuracy. Next, I further improve the performance of story detection with a novel algorithm which aggregates the triplets producing generalized concepts and relations. A major challenge in automated text analysis is that different words are used for related concepts. Analyzing text at the surface level would treat related concepts (i.e. actors, actions, targets, and victims) as different objects, potentially missing common narrative patterns. The algorithm clusters <Subject, Verb, Object> triplets into generalized concepts by utilizing syntactic criteria based on common contexts and semantic corpus-based statistical criteria based on "contextual synonyms''. Generalized concepts representation of text (1) overcomes surface level differences (which arise when different keywords are used for related concepts) without drift, (2) leads to a higher-level semantic network representation of related stories, and (3) when used as features, they yield a significant (36%) boost in performance for the story detection task. Finally, I implement co-clustering based on generalized concepts/relations to automatically detect story forms. Overlapping generalized concepts and relationships correspond to archetypes/targets and actions that characterize story forms. I perform co-clustering of stories using standard unigrams/bigrams and generalized concepts. I show that the residual error of factorization with concept-based features is significantly lower than the error with standard keyword-based features. I also present qualitative evaluations by a subject matter expert, which suggest that concept-based features yield more coherent, distinctive and interesting story forms compared to those produced by using standard keyword-based features. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2016

abstract: Automating aspects of biocuration through biomedical information extraction could significantly impact biomedical research by enabling greater biocuration throughput and improving the feasibility of a wider scope. An important step in biomedical information extraction systems is named entity recognition (NER), where mentions of entities such as proteins and diseases are located within natural-language text and their semantic type is determined. This step is critical for later tasks in an information extraction pipeline, including normalization and relationship extraction. BANNER is a benchmark biomedical NER system using linear-chain conditional random fields and the rich feature set approach. A case study with BANNER locating genes and proteins in biomedical literature is described. The first corpus for disease NER adequate for use as training data is introduced, and employed in a case study of disease NER. The first corpus locating adverse drug reactions (ADRs) in user posts to a health-related social website is also described, and a system to locate and identify ADRs in social media text is created and evaluated. The rich feature set approach to creating NER feature sets is argued to be subject to diminishing returns, implying that additional improvements may require more sophisticated methods for creating the feature set. This motivates the first application of multivariate feature selection with filters and false discovery rate analysis to biomedical NER, resulting in a feature set at least 3 orders of magnitude smaller than the set created by the rich feature set approach. Finally, two novel approaches to NER by modeling the semantics of token sequences are introduced. The first method focuses on the sequence content by using language models to determine whether a sequence resembles entries in a lexicon of entity names or text from an unlabeled corpus more closely. The second method models the distributional semantics of token sequences, determining the similarity between a potential mention and the token sequences from the training data by analyzing the contexts where each sequence appears in a large unlabeled corpus. The second method is shown to improve the performance of BANNER on multiple data sets. / Dissertation/Thesis / Ph.D. Computer Science 2013

Purpose. This work addresses an escalating problem within the realm of scientific publishing, that stems from accelerated publication rates of article formats difficult to process automatically. The amount of manual labour required to organise a comprehensive corpus of relevant literature has long been impractical. This has, in effect, reduced research efficiency and delayed scientific advancement. Two complementary approaches meant to alleviate this problem are detailed and improved upon beyond the current state-of-the-art, namely logical structure recovery of articles and keyphrase extraction. Methodology. The first approach targets the issue of flat-format publishing. It performs a structural analysis of the camera-ready PDF article and recognises its fine-grained organisation over logical units. The second approach is the application of a keyphrase extraction algorithm that relies on rhetorical information from the recovered structure to better contour an article’s true points of focus. A recount of the scientific article’s function, content and structure is provided, along with insights into how different logical components such as section headings or the bibliography can be automatically identified and utilised for higher-quality keyphrase extraction. Findings. Structure recovery can be carried out independently of an article’s formatting specifics, by exploiting conventional dependencies between logical components. In addition, access to an article’s logical structure is beneficial across term extraction approaches, reducing input noise and facilitating the emphasis of regions of interest. Value. The first part of this work details a novel method for recovering the rhetorical structure of scientific articles that is competitive with state-of-the-art machine learning techniques, yet requires no layout-specific tuning or prior training. The second part showcases a keyphrase extraction algorithm that outperforms other solutions in an established benchmark, yet does not rely on collection statistics or external knowledge sources in order to be proficient.

More and more institutes and health agencies choose knowledge graphs over traditional relational databases to store semantic data. The knowledge graphs, using some form of ontology as a framework, can store domain-specific information and derive new knowledge using a reasoner. However, much of the data that must be moved to the graphs is either inside a relational database, or inside a semi-structured report. While there has been much progress in developing tools that export data from relational databases to graphs, there is a lack of progress in semantic extraction from domain-specific unstructured texts. In this thesis, a system architecture is proposed for semantic extraction from semi-structured legacy surveillance reports of infectious diseases in animals and humans in Sweden. The results were mostly positive since the system could identify 17 out of the 20 different types of relations.

Information Extraction (IE) is the process of analyzing documents and identifying desired pieces of information within them. Many IE systems have been developed over the last couple of decades, but there is still room for improvement as IE remains an open problem for researchers. This work discusses the development of a hybrid IE system that attempts to combine the strengths of rule-based and statistical IE systems while avoiding their unique pitfalls in order to achieve high performance for any type of information on any type of document. Test results show that this system operates competitively in cases where target information belongs to a highly-structured data type and when critical contextual information is in close proximity to the target.

In my dissertation, I elaborate on the work that I have done to extract temporally-anchored spatial knowledge from text, including both intra- and inter-sentential knowledge. I also detail multiple approaches to infer spatial timeline of a person from biographies and social media. I present and analyze two strategies to annotate information regarding whether a given entity is or is not located at some location, and for how long with respect to an event. Specifically, I leverage semantic roles or syntactic dependencies to generate potential spatial knowledge and then crowdsource annotations to validate the potential knowledge. The resulting annotations indicate how long entities are or are not located somewhere, and temporally anchor this spatial information. I present an in-depth corpus analysis and experiments comparing the spatial knowledge generated by manipulating roles or dependencies. In my work, I also explore research methodologies that go beyond single sentences and extract spatio-temporal information from text. Spatial timelines refer to a chronological order of locations where a target person is or is not located. I present corpus and experiments to extract spatial timelines from Wikipedia biographies. I present my work on determining locations and the order in which they are actually visited by a person from their travel experiences. Specifically, I extract spatio-temporal graphs that capture the order (edges) of locations (nodes) visited by a person. Further, I detail my experiments that leverage both text and images to extract spatial timeline of a person from Twitter.

The importance of Clinical Data Warehouses (CDW) has increased significantly in recent years as they support or enable many applications such as clinical trials, data mining, and decision making. CDWs integrate Electronic Health Records which still contain a large amount of text data, such as discharge letters or reports on diagnostic findings in addition to structured and coded data like ICD-codes of diagnoses. Existing CDWs hardly support features to gain information covered in texts. Information extraction methods offer a solution for this problem but they have a high and long development effort, which can only be carried out by computer scientists. Moreover, such systems only exist for a few medical domains. This paper presents a method empowering clinicians to extract information from texts on their own. Medical concepts can be extracted ad hoc from e.g. discharge letters, thus physicians can work promptly and autonomously. The proposed system achieves these improvements by efficient data storage, preprocessing, and with powerful query features. Negations in texts are recognized and automatically excluded, as well as the context of information is determined and undesired facts are filtered, such as historical events or references to other persons (family history). Context-sensitive queries ensure the semantic integrity of the concepts to be extracted. A new feature not available in other CDWs is to query numerical concepts in texts and even filter them (e.g. BMI > 25). The retrieved values can be extracted and exported for further analysis. This technique is implemented within the efficient architecture of the PaDaWaN CDW and evaluated with comprehensive and complex tests. The results outperform similar approaches reported in the literature. Ad hoc IE determines the results in a few (milli-) seconds and a user friendly GUI enables interactive working, allowing flexible adaptation of the extraction. In addition, the applicability of this system is demonstrated in three real-world applications at the Würzburg University Hospital (UKW). Several drug trend studies are replicated: Findings of five studies on high blood pressure, atrial fibrillation and chronic renal failure can be partially or completely confirmed in the UKW. Another case study evaluates the prevalence of heart failure in inpatient hospitals using an algorithm that extracts information with ad hoc IE from discharge letters and echocardiogram report (e.g. LVEF < 45 ) and other sources of the hospital information system. This study reveals that the use of ICD codes leads to a significant underestimation (31%) of the true prevalence of heart failure. The third case study evaluates the consistency of diagnoses by comparing structured ICD-10-coded diagnoses with the diagnoses described in the diagnostic section of the discharge letter. These diagnoses are extracted from texts with ad hoc IE, using synonyms generated with a novel method. The developed approach can extract diagnoses from the discharge letter with a high accuracy and furthermore it can prove the degree of consistency between the coded and reported diagnoses. / Die Bedeutung von Clinical Data Warehouses (CDW) hat in den letzten Jahren stark zugenommen, da sie viele Anwendungen wie klinische Studien, Data Mining und Entscheidungsfindung unterstützen oder ermöglichen. CDWs integrieren elektronische Patientenakten, die neben strukturierten und kodierten Daten wie ICD-Codes von Diagnosen immer noch sehr vielen Textdaten enthalten, sowie Arztbriefe oder Befundberichte. Bestehende CDWs unterstützen kaum Funktionen, um die in den Texten enthaltenen Informationen zu nutzen. Informationsextraktionsmethoden bieten zwar eine Lösung für dieses Problem, erfordern aber einen hohen und langen Entwicklungsaufwand, der nur von Informatikern durchgeführt werden kann. Außerdem gibt es solche Systeme nur für wenige medizinische Bereiche. Diese Arbeit stellt eine Methode vor, die es Ärzten ermöglicht, Informationen aus Texten selbstständig zu extrahieren. Medizinische Konzepte können ad hoc aus Texten (z. B. Arztbriefen) extrahiert werden, so dass Ärzte unverzüglich und autonom arbeiten können. Das vorgestellte System erreicht diese Verbesserungen durch effiziente Datenspeicherung, Vorverarbeitung und leistungsstarke Abfragefunktionen. Negationen in Texten werden erkannt und automatisch ausgeschlossen, ebenso wird der Kontext von Informationen bestimmt und unerwünschte Fakten gefiltert, wie z. B. historische Ereignisse oder ein Bezug zu anderen Personen (Familiengeschichte). Kontextsensitive Abfragen gewährleisten die semantische Integrität der zu extrahierenden Konzepte. Eine neue Funktion, die in anderen CDWs nicht verfügbar ist, ist die Abfrage numerischer Konzepte in Texten und sogar deren Filterung (z. B. BMI > 25). Die abgerufenen Werte können extrahiert und zur weiteren Analyse exportiert werden. Diese Technik wird innerhalb der effizienten Architektur des PaDaWaN-CDW implementiert und mit umfangreichen und aufwendigen Tests evaluiert. Die Ergebnisse übertreffen ähnliche Ansätze, die in der Literatur beschrieben werden. Ad hoc IE ermittelt die Ergebnisse in wenigen (Milli-)Sekunden und die benutzerfreundliche Oberfläche ermöglicht interaktives Arbeiten und eine flexible Anpassung der Extraktion. Darüber hinaus wird die Anwendbarkeit dieses Systems in drei realen Anwendungen am Universitätsklinikum Würzburg (UKW) demonstriert: Mehrere Medikationstrendstudien werden repliziert: Die Ergebnisse aus fünf Studien zu Bluthochdruck, Vorhofflimmern und chronischem Nierenversagen können in dem UKW teilweise oder vollständig bestätigt werden. Eine weitere Fallstudie bewertet die Prävalenz von Herzinsuffizienz in stationären Patienten in Krankenhäusern mit einem Algorithmus, der Informationen mit Ad-hoc-IE aus Arztbriefen, Echokardiogrammbericht und aus anderen Quellen des Krankenhausinformationssystems extrahiert (z. B. LVEF < 45). Diese Studie zeigt, dass die Verwendung von ICD-Codes zu einer signifikanten Unterschätzung (31%) der tatsächlichen Prävalenz von Herzinsuffizienz führt. Die dritte Fallstudie bewertet die Konsistenz von Diagnosen, indem sie strukturierte ICD-10-codierte Diagnosen mit den Diagnosen, die im Diagnoseabschnitt des Arztbriefes beschriebenen, vergleicht. Diese Diagnosen werden mit Ad-hoc-IE aus den Texten gewonnen, dabei werden Synonyme verwendet, die mit einer neuartigen Methode generiert werden. Der verwendete Ansatz kann Diagnosen mit hoher Genauigkeit aus Arztbriefen extrahieren und darüber hinaus den Grad der Übereinstimmung zwischen den kodierten und beschriebenen Diagnosen bestimmen.

Family history research on the web is increasing in popularity, and many competing genealogical websites host large amounts of data-rich, unstructured, primary genealogical records. It is labor-intensive, however, even after making these records machine-readable, for humans to make these records easily searchable. What we need are computer tools that can automatically produce indices and databases from these genealogical records and can automatically identify individuals and events, determine relationships, and put families together. We propose here a possible solution—specialized ontologies, built specifically for extracting information from primary genealogical records, with expert logic and rules to infer genealogical facts and assemble relationship links between persons with respect to the genealogical events in their lives. The deliverables of this solution are extraction ontologies that can extract from parish or town records, annotated versions of original documents, data files of individuals and events, and rules to infer family relationships from stored data. The solution also provides for the ability to query over the rules and data files and to obtain query-result justification linking back to primary genealogical records. An evaluation of the prototype solution shows that the extraction has excellent recall and precision results and that inferred facts are correct.

Event extraction refers to extracting specific knowledge of incidents from natural language text and consolidating it into a structured form. Some important applications of event extraction include search, retrieval, question answering and event forecasting. However, before events can be extracted it is imperative to detect events i.e. identify which documents from a large collection contain events of interest and from those extracting the sentences that might contain the event related information. This task is challenging because it is easier to obtain labels at the document level than finegrained annotations at the sentence level. Current approaches for this task are suboptimal because they directly aggregate sentence probabilities estimated by a classifier to obtain document probabilities resulting in error propagation. To alleviate this problem we propose to leverage recent advances in representation learning by using attention mechanisms. Specifically, for event detection we propose a method to compute document embeddings from sentence embeddings by leveraging attention and training a document classifier on those embeddings to mitigate the error propagation problem. However, we find that existing attention mechanisms are inept for this task, because either they are suboptimal or they use a large number of parameters. To address this problem we propose a lean attention mechanism which is effective for event detection. Current approaches for event extraction rely on finegrained labels in specific domains. Extending extraction to new domains is challenging because of difficulty of collecting finegrained data. Machine reading comprehension(MRC) based approaches, that enable zero-shot extraction struggle with syntactically complex sentences and long-range dependencies. To mitigate this problem, we propose a syntactic sentence simplification approach that is guided by MRC model to improve its performance on event extraction. / Doctor of Philosophy / Event extraction is the task of extracting events of societal importance from natural language texts. The task has a wide range of applications from search, retrieval, question answering to forecasting population level events like civil unrest, disease occurrences with reasonable accuracy. Before events can be extracted it is imperative to identify the documents that are likely to contain the events of interest and extract the sentences that mention those events. This is termed as event detection. Current approaches for event detection are suboptimal. They assume that events are neatly partitioned into sentences and obtain document level event probabilities directly from predicted sentence level probabilities. In this dissertation, under the same assumption by leveraging representation learning we mitigate some of the shortcomings of the previous event detection methods. Current approaches to event extraction are only limited to restricted domains and require finegrained labeled corpora for their training. One way to extend event extraction to new domains in by enabling zero-shot extraction. Machine reading comprehension (MRC) based approach provides a promising way forward for zero-shot extraction. However, this approach suffers from the long-range dependency problem and faces difficulty in handling syntactically complex sentences with multiple clauses. To mitigate this problem we propose a syntactic sentence simplification algorithm that is guided by the MRC system to improves its performance.