Journal of Computational Social Science最新文献

Since early 2020, the global coronavirus pandemic has strained economic activities and traditional lifestyles. For such emergencies, our paper proposes a social sentiment estimation model that changes in response to infection conditions and state government orders. By designing mediation keywords that do not directly evoke coronavirus, it is possible to observe sentiment waveforms that vary as confirmed cases increase or decrease and as behavioral restrictions are ordered or lifted over a long period. The model demonstrates guaranteed performance with transformer-based neural network models and has been validated in New York City, Los Angeles, and Chicago, given that coronavirus infections explode in overcrowded cities. The time-series of the extracted social sentiment reflected the infection conditions of each city during the 2-year period from pre-pandemic to the new normal and shows a concurrency of waveforms common to the three cities. The methods of this paper could be applied not only to analysis of the COVID-19 pandemic but also to analyses of a wide range of emergencies and they could be a policy support tool that complements traditional surveys in the future.

One of the first steps in many text-based social science studies is to retrieve documents that are relevant for an analysis from large corpora of otherwise irrelevant documents. The conventional approach in social science to address this retrieval task is to apply a set of keywords and to consider those documents to be relevant that contain at least one of the keywords. But the application of incomplete keyword lists has a high risk of drawing biased inferences. More complex and costly methods such as query expansion techniques, topic model-based classification rules, and active as well as passive supervised learning could have the potential to more accurately separate relevant from irrelevant documents and thereby reduce the potential size of bias. Yet, whether applying these more expensive approaches increases retrieval performance compared to keyword lists at all, and if so, by how much, is unclear as a comparison of these approaches is lacking. This study closes this gap by comparing these methods across three retrieval tasks associated with a data set of German tweets (Linder in SSRN, 2017. 10.2139/ssrn.3026393), the Social Bias Inference Corpus (SBIC) (Sap et al. in Social bias frames: reasoning about social and power implications of language. In: Jurafsky et al. (eds) Proceedings of the 58th annual meeting of the association for computational linguistics. Association for Computational Linguistics, p 5477-5490, 2020. 10.18653/v1/2020.aclmain.486), and the Reuters-21578 corpus (Lewis in Reuters-21578 (Distribution 1.0). [Data set], 1997. http://www.daviddlewis.com/resources/testcollections/reuters21578/). Results show that query expansion techniques and topic model-based classification rules in most studied settings tend to decrease rather than increase retrieval performance. Active supervised learning, however, if applied on a not too small set of labeled training instances (e.g. 1000 documents), reaches a substantially higher retrieval performance than keyword lists.

As part of the "text-as-data" movement, Natural Language Processing (NLP) provides a computational way to examine political polarization. We conducted a methodological scoping review of studies published since 2010 (n = 154) to clarify how NLP research has conceptualized and measured political polarization, and to characterize the degree of integration of the two different research paradigms that meet in this research area. We identified biases toward US context (59%), Twitter data (43%) and machine learning approach (33%). Research covers different layers of the political public sphere (politicians, experts, media, or the lay public), however, very few studies involved more than one layer. Results indicate that only a few studies made use of domain knowledge and a high proportion of the studies were not interdisciplinary. Those studies that made efforts to interpret the results demonstrated that the characteristics of political texts depend not only on the political position of their authors, but also on other often-overlooked factors. Ignoring these factors may lead to overly optimistic performance measures. Also, spurious results may be obtained when causal relations are inferred from textual data. Our paper provides arguments for the integration of explanatory and predictive modeling paradigms, and for a more interdisciplinary approach to polarization research.

Supplementary information: The online version contains supplementary material available at 10.1007/s42001-022-00196-2.

Intimate partner violence (IPV) is a significant public health problem that adversely affects the well-being of victims. IPV is often under-reported and non-physical forms of violence may not be recognized as IPV, even by victims. With the increasing popularity of social media and due to the anonymity provided by some of these platforms, people feel comfortable sharing descriptions of their relationship problems in social media. The content generated in these platforms can be useful in identifying IPV and characterizing the prevalence, causes, consequences, and correlates of IPV in broad populations. However, these descriptions are in the form of free text and no corpus of labeled data is available to perform large-scale computational and statistical analyses. Here, we use data from established questionnaires that are used to collect self-report data on IPV to train machine learning models to predict IPV from free text. Using Universal Sentence Encoder (USE) along with multiple machine learning algorithms (random forest, SVM, logistic regression, Naïve Bayes), we develop DetectIPV, a tool for detecting IPV in free text. Using DetectIPV, we comprehensively characterize the predictability of different types of violence (physical abuse, emotional abuse, sexual abuse) from free text. Our results show that a general model that is trained using examples of all violence types can identify IPV from free text with area under the ROC curve (AUROC) 89%. We also train type-specific models and observe that physical abuse can be identified with greatest accuracy (AUROC 98%), while sexual abuse can be identified with high precision but relatively low recall. While our results indicate that the prediction of emotional abuse is the most challenging, DetectIPV can identify emotional abuse with AUROC above 80%. These results establish DetectIPV as a tool that can be used to reliably detect IPV in the context of various applications, ranging from flagging social media posts to detecting IPV in large text corpuses for research purposes. DetectIPV is available as a web service at https://www.ipvlab.case.edu/ipvdetect/.