arXiv subjects

Jana Diesner

Publications and source records attributed to Jana Diesner.

33 records · Page 2Linked to original sources

arXiv2023

StereoMap: Quantifying the Awareness of Human-like Stereotypes in Large Language Models

Large Language Models (LLMs) have been observed to encode and perpetuate harmful associations present in the training data. We propose a theoretically grounded framework called StereoMap to gain insights into their perceptions of how demographic groups have been viewed by society. The framework is grounded in the Stereotype Content Model (SCM); a well-established theory from psychology. According to SCM, stereotypes are not all alike. Instead, the dimensions of Warmth and Competence serve as the factors that delineate the nature of stereotypes. Based on the SCM theory, StereoMap maps LLMs' perceptions of social groups (defined by socio-demographic features) using the dimensions of Warmth and Competence. Furthermore, the framework enables the investigation of keywords and verbalizations of reasoning of LLMs' judgments to uncover underlying factors influencing their perceptions. Our results show that LLMs exhibit a diverse range of perceptions towards these groups, characterized by mixed evaluations along the dimensions of Warmth and Competence. Furthermore, analyzing the reasonings of LLMs, our findings indicate that LLMs demonstrate an awareness of social disparities, often stating statistical data and research findings to support their reasoning. This study contributes to the understanding of how LLMs perceive and represent social groups, shedding light on their potential biases and the perpetuation of harmful associations.

cs.CL↗

arXiv2023

What should I Ask: A Knowledge-driven Approach for Follow-up Questions Generation in Conversational Surveys

Generating follow-up questions on the fly could significantly improve conversational survey quality and user experiences by enabling a more dynamic and personalized survey structure. In this paper, we proposed a novel task for knowledge-driven follow-up question generation in conversational surveys. We constructed a new human-annotated dataset of human-written follow-up questions with dialogue history and labeled knowledge in the context of conversational surveys. Along with the dataset, we designed and validated a set of reference-free Gricean-inspired evaluation metrics to systematically evaluate the quality of generated follow-up questions. We then propose a two-staged knowledge-driven model for the task, which generates informative and coherent follow-up questions by using knowledge to steer the generation process. The experiments demonstrate that compared to GPT-based baseline models, our two-staged model generates more informative, coherent, and clear follow-up questions.

cs.CL↗

arXiv2023

Examining the Causal Effect of First Names on Language Models: The Case of Social Commonsense Reasoning

As language models continue to be integrated into applications of personal and societal relevance, ensuring these models' trustworthiness is crucial, particularly with respect to producing consistent outputs regardless of sensitive attributes. Given that first names may serve as proxies for (intersectional) socio-demographic representations, it is imperative to examine the impact of first names on commonsense reasoning capabilities. In this paper, we study whether a model's reasoning given a specific input differs based on the first names provided. Our underlying assumption is that the reasoning about Alice should not differ from the reasoning about James. We propose and implement a controlled experimental framework to measure the causal effect of first names on commonsense reasoning, enabling us to distinguish between model predictions due to chance and caused by actual factors of interest. Our results indicate that the frequency of first names has a direct effect on model prediction, with less frequent names yielding divergent predictions compared to more frequent names. To gain insights into the internal mechanisms of models that are contributing to these behaviors, we also conduct an in-depth explainable analysis. Overall, our findings suggest that to ensure model robustness, it is essential to augment datasets with more diverse first names during the configuration stage.

cs.CL↗

arXiv2022

PyTAIL: Interactive and Incremental Learning of NLP Models with Human in the Loop for Online Data

Online data streams make training machine learning models hard because of distribution shift and new patterns emerging over time. For natural language processing (NLP) tasks that utilize a collection of features based on lexicons and rules, it is important to adapt these features to the changing data. To address this challenge we introduce PyTAIL, a python library, which allows a human in the loop approach to actively train NLP models. PyTAIL enhances generic active learning, which only suggests new instances to label by also suggesting new features like rules and lexicons to label. Furthermore, PyTAIL is flexible enough for users to accept, reject, or update rules and lexicons as the model is being trained. Finally, we simulate the performance of PyTAIL on existing social media benchmark datasets for text classification. We compare various active learning strategies on these benchmarks. The model closes the gap with as few as 10% of the training data. Finally, we also highlight the importance of tracking evaluation metric on remaining data (which is not yet merged with active learning) alongside the test dataset. This highlights the effectiveness of the model in accurately annotating the remaining dataset, which is especially suitable for batch processing of large unlabelled corpora. PyTAIL will be available at https://github.com/socialmediaie/pytail.

cs.CL↗

arXiv2022

What Changed? Investigating Debiasing Methods using Causal Mediation Analysis

Previous work has examined how debiasing language models affect downstream tasks, specifically, how debiasing techniques influence task performance and whether debiased models also make impartial predictions in downstream tasks or not. However, what we don't understand well yet is why debiasing methods have varying impacts on downstream tasks and how debiasing techniques affect internal components of language models, i.e., neurons, layers, and attentions. In this paper, we decompose the internal mechanisms of debiasing language models with respect to gender by applying causal mediation analysis to understand the influence of debiasing methods on toxicity detection as a downstream task. Our findings suggest a need to test the effectiveness of debiasing methods with different bias metrics, and to focus on changes in the behavior of certain components of the models, e.g.,first two layers of language models, and attention heads.

cs.CL↗

arXiv2021

WikiCSSH: Extracting and Evaluating Computer Science Subject Headings from Wikipedia

Hierarchical domain-specific classification schemas (or subject heading vocabularies) are often used to identify, classify, and disambiguate concepts that occur in scholarly articles. In this work, we develop, apply, and evaluate a human-in-the-loop workflow that first extracts an initial category tree from crowd-sourced Wikipedia data, and then combines community detection, machine learning, and hand-crafted heuristics or rules to prune the initial tree. This work resulted in WikiCSSH; a large-scale, hierarchically organized vocabulary for the domain of computer science (CS). Our evaluation suggests that WikiCSSH outperforms alternative CS vocabularies in terms of vocabulary size as well as the performance of lexicon-based key-phrase extraction from scholarly data. WikiCSSH can further distinguish between coarse-grained versus fine-grained CS concepts. The outlined workflow can serve as a template for building hierarchically-organized subject heading vocabularies for other domains that are covered in Wikipedia.

cs.SI↗

arXiv2021

Formational bounds of link prediction in collaboration networks

Link prediction in collaboration networks is often solved by identifying structural properties of existing nodes that are disconnected at one point in time, and that share a link later on. The maximally possible recall rate or upper bound of this approach's success is capped by the proportion of links that are formed among existing nodes embedded in these properties. Consequentially, sustained ties as well as links that involve one or two new network participants are typically not predicted. The purpose of this study is to highlight formational constraints that need to be considered to increase the practical value of link prediction methods for collaboration networks. In this study, we identify the distribution of basic link formation types based on four large-scale, over-time collaboration networks, showing that current link predictors can maximally anticipate around 25% of links that involve at least one prior network member. This implies that for collaboration networks, increasing the accuracy of computational link prediction solutions may not be a reasonable goal when the ratio of collaboration ties that are eligible to the classic link prediction process is low.

cs.SI↗

arXiv2021

Over-time measurement of triadic closure in coauthorship networks

Applying the concept of triadic closure to coauthorship networks means that scholars are likely to publish a joint paper if they have previously coauthored with the same people. Prior research has identified moderate to high (20% to 40%) closure rates; suggesting that this mechanism is a reasonable explanation for tie formation between future coauthors. We show how calculating triadic closure based on prior operationalizations of closure, namely Newman's measure for one-mode networks (NCC) and Opsahl's measure for two-mode networks (OCC), may lead to higher amounts of closure as compared to measuring closure over time via a metric that we introduce and test in this paper. Based on empirical experiments using four large-scale, longitudinal datasets, we find a lower bound of about 1~3% closure rates and an upper bound of about 4~7%. These results motivate research on new explanatory factors for the formation of coauthorship links.

cs.SI↗

arXiv2020

Multilevel Structural Evaluation of Signed Directed Social Networks based on Balance Theory

Balance theory explains the forces behind the structure of social systems, which are commonly modeled as static undirected signed networks. We expand this modeling approach to incorporate directionality of edges, and consider three levels of analysis: triads, subgroups, and the whole network. For triad-level balance, we operationalize a new measure by utilizing semicycles that satisfy the condition of transitivity. For subgroup-level balance, we propose measures of cohesiveness (intra-group solidarity) and divisiveness (inter-group antagonism) to capture balance within and among subgroups of the network using the most fitting partition of nodes into two groups. For network-level balance, we re-purpose the normalized line index to incorporate directionality, and provide the proportion of edges whose position suits balance. Through extensive computational analysis, we quantify and analyze patterns of social structure in triads, subgroups, and the whole network across a range of social settings from college students and Wikipedia editors to philosophers and Bitcoin traders. We then apply our multilevel framework of analysis to examine balance in temporal and multilayer networks, which demonstrates the generalizability of our approach to evaluating balance, and leads to new observations on balance with respect to time and layer dimensions. Our complementary findings on a variety of social networks highlight the need to evaluate balance at different levels. We propose a comprehensive yet parsimonious approach to address this need.

cs.SI↗

arXiv2020

Structural balance in signed digraphs: considering transitivity to measure balance in graphs constructed by using different link signing methods

Structural balance theory assumes triads in networks to gravitate towards stable configurations. The theory has been verified for undirected graphs. Since real-world networks are often directed, we introduce a novel method for considering both transitivity and sign consistency for calculating balance in signed digraphs. We test our approach on graphs that we constructed by using different methods for identifying edge signs: natural language processing to infer signs from underlying text data, and self-reported survey data. Our results show that for various social contexts and edge sign detection methods, balance is moderately high, ranging from 67.5% to 92.4%.

cs.SI↗

arXiv2020

An Empirical Methodology for Detecting and Prioritizing Needs during Crisis Events

In times of crisis, identifying the essential needs is a crucial step to providing appropriate resources and services to affected entities. Social media platforms such as Twitter contain vast amount of information about the general public's needs. However, the sparsity of the information as well as the amount of noisy content present a challenge to practitioners to effectively identify shared information on these platforms. In this study, we propose two novel methods for two distinct but related needs detection tasks: the identification of 1) a list of resources needed ranked by priority, and 2) sentences that specify who-needs-what resources. We evaluated our methods on a set of tweets about the COVID-19 crisis. For task 1 (detecting top needs), we compared our results against two given lists of resources and achieved 64% precision. For task 2 (detecting who-needs-what), we compared our results on a set of 1,000 annotated tweets and achieved a 68% F1-score.

cs.CL↗

arXiv2019

REO-Relevance, Extraness, Omission: A Fine-grained Evaluation for Image Captioning

Popular metrics used for evaluating image captioning systems, such as BLEU and CIDEr, provide a single score to gauge the system's overall effectiveness. This score is often not informative enough to indicate what specific errors are made by a given system. In this study, we present a fine-grained evaluation method REO for automatically measuring the performance of image captioning systems. REO assesses the quality of captions from three perspectives: 1) Relevance to the ground truth, 2) Extraness of the content that is irrelevant to the ground truth, and 3) Omission of the elements in the images and human references. Experiments on three benchmark datasets demonstrate that our method achieves a higher consistency with human judgments and provides more intuitive evaluation results than alternative metrics.

cs.CL↗

arXiv2019

TIGEr: Text-to-Image Grounding for Image Caption Evaluation

This paper presents a new metric called TIGEr for the automatic evaluation of image captioning systems. Popular metrics, such as BLEU and CIDEr, are based solely on text matching between reference captions and machine-generated captions, potentially leading to biased evaluations because references may not fully cover the image content and natural language is inherently ambiguous. Building upon a machine-learned text-image grounding model, TIGEr allows to evaluate caption quality not only based on how well a caption represents image content, but also on how well machine-generated captions match human-generated captions. Our empirical tests show that TIGEr has a higher consistency with human judgments than alternative existing metrics. We also comprehensively assess the metric's effectiveness in caption evaluation by measuring the correlation between human judgments and metric scores.

cs.CL↗

arXiv2015

Coauthorship networks: A directed network approach considering the order and number of coauthors

In many scientific fields, the order of coauthors on a paper conveys information about each individual's contribution to a piece of joint work. We argue that in prior network analyses of coauthorship networks, the information on ordering has been insufficiently considered because ties between authors are typically symmetrized. This is basically the same as assuming that each co-author has contributed equally to a paper. We introduce a solution to this problem by adopting a coauthorship credit allocation model proposed by Kim and Diesner (2014), which in its core conceptualizes co-authoring as a directed, weighted, and self-looped network. We test and validate our application of the adopted framework based on a sample data of 861 authors who have published in the journal Psychometrika. Results suggest that this novel sociometric approach can complement traditional measures based on undirected networks and expand insights into coauthoring patterns such as the hierarchy of collaboration among scholars. As another form of validation, we also show how our approach accurately detects prominent scholars in the Psychometric Society affiliated with the journal.

cs.DL↗

arXiv2015

Distortive Effects of Initial-Based Name Disambiguation on Measurements of Large-Scale Coauthorship Networks

Scholars have often relied on name initials to resolve name ambiguities in large-scale coauthorship network research. This approach bears the risk of incorrectly merging or splitting author identities. The use of initial-based disambiguation has been justified by the assumption that such errors would not affect research findings too much. This paper tests this assumption by analyzing coauthorship networks from five academic fields - biology, computer science, nanoscience, neuroscience, and physics - and an interdisciplinary journal, PNAS. Name instances in datasets of this study were disambiguated based on heuristics gained from previous algorithmic disambiguation solutions. We use disambiguated data as a proxy of ground-truth to test the performance of three types of initial-based disambiguation. Our results show that initial-based disambiguation can misrepresent statistical properties of coauthorship networks: it deflates the number of unique authors, number of component, average shortest paths, clustering coefficient, and assortativity, while it inflates average productivity, density, average coauthor number per author, and largest component size. Also, on average, more than half of top 10 productive or collaborative authors drop off the lists. Asian names were found to account for the majority of misidentification by initial-based disambiguation due to their common surname and given name initials.

cs.DL↗