Felix Hamborg

Jonas Golde, Patrick Haller, Felix Hamborg et al.

Most NLP tasks are modeled as supervised learning and thus require labeled training data to train effective models. However, manually producing such data at sufficient quality and quantity is known to be costly and time-intensive. Current research addresses this bottleneck by exploring a novel paradigm called zero-shot learning via dataset generation. Here, a powerful LLM is prompted with a task description to generate labeled data that can be used to train a downstream NLP model. For instance, an LLM might be prompted to "generate 500 movie reviews with positive overall sentiment, and another 500 with negative sentiment." The generated data could then be used to train a binary sentiment classifier, effectively leveraging an LLM as a teacher to a smaller student model. With this demo, we introduce Fabricator, an open-source Python toolkit for dataset generation. Fabricator implements common dataset generation workflows, supports a wide range of downstream NLP tasks (such as text classification, question answering, and entity recognition), and is integrated with well-known libraries to facilitate quick experimentation. With Fabricator, we aim to support researchers in conducting reproducible dataset generation experiments using LLMs and help practitioners apply this approach to train models for downstream tasks.

Anastasia Zhukova, Felix Hamborg, Karsten Donnay et al.

Cross-document coreference resolution (CDCR) identifies and links mentions of the same entities and events across related documents, enabling content analysis that aggregates information at the level of discourse participants. However, existing datasets primarily focus on event resolution and employ a narrow definition of coreference, which limits their effectiveness in analyzing diverse and polarized news coverage where wording varies widely. This paper proposes a revised CDCR annotation scheme of the NewsWCL50 dataset, treating coreference chains as discourse elements (DEs) and conceptual units of analysis. The approach accommodates both identity and near-identity relations, e.g., by linking "the caravan" - "asylum seekers" - "those contemplating illegal entry", allowing models to capture lexical diversity and framing variation in media discourse, while maintaining the fine-grained annotation of DEs. We reannotate the NewsWCL50 and a subset of ECB+ using a unified codebook and evaluate the new datasets through lexical diversity metrics and a same-head-lemma baseline. The results show that the reannotated datasets align closely, falling between the original ECB+ and NewsWCL50, thereby supporting balanced and discourse-aware CDCR research in the news domain.

Felix Hamborg, Corinna Breitinger, Bela Gipp

Event extraction from news articles is a commonly required prerequisite for various tasks, such as article summarization, article clustering, and news aggregation. Due to the lack of universally applicable and publicly available methods tailored to news datasets, many researchers redundantly implement event extraction methods for their own projects. The journalistic 5W1H questions are capable of describing the main event of an article, i.e., by answering who did what, when, where, why, and how. We provide an in-depth description of an improved version of Giveme5W1H, a system that uses syntactic and domain-specific rules to automatically extract the relevant phrases from English news articles to provide answers to these 5W1H questions. Given the answers to these questions, the system determines an article's main event. In an expert evaluation with three assessors and 120 articles, we determined an overall precision of p=0.73, and p=0.82 for answering the first four W questions, which alone can sufficiently summarize the main event reported on in a news article. We recently made our system publicly available, and it remains the only universal open-source 5W1H extractor capable of being applied to a wide range of use cases in news analysis.

Moritz Schubotz, Philipp Scharpf, Kaushal Dudhat et al.

We present an open source math-aware Question Answering System based on Ask Platypus. Our system returns as a single mathematical formula for a natural language question in English or Hindi. This formulae originate from the knowledge-base Wikidata. We translate these formulae to computable data by integrating the calculation engine sympy into our system. This way, users can enter numeric values for the variables occurring in the formula. Moreover, the system loads numeric values for constants occurring in the formula from Wikidata. In a user study, our system outperformed a commercial computational mathematical knowledge engine by 13%. However, the performance of our system heavily depends on the size and quality of the formula data available in Wikidata. Since only a few items in Wikidata contained formulae when we started the project, we facilitated the import process by suggesting formula edits to Wikidata editors. With the simple heuristic that the first formula is significant for the article, 80% of the suggestions were correct.

Jonas Golde, Felix Hamborg, Alan Akbik

Few-shot named entity recognition (NER) detects named entities within text using only a few annotated examples. One promising line of research is to leverage natural language descriptions of each entity type: the common label PER might, for example, be verbalized as ''person entity.'' In an initial label interpretation learning phase, the model learns to interpret such verbalized descriptions of entity types. In a subsequent few-shot tagset extension phase, this model is then given a description of a previously unseen entity type (such as ''music album'') and optionally a few training examples to perform few-shot NER for this type. In this paper, we systematically explore the impact of a strong semantic prior to interpret verbalizations of new entity types by massively scaling up the number and granularity of entity types used for label interpretation learning. To this end, we leverage an entity linking benchmark to create a dataset with orders of magnitude of more distinct entity types and descriptions as currently used datasets. We find that this increased signal yields strong results in zero- and few-shot NER in in-domain, cross-domain, and even cross-lingual settings. Our findings indicate significant potential for improving few-shot NER through heuristical data-based optimization.

Anastasia Zhukova, Terry Ruas, Felix Hamborg et al.

In a world overwhelmed with news, determining which information comes from reliable sources or how neutral is the reported information in the news articles poses a challenge to news readers. In this paper, we propose a methodology for automatically identifying bias by commission, omission, and source selection (COSS) as a joint three-fold objective, as opposed to the previous work separately addressing these types of bias. In a pipeline concept, we describe the goals and tasks of its steps toward bias identification and provide an example of a visualization that leverages the extracted features and patterns of text reuse.

David Schulte, Felix Hamborg, Alan Akbik

Intermediate task transfer learning can greatly improve model performance. If, for example, one has little training data for emotion detection, first fine-tuning a language model on a sentiment classification dataset may improve performance strongly. But which task to choose for transfer learning? Prior methods producing useful task rankings are infeasible for large source pools, as they require forward passes through all source language models. We overcome this by introducing Embedding Space Maps (ESMs), light-weight neural networks that approximate the effect of fine-tuning a language model. We conduct the largest study on NLP task transferability and task selection with 12k source-target pairs. We find that applying ESMs on a prior method reduces execution time and disk space usage by factors of 10 and 278, respectively, while retaining high selection performance (avg. regret@5 score of 2.95).

Franziska Weeber, Felix Hamborg, Karsten Donnay et al.

Large amounts of annotated data have become more important than ever, especially since the rise of deep learning techniques. However, manual annotations are costly. We propose a tool that enables researchers to create large, high-quality, annotated datasets with only a few manual annotations, thus strongly reducing annotation cost and effort. For this purpose, we combine an active learning (AL) approach with a pre-trained language model to semi-automatically identify annotation categories in the given text documents. To highlight our research direction's potential, we evaluate the approach on the task of identifying frames in news articles. Our preliminary results show that employing AL strongly reduces the number of annotations for correct classification of even these complex and subtle frames. On the framing dataset, the AL approach needs only 16.3\% of the annotations to reach the same performance as a model trained on the full dataset.

Timo Spinde, Christina Kreuter, Wolfgang Gaissmaier et al.

Media coverage possesses a substantial effect on the public perception of events. The way media frames events can significantly alter the beliefs and perceptions of our society. Nevertheless, nearly all media outlets are known to report news in a biased way. While such bias can be introduced by altering the word choice or omitting information, the perception of bias also varies largely depending on a reader's personal background. Therefore, media bias is a very complex construct to identify and analyze. Even though media bias has been the subject of many studies, previous assessment strategies are oversimplified, lack overlap and empirical evaluation. Thus, this study aims to develop a scale that can be used as a reliable standard to evaluate article bias. To name an example: Intending to measure bias in a news article, should we ask, "How biased is the article?" or should we instead ask, "How did the article treat the American president?". We conducted a literature search to find 824 relevant questions about text perception in previous research on the topic. In a multi-iterative process, we summarized and condensed these questions semantically to conclude a complete and representative set of possible question types about bias. The final set consisted of 25 questions with varying answering formats, 17 questions using semantic differentials, and six ratings of feelings. We tested each of the questions on 190 articles with overall 663 participants to identify how well the questions measure an article's perceived bias. Our results show that 21 final items are suitable and reliable for measuring the perception of media bias. We publish the final set of questions on http://bias-question-tree.gipplab.org/.

Timo Spinde, Lada Rudnitckaia, Felix Hamborg et al.

Slanted news coverage, also called media bias, can heavily influence how news consumers interpret and react to the news. To automatically identify biased language, we present an exploratory approach that compares the context of related words. We train two word embedding models, one on texts of left-wing, the other on right-wing news outlets. Our hypothesis is that a word's representations in both word embedding spaces are more similar for non-biased words than biased words. The underlying idea is that the context of biased words in different news outlets varies more strongly than the one of non-biased words, since the perception of a word as being biased differs depending on its context. While we do not find statistical significance to accept the hypothesis, the results show the effectiveness of the approach. For example, after a linear mapping of both word embeddings spaces, 31% of the words with the largest distances potentially induce bias. To improve the results, we find that the dataset needs to be significantly larger, and we derive further methodology as future research direction. To our knowledge, this paper presents the first in-depth look at the context of bias words measured by word embeddings.

Anastasia Zhukova, Felix Hamborg, Bela Gipp

Named entity recognition (NER) is an important task that aims to resolve universal categories of named entities, e.g., persons, locations, organizations, and times. Despite its common and viable use in many use cases, NER is barely applicable in domains where general categories are suboptimal, such as engineering or medicine. To facilitate NER of domain-specific types, we propose ANEA, an automated (named) entity annotator to assist human annotators in creating domain-specific NER corpora for German text collections when given a set of domain-specific texts. In our evaluation, we find that ANEA automatically identifies terms that best represent the texts' content, identifies groups of coherent terms, and extracts and assigns descriptive labels to these groups, i.e., annotates text datasets into the domain (named) entities.

Felix Hamborg, Kim Heinser, Anastasia Zhukova et al.

Media bias and its extreme form, fake news, can decisively affect public opinion. Especially when reporting on policy issues, slanted news coverage may strongly influence societal decisions, e.g., in democratic elections. Our paper makes three contributions to address this issue. First, we present a system for bias identification, which combines state-of-the-art methods from natural language understanding. Second, we devise bias-sensitive visualizations to communicate bias in news articles to non-expert news consumers. Third, our main contribution is a large-scale user study that measures bias-awareness in a setting that approximates daily news consumption, e.g., we present respondents with a news overview and individual articles. We not only measure the visualizations' effect on respondents' bias-awareness, but we can also pinpoint the effects on individual components of the visualizations by employing a conjoint design. Our bias-sensitive overviews strongly and significantly increase bias-awareness in respondents. Our study further suggests that our content-driven identification method detects groups of similarly slanted news articles due to substantial biases present in individual news articles. In contrast, the reviewed prior work rather only facilitates the visibility of biases, e.g., by distinguishing left- and right-wing outlets.

Felix Hamborg, Timo Spinde, Kim Heinser et al.

Slanted news coverage strongly affects public opinion. This is especially true for coverage on politics and related issues, where studies have shown that bias in the news may influence elections and other collective decisions. Due to its viable importance, news coverage has long been studied in the social sciences, resulting in comprehensive models to describe it and effective yet costly methods to analyze it, such as content analysis. We present an in-progress system for news recommendation that is the first to automate the manual procedure of content analysis to reveal person-targeting biases in news articles reporting on policy issues. In a large-scale user study, we find very promising results regarding this interdisciplinary research direction. Our recommender detects and reveals substantial frames that are actually present in individual news articles. In contrast, prior work rather only facilitates the visibility of biases, e.g., by distinguishing left- and right-wing outlets. Further, our study shows that recommending news articles that differently frame an event significantly improves respondents' awareness of bias.

Anastasia Zhukova, Felix Hamborg, Karsten Donnay et al.

Datasets and methods for cross-document coreference resolution (CDCR) focus on events or entities with strict coreference relations. They lack, however, annotating and resolving coreference mentions with more abstract or loose relations that may occur when news articles report about controversial and polarized events. Bridging and loose coreference relations trigger associations that may lead to exposing news readers to bias by word choice and labeling. For example, coreferential mentions of "direct talks between U.S. President Donald Trump and Kim" such as "an extraordinary meeting following months of heated rhetoric" or "great chance to solve a world problem" form a more positive perception of this event. A step towards bringing awareness of bias by word choice and labeling is the reliable resolution of coreferences with high lexical diversity. We propose an unsupervised method named XCoref, which is a CDCR method that capably resolves not only previously prevalent entities, such as persons, e.g., "Donald Trump," but also abstractly defined concepts, such as groups of persons, "caravan of immigrants," events and actions, e.g., "marching to the U.S. border." In an extensive evaluation, we compare the proposed XCoref to a state-of-the-art CDCR method and a previous method TCA that resolves such complex coreference relations and find that XCoref outperforms these methods. Outperforming an established CDCR model shows that the new CDCR models need to be evaluated on semantically complex mentions with more loose coreference relations to indicate their applicability of models to resolve mentions in the "wild" of political news articles.

Anastasia Zhukova, Felix Hamborg, Bela Gipp

Established cross-document coreference resolution (CDCR) datasets contain event-centric coreference chains of events and entities with identity relations. These datasets establish strict definitions of the coreference relations across related tests but typically ignore anaphora with more vague context-dependent loose coreference relations. In this paper, we qualitatively and quantitatively compare the annotation schemes of ECB+, a CDCR dataset with identity coreference relations, and NewsWCL50, a CDCR dataset with a mix of loose context-dependent and strict coreference relations. We propose a phrasing diversity metric (PD) that encounters for the diversity of full phrases unlike the previously proposed metrics and allows to evaluate lexical diversity of the CDCR datasets in a higher precision. The analysis shows that coreference chains of NewsWCL50 are more lexically diverse than those of ECB+ but annotating of NewsWCL50 leads to the lower inter-coder reliability. We discuss the different tasks that both CDCR datasets create for the CDCR models, i.e., lexical disambiguation and lexical diversity. Finally, to ensure generalizability of the CDCR models, we propose a direction for CDCR evaluation that combines CDCR datasets with multiple annotation schemes that focus of various properties of the coreference chains.

Anastasia Zhukova, Felix Hamborg, Karsten Donnay et al.

Unsupervised concept identification through clustering, i.e., identification of semantically related words and phrases, is a common approach to identify contextual primitives employed in various use cases, e.g., text dimension reduction, i.e., replace words with the concepts to reduce the vocabulary size, summarization, and named entity resolution. We demonstrate the first results of an unsupervised approach for the identification of groups of persons as actors extracted from a set of related articles. Specifically, the approach clusters mentions of groups of persons that act as non-named entity actors in the texts, e.g., "migrant families" = "asylum-seekers." Compared to our baseline, the approach keeps the mentions of the geopolitical entities separated, e.g., "Iran leaders" != "European leaders," and clusters (in)directly related mentions with diverse wording, e.g., "American officials" = "Trump Administration."

Felix Hamborg, Karsten Donnay, Bela Gipp

Extensive research on target-dependent sentiment classification (TSC) has led to strong classification performances in domains where authors tend to explicitly express sentiment about specific entities or topics, such as in reviews or on social media. We investigate TSC in news articles, a much less researched domain, despite the importance of news as an essential information source in individual and societal decision making. This article introduces NewsTSC, a manually annotated dataset to explore TSC on news articles. Investigating characteristics of sentiment in news and contrasting them to popular TSC domains, we find that sentiment in the news is expressed less explicitly, is more dependent on context and readership, and requires a greater degree of interpretation. In an extensive evaluation, we find that the state of the art in TSC performs worse on news articles than on other domains (average recall AvgRec = 69.8 on NewsTSC compared to AvgRev = [75.6, 82.2] on established TSC datasets). Reasons include incorrectly resolved relation of target and sentiment-bearing phrases and off-context dependence. As a major improvement over previous news TSC, we find that BERT's natural language understanding capabilities capture the less explicit sentiment used in news articles.

Lukas Gebhard, Felix Hamborg

News articles covering policy issues are an essential source of information in the social sciences and are also frequently used for other use cases, e.g., to train NLP language models. To derive meaningful insights from the analysis of news, large datasets are required that represent real-world distributions, e.g., with respect to the contained outlets' popularity, topically, or across time. Information on the political leanings of media publishers is often needed, e.g., to study differences in news reporting across the political spectrum, which is one of the prime use cases in the social sciences when studying media bias and related societal issues. Concerning these requirements, existing datasets have major flaws, resulting in redundant and cumbersome effort in the research community for dataset creation. To fill this gap, we present POLUSA, a dataset that represents the online media landscape as perceived by an average US news consumer. The dataset contains 0.9M articles covering policy topics published between Jan. 2017 and Aug. 2019 by 18 news outlets representing the political spectrum. Each outlet is labeled by its political leaning, which we derive using a systematic aggregation of eight data sources. The news dataset is balanced with respect to publication date and outlet popularity. POLUSA enables studying a variety of subjects, e.g., media effects and political partisanship. Due to its size, the dataset allows to utilize data-intense deep learning methods.

Philipp Scharpf, Moritz Schubotz, Abdou Youssef et al.

In this paper, we show how selecting and combining encodings of natural and mathematical language affect classification and clustering of documents with mathematical content. We demonstrate this by using sets of documents, sections, and abstracts from the arXiv preprint server that are labeled by their subject class (mathematics, computer science, physics, etc.) to compare different encodings of text and formulae and evaluate the performance and runtimes of selected classification and clustering algorithms. Our encodings achieve classification accuracies up to $82.8\%$ and cluster purities up to $69.4\%$ (number of clusters equals number of classes), and $99.9\%$ (unspecified number of clusters) respectively. We observe a relatively low correlation between text and math similarity, which indicates the independence of text and formulae and motivates treating them as separate features of a document. The classification and clustering can be employed, e.g., for document search and recommendation. Furthermore, we show that the computer outperforms a human expert when classifying documents. Finally, we evaluate and discuss multi-label classification and formula semantification.

Felix Hamborg, Philipp Meschenmoser, Moritz Schubotz et al.

In scientific publications, citations allow readers to assess the authenticity of the presented information and verify it in the original context. News articles, however, do not contain citations and only rarely refer readers to further sources. Readers often cannot assess the authenticity of the presented information as its origin is unclear. We present NewsDeps, the first approach that analyzes and visualizes where information in news articles stems from. NewsDeps employs methods from natural language processing and plagiarism detection to measure article similarity. We devise a temporal-force-directed graph that places articles as nodes chronologically. The graph connects articles by edges varying in width depending on the articles' similarity. We demonstrate our approach in a case study with two real-world scenarios. We find that NewsDeps increases efficiency and transparency in news consumption by revealing which previously published articles are the primary sources of each given article.