Andreas Spitz

Celina Kacperski, Roberto Ulloa, Peter Selb et al.

As citizens increasingly encounter political information in digital environments, understanding whether this engagement shapes their policy views has become a central concern. Drawing on dual-process theories of persuasion, we argue that motivational activation is an enabling condition for policy support change in high-choice online environments. We test this in a three-wave field experiment with German participants (n = 791) across three policy topics (basic child support, renewable energy transition, cannabis legalization), in which participants were randomly assigned to a control group, and two encouragement conditions: a verbal encouragement, or a monetary incentive tied to a knowledge test. Browsing behavior was passively tracked via digital trace data over a 20-hour window. We find that self-directed online information search produced changes in policy support for child support and cannabis legalization but not for the energy transition, with monetary incentives producing significant effects rather than verbal prompts. We discuss motivational salience, issue malleability, and search-environment quality as joint conditions under which political information engagement can produce detectable changes in policy support.

Jonathan Külz, Andreas Spitz, Ahmad Abu-Akel et al.

There is a widespread belief that the tone of US political language has become more negative recently, in particular when Donald Trump entered politics. At the same time, there is disagreement as to whether Trump changed or merely continued previous trends. To date, data-driven evidence regarding these questions is scarce, partly due to the difficulty of obtaining a comprehensive, longitudinal record of politicians' utterances. Here we apply psycholinguistic tools to a novel, comprehensive corpus of 24 million quotes from online news attributed to 18,627 US politicians in order to analyze how the tone of US politicians' language evolved between 2008 and 2020. We show that, whereas the frequency of negative emotion words had decreased continuously during Obama's tenure, it suddenly and lastingly increased with the 2016 primary campaigns, by 1.6 pre-campaign standard deviations, or 8% of the pre-campaign mean, in a pattern that emerges across parties. The effect size drops by 40% when omitting Trump's quotes, and by 50% when averaging over speakers rather than quotes, implying that prominent speakers, and Trump in particular, have disproportionately, though not exclusively, contributed to the rise in negative language. This work provides the first large-scale data-driven evidence of a drastic shift toward a more negative political tone following Trump's campaign start as a catalyst, with important implications for the debate about the state of US politics.

Vuk Vuković, Akhil Arora, Huan-Cheng Chang et al.

The use of attributed quotes is the most direct and least filtered pathway of information propagation in news. Consequently, quotes play a central role in the conception, reception, and analysis of news stories. Since quotes provide a more direct window into a speaker's mind than regular reporting, they are a valuable resource for journalists and researchers alike. While substantial research efforts have been devoted to methods for the automated extraction of quotes from news and their attribution to speakers, few comprehensive corpora of attributed quotes from contemporary sources are available to the public. Here, we present an adaptive web interface for searching Quotebank, a massive collection of quotes from the news, which we make available at https://quotebank.dlab.tools.

Marko Čuljak, Andreas Spitz, Robert West et al.

Named entity linking (NEL) in news is a challenging endeavour due to the frequency of unseen and emerging entities, which necessitates the use of unsupervised or zero-shot methods. However, such methods tend to come with caveats, such as no integration of suitable knowledge bases (like Wikidata) for emerging entities, a lack of scalability, and poor interpretability. Here, we consider person disambiguation in Quotebank, a massive corpus of speaker-attributed quotations from the news, and investigate the suitability of intuitive, lightweight, and scalable heuristics for NEL in web-scale corpora. Our best performing heuristic disambiguates 94% and 63% of the mentions on Quotebank and the AIDA-CoNLL benchmark, respectively. Additionally, the proposed heuristics compare favourably to the state-of-the-art unsupervised and zero-shot methods, Eigenthemes and mGENRE, respectively, thereby serving as strong baselines for unsupervised and zero-shot entity linking.

Silke Husse, Andreas Spitz

The awareness and mitigation of biases are of fundamental importance for the fair and transparent use of contextual language models, yet they crucially depend on the accurate detection of biases as a precursor. Consequently, numerous bias detection methods have been proposed, which vary in their approach, the considered type of bias, and the data used for evaluation. However, while most detection methods are derived from the word embedding association test for static word embeddings, the reported results are heterogeneous, inconsistent, and ultimately inconclusive. To address this issue, we conduct a rigorous analysis and comparison of bias detection methods for contextual language models. Our results show that minor design and implementation decisions (or errors) have a substantial and often significant impact on the derived bias scores. Overall, we find the state of the field to be both worse than previously acknowledged due to systematic and propagated errors in implementations, yet better than anticipated since divergent results in the literature homogenize after accounting for implementation errors. Based on our findings, we conclude with a discussion of paths towards more robust and consistent bias detection methods.

Thilo Spinner, Rebecca Kehlbeck, Rita Sevastjanova et al.

The growing popularity of generative language models has amplified interest in interactive methods to guide model outputs. Prompt refinement is considered one of the most effective means to influence output among these methods. We identify several challenges associated with prompting large language models, categorized into data- and model-specific, linguistic, and socio-linguistic challenges. A comprehensive examination of model outputs, including runner-up candidates and their corresponding probabilities, is needed to address these issues. The beam search tree, the prevalent algorithm to sample model outputs, can inherently supply this information. Consequently, we introduce an interactive visual method for investigating the beam search tree, facilitating analysis of the decisions made by the model during generation. We quantitatively show the value of exposing the beam search tree and present five detailed analysis scenarios addressing the identified challenges. Our methodology validates existing results and offers additional insights.

Mats Faulborn, Indira Sen, Max Pellert et al.

Prompt-based language models like GPT4 and LLaMa have been used for a wide variety of use cases such as simulating agents, searching for information, or for content analysis. For all of these applications and others, political biases in these models can affect their performance. Several researchers have attempted to study political bias in language models using evaluation suites based on surveys, such as the Political Compass Test (PCT), often finding a particular leaning favored by these models. However, there is some variation in the exact prompting techniques, leading to diverging findings, and most research relies on constrained-answer settings to extract model responses. Moreover, the Political Compass Test is not a scientifically valid survey instrument. In this work, we contribute a political bias measured informed by political science theory, building on survey design principles to test a wide variety of input prompts, while taking into account prompt sensitivity. We then prompt 11 different open and commercial models, differentiating between instruction-tuned and non-instruction-tuned models, and automatically classify their political stances from 88,110 responses. Leveraging this dataset, we compute political bias profiles across different prompt variations and find that while PCT exaggerates bias in certain models like GPT3.5, measures of political bias are often unstable, but generally more left-leaning for instruction-tuned models. Code and data are available on: https://github.com/MaFa211/theory_grounded_pol_bias

Julian Schelb, Roberto Ulloa, Andreas Spitz

Researchers in the political and social sciences often rely on classification models to analyze trends in information consumption by examining browsing histories of millions of webpages. Automated scalable methods are necessary due to the impracticality of manual labeling. In this paper, we model the detection of topic-related content as a binary classification task and compare the accuracy of fine-tuned pre-trained encoder models against in-context learning strategies. Using only a few hundred annotated data points per topic, we detect content related to three German policies in a database of scraped webpages. We compare multilingual and monolingual models, as well as zero and few-shot approaches, and investigate the impact of negative sampling strategies and the combination of URL & content-based features. Our results show that a small sample of annotated data is sufficient to train an effective classifier. Fine-tuning encoder-based models yields better results than in-context learning. Classifiers using both URL & content-based features perform best, while using URLs alone provides adequate results when content is unavailable.

Julian Schelb, Orr Borin, David Garcia et al.

Generative language models are increasingly being subjected to psychometric questionnaires intended for human testing, in efforts to establish their traits, as benchmarks for alignment, or to simulate participants in social science experiments. While this growing body of work sheds light on the likeness of model responses to those of humans, concerns are warranted regarding the rigour and reproducibility with which these experiments may be conducted. Instabilities in model outputs, sensitivity to prompt design, parameter settings, and a large number of available model versions increase documentation requirements. Consequently, generalization of findings is often complex and reproducibility is far from guaranteed. In this paper, we present R.U.Psycho, a framework for designing and running robust and reproducible psychometric experiments on generative language models that requires limited coding expertise. We demonstrate the capability of our framework on a variety of psychometric questionnaires, which lend support to prior findings in the literature. R.U.Psycho is available as a Python package at https://github.com/julianschelb/rupsycho.

Mengying Wang, Andreas Spitz

Writing assistants and large language models see widespread use in the creation of text content. While their effectiveness for individual users has been evaluated in the literature, little is known about their proclivity to change language or reduce its richness when adopted by a large user base. In this paper, we take a first step towards quantifying this risk by measuring the semantic and vocabulary change enacted by the use of rephrasing tools on a multi-domain corpus of human-generated text.

Cong Tran, Won-Yong Shin, Andreas Spitz

Since the structure of complex networks is often unknown, we may identify the most influential seed nodes by exploring only a part of the underlying network, given a small budget for node queries. We propose IM-META, a solution to influence maximization (IM) in networks with unknown topology by retrieving information from queries and node metadata. Since using such metadata is not without risk due to the noisy nature of metadata and uncertainties in connectivity inference, we formulate a new IM problem that aims to find both seed nodes and queried nodes. In IM-META, we develop an effective method that iteratively performs three steps: 1) we learn the relationship between collected metadata and edges via a Siamese neural network, 2) we select a number of inferred confident edges to construct a reinforced graph, and 3) we identify the next node to query by maximizing the inferred influence spread using our topology-aware ranking strategy. Through experimental evaluation of IM-META on four real-world datasets, we demonstrate a) the speed of network exploration via node queries, b) the effectiveness of each module, c) the superiority over benchmark methods, d) the robustness to more difficult settings, e) the hyperparameter sensitivity, and f) the scalability.

Cong Tran, Won-Yong Shin, Andreas Spitz et al.

Most network data are collected from partially observable networks with both missing nodes and missing edges, for example, due to limited resources and privacy settings specified by users on social media. Thus, it stands to reason that inferring the missing parts of the networks by performing network completion should precede downstream applications. However, despite this need, the recovery of missing nodes and edges in such incomplete networks is an insufficiently explored problem due to the modeling difficulty, which is much more challenging than link prediction that only infers missing edges. In this paper, we present DeepNC, a novel method for inferring the missing parts of a network based on a deep generative model of graphs. Specifically, our method first learns a likelihood over edges via an autoregressive generative model, and then identifies the graph that maximizes the learned likelihood conditioned on the observable graph topology. Moreover, we propose a computationally efficient DeepNC algorithm that consecutively finds individual nodes that maximize the probability in each node generation step, as well as an enhanced version using the expectation-maximization algorithm. The runtime complexities of both algorithms are shown to be almost linear in the number of nodes in the network. We empirically demonstrate the superiority of DeepNC over state-of-the-art network completion approaches.

Andreas Spitz, Satya Almasian, Michael Gertz

The recent introduction of entity-centric implicit network representations of unstructured text offers novel ways for exploring entity relations in document collections and streams efficiently and interactively. Here, we present TopExNet as a tool for exploring entity-centric network topics in streams of news articles. The application is available as a web service at https://topexnet.ifi.uni-heidelberg.de/ .

Gloria Feher, Andreas Spitz, Michael Gertz

Word embeddings have gained significant attention as learnable representations of semantic relations between words, and have been shown to improve upon the results of traditional word representations. However, little effort has been devoted to using embeddings for the retrieval of entity associations beyond pairwise relations. In this paper, we use popular embedding methods to train vector representations of an entity-annotated news corpus, and evaluate their performance for the task of predicting entity participation in news events versus a traditional word cooccurrence network as a baseline. To support queries for events with multiple participating entities, we test a number of combination modes for the embedding vectors. While we find that even the best combination modes for word embeddings do not quite reach the performance of the full cooccurrence network, especially for rare entities, we observe that different embedding methods model different types of relations, thereby indicating the potential for ensemble methods.

Satya Almasian, Andreas Spitz, Michael Gertz

Learned vector representations of words are useful tools for many information retrieval and natural language processing tasks due to their ability to capture lexical semantics. However, while many such tasks involve or even rely on named entities as central components, popular word embedding models have so far failed to include entities as first-class citizens. While it seems intuitive that annotating named entities in the training corpus should result in more intelligent word features for downstream tasks, performance issues arise when popular embedding approaches are naively applied to entity annotated corpora. Not only are the resulting entity embeddings less useful than expected, but one also finds that the performance of the non-entity word embeddings degrades in comparison to those trained on the raw, unannotated corpus. In this paper, we investigate approaches to jointly train word and entity embeddings on a large corpus with automatically annotated and linked entities. We discuss two distinct approaches to the generation of such embeddings, namely the training of state-of-the-art embeddings on raw-text and annotated versions of the corpus, as well as node embeddings of a co-occurrence graph representation of the annotated corpus. We compare the performance of annotated embeddings and classical word embeddings on a variety of word similarity, analogy, and clustering evaluation tasks, and investigate their performance in entity-specific tasks. Our findings show that it takes more than training popular word embedding models on an annotated corpus to create entity embeddings with acceptable performance on common test cases. Based on these results, we discuss how and when node embeddings of the co-occurrence graph representation of the text can restore the performance.

Cong Tran, Won-Yong Shin, Andreas Spitz

The discovery of community structures in social networks has gained significant attention since it is a fundamental problem in understanding the networks' topology and functions. However, most social network data are collected from partially observable networks with both missing nodes and edges. In this paper, we address a new problem of detecting overlapping community structures in the context of such an incomplete network, where communities in the network are allowed to overlap since nodes belong to multiple communities at once. To solve this problem, we introduce KroMFac, a new framework that conducts community detection via regularized nonnegative matrix factorization (NMF) based on the Kronecker graph model. Specifically, from an inferred Kronecker generative parameter matrix, we first estimate the missing part of the network. As our major contribution to the proposed framework, to improve community detection accuracy, we then characterize and select influential nodes (which tend to have high degrees) by ranking, and add them to the existing graph. Finally, we uncover the community structures by solving the regularized NMF-aided optimization problem in terms of maximizing the likelihood of the underlying graph. Furthermore, adopting normalized mutual information (NMI), we empirically show superiority of our KroMFac approach over two baseline schemes by using both synthetic and real-world networks.

Erich Schubert, Andreas Spitz, Michael Weiler et al.

Many word clouds provide no semantics to the word placement, but use a random layout optimized solely for aesthetic purposes. We propose a novel approach to model word significance and word affinity within a document, and in comparison to a large background corpus. We demonstrate its usefulness for generating more meaningful word clouds as a visual summary of a given document. We then select keywords based on their significance and construct the word cloud based on the derived affinity. Based on a modified t-distributed stochastic neighbor embedding (t-SNE), we generate a semantic word placement. For words that cooccur significantly, we include edges, and cluster the words according to their cooccurrence. For this we designed a scalable and memory-efficient sketch-based approach usable on commodity hardware to aggregate the required corpus statistics needed for normalization, and for identifying keywords as well as significant cooccurences. We empirically validate our approch using a large Wikipedia corpus.