This article introduces embComp, a novel approach for comparing two embeddings that capture the similarity between objects, such as word and document embeddings. We survey scenarios where comparing these embedding spaces is useful. From those scenarios, we derive common tasks, introduce visual analysis methods that support these tasks, and combine them into a comprehensive system. One of embComp's central features are overview visualizations that are based on metrics for measuring differences in the local structure around objects. Summarizing these local metrics over the embeddings provides global overviews of similarities and differences. Detail views allow comparison of the local structure around selected objects and relating this local information to the global views. Integrating and connecting all of these components, embComp supports a range of analysis workflows that help understand similarities and differences between embedding spaces. We assess our approach by applying it in several use cases, including understanding corpora differences via word vector embeddings, and understanding algorithmic differences in generating embeddings.

Most visual analytics systems assume that all foraging for data happens before the analytics process; once analysis begins, the set of data attributes considered is fixed. Such separation of data construction from analysis precludes iteration that can enable foraging informed by the needs that arise in-situ during the analysis. The separation of the foraging loop from the data analysis tasks can limit the pace and scope of analysis. In this paper, we present CAVA, a system that integrates data curation and data augmentation with the traditional data exploration and analysis tasks, enabling information foraging in-situ during analysis. Identifying attributes to add to the dataset is difficult because it requires human knowledge to determine which available attributes will be helpful for the ensuing analytical tasks. CAVA crawls knowledge graphs to provide users with a a broad set of attributes drawn from external data to choose from. Users can then specify complex operations on knowledge graphs to construct additional attributes. CAVA shows how visual analytics can help users forage for attributes by letting users visually explore the set of available data, and by serving as an interface for query construction. It also provides visualizations of the knowledge graph itself to help users understand complex joins such as multi-hop aggregations. We assess the ability of our system to enable users to perform complex data combinations without programming in a user study over two datasets. We then demonstrate the generalizability of CAVA through two additional usage scenarios. The results of the evaluation confirm that CAVA is effective in helping the user perform data foraging that leads to improved analysis outcomes, and offer evidence in support of integrating data augmentation as a part of the visual analytics pipeline.

Abstract Machine learning practitioners often compare the results of different classifiers to help select, diagnose and tune models. We present Boxer, a system to enable such comparison. Our system facilitates interactive exploration of the experimental results obtained by applying multiple classifiers to a common set of model inputs. The approach focuses on allowing the user to identify interesting subsets of training and testing instances and comparing performance of the classifiers on these subsets. The system couples standard visual designs with set algebra interactions and comparative elements. This allows the user to compose and coordinate views to specify subsets and assess classifier performance on them. The flexibility of these compositions allow the user to address a wide range of scenarios in developing and assessing classifiers. We demonstrate Boxer in use cases including model selection, tuning, fairness assessment, and data quality diagnosis.

Magic lens based focus+context techniques are powerful means for exploring document spatializations. Typically, they only offer additional summarized or abstracted views on focused documents. As a consequence, users might miss important information that is either not shown in aggregated form or that never happens to get focused. In this work, we present the design process and user study results for improving a magic lens based document exploration approach with exemplary visual quality cues to guide users in steering the exploration and support them in interpreting the summarization results. We contribute a thorough analysis of potential sources of information loss involved in these techniques, which include the visual spatialization of text documents, user-steered exploration, and the visual summarization. With lessons learned from previous research, we highlight the various ways those information losses could hamper the exploration. Furthermore, we formally define measures for the aforementioned different types of information losses and bias. Finally, we present the visual cues to depict these quality measures that are seamlessly integrated into the exploration approach. These visual cues guide users during the exploration and reduce the risk of misinterpretation and accelerate insight generation. We conclude with the results of a controlled user study and discuss the benefits and challenges of integrating quality guidance in exploration techniques.

Magic lens based focus+context techniques are powerful means for exploring document spatializations. Typically, they only offer additional summarized or abstracted views on focused documents. As a consequence, users might miss important information that is either not shown in aggregated form or that never happens to get focused. In this work, we present the design process and user study results for improving a magic lens based document exploration approach with exemplary visual quality cues to guide users in steering the exploration and support them in interpreting the summarization results. We contribute a thorough analysis of potential sources of information loss involved in these techniques, which include the visual spatialization of text documents, user-steered exploration, and the visual summarization. With lessons learned from previous research, we highlight the various ways those information losses could hamper the exploration. Furthermore, we formally define measures for the aforementioned different types of information losses and bias. Finally, we present the visual cues to depict these quality measures that are seamlessly integrated into the exploration approach. These visual cues guide users during the exploration and reduce the risk of misinterpretation and accelerate insight generation. We conclude with the results of a controlled user study and discuss the benefits and challenges of integrating quality guidance in exploration techniques.

Abstract Many visual analytics systems allow users to interact with machine learning models towards the goals of data exploration and insight generation on a given dataset. However, in some situations, insights may be less important than the production of an accurate predictive model for future use. In that case, users are more interested in generating of diverse and robust predictive models, verifying their performance on holdout data, and selecting the most suitable model for their usage scenario. In this paper, we consider the concept of Exploratory Model Analysis (EMA), which is defined as the process of discovering and selecting relevant models that can be used to make predictions on a data source. We delineate the differences between EMA and the well‐known term exploratory data analysis in terms of the desired outcome of the analytic process: insights into the data or a set of deployable models. The contributions of this work are a visual analytics system workflow for EMA, a user study, and two use cases validating the effectiveness of the workflow. We found that our system workflow enabled users to generate complex models, to assess them for various qualities, and to select the most relevant model for their task.

Abstract Word vector embeddings are an emerging tool for natural language processing. They have proven beneficial for a wide variety of language processing tasks. Their utility stems from the ability to encode word relationships within the vector space. Applications range from components in natural language processing systems to tools for linguistic analysis in the study of language and literature. In many of these applications, interpreting embeddings and understanding the encoded grammatical and semantic relations between words is useful, but challenging. Visualization can aid in such interpretation of embeddings. In this paper, we examine the role for visualization in working with word vector embeddings. We provide a literature survey to catalogue the range of tasks where the embeddings are employed across a broad range of applications. Based on this survey, we identify key tasks and their characteristics. Then, we present visual interactive designs that address many of these tasks. The designs integrate into an exploration and analysis environment for embeddings. Finally, we provide example use cases for them and discuss domain user feedback.

The increasingly large number of available writings describing technical and scientific progress, calls for advanced analytic tools for their efficient analysis. This is true for many application scenarios in science and industry and for different types of writings, comprising patents and scientific articles. Despite important differences between patents and scientific articles, both have a variety of common characteristics that lead to similar search and analysis tasks. However, the analysis and visualization of these documents is not a trivial task due to the complexity of the documents as well as the large number of possible relations between their multivariate attributes. In this survey, we review interactive analysis and visualization approaches of patents and scientific articles, ranging from exploration tools to sophisticated mining methods. In a bottom-up approach, we categorize them according to two aspects: (a) data type (text, citations, authors, metadata, and combinations thereof), and (b) task (finding and comparing single entities, seeking elementary relations, finding complex patterns, and in particular temporal patterns, and investigating connections between multiple behaviours). Finally, we identify challenges and research directions in this area that ask for future investigations.

We have created and made available to all a dataset with information about every paper that has appeared at the IEEE Visualization (VIS) set of conferences: InfoVis, SciVis, VAST, and Vis. The information about each paper includes its title, abstract, authors, and citations to other papers in the conference series, among many other attributes. This article describes the motivation for creating the dataset, as well as our process of coalescing and cleaning the data, and a set of three visualizations we created to facilitate exploration of the data. This data is meant to be useful to the broad data visualization community to help understand the evolution of the field and as an example document collection for text data visualization research.

The analysis of inherent structures of movies plays an important role in studying stylistic devices and specific, content-related questions. Examples are the analysis of personal constellations in movie scenes, dialogue-based content analysis, or the investigation of image-based features. We provide a visual analytics approach that supports the analytical reasoning process to derive higher level insights about the content on a semantic level. Combining automatic methods for semantic scene analysis based on script and subtitle text, we perform a low-level analysis of the data automatically. Our approach features an interactive visualization that allows a multilayer interpretation of descriptive features to characterize movie content. For semantic analysis, we extract scene information from movie scripts and match them with the corresponding subtitles. With text- and image-based query techniques, we facilitate an interactive comparison of different movie scenes on an image and on a semantic level. We demonstrate how our approach can be applied for content analysis on a popular Hollywood movie.

The exploration and analysis of scientific literature collections is an important task for effective knowledge management. Past interest in such document sets has spurred the development of numerous visualization approaches for their interactive analysis. They either focus on the textual content of publications, or on document metadata including authors and citations. Previously presented approaches for citation analysis aim primarily at the visualization of the structure of citation networks and their exploration. We extend the state-of-the-art by presenting an approach for the interactive visual analysis of the contents of scientific documents, and combine it with a new and flexible technique to analyze their citations. This technique facilitates user-steered aggregation of citations which are linked to the content of the citing publications using a highly interactive visualization approach. Through enriching the approach with additional interactive views of other important aspects of the data, we support the exploration of the dataset over time and enable users to analyze citation patterns, spot trends, and track long-term developments. We demonstrate the strengths of our approach through a use case and discuss it based on expert user feedback.

We present a new visualization approach for displaying eye tracking data from multiple participants. We aim to show the spatio-temporal data of the gaze points in the context of the underlying image or video stimulus without occlusion. Our technique, denoted as gaze stripes, does not require the explicit definition of areas of interest but directly uses the image data around the gaze points, similar to thumbnails for images. A gaze stripe consists of a sequence of such gaze point images, oriented along a horizontal timeline. By displaying multiple aligned gaze stripes, it is possible to analyze and compare the viewing behavior of the participants over time. Since the analysis is carried out directly on the image data, expensive post-processing or manual annotation are not required. Therefore, not only patterns and outliers in the participants' scanpaths can be detected, but the context of the stimulus is available as well. Furthermore, our approach is especially well suited for dynamic stimuli due to the non-aggregated temporal mapping. Complementary views, i.e., markers, notes, screenshots, histograms, and results from automatic clustering, can be added to the visualization to display analysis results. We illustrate the usefulness of our technique on static and dynamic stimuli. Furthermore, we discuss the limitations and scalability of our approach in comparison to established visualization techniques.

Word clouds provide a simple and effective means to visually communicate the most frequent words of text documents. However, only few word cloud visualizations support the contrastive analysis of multiple texts. This paper introduces Concentri Cloud, a layered word cloud layout that merges the words from several text documents into a single visualization. The weighted words are arranged in a concentric layout, with those representing the individual documents on the outer circle and the merged ones on inner circles. Interaction techniques allow to analyze the word cloud composition and to provide details on demand. The approach has been implemented and tested on several examples. A qualitative evaluation indicates the general value of Concentri Cloud and reveals benefits and limitations.

We introduce a new design for the visual analysis of eye tracking data recorded from dynamic stimuli such as video. ISeeCube includes multiple coordinated views to support different aspects of various analysis tasks. It combines methods for the spatiotemporal analysis of gaze data recorded from unlabeled videos as well as the possibility to annotate and investigate dynamic Areas of Interest (AOIs). A static overview of the complete data set is provided by a space-time cube visualization that shows gaze points with density-based color mapping and spatiotemporal clustering of the data. A timeline visualization supports the analysis of dynamic AOIs and the viewers' attention on them. AOI-based scanpaths of different viewers can be clustered by their Levenshtein distance, an attention map, or the transitions between AOIs. With the provided visual analytics techniques, the exploration of eye tracking data recorded from several viewers is supported for a wide range of analysis tasks.

We introduce a new design for the visual analysis of eye tracking data recorded from dynamic stimuli such as video. ISeeCube includes multiple coordinated views to support different aspects of various analysis tasks. It combines methods for the spatiotemporal analysis of gaze data recorded from unlabeled videos as well as the possibility to annotate and investigate dynamic Areas of Interest (AOIs). A static overview of the complete data set is provided by a space-time cube visualization that shows gaze points with density-based color mapping and spatiotemporal clustering of the data. A timeline visualization supports the analysis of dynamic AOIs and the viewers' attention on them. AOI-based scanpaths of different viewers can be clustered by their Levenshtein distance, an attention map, or the transitions between AOIs. With the provided visual analytics techniques, the exploration of eye tracking data recorded from several viewers is supported for a wide range of analysis tasks.

The number of microblog posts published daily has reached a level that hampers the effective retrieval of relevant messages, and the amount of information conveyed through services such as Twitter is still increasing. Analysts require new methods for monitoring their topic of interest, dealing with the data volume and its dynamic nature. It is of particular importance to provide situational awareness for decision making in time-critical tasks. Current tools for monitoring microblogs typically filter messages based on user-defined keyword queries and metadata restrictions. Used on their own, such methods can have drawbacks with respect to filter accuracy and adaptability to changes in trends and topic structure. We suggest ScatterBlogs2, a new approach to let analysts build task-tailored message filters in an interactive and visual manner based on recorded messages of well-understood previous events. These message filters include supervised classification and query creation backed by the statistical distribution of terms and their co-occurrences. The created filter methods can be orchestrated and adapted afterwards for interactive, visual real-time monitoring and analysis of microblog feeds. We demonstrate the feasibility of our approach for analyzing the Twitter stream in emergency management scenarios.

Performing exhaustive searches over a large number of text documents can be tedious, since it is very hard to formulate search queries or define filter criteria that capture an analyst's information need adequately. Classification through machine learning has the potential to improve search and filter tasks encompassing either complex or very specific information needs, individually. Unfortunately, analysts who are knowledgeable in their field are typically not machine learning specialists. Most classification methods, however, require a certain expertise regarding their parametrization to achieve good results. Supervised machine learning algorithms, in contrast, rely on labeled data, which can be provided by analysts. However, the effort for labeling can be very high, which shifts the problem from composing complex queries or defining accurate filters to another laborious task, in addition to the need for judging the trained classifier's quality. We therefore compare three approaches for interactive classifier training in a user study. All of the approaches are potential candidates for the integration into a larger retrieval system. They incorporate active learning to various degrees in order to reduce the labeling effort as well as to increase effectiveness. Two of them encompass interactive visualization for letting users explore the status of the classifier in context of the labeled documents, as well as for judging the quality of the classifier in iterative feedback loops. We see our work as a step towards introducing user controlled classification methods in addition to text search and filtering for increasing recall in analytics scenarios involving large corpora.