TY - THES A1 - Krug, Markus T1 - Techniques for the Automatic Extraction of Character Networks in German Historic Novels T1 - Techniken zur automatischen Extraktion von Figurennetzwerken aus deutschen Romanen N2 - Recent advances in Natural Language Preprocessing (NLP) allow for a fully automatic extraction of character networks for an incoming text. These networks serve as a compact and easy to grasp representation of literary fiction. They offer an aggregated view of the text, which can be used during distant reading approaches for the analysis of literary hypotheses. In their core, the networks consist of nodes, which represent literary characters, and edges, which represent relations between characters. For an automatic extraction of such a network, the first step is the detection of the references of all fictional entities that are of importance for a text. References to the fictional entities appear in the form of names, noun phrases and pronouns and prior to this work, no components capable of automatic detection of character references were available. Existing tools are only capable of detecting proper nouns, a subset of all character references. When evaluated on the task of detecting proper nouns in the domain of literary fiction, they still underperform at an F1-score of just about 50%. This thesis uses techniques from the field of semi-supervised learning, such as Distant supervision and Generalized Expectations, and improves the results of an existing tool to about 82%, when evaluated on all three categories in literary fiction, but without the need for annotated data in the target domain. However, since this quality is still not sufficient, the decision to annotate DROC, a corpus comprising 90 fragments of German novels was made. This resulted in a new general purpose annotation environment titled as ATHEN, as well as annotated data that spans about 500.000 tokens in total. Using this data, the combination of supervised algorithms and a tailored rule based algorithm, which in combination are able to exploit both - local consistencies as well as global consistencies - yield an algorithm with an F1-score of about 93%. This component is referred to as the Kallimachos tagger. A character network can not directly display references however, instead they need to be clustered so that all references that belong to a real world or fictional entity are grouped together. This process widely known as coreference resolution is a hard problem in the focus of research for more than half a century. This work experimented with adaptations of classical feature based machine learning, with a dedicated rule based algorithm and with modern techniques of Deep Learning, but no approach can surpass 55% B-Cubed F1, when evaluated on DROC. Due to this barrier, many researchers do not use a fully-fledged coreference resolution when they extract character networks, but only focus on a more forgiving subset- the names. For novels such as Alice's Adventures in Wonderland by Lewis Caroll, this would however only result in a network in which many important characters are missing. In order to integrate important characters into the network that are not named by the author, this work makes use of automatic detection of speaker and addressees for direct speech utterances (all entities involved in a dialog are considered to be of importance). This problem is by itself not an easy task, however the most successful system analysed in this thesis is able to correctly determine the speaker to about 85% of the utterances as well as about 65% of the addressees. This speaker information can not only help to identify the most dominant characters, but also serves as a way to model the relations between entities. During the span of this work, components have been developed to model relations between characters using speaker attribution, using co-occurrences as well as by the usage of true interactions, for which yet again a dataset was annotated using ATHEN. Furthermore, since relations between characters are usually typed, a component for the extraction of a typed relation was developed. Similar to the experiments for the character reference detection, a combination of a rule based and a Maximum Entropy classifier yielded the best overall results, with the extraction of family relations showing a score of about 80% and the quality of love relations with a score of about 50%. For family relations, a kernel for a Support Vector Machine was developed that even exceeded the scores of the combined approach but is behind on the other labels. In addition, this work presents new ways to evaluate automatically extracted networks without the need of domain experts, instead it relies on the usage of expert summaries. It also refrains from the uses of social network analysis for the evaluation, but instead presents ranked evaluations using Precision@k and the Spearman Rank correlation coefficient for the evaluation of the nodes and edges of the network. An analysis using these metrics showed, that the central characters of a novel are contained with high probability but the quality drops rather fast if more than five entities are analyzed. The quality of the edges is mainly dominated by the quality of the coreference resolution and the correlation coefficient between gold edges and system edges therefore varies between 30 and 60%. All developed components are aggregated alongside a large set of other preprocessing modules in the Kallimachos pipeline and can be reused without any restrictions. N2 - Techniken zur automatischen Extraktion von Figurennetzwerken aus deutschen Romanen KW - Textanalyse KW - Character Networks KW - Coreference KW - Character Reference Detection KW - Relation Detection KW - Quotation Attribution KW - Netzwerkanalyse KW - Digital Humanities KW - Netzwerk Y1 - 2020 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:20-opus-209186 ER - TY - THES A1 - Reger, Isabella T1 - Figurennetzwerke als Ähnlichkeitsmaß T1 - Character networks as a measure of similarity N2 - Die vorliegende Arbeit lässt sich dem Bereich der quantitativen Literaturanalyse zuordnen und verfolgt das Ziel, mittels computergestützter Verfahren zu untersuchen, inwieweit sich Romane hinsichtlich ihrer Figurenkonstellation ähneln. Dazu wird die Figurenkonstellation, als wichtiges strukturgebendes Ordnungsprinzip eines Romans, als soziales Netzwerk der Figuren operationalisiert. Solche Netzwerke können unter Anwendung von Verfahren des Natural Language Processing automatisch aus dem Text erstellt werden. Als Datengrundlage dient ein Korpus von deutschsprachigen Romanen aus dem 19. Jahrhundert, das mit automatischen Verfahren zur Figurenerkennung und Koreferenzauflösung prozessiert und manuell nachkorrigiert wurde, um eine möglichst saubere Datenbasis zu schaffen. Ausgehend von der intensiven vergleichenden Betrachtung der Figurenkonstellationen von Fontanes "Effi Briest" und Flauberts "Madame Bovary" wurde in einer manuell erstellten Distanzmatrix die menschliche Intuition solcher Ähnlichkeit zwischen allen Romanen des Korpus festgehalten, basierend auf der Lektüre von Zusammenfassungen der Romane. Diese Daten werden als Evaluationsgrundlage genutzt. Mit Hilfe von Methoden der sozialen Netzwerkanalyse können strukturelle Eigenschaften dieser Netzwerke als Features erhoben werden. Diese wurden anschließend zur Berechnung der Kosinusdistanz zwischen den Romanen verwendet. Obwohl die automatisch erstellten Netzwerke die Figurenkonstellationen der Romane im Allgemeinen gut widerspiegeln und die Netzwerkfeatures sinnvoll interpretierbar sind, war die Korrelation mit der Evaluationsgrundlage niedrig. Dies legt die Vermutung nahe, dass neben der Struktur der Figurenkonstellation auch wiederkehrende Themen und Motive die Erstellung der Evaluationsgrundlage unterbewusst beeinflusst haben. Daher wurde Topic Modeling angewendet, um wichtige zwischenmenschliche Motive zu modellieren, die für die Figurenkonstellation von Bedeutung sein können. Die Netzwerkfeatures und die Topic-Verteilung wurden in Kombination zur Distanzberechnung herangezogen. Außerdem wurde versucht, jeder Kante des Figurennetzwerks ein Topic zuzuordnen, das diese Kante inhaltlich beschreibt. Hier zeigte sich, dass einerseits Topics, die sehr spezifisch für bestimmte Texte sind, und andererseits Topics, die über alle Texte hinweg stark vertreten sind, das Ergebnis bestimmen, sodass wiederum keine, bzw. nur eine sehr schwache Korrelation mit der Evaluationsgrundlage gefunden werden konnte. Der Umstand, dass keine Verbindung zwischen den berechneten Distanzen und der Evaluationsgrundlage gefunden werden konnte, obwohl die einzelnen Features sinnvoll interpretierbar sind, lässt Zweifel an der Evaluationsmatrix aufkommen. Diese scheint stärker als zu Beginn angenommen unterbewusst von thematischen und motivischen Ähnlichkeiten zwischen den Romanen beeinflusst zu sein. Auch die Qualität der jeweiligen Zusammenfassung hat hier einen nicht unwesentlichen Einfluss. Daher wäre eine weniger subjektiv geprägte Möglichkeit der Auswertung von Nöten, beispielsweise durch die parallele Einschätzung mehrerer Annotatoren. Auch die weitere Verbesserung von NLP-Verfahren für literarische Texte in deutscher Sprache ist ein Desideratum für anknüpfende Forschungsansätze. N2 - This thesis is a work in the field of Digital Literary Studies with the goal of computationally analyzing the similarity of novels with regard to their character constellation. The character constellation, as an important structural arrangement in a novel, is operationalized as a social network of these characters. Such networks can be generated automatically from a literary text using Natural Language Processing techniques. The work is based on a corpus of German novels of the 19th century that have been preprocessed using automatic methods of character identification and coreference resolution. The results have been manually corrected in order to ensure the best possible data quality. Starting from an intensive comparative contemplation of the character constellations in “Effi Briest” by Theodor Fontane and “Madame Bovary” by Gustave Flaubert, a distance matrix capturing the human intuition of similarity between character constellations of different novels was devised manually, based on summaries of the respective novels. This distance matrix serves as a means of evaluation. Using methods of social network analysis, structural properties of character networks can be modeled as features and used to compute cosine distances between the novels. The automatically generated networks are generally an adequate representation of the character constellations of the novels and the network features are meaningfully interpretable. Nevertheless, the correlation with the evaluation matrix was low. This raises the assumption that, apart from the structure of the character constellation, other properties such as recurring themes or motives may have subconsciously influenced the manual creation of the evaluation matrix. Therefore, topic modeling was used to represent important interpersonal motives that might be of importance for the character constellation. The network features and the topic distribution were used in combination for the computation of distances. Moreover, each edge in a character network was associated with a topic, trying to describe the kind of the relation. It could be observed that the result is heavily influenced on the one hand by topics that are very specific for a single text and on the other hand by topics featured strongly in all texts, so that, again, no or only a very weak correlation with the evaluation matrix could be found. The fact that there was no obvious relation between the computed distances and the manual distance matrix, even though the individual features are meaningfully interpretable, raises doubts about the evaluation matrix. It seems to be more strongly influenced by subconscious thematic and motivic similarities between the novels than assumed. The quality and extent of the respective summaries used in the creation of the manual matrix also have a substantial influence. Hence, a less subjective way of evaluation is needed, for example by parallel estimation of the distances by several annotators. Further improvement of the quality of NLP methods for literary texts in German is also a desideratum for following research. KW - Digital Humanities KW - Netzwerkanalyse KW - Literaturwissenschaft KW - Figurennetzwerke KW - Topic Modeling KW - Figurenkonstellation KW - Digitale Textanalyse Y1 - 2016 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:20-opus-149106 ER -