TY - THES A1 - Sidarenka, Uladzimir T1 - Sentiment analysis of German Twitter T1 - Sentimentanalyse des deutschen Twitters N2 - The immense popularity of online communication services in the last decade has not only upended our lives (with news spreading like wildfire on the Web, presidents announcing their decisions on Twitter, and the outcome of political elections being determined on Facebook) but also dramatically increased the amount of data exchanged on these platforms. Therefore, if we wish to understand the needs of modern society better and want to protect it from new threats, we urgently need more robust, higher-quality natural language processing (NLP) applications that can recognize such necessities and menaces automatically, by analyzing uncensored texts. Unfortunately, most NLP programs today have been created for standard language, as we know it from newspapers, or, in the best case, adapted to the specifics of English social media. This thesis reduces the existing deficit by entering the new frontier of German online communication and addressing one of its most prolific forms—users’ conversations on Twitter. In particular, it explores the ways and means by how people express their opinions on this service, examines current approaches to automatic mining of these feelings, and proposes novel methods, which outperform state-of-the-art techniques. For this purpose, I introduce a new corpus of German tweets that have been manually annotated with sentiments, their targets and holders, as well as lexical polarity items and their contextual modifiers. Using these data, I explore four major areas of sentiment research: (i) generation of sentiment lexicons, (ii) fine-grained opinion mining, (iii) message-level polarity classification, and (iv) discourse-aware sentiment analysis. In the first task, I compare three popular groups of lexicon generation methods: dictionary-, corpus-, and word-embedding–based ones, finding that dictionary-based systems generally yield better polarity lists than the last two groups. Apart from this, I propose a linear projection algorithm, whose results surpass many existing automatically-generated lexicons. Afterwords, in the second task, I examine two common approaches to automatic prediction of sentiment spans, their sources, and targets: conditional random fields (CRFs) and recurrent neural networks, obtaining higher scores with the former model and improving these results even further by redefining the structure of CRF graphs. When dealing with message-level polarity classification, I juxtapose three major sentiment paradigms: lexicon-, machine-learning–, and deep-learning–based systems, and try to unite the first and last of these method groups by introducing a bidirectional neural network with lexicon-based attention. Finally, in order to make the new classifier aware of microblogs' discourse structure, I let it separately analyze the elementary discourse units of each tweet and infer the overall polarity of a message from the scores of its EDUs with the help of two new approaches: latent-marginalized CRFs and Recursive Dirichlet Process. N2 - Die enorme Popularität von Online-Kommunikationsdiensten in den letzten Jahrzehnten hat nicht unser Leben massiv geändert (sodass Nachrichten sich wie Fegefeuer übers Internet ausbreiten, Präsidenten ihre Entscheidungen auf Twitter ankündigen, und Ergebnisse politischer Wahlen auf Facebook entschieden werden) sondern auch zu einem dramatischen Anstieg der Datenmenge geführt, die über solche Plattformen ausgetauscht werden. Deswegen braucht man heutzutage dringend zuverlässige, qualitätvolle NLP-Programme, um neue gesellschaftliche Bedürfnisse und Risiken in unzensierten Nutzernachrichten automatisch erkennen und abschätzen zu können. Leider sind die meisten modernen NLP-Anwendungen entweder auf die Analyse der Standardsprache (wie wir sie aus Zeitungstexten kennen) ausgerichtet oder im besten Fall an die Spezifika englischer Social Media angepasst. Diese Dissertation reduziert den bestehenden Rückstand, indem sie das "Neuland" der deutschen Online-Kommunikation betritt und sich einer seiner produktivsten Formen zuwendet—den User-Diskussionen auf Twitter. Diese Arbeit erforscht insbesondere die Art und Weise, wie Leute ihre Meinungen auf diesem Online-Service äußern, analysiert existierende Verfahren zur automatischen Erkennung ihrer Gefühle und schlägt neue Verfahren vor, die viele heutige State-of-the-Art-Systeme übertreffen. Zu diesem Zweck stelle ich ein neues Korpus deutscher Tweets vor, die manuell von zwei menschlichen Experten mit Sentimenten (polaren Meinungen), ihren Quellen (sources) und Zielen (targets) sowie lexikalischen polaren Termen und deren kontextuellen Modifizierern annotiert wurden. Mithilfe dieser Daten untersuche ich vier große Teilgebiete der Sentimentanalyse: (i) automatische Generierung von Sentiment-Lexika, (ii) aspekt-basiertes Opinion-Mining, (iii) Klassifizierung der Polarität von ganzen Nachrichten und (iv) diskurs-bewusste Sentimentanalyse. In der ersten Aufgabe vergleiche ich drei populäre Gruppen von Lexikongenerierungsmethoden: wörterbuch-, corpus- und word-embedding-basierte Verfahren, und komme zu dem Schluss, dass wörterbuch-basierte Ansätze generell bessere Polaritätslexika liefern als die letzten zwei Gruppen. Abgesehen davon, schlage ich einen neuen Linearprojektionsalgorithmus vor, dessen Resultate deutlich besser als viele automatisch generierte Polaritätslisten sind. Weiterhin, in der zweiten Aufgabe, untersuche ich zwei gängige Herangehensweisen an die automatische Erkennung der Textspannen von Sentimenten, Sources und Targets: Conditional Random Fields (CRFs) und rekurrente neuronale Netzwerke. Ich erziele bessere Ergebnisse mit der ersten Methode und verbessere diese Werte noch weiter durch alternative Topologien der CRF-Graphen. Bei der Analyse der Nachrichtenpolarität stelle ich drei große Sentiment-Paradigmen gegenüber: lexikon-, Machine-Learning–, und Deep-Learning–basierte Systeme, und versuche die erste und die letzte dieser Gruppen in einem Verfahren zu vereinigen, indem ich eine neue neuronale Netzwerkarchitektur vorschlage: bidirektionales rekurrentes Netzwerk mit lexikon-basierter Attention (LBA). Im letzten Kapitel unternehme ich einen Versuch, die Prädiktion der Gesamtpolarität von Tweets über die Diskursstruktur der Nachrichten zu informieren. Zu diesem Zweck wende ich den vorgeschlagenen LBA-Klassifikator separat auf jede einzelne elementare Diskurs-Einheit (EDU) eines Microblogs an und induziere die allgemeine semantische Ausrichtung dieser Nachricht mithilfe von zwei neuen Methoden: latenten marginalisierten CRFs und rekursivem Dirichlet-Prozess. KW - sentiment analysis KW - opinion mining KW - social media KW - Twitter KW - natural language processing KW - discourse analysis KW - NLP KW - computational linguistics KW - machine learning KW - Sentimentanalyse KW - Computerlinguistik KW - Meinungsforschung Y1 - 2019 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:kobv:517-opus4-437422 ER - TY - JOUR A1 - Wilksch, Moritz A1 - Abramova, Olga T1 - PyFin-sentiment BT - towards a machine-learning-based model for deriving sentiment from financial tweets JF - International journal of information management data insights N2 - Responding to the poor performance of generic automated sentiment analysis solutions on domain-specific texts, we collect a dataset of 10,000 tweets discussing the topics of finance and investing. We manually assign each tweet its market sentiment, i.e., the investor’s anticipation of a stock’s future return. Using this data, we show that all existing sentiment models trained on adjacent domains struggle with accurate market sentiment analysis due to the task’s specialized vocabulary. Consequently, we design, train, and deploy our own sentiment model. It outperforms all previous models (VADER, NTUSD-Fin, FinBERT, TwitterRoBERTa) when evaluated on Twitter posts. On posts from a different platform, our model performs on par with BERT-based large language models. We achieve this result at a fraction of the training and inference costs due to the model’s simple design. We publish the artifact as a python library to facilitate its use by future researchers and practitioners. KW - sentiment analysis KW - financial market sentiment KW - opinion mining KW - machine learning KW - deep learning Y1 - 2023 U6 - https://doi.org/10.1016/j.jjimei.2023.100171 SN - 2667-0968 VL - 3 IS - 1 PB - Elsevier CY - Amsterdam ER - TY - JOUR A1 - Munnes, Stefan A1 - Harsch, Corinna A1 - Knobloch, Marcel A1 - Vogel, Johannes S. A1 - Hipp, Lena A1 - Schilling, Erik T1 - Examining Sentiment in Complex Texts. A Comparison of Different Computational Approaches JF - Frontiers in Big Data N2 - Can we rely on computational methods to accurately analyze complex texts? To answer this question, we compared different dictionary and scaling methods used in predicting the sentiment of German literature reviews to the "gold standard " of human-coded sentiments. Literature reviews constitute a challenging text corpus for computational analysis as they not only contain different text levels-for example, a summary of the work and the reviewer's appraisal-but are also characterized by subtle and ambiguous language elements. To take the nuanced sentiments of literature reviews into account, we worked with a metric rather than a dichotomous scale for sentiment analysis. The results of our analyses show that the predicted sentiments of prefabricated dictionaries, which are computationally efficient and require minimal adaption, have a low to medium correlation with the human-coded sentiments (r between 0.32 and 0.39). The accuracy of self-created dictionaries using word embeddings (both pre-trained and self-trained) was considerably lower (r between 0.10 and 0.28). Given the high coding intensity and contingency on seed selection as well as the degree of data pre-processing of word embeddings that we found with our data, we would not recommend them for complex texts without further adaptation. While fully automated approaches appear not to work in accurately predicting text sentiments with complex texts such as ours, we found relatively high correlations with a semiautomated approach (r of around 0.6)-which, however, requires intensive human coding efforts for the training dataset. In addition to illustrating the benefits and limits of computational approaches in analyzing complex text corpora and the potential of metric rather than binary scales of text sentiment, we also provide a practical guide for researchers to select an appropriate method and degree of pre-processing when working with complex texts. KW - sentiment analysis KW - German literature KW - dictionary KW - word embeddings KW - automated text analysis KW - computer-assisted text analysis KW - scaling method Y1 - 2022 U6 - https://doi.org/10.3389/fdata.2022.886362 SN - 2624-909X VL - 5 PB - Frontiers Media CY - Lausanne ER - TY - CHAP A1 - Hagemann, Linus A1 - Abramova, Olga T1 - Crafting audience engagement in social media conversations BT - evidence from the U.S. 2020 presidential elections T2 - Proceedings of the 55th Hawaii International Conference on System Sciences N2 - Observing inconsistent results in prior studies, this paper applies the elaboration likelihood model to investigate the impact of affective and cognitive cues embedded in social media messages on audience engagement during a political event. Leveraging a rich dataset in the context of the 2020 U.S. presidential elections containing more than 3 million tweets, we found the prominence of both cue types. For the overall sample, positivity and sentiment are negatively related to engagement. In contrast, the post-hoc sub-sample analysis of tweets from famous users shows that emotionally charged content is more engaging. The role of sentiment decreases when the number of followers grows and ultimately becomes insignificant for Twitter participants with a vast number of followers. Prosocial orientation (“we-talk”) is consistently associated with more likes, comments, and retweets in the overall sample and sub-samples. KW - mediated conversation KW - big data KW - engagement KW - sentiment analysis KW - social media Y1 - 2022 SN - 978-0-9981331-5-7 SP - 3222 EP - 3231 PB - HICSS Conference Office University of Hawaii at Manoa CY - Honolulu ER - TY - JOUR A1 - Stede, Manfred T1 - Automatic argumentation mining and the role of stance and sentiment JF - Journal of argumentation in context N2 - Argumentation mining is a subfield of Computational Linguistics that aims (primarily) at automatically finding arguments and their structural components in natural language text. We provide a short introduction to this field, intended for an audience with a limited computational background. After explaining the subtasks involved in this problem of deriving the structure of arguments, we describe two other applications that are popular in computational linguistics: sentiment analysis and stance detection. From the linguistic viewpoint, they concern the semantics of evaluation in language. In the final part of the paper, we briefly examine the roles that these two tasks play in argumentation mining, both in current practice, and in possible future systems. KW - argumentation structure KW - argumentation mining KW - sentiment analysis KW - stance detection Y1 - 2020 U6 - https://doi.org/10.1075/jaic.00006.ste SN - 2211-4742 SN - 2211-4750 VL - 9 IS - 1 SP - 19 EP - 41 PB - John Benjamins Publishing Co. CY - Amsterdam ER -