What should I read next? A Visual Publication Recommender System André Calero Valdez Simon Bruns Christoph Greven Ulrik Schroeder Martina Ziefle
Motivation - Challenges ● Information overload 679,858 papers per year in 2009 / 365 days / 24 hours / 60 minutes o = 1.29 papers per minute (medicine) ● Identify relevant publications ● External Publication Sources 02
Motivation - User Researcher ● Scientific Output ● Profile (Keywords) ● Colleagues at one institute 01
Recommender System Characteristics ● Ranking of top items ● Inherent issues Trustworthiness o Understanding o Transparency 03
Recommender System Transparency ● Comprehensive Items and Connections ● Focus on UI solution ● User centered system ○ User in control ○ Part of the system 04
Recommender System Item Scoring ● Keywords well-known to users ● TF-ICF ○ Iteration of TF-IDF ○ Generates Keyword relevance (PDF) 06
Recommender System Exploration ● Graph ○ Publication Network ○ Explorative Design ● Publication and Topic Nodes ● Semantic connections 05
Recommender System Exploration ● Filter Control ○ Input Interface ○ Choose from different Filter types ● Filter Interaction ○ Modification Interface ● Information Tab ○ Detailed Information on each graph node (topic & publication)
Us User Study - Sa Sample ● 16 Participants ○ Researchers ○ 50% Male/Female ratio ○ Asses track record
Us User Study - Met Metho hod Measured Variables • SUS, NPS and ResQue • Trust in System • Accuracy of Recommendations • Additional Benefits: • Structure and Overview • Research Interests
Us User Study - Re Results ● Trust is a major factor ○ Strong correlation ■ Accuracy ■ Structure & Overview ○ Weaker correlation ■ NPS ■ Research Interest of Colleagues
Summary & Contact TIGRS – Publication recommendation System § Graph-based visualization § Exploration instead of linear search § Positive User Evaluation § Trust is important § Overview over colleagues research interests as secondary benefit André Calero Valdez § calero-valdez@comm.rwth-aachen.de § 0241 – 80 239480 9/26/18 Summary 12
Recommend
More recommend
