Search

We apply a dynamic influence model to the opinions of the US federal courts to examine the role of the US Supreme Court in influencing the direction of legal discourse in the federal courts. We propose two mechanisms for how the Court affects innovation in legal language: a selection mechanism where the Court's influence primarily derives from its discretionary jurisdiction, and an authorship mechanism in which the Court's influence derives directly from its own innovations. To test these alternative hypotheses, we develop a novel influence measure based on a dynamic topic model that separates the Court's own language innovations from those of the lower courts. Applying this measure to the US federal courts, we find that the Supreme Court primarily exercises influence through the selection mechanism, with modest additional influence attributable to the authorship mechanism. This article is part of the theme issue 'A complexity science approach to law and governance'. [ABSTRACT FROM AUTHOR]

As one of the oldest continuously publishing journals in statistics (published since 1901), Biometrika provides a unique window onto the history of statistics and its epistemic development throughout the 20th and the beginning of the 21st centuries. While the early history of the discipline, with the works of key figures, such as Karl Pearson, Francis Galton, or Ronald Fisher, is relatively well known, the later (and longer) episodes of its intellectual development remain understudied. By applying digital tools to the full-text corpus of the journal articles (N  =   5,596), the objective of this study is to provide a novel quantitative exploration of the history of the statistical sciences via an all-encompassing view of 120 years of Biometrika. To this aim, topic-modelling analyses are used and provide insights into the epistemic content of the journal and its evolution. Striking changes in the thematic content of the journal are documented and quantified for the first time, from the decline of Pearsonian and Weldonian biometrical research and the journal's tight connection to biology in the 1930s to the rise of modern statistical methods beginning in the 1960s and 1970s. Newly developed approaches are used to infer author networks from publication topics. The resulting network of authors shows the existence of several communities, well-aligned with topic clusters and their evolution through time. It also highlights the role of specific figures over more than a century of publishing history and provides a first window onto the foundation, development, and diverse applications of the statistical sciences. [ABSTRACT FROM AUTHOR]

Between 2015 and 2017, the Internet Research Agency (IRA) – a Kremlin-backed "troll farm" based in St. Petersburg – executed a propaganda campaign on Twitter to target US voters. Scholarship has expended relatively little effort to study the role of Islamophobia in the IRA's propaganda campaign. Following critical disinformation research, this article demonstrates that Islamophobia, affect, and white identity played a crucial role in the IRA's targeting of right-wing US voters. With an official release of tweets and associated visual content from Twitter, we use topic modeling and visual analysis to explore both how, and to what extent, the IRA used Islamophobia in its propaganda. To do so, we develop a multimodal distant reading technique to study how the IRA aligned users with contemporary far right social movements by deploying racial and emotional appeals that center on narrating a transnational white identity under threat from Islam and Muslims. [ABSTRACT FROM AUTHOR]

As an extension to the standard symmetric latent Dirichlet allocation topic model, we implement asymmetric Beta-Liouville as a conjugate prior to the multinomial and therefore propose the maximum a posteriori for latent Beta-Liouville allocation as an alternative to maximum likelihood estimator for models such as probabilistic latent semantic indexing, unigrams, and mixture of unigrams. Since most Bayesian posteriors, for complex models, are intractable in general, we propose a point estimate (the mode) that offers a much tractable solution. The maximum a posteriori hypotheses using point estimates are much easier than full Bayesian analysis that integrates over the entire parameter space. We show that the proposed maximum a posteriori reduces the three-level hierarchical latent Beta-Liouville allocation to two-level topic mixture as we marginalize out the latent variables. In each document, the maximum a posteriori provides a soft assignment and constructs dense expectation–maximization probabilities over each word (responsibilities) for accurate estimates. For simplicity, we present a stochastic at word-level online expectation–maximization algorithm as an optimization method for maximum a posteriori latent Beta-Liouville allocation estimation whose unnormalized reparameterization is equivalent to a stochastic collapsed variational Bayes. This implicit connection between the collapsed space and expectation–maximization-based maximum a posteriori latent Beta-Liouville allocation shows its flexibility and helps in providing alternative to model selection. We characterize efficiency in the proposed approach for its ability to simultaneously stream both large-scale data and parameters seamlessly. The performance of the model using predictive perplexities as evaluation method shows the robustness of the proposed technique with text document datasets. [ABSTRACT FROM AUTHOR]

To address the limitations of existing methods of short-text entity disambiguation, specifically in terms of their insufficient feature extraction and reliance on massive training samples, we propose an entity disambiguation model called COLBERT, which fuses LDA-based topic features and BERT-based semantic features, as well as using contrastive learning, to enhance the disambiguation process. Experiments on a publicly available Chinese short-text entity disambiguation dataset show that the proposed model achieves an F1-score of 84.0%, which outperforms the benchmark method by 0.6%. Moreover, our model achieves an F1-score of 74.5% with a limited number of training samples, which is 2.8% higher than the benchmark method. These results demonstrate that our model achieves better effectiveness and robustness and can reduce the burden of data annotation as well as training costs. [ABSTRACT FROM AUTHOR]

Topic models have been widely used by researchers across disciplines to automatically analyze large textual data. However, they often fail to automate content analysis, because the algorithms cannot accurately classify individual sentences into pre-defined topics. Aiming to make topic classification more theoretically grounded and content analysis in general more topic-specific, we have developed Seeded Sequential Latent Dirichlet allocation (LDA), extending the existing LDA algorithm, and implementing it in a widely accessible open-source package. Taking a large corpus of speeches delivered by delegates at the United Nations General Assembly as an example, we explain how our algorithm differs from the original algorithm; why it can classify sentences more accurately; how it accepts pre-defined topics in deductive or semi-deductive analysis; how such ex-ante topic mapping differs from ex-post topic mapping; how it enables topic-specific framing analysis in applied research. We also offer practical guidance on how to determine the optimal number of topics and select seed words for the algorithm. [ABSTRACT FROM AUTHOR]

Cultural ecosystem services (CES) are of significant interest due to their effectiveness in assessing the impact of ecosystem degradation on human well-being. The evaluation of spatial coordination between the supply and demand of urban CES is crucial for guiding the sustainable development of urban areas. However, current research and survey methods are time-consuming and laborious, lacking in full integration of beneficiary opinions, and insufficient in examining multiple CES in an integrated manner. In this study, we have developed a novel framework that integrates an investigation-driven approach with a natural language processing topic model to assess the supply and demand of CES and their spatial correspondence. Using the main urban area of Xi'an as a case study, we analyzed the spatial coherence of supply and demand including aesthetic, cultural, educational, recreational, and spiritual services. Our findings demonstrate that the fusion of multiple models can effectively evaluate the CES supply from a natural perspective of nature and the demand for CES from a beneficiary perspective. High-supply areas for CES are concentrated in the center region of the study area and in proximity to roads and rivers. Aesthetic service supply was higher than that of other service types, followed by recreational and spiritual services supply. Demand is notably high in the city's hinterland, with substantial interest in recreational and aesthetic services. We observed mismatches between supply and demand across service categories ranging from 33.58% to 67.64%, with supply exceeding demand in most regions. Therefore, tailored measures should be implemented to address the supply and demand mismatch of CES based on local conditions. Our research introduces a new technical tool for analyzing the supply-demand matching relationship of CES and establishes a theoretical foundation for sustainable urban management. • Topic model can effectively study cultural ecosystem services (CES) demand. • The distribution of supply and demand for CES are highly localized. • A framework for supply and demand assessment by urban CES. • The framework addresses the difficulty of measuring services at the city scale. [ABSTRACT FROM AUTHOR]

Benchmarking is one of the key management approaches to support companies to continuously improve and gain competitive advantage. However, previous benchmarking management suffered from subjectivity and one-sidedness, and the benchmarking evaluation was only to identify the benchmark and its own deficiencies, with a high degree of independence between the sub-processes. The framework is the first to jointly apply topic mining and relation extraction techniques to establish an objective index system and mine the mechanistic relations among the indexes, and finally provides a concise and interactive display of the results based on knowledge graph visualization. In this study, 50 oil and gas companies are taken as examples to carry out the case study of sustainable development benchmarking management, and a sustainable development program with strong logic and systematicity is proposed at both macro and micro levels. This study results verify that the framework can effectively improve the scientific, comprehensive and systematic nature of sustainable development benchmarking management. At the same time, the framework extends the role of the index system so that it can provide the necessary mechanistic guidance for subsequent sustainable development programmatic to achieve programmatic with a high degree of synergy. In addition, the framework has a high degree of coherence with the United Nations Sustainable Development Goals (SDGs), and the proposals put forward can not only be used to improve the sustainable development capability of companies themselves, but also contribute to the realization of the SDGs. [Display omitted] • A novel sustainable development benchmarking framework for energy companies. • A first joint application of topic mining and relation extraction for benchmarking. • Energy companies should promote localization and diversification of the workforce. • Energy companies should expand joint venture projects with developing country. [ABSTRACT FROM AUTHOR]

Stigma surrounding suicide is a massive problem in Indonesia. Thus, it is important to study how conversations about suicide take place. We take a machine learning approach and study tweets with suicide keywords to understand how people converse about suicide or express suicide ideation. Tweets with suicide-related keywords were extracted from May to June 2023. 20,057 tweets were subject to topic modelling with an 11-topic solution. While most topics contain negative messages, no purely stigmatizing topics emerge, despite prior research suggesting overwhelming stigma. Various kinds of existential, emotional, and social tweets about suicide take place among Indonesian users, indicating that Indonesian Twitter users utilize the platform to express their thoughts and emotions. Notably, religious-spiritual keywords are highly prevalent, suggesting that in a highly religious society, there is a need for policy makers and awareness campaigns to frame their positive messaging within the society’s religious context. [ABSTRACT FROM AUTHOR]

The Transformer-based summarization models rely solely on the attention mechanism for document encoding, making it difficult to accurately capture long-range dependencies in long documents due to the presence of attention redundancy. To address this issue, we propose an extractive summarization framework guided by a topic model (TopicSum) that utilizes a heterogeneous graph neural network to leverage the topic information as document-level features during the sentence selection process, thereby capturing the long-range dependencies among sentences. The sentence-level features in this topic model align with the basic unit of the extractive summarization task. Additionally, a memory mechanism is employed to dynamically store and update the memory module, reducing the potential of repetitive information guiding sentence selection. We evaluated the model on three large document datasets, namely Pubmed, arXiv, and GovReport, and achieved significantly higher Rouge scores than previous works, including extractive and abstractive models. Furthermore, our experiments demonstrate that recent highly regarded large language models such as ChatGPT are insufficient to handle the long document summarization task directly. The proposed approach in this paper exhibits sufficient competitiveness in terms of both generation quality and deployment conditions. [ABSTRACT FROM AUTHOR]

Modelling short text is challenging due to the small number of word co-occurrence and insufficient semantic information that affects downstream Natural Language Processing (NLP) tasks, for example, text classification. Gathering information from external sources is expensive and may increase noise. For efficient short text classification without depending on external knowledge sources, we propose Expressive Short text Classification (EStC). EStC consists of a novel document context-aware semantically enriched topic model called the Short text Topic Model (StTM) that captures words, topics and documents semantics in a joint learning framework. In StTM, the probability of predicting a context word involves the topic distribution of word embeddings and the document vector as the global context, which obtains by weighted averaging of word embeddings on the fly simultaneously with the topic distribution of words without requiring an additional inference method for the document embedding. EStC represents documents in an expressive (number of topics × number of word embedding features) embedding space and uses a linear support vector machine (SVM) classifier for their classification. Experimental results demonstrate that EStC outperforms many state-of-the-art language models in short text classification using several publicly available short text data sets. [ABSTRACT FROM AUTHOR]

Currently, Android malware detection methods always focus on one kind of app feature, such as structural, semantic, or other statistical features. This paper proposes a novel Android malware detection method that integrates multiple features of Android applications. First, to effectively extract the structural and semantic features, we propose a new type of call graph named the class-set call graph (CSCG) that uses the sets of Java classes as nodes and the call relationships between class sets as edges, and we design a dynamic adaptive CSCG construction method that can automatically determine the node size for applications with different scales. The topic model is used to mine the source code semantics from the class sets as the node features. Then, we use a graph attention network (GAT) with max pooling to extract the CSCG feature that encompass both the semantic and structural features of the Android application. Furthermore, we construct a deep multimodal feature fusion network to fuse the CSCG features with permission features. Experimental results show that our method achieves a detection accuracy of 97.28%–99.54% on the three constructed datasets, which is better than the existing methods. • Proposing multimodal features deep fusion method for Android malware detection. • Designing class-set call graph to integrates structural and semantic features. • Adaptive class merging method for class-set call graph construction. • Introducing GAT and max pooling to extract salient graph features. • Proposing fusion network for graph features and permission features. [ABSTRACT FROM AUTHOR]