Topic Models

In this paper, we focus on cross-modal (visual and textual) e-commerce search within the fashion domain. Particularly, we investigate two tasks: 1) given a query image, we retrieve textual descriptions that correspond to the visual attributes in the query; and 2) given a textual query that may express an interest in specific visual product characteristics, we retrieve relevant images that exhibit the required visual attributes. To this end, we introduce a new dataset that consists of 53,689 images coupled with textual descriptions. The images contain fashion garments that display a great variety of visual attributes, such as different shapes, colors and textures in natural language. Unlike previous datasets, the text provides a rough and noisy description of the item in the image. We extensively analyze this dataset in the context of cross-modal e-commerce search. We investigate two state-of-the-art latent variable models to bridge between textual and visual data: bilingual latent Dirichlet allocation and canonical correlation analysis. We use state-of-the-art visual and textual features and report promising results.

When did European modes of political thought diverge from those that existed in other world regions? We compare Muslim and Christian political advice texts from the medieval period using automated text analysis to identify four major and 60 granular themes common to Muslim and Christian polities, and examine how emphasis on these topics evolves over time. For Muslim texts, we identify an inflection point in political discourse between the eleventh to thirteenth centuries, a juncture that historians suggest is an ideational watershed brought about by the Turkic and Mongol invaders. For Christian texts, we identify a decline in the relevance of religious appeals from the Middle Ages to the Renaissance. Our findings also suggest that Machiavelli’s Prince was less a turn away from religious discourse on statecraft than the culmination of centuries-long developments in European advice literature.

In this paper, we present a framework that combines automatic and manual approaches to discover themes in disaster-related tweets.  As case study, we decided to focus on tweets related to typhoon Haiyan, which caused billions of dollars in damages.  We collected tweets from November 2013 to March 2014 and used the local typhoon name “Yolanda” as filter.  Data association was used to expand the tweet set and k-means clustering was then applied.  Clusters with high number of instances were subjected to open coding for labeling.  The Silhouette indices ranged from 0.27 to 0.50.  Analyses reveal that the use of automated Natural Language Processing (NLP) approach has the potential to deal with huge volumes of tweets by clustering frequently occurring words and phrases.  This complements the manual approach to surface themes from a more manageable set of tweet pool, allowing for a more nuanced analysis of tweets from a human expert.  As application, the themes extracted were used as categories in a classifier system.  Future work could explore on using topic models and focusing on specific content or issues, such as natural calamities and citizen’s participation in addressing these.

Increasingly, management researchers are using topic modeling, a new method borrowed from computer science, to reveal phenomenon-based constructs and grounded conceptual relationships in textual data. By conceptualizing topic modeling as the process of rendering constructs and conceptual relationships from textual data, we demonstrate how this new method can advance management scholarship without turning topic modeling into a black box of complex computer-driven algorithms. We begin by comparing features of topic modeling to related techniques (content analysis, grounded theorizing, and natural language processing). We then walk through the steps of rendering with topic modeling and apply rendering to management articles that draw on topic modeling. Doing so enables us to identify and discuss how topic modeling has advanced management theory in five areas: detecting novelty and emergence, developing inductive classification systems, understanding online audiences and products, analyzing frames and social movements, and understanding cultural dynamics. We conclude with a review of new topic modeling trends and revisit the role of researcher interpretation in a world of computer-driven textual analysis. We would like to thank the editors of the Academy of Management Annals for their support and helpful comments. We also thank the participants in our various topic modeling presentations and reviewers and division organizers (specifically Peer Fiss and Renate Meyer) at the Academy of Management meetings. In addition, we would like to recognize Marc-David Seidel and Christopher Steele from the Interpretive Data Science (IDeaS) group for their role in germinating these ideas, Mike Pfarrer for his comments on a later draft of the paper, and Kara Gehman for her fine-grained edits on next-to-final drafts. Finally, we wish to express our appreciation to our life partners for not only putting up with but also actively discussing this paper as it evolved.

The term digital transformation (DT) captures the changes in our personal, social, economic, political, cultural and academic life brought about by the rapid diffusion of information and communication technologies. With the increasing digital connectivity, the rise of platform economies and the cost reductions in artificial neural networks, the collection, processing and analysis of large amounts of Big Data have become feasible and widespread.

The combination of Big Data and artificial intelligence (BDAI) is exerting a great impact on the natural, life, physical and engineering sciences in which research has generally been conducted with a more positivist attitude. Here, BDAI reconfigures the scientific research process as it reduces the reliance on a priori theories: Making plausible, but not yet proven statements based on interpreting collected data (so-called “abductive reasoning”) now plays a prominent role in the formulation of new hypotheses and theory-building. As has been argued convincingly, this represents a significant epistemological change (Kitchin, 2014). In contrast, for the social sciences and humanities, representing areas of scholarship that are highly diverse in their philosophical underpinnings, it is less clear what the consequences of the digital transformation will be.

The contributions gathered in this collection shed light on this question. They are the outcome of the workshop “The Digital Transformation – Implications for the Social Sciences and the Humanities” that took place at the German Institute for Japanese Studies (DIJ) on 24 and 25 September 2019. Co-organized by the Nippon Institute for Research Advancement (NIRA) and the German Centre for Research and Innovation Tokyo (DWIH), the workshop brought together scholars from Japan and from abroad, many of the latter being specialists on Japan. This is why the Japanese case is given special attention in many contributions. Even there, however, the insights are not limited to Japan, but address phenomena and implications of the DT of general relevance, even though they may manifest themselves elsewhere under different conditions.

A huge amount of potentially dangerous COVID-19 misinformation is appearing online. Here we use machine learning to quantify COVID-19 content among online opponents of establishment health guidance, in particular vaccinations ("anti-vax"). We find that the anti-vax community is developing a less focused debate around COVID-19 than its counterpart, the pro-vaccination ("pro-vax") community. However, the anti-vax community exhibits a broader range of "flavors" of COVID-19 topics, and hence can appeal to a broader cross-section of individuals seeking COVID-19 guidance online, e.g. individuals wary of a mandatory fast-tracked COVID-19 vaccine or those seeking alternative remedies. Hence the anti-vax community looks better positioned to attract fresh support going forward than the pro-vax community. This is concerning since a widespread lack of adoption of a COVID-19 vaccine will mean the world falls short of providing herd immunity, leaving countries open to future COVID-19 resurgences. We provide a mechanistic model that interprets these results and could help in assessing the likely efficacy of intervention strategies. Our approach is scalable and hence tackles the urgent problem facing social media platforms of having to analyze huge volumes of online health misinformation and disinformation.

In this work we analyse the evolution of the vulnerability concept in the research streams of climate change adaptation (CCA) and disaster risk reduction (DRR). We combine a traditional literature review with data mining procedures applied to bibliographic databases to reconstruct the history of the concept within various research topics, showing its evolution and convergences over time. To do that, we integrate different methods combining machine learning algorithms with network and cluster analyses to examine a set of 3757 articles, analysing their distinctive features and similarities on the basis of their contents as well as co-authorships. Bibliometric analyses enable the identification of different communities of articles, pinpointing key papers and authors, while literature review makes it possible to assess the concept of vulnerability evolved within and beyond research communities and scientific networks. Moreover, this work examines the role played by documents published by UN institutions (UNDRO, UNISDR, IPCC) in contributing to the evolution of vulnerability and related concepts. Results show that signs of convergence are evident between the two research streams, and that the IPCC reports have played a major role in proposing solutions for unifying definitions of vulnerability. We observe that the phases of preparation of the IPCC reports are very rich in methodological and terminological developments, while after publication, the literature shows evident signs of propagation of the proposed concepts. The DRR research stream developed before the research stream on CCA, but the latter flourished rapidly and became much larger in terms of number of publications. Nevertheless, in terms of contents, adaptation studies and the IPCC have shown increasing adoption of the concepts developed within the disaster research stream, in particular with regard to the interpretation of vulnerability as one of the dimensions of risk.

An important text mining problem is to find, in a large collection of texts, documents related to specific topics and then discern further structure among the found texts. This problem is especilly important for social sciences, where the purpose is to find the representative documents for subsequent qualitative interpretation. To solvw this problem, we propose an interval semi-supervised LDA approach, in which certain predefined sets of keywords (that define the topics researchers are interested in) are restricted to specific intervals of topic assignments. We present a case study on a Russian LiveJournal dataset aimed at ethnicity discource analysis

This paper suggests the use of automatic topic modeling for large-scale corpora of privacy policies using unsupervised learning techniques. The advantages of using unsupervised learning for this task are numerous. The primary advantages include the ability to analyze any new corpus with a fraction of the effort required by supervised learning, the ability to study changes in topics of interest along time, and the ability to identify finer-grained topics of interest in these privacy policies. Based on general principles of document analysis we synthesize a cohesive framework for privacy policy topic modeling and apply it over a corpus of 4,982 privacy policies of mobile applications crawled from the Google Play Store. The results demonstrate that even with this relatively moderate-size corpus quite comprehensive insights can be attained regarding the focus and scope of current privacy policy documents. The topics extracted, their structure and the applicability of the unsupervised approach for that matter are validated through an extensive comparison to similar findings reported in prior work that uses supervised learning (which heavily depends on manual annotation of experts). The comparison suggests a substantial overlap between the topics found and those reported in prior work, and also unveils some new topics of interest.
CCS CONCEPTS • Information systems → Document topic models; • Theory of computation → Unsupervised learning and clustering.

In this article we apply the method of quantitative textual analysis known as 'topic modelling' to a significant Australian legal text corpus: that of judgments of the High Court of Australia from 1903 to 2015. 1 The High Court of Australia has been a perennial topic for study and analysis. It is the highest court in the Australian judicial hierarchy and the site of many of the most significant contests of legal doctrine and practice in Australian history. We find that the topic models generated by this research enable the development of a range of unique, novel and robust observations of the High Court's judicial workload and the shifting make-up of its legal subject matter over time. Moreover, this article reveals the feasibility and value of topic modelling as a method for the study of legal texts and practices that might fruitfully complement other methods of legal scholarship. To our knowledge this is the first time topic modelling has been applied in this way to the entirety of this legal textual corpus; 2 consequently, in this article we have focused more on presenting the method, deferring for the present

The urban policy mobility literature describes the widespread circulation of policy ideas while highlighting their mutations along the way. At the same time, the literature often analyzes the localization of such ideas by examining their adoption in one or several cities. To better understand policy replications and mutations, we develop theoretical and methodological strategies that provide sensitivity to both local distinctiveness and global variability. We build on the Urban Policy Mobility literature and combine it with ecological theories of conceptual spaces to develop the concept of Urban Model Spaces – a matrix of discursive possibilities evolving from the accumulated replications and localizations of a model. We articulate it via three core properties central to Urban Policy Mobility - Temporality, Scale, and Position and test how they shape the emergence of policy discourses. To demonstrate the concept we analyze public art policy and the funding mechanism of the Percent for Art ordinance from 26 cities combining Structural Topic Modeling and regression analysis.

Purpose À The authors apply topic sentiment analysis (several relatively new text analysis methods) to the study of public opinion as expressed in social media by comparing reactions to the Trayvon Martin controversy in spring 2012 by commenters on the partisan news websites the Huffington Post and Daily Caller.