Abstract
In this paper, we present an overview of the MOVING platform, a user-driven approach that enables young researchers, decision makers, and public administrators to use machine learning and data mining tools to search, organize, and manage large-scale information sources on the web such as scientific publications, videos of research talks, and social media. In order to provide a concise overview of the platform, we focus on its front end, which is the MOVING web application. By presenting the main components of the web application, we illustrate what functionalities and capabilities the platform offer its end-users, rather than delving into the data analysis and machine learning technologies that make these functionalities possible.
Introduction: Navigating the Scientific Infodemic
Modern researchers, auditors, and administrators are increasingly overwhelmed by a scientific "infodemic." With millions of new publications, thousands of hours of recorded research talks, and a non-stop stream of social media updates, the primary challenge has shifted from simply finding information to effectively organizing and making sense of it.
The MOVING platform addresses a critical gap in current workflows: while sophisticated data mining tools exist, many professionals lack the "information literacy" required to integrate them into their daily routines. MOVING is designed as an all-in-one ecosystem that transforms users into data-savvy professionals by integrating advanced search and visualization tools directly with a social learning environment, including Massive Open Online Courses (MOOCs) and adaptive training.
System Architecture
The MOVING platform utilizes a robust multi-tier architecture to bridge the gap between raw web data and actionable user insights. This architecture is divided into three primary layers: data acquisition, data processing, and the web application interface.
Data Acquisition and Processing
The system employs specialized web crawlers and metadata injection techniques to index a massive repository of over 22 million documents. Key technical innovations within the processing layer include:
- Author Name Disambiguation (AND): A novel method using agglomerative clustering on features such as institutional affiliations and co-author networks to distinguish between different researchers sharing the same name.
- Video Analysis & NER&L: Automated tools analyze video content and perform Named Entity Recognition and Linking (NER&L) to enrich the search index with semantic metadata.
Advanced Search and Adaptive Training
Moving beyond traditional list-based search results, the platform provides interactive tools designed to reveal the underlying structure of information. The "user-centric" core of the platform is the Adaptive Training Support (ATS), which tracks user interactions to provide personalized feedback and "nudges."
Key User Features
- Concept Graph: As shown in the interface, this interactive network visualization displays connections between documents, authors, and concepts, allowing users to explore the research landscape non-linearly.
- uRank: A dynamic ranking tool that enables users to re-sort results in real-time by adjusting the weight of specific keywords.
- ATS Widgets: The platform monitors feature usage (e.g., via the UCIVIT tool). If a user relies solely on simple keyword searches, a widget displays a bar chart of their behavior, encouraging them to explore more powerful query methods like the Concept Graph or faceted search.
Data Management and Scale
The platform's live index serves as a testament to its scalability. At the time of writing, the repository included over 22 million records, encompassing scientific articles, books, laws, and regulations. It also integrates multimedia content from sources like videolectures.net and real-time social media streams. User interaction data is logged via WevQuery, which powers a recommender system using HCF-IDF (Hierarchical Concept Frequency - Inverse Document Frequency) to suggest relevant resources based on semantic profiles rather than simple keyword matching.
Conclusion and Future Work
The MOVING platform successfully demonstrates the feasibility of an integrated research and learning environment. By "learning while working," users are motivated to improve their information literacy through direct feedback and gamified progress tracking (such as sunburst visualizations of their curriculum mastery).
While the current architecture is robust and has been successfully deployed in MOOC environments, future work will focus on deeper quantitative benchmarking of individual machine learning components and long-term studies to evaluate the platform's impact on professional data-savviness over extended periods. The ultimate goal remains the expansion of community features to foster open collaboration across diverse scientific disciplines.
Summary updated