Journal of Computational Social Science最新文献

Word embeddings have become powerful tools for detecting social biases encoded in language, yet research on measuring race bias through embeddings remains underdeveloped compared to studies on gender bias. This gap largely stems from the complexity of constructing race dimensions, which involve socially contested meanings and less clear semantic oppositions. Existing studies on race bias often rely on intuition and context-specific approaches when choosing anchor terms. In this paper, we address this methodological gap by providing statistical metrics to evaluate the quality and adaptability of race categories in embeddings. We apply these metrics to race categories across three embeddings-Google News (U.S.-centric), South African News (South African context), and Wikipedia (neutral, general-purpose). We find that names are effective for constructing race dimensions, with sub-Saharan African/European name categories producing more stable and generalisable dimensions than other categories, while American names were less generalisable. Validation shows that SSA/European name embeddings correlate most strongly with human ratings and demonstrate that our metrics capture human-perceived semantic structure of race. This research provides a framework for constructing robust race dimensions for measuring race bias in word embeddings.

Supplementary information: The online version contains supplementary material available at 10.1007/s42001-025-00449-w.

Many important social phenomena are characterized by repeated interactions among individuals over time such as email exchanges in an organization or face-to-face interactions in a classroom. To understand the underlying mechanisms of social interaction dynamics, statistical simulation techniques for network data at fine temporal granularity are crucial. This article makes two contributions to the field. First, we present statistical frameworks to simulate relational event networks under dyadic and actor-oriented relational event models implemented in an R package remulate. Second, we show how this simulation framework can address key challenges in temporal social network analysis through five case studies. The first study illustrates the necessity of simulation based techniques for model assessment, using a network of criminal gangs. The second shows how simulation supports social theory development which is illustrated via optimal distinctiveness theory. The third explores simulation for understanding the effects of network interventions. In the fourth study, we illustrate how simulation-based analysis can be used to assess the sensitivity of relational event models. The fifth study demonstrates how simulation frameworks can be used to make predictions about future relational dynamics. Through these case studies and software, researchers will be able to better understand social interaction dynamics using relational event data from real-life networks.

This paper studies the performance of open-source Large Language Models (LLMs) in text classification tasks typical for political science research. By examining tasks like stance, topic, and relevance classification, we aim to guide scholars in making informed decisions about their use of LLMs for text analysis and to establish a baseline performance benchmark that demonstrates the models' effectiveness. Specifically, we conduct an assessment of both zero-shot and fine-tuned LLMs across a range of text annotation tasks using news articles and tweets datasets. Our analysis shows that fine-tuning improves the performance of open-source LLMs, allowing them to match or even surpass zero-shot GPT $-$ 3.5 and GPT-4, though still lagging behind fine-tuned GPT $-$ 3.5. We further establish that fine-tuning is preferable to few-shot training with a relatively modest quantity of annotated text. Our findings show that fine-tuned open-source LLMs can be effectively deployed in a broad spectrum of text annotation applications. We provide a Python notebook facilitating the application of LLMs in text annotation for other researchers.

Supplementary information: The online version contains supplementary material available at 10.1007/s42001-024-00345-9.

The Covid-19 pandemic brought unprecedented changes to business ownership in the UK which affects a generation of entrepreneurs and their employees. Nonetheless, the impact remains poorly understood. This is because research on capital accumulation has typically lacked high-quality, individualized, population-level data. We overcome these barriers to examine who benefits from economic crises through a computationally orientated lens of firm creation. Leveraging a comprehensive cache of administrative data on every UK firm and all nine million people running them, combined with probabilistic algorithms, we conduct individual-level analyzis to understand who became Covid entrepreneurs. Using these techniques, we explore characteristics of entrepreneurs-such as age, gender, region, business experience, and industry-which potentially predict Covid entrepreneurship. By employing an automated time series model selection procedure to generate counterfactuals, we show that Covid entrepreneurs were typically aged 35-49 (40.4%), men (73.1%), and had previously held roles in existing firms (59.4%). For most industries, growth was disproportionately concentrated around London. It was therefore existing corporate elites who were most able to capitalize on the Covid crisis and not, as some hypothesized, young entrepreneurs who were setting up their first businesses. In this respect, the pandemic will likely impact future wealth inequalities. Our work offers methodological guidance for future policymakers during economic crises and highlights the long-term consequences for capital and wealth inequality.

Online sexual predators target children by building trust, creating dependency, and arranging meetings for sexual purposes. This poses a significant challenge for online communication platforms that strive to monitor and remove such content and terminate predators' accounts. However, these platforms can only take such actions if sexual predators explicitly violate the terms of service, not during the initial stages of relationship-building. This study designed and evaluated a strategy to prevent sexual predation and victimization by delivering warnings and raising awareness among high-risk individuals based on the routine activity theory in criminal psychology. We identified high-risk users as those with a high probability of committing or being subjected to violations, using a machine learning model that analyzed social networks and monitoring data from the platform. We conducted a randomized controlled trial on a Japanese avatar-based communication application, Pigg Party. High-risk players in the intervention group received warnings and awareness-building messages, while those in the control group did not receive the messages, regardless of their risk level. The trial involved 12,842 high-risk players in the intervention group and 12,844 in the control group for 138 days. The intervention successfully reduced violations and being violated among women for 12 weeks, although the impact on men was limited. These findings contribute to efforts to combat online sexual abuse and advance understanding of criminal psychology.

Supplementary information: The online version contains supplementary material available at 10.1007/s42001-025-00399-3.

School curricula guide the daily learning activities of millions of students. They embody the understanding of the education experts who designed them of how to organize the knowledge that students should acquire in a way that is optimal for learning. This can be viewed as a learning 'theory' which is, nevertheless, rarely put to the test. Here, we model a data set obtained from a Computer-Based Formative Assessment system used by thousands of students. The student-item response matrix is highly sparse and admits a natural representation as a bipartite graph, in which nodes stand for students or items and an edge between a student and an item represents a response of the student to that item. To predict unobserved edge labels (correct/incorrect responses) we resort to a graph neural network (GNN), a machine learning method for graph-structured data. Nodes and edges are represented as multidimensional embeddings. After fitting the model, the learned item embeddings reflect properties of the curriculum, such as item difficulty and the structure of school subject domains and competences. Simulations show that the GNN is particularly advantageous over a classical model when group patterns are present in the connections between students and items, such that students from a particular group have a higher probability of successfully answering items from a specific set. In sum, important aspects of the structure of the school curriculum are reflected in response patterns from educational assessments and can be partially retrieved by our graph-based neural model.

A network analysis approach, complemented by machine learning (ML) techniques, is applied to analyse the factors influencing Bilateral Investment Treaties (BITs) at the country level. Using the Electronic Database of Investment Treaties, BITs with health safeguards from 167 countries were charted, resulting in 534 connections with countries as nodes and their BITs as edges. Network analysis found that, on average, a country established BITs with six other nations. Additionally, we used node embedding techniques to generate features from the network, such as the Jaccard coefficient, resource allocation, and Adamic Adar for downstream link prediction. This study employed five tree-based ML models to predict future BIT formations with health inclusion. The eXtreme Gradient Boosting model proved to be superior, achieving a 64.02% accuracy rate. Notably, the Common Neighbor centrality feature and the Capital Account Balance Ratio emerged as influential factors in creating new BITs with health inclusions. Beyond economic considerations, our study highlighted a vital intersection: the nexus between BITs, economic growth, and public health policies. In essence, this research underscores the importance of safeguarding public health in BITs and showcases the potential of ML in understanding the intricacies of international treaties.

Epidemics, apart from affecting the health of populations, can have large impacts on their social and economic behavior and subsequently feed back to and influence the spreading of the disease. This calls for systematic investigation which factors affect significantly and either beneficially or adversely the disease spreading and regional socio-economics. Based on our recently developed hybrid agent-based socio-economy and epidemic spreading model we perform extensive exploration of its six-dimensional parameter space of the socio-economic part of the model, namely, the attitudes towards the spread of the pandemic, health and the economic situation for both, the population and government agents who impose regulations. We search for significant patterns from the resulting simulated data using basic classification tools, such as self-organizing maps and principal component analysis, and we monitor different quantities of the model output, such as infection rates, the propagation speed of the epidemic, economic activity, government regulations, and the compliance of population on government restrictions. Out of these, the ones describing the epidemic spreading were resulting in the most distinctive clustering of the data, and they were selected as the basis of the remaining analysis. We relate the found clusters to three distinct types of disease spreading: wave-like, chaotic, and transitional spreading patterns. The most important value parameter contributing to phase changes and the speed of the epidemic was found to be the compliance of the population agents towards the government regulations. We conclude that in compliant populations, the infection rates are significantly lower and the infection spreading is slower, while the population agents' health and economical attitudes show a weaker effect.

Large relational-event history data stemming from large networks are becoming increasingly available due to recent technological developments (e.g. digital communication, online databases, etc). This opens many new doors to learn about complex interaction behavior between actors in temporal social networks. The relational event model has become the gold standard for relational event history analysis. Currently, however, the main bottleneck to fit relational events models is of computational nature in the form of memory storage limitations and computational complexity. Relational event models are therefore mainly used for relatively small data sets while larger, more interesting datasets, including multilevel data structures and relational event data streams, cannot be analyzed on standard desktop computers. This paper addresses this problem by developing approximation algorithms based on meta-analysis methods that can fit relational event models significantly faster while avoiding the computational issues. In particular, meta-analytic approximations are proposed for analyzing streams of relational event data, multilevel relational event data and potentially combinations thereof. The accuracy and the statistical properties of the methods are assessed using numerical simulations. Furthermore, real-world data are used to illustrate the potential of the methodology to study social interaction behavior in an organizational network and interaction behavior among political actors. The algorithms are implemented in the publicly available R package 'remx'.