EPJ Data Science最新文献

Abstract

Artificial Intelligence (AI) technologies have exposed more and more ethical issues while providing services to people. It is challenging for people to realize the occurrence of AI ethical issues in most cases. The lower the public awareness, the more difficult it is to address AI ethical issues. Many previous studies have explored public reactions and opinions on AI ethical issues through questionnaires and social media platforms like Twitter. However, these approaches primarily focus on categorizing popular topics and sentiments, overlooking the public’s potential lack of knowledge underlying these issues. Few studies revealed the holistic knowledge structure of AI ethical topics and the relations among the subtopics. As the world’s largest online encyclopedia, Wikipedia encourages people to jointly contribute and share their knowledge by adding new topics and following a well-accepted hierarchical structure. Through public viewing and editing, Wikipedia serves as a proxy for knowledge transmission. This study aims to analyze how the public comprehend the body of knowledge of AI ethics. We adopted the community detection approach to identify the hierarchical community of the AI ethical topics, and further extracted the AI ethics-related entities, which are proper nouns, organizations, and persons. The findings reveal that the primary topics at the top-level community, most pertinent to AI ethics, predominantly revolve around knowledge-based and ethical issues. Examples include transitions from Information Theory to Internet Copyright Infringement. In summary, this study contributes to three points, (1) to present the holistic knowledge structure of AI ethics, (2) to evaluate and improve the existing body of knowledge of AI ethics, (3) to enhance public perception of AI ethics to mitigate the risks associated with AI technologies.

For U.S. presidential elections, most states use the so-called winner-take-all system, in which the state’s presidential electors are awarded to the winning political party in the state after a popular vote phase, regardless of the actual margin of victory. Therefore, election campaigns are especially intense in states where there is no clear direction on which party will be the winning party. These states are often referred to as swing states. To measure the impact of such an election law on the campaigns, we analyze the Twitter activity surrounding the 2020 US preelection debate, with a particular focus on the spread of disinformation. We find that about 88% of the online traffic was associated with swing states. In addition, the sharing of links to unreliable news sources is significantly more prevalent in tweets associated with swing states: in this case, untrustworthy tweets are predominantly generated by automated accounts. Furthermore, we observe that the debate is mostly led by two main communities, one with a predominantly Republican affiliation and the other with accounts of different political orientations. Most of the disinformation comes from the former.

Despite the historically documented regularity in human mobility patterns, the relaxation of spatial and temporal constraints, brought by the widespread adoption of telecommuting and e-commerce during the COVID-19 pandemic, as well as a growing desire for flexible work arrangements in a post-pandemic work, indicates a potential reshaping of these patterns. In this paper, we investigate the multifaceted impacts of relaxed spatio-temporal constraints on human mobility, using well-established metrics from the travel behavior literature. Further, we introduce a novel metric for schedule regularity, accounting for specific day-of-week characteristics that previous approaches overlooked. Building on the large body of literature on the impacts of COVID-19 on human mobility, we make use of passively tracked Point of Interest (POI) data for approximately 21,700 smartphone users in the US, and analyze data between January 2020 and September 2022 to answer two key questions: (1) has the COVID-19 pandemic and its associated relaxation of spatio-temporal activity patterns reshaped the different aspects of human mobility, and (2) have we achieved a state of stable post-pandemic “new normal”? We hypothesize that the relaxation of the spatiotemporal constraints around key activities will result in people exhibiting less regular schedules. Findings reveal a complex landscape: while some mobility indicators have reverted to pre-pandemic norms, such as trip frequency and travel distance, others, notably at-home dwell-time, persist at altered levels, suggesting a recalibration rather than a return to past behaviors. Most notably, our analysis reveals a paradox: despite the documented large-scale shift towards flexible work arrangements, schedule habits have strengthened rather than relaxed, defying our initial hypotheses and highlighting a desire for regularity. The study’s results contribute to a deeper understanding of the post-pandemic “new normal”, offering key insights on how multiple facets of travel behavior were reshaped, if at all, by the COVID-19 pandemic, and will help inform transportation planning in a post-pandemic world.

Abstract

Automated positioning devices can generate large datasets with information on the movement of humans, animals and objects, revealing patterns of movement, hot spots and overlaps among others. However, in the case of Automated Information Systems (AIS), attached to vessels, observed strange behaviors in the tracking datasets may come from intentional manipulation of the electronic devices. Thus, the analysis of anomalies can provide valuable information on suspicious behavior. Here, we analyze anomalies of fishing vessel trajectories obtained with the Automatic Identification System. The map of silent anomalies, those that occur when positioning data are absent for more than 24 hours, shows that they are most likely to occur closer to land, with 87.1% of anomalies observed within 100 km of the coast. This behavior suggests the potential of identifying silence anomalies as a proxy for illegal activities. With the increasing availability of high-resolution positioning of vessels and the development of powerful statistical analytical tools, we provide hints on the automatic detection of illegal activities that may help optimize the management of fishing resources.

Modeling human mobility helps to understand how people are accessing resources and physically contacting with each other in cities, and thus contributes to various applications such as urban planning, epidemic control, and location-based advertisement. Next location prediction is one decisive task in individual human mobility modeling and is usually viewed as sequence modeling, solved with Markov or RNN-based methods. However, the existing models paid little attention to the logic of individual travel decisions and the reproducibility of the collective behavior of population. To this end, we propose a Causal and Spatial-constrained Long and Short-term Learner (CSLSL) for next location prediction. CSLSL utilizes a causal structure based on multi-task learning to explicitly model the “when→what→where”, a.k.a. “time→activity→location” decision logic. We next propose a spatial-constrained loss function as an auxiliary task, to ensure the consistency between the predicted and actual spatial distribution of travelers’ destinations. Moreover, CSLSL adopts modules named Long and Short-term Capturer (LSC) to learn the transition regularities across different time spans. Extensive experiments on three real-world datasets show promising performance improvements of CSLSL over baselines and confirm the effectiveness of introducing the causality and consistency constraints. The implementation is available at https://github.com/urbanmobility/CSLSL.

Abstract

This paper examines the phenomenon of residential segregation in Berlin over time using a dynamic clustering analysis approach. Previous research has examined the phenomenon of residential segregation in Berlin at a high spatial and temporal aggregation and statically, i.e. not over time. We propose a methodology to investigate the existence of clusters of residential areas according to migration background, age group, gender, and socio-economic dimension over time. To this end, we have developed a sequential mixed methods approach that includes a multivariate kernel density estimation technique to estimate the density of subpopulations and a dynamic cluster analysis to discover spatial patterns of residential segregation over time (2009-2020). The dynamic analysis shows the emergence of clusters on the dimensions of migration background, age group, gender and socio-economic variables. We also identified a structural change in 2015, resulting in a new cluster in Berlin that reflects the changing distribution of subpopulations with a particular migratory background. Finally, we discuss the findings of this study with previous research and suggest possibilities for policy applications and future research using a dynamic clustering approach for analyzing changes in residential segregation at the city level.

Abstract

The same individuals can express very different emotions in online social media with respect to face-to-face interactions, partially because of intrinsic limitations of the digital environments and partially because of their algorithmic design, which is optimized to maximize engagement. Such differences become even more pronounced for topics concerning socially sensitive and polarizing issues, such as massive pharmaceutical interventions. Here, we investigate how online emotional responses change during the large-scale COVID-19 vaccination campaign with respect to a baseline in which no specific contentious topic dominates. We show that the online discussions during the pandemic generate a vast spectrum of emotional response compared to the baseline, especially when we take into account the characteristics of the users and the type of information shared in the online platform. Furthermore, we analyze the role of the political orientation of shared news, whose circulation seems to be driven not only by their actual informational content but also by the social need to strengthen one’s affiliation to, and positioning within, a specific online community by means of emotionally arousing posts. Our findings stress the importance of better understanding the emotional reactions to contentious topics at scale from digital signatures, while providing a more quantitative assessment of the ongoing online social dynamics to build a faithful picture of offline social implications.

Artificial intelligence (AI)-powered recommender systems play a crucial role in determining the content that users are exposed to on social media platforms. However, the behavioural patterns of these systems are often opaque, complicating the evaluation of their impact on the dissemination and consumption of disinformation and misinformation. To begin addressing this evidence gap, this study presents a measurement approach that uses observed digital traces to infer the status of algorithmic amplification of low-credibility content on Twitter over a 14-day period in January 2023. Using an original dataset of ≈ 2.7 million posts on COVID-19 and climate change published on the platform, this study identifies tweets sharing information from low-credibility domains, and uses a bootstrapping model with two stratifications, a tweet’s engagement level and a user’s followers level, to compare any differences in impressions generated between low-credibility and high-credibility samples. Additional stratification variables of toxicity, political bias, and verified status are also examined. This analysis provides valuable observational evidence on whether the Twitter algorithm favours the visibility of low-credibility content, with results indicating that, on aggregate, tweets containing low-credibility URL domains perform better than tweets that do not across both datasets. However, this effect is largely attributable to a difference in high-engagement, high-followers tweets, which are very impactful in terms of impressions generation, and are more likely receive amplified visibility when containing low-credibility content. Furthermore, high toxicity tweets and those with right-leaning bias see heightened amplification, as do low-credibility tweets from verified accounts. Ultimately, this suggests that Twitter’s recommender system may have facilitated the diffusion of false content by amplifying the visibility of low-credibility content with high-engagement generated by very influential users.

In this paper, we engage with and expand on the keynote talk about the “Right to Audit” given by Prof. Christian Sandvig at the International Conference on Computational Social Science 2021 through a critical reflection on power asymmetries in the algorithm auditing field. We elaborate on the challenges and asymmetries mentioned by Sandvig — such as those related to legal issues and the disparity between early-career and senior researchers. We also contribute a discussion of the asymmetries that were not covered by Sandvig but that we find critically important: those related to other disparities between researchers, incentive structures related to the access to data from companies, targets of auditing and users and their rights. We also discuss the implications these asymmetries have for algorithm auditing research such as the Western-centrism and the lack of the diversity of perspectives. While we focus on the field of algorithm auditing specifically, we suggest some of the discussed asymmetries affect Computational Social Science more generally and need to be reflected on and addressed.

Higher-order networks are widely used to describe complex systems in which interactions can involve more than two entities at once. In this paper, we focus on inclusion within higher-order networks, referring to situations where specific entities participate in an interaction, and subsets of those entities also interact with each other. Traditional modeling approaches to higher-order networks tend to either not consider inclusion at all (e.g., hypergraph models) or explicitly assume perfect and complete inclusion (e.g., simplicial complex models). To allow for a more nuanced assessment of inclusion in higher-order networks, we introduce the concept of “simpliciality” and several corresponding measures. Contrary to current modeling practice, we show that empirically observed systems rarely lie at either end of the simpliciality spectrum. In addition, we show that generative models fitted to these datasets struggle to capture their inclusion structure. These findings suggest new modeling directions for the field of higher-order network science.