Sebastian Dungs

A Survey on Visual Information Search Behavior and Requirements of Radiologists

OBJECTIVES The main objective of this study is to learn more on the image use and search requirements of radiologists. These requirements will then be taken into account to develop a new search system for images and associated meta data search in the Khresmoi project. METHODS Observations of the radiology workflow, case discussions and a literature review were performed to construct a survey form that was given online and in paper form to radiologists. Eye tracking was performed on a radiology viewing station to analyze typical tasks and to complement the survey. RESULTS In total 34 radiologists answered the survey online or on paper. Image search was mentioned as a frequent and common task, particularly for finding cases of interest for differential diagnosis. Sources of information besides the Internet are books and discussions with colleagues. Search for images is unsuccessful in around 25% of the cases, stopping the search after around 10 minutes. The most common reason for failure is that target images are considered rare. Important additions for search requested in the survey are filtering by pathology and modality, as well as search for visually similar images and cases. Few radiologists are familiar with visual retrieval but they desire the option to upload images for searching similar ones. CONCLUSIONS Image search is common in radiology but few radiologists are fully aware of visual information retrieval. Taking into account the many unsuccessful searches and time spent for this, a good image search could improve the situation and help in clinical practice.

An Eye-Tracking Study of Query Reformulation

Information about a users domain knowledge and interest can be important signals for many information retrieval tasks such as query suggestion or result ranking. State-of-the-art user models rely on coarse-grained representations of the users previous knowledge about a topic or domain. In this paper, we study query refinement using eye-tracking in order to gain precise and detailed insight into which terms the user was exposed to in a search session and which ones they showed a particular interest in. We measure fixations on the term level, allowing for a detailed model of user attention. To allow for a wide-spread exploitation of our findings, we generalize from the restrictive eye-gaze tracking to using more accessible signals: mouse cursor traces. Based on the public API of a popular search engine, we demonstrate how query suggestion candidates can be ranked according to traces of user attention and interest, resulting in significantly better performance than achieved by an attention-oblivious industry solution. Our experiments suggest that modelling term-level user attention can be achieved with great reliability and holds significant potential for supporting a range of traditional IR tasks.

Advanced Hidden Markov Models for Recognizing Search Phases

Although cognitive IR approaches usually distinguish between different search phases, there are no automatic methods for recognizing these phases. In this paper, we use constrained Hidden Markov Models (HMM), for addressing this issue. Especially, we develop a hybrid form of HMM combining both discrete and continuous signal values, which improves the recognition process. Furthermore, we show how the new model can be used for predicting the time to the next relevant document, which is a prerequisite for the application of the interactive probability ranking principle.

Khresmoi Professional: Multilingual, Multimodal Professional Medical Search

In this demonstration we present the Khresmoi medical search and access system. The system uses a component based architecture housed in the cloud to support target users medical information needs. This includes web systems, computer applications and mobile applications to support the multilingual and multimodal information needs of three test target groups: the general public, general practitioners GPs and radiologists. This demonstration presents the systems for GPs and radiologists providing multilingual text and image based including 2D and 3D radiology images search functionality.

Is It Really Fake? – Towards an Understanding of Fake News in Social Media Communication

This paper outlines the development of Fake News and seeks to clarify different perspectives regarding the term within Social Media communication. Current information systems, such as Social Media platforms, allow real-time communication, enabling people to produce and spread false information and rumors within a few seconds, potentially reaching a wide audience. This, in turn, could have negative impacts on politics, society, and business. To demystify Fake News and create a common understanding, we analyzed the literature on Fake News and summarized existing articles as well as strategies tested to detect Fake News. We conclude that detection methods mostly perform binary classifications based on linguistic features without providing explanations or further information to the user.

ezDL: An Interactive IR Framework, Search Tool, and Evaluation System

ezDL is an open-source IR frontend system supporting proactivity, higher level search activities, the digital library life cycle, and collaboration of searchers. The ezDL framework is based on an extensible, service-oriented architecture, with user clients running on the desktop, in a browser or as a smartphone app. For performing user-centered evaluations, ezDL has a builtin evaluation mode that addresses many of the major challenges inherent in setting up evaluation tasks and tracking user activity during the experiments.