Semantic and Time-Dependent Expertise Profiling Models in Community-Driven Knowledge Curation Platforms

Abstract

Abstract:Online collaboration and web-based knowledge sharing have gained momentum as major components of the Web 2.0 movement. Consequently, knowledge embedded in such platforms is no longer static and continuously evolves through experts' micro-contributions. Traditional Information Retrieval and Social Network Analysis techniques take a document-centric approach to expertise modeling by creating a macro-perspective of knowledge embedded in large corpus of static documents. However, as knowledge in collaboration platforms changes dynamically, the traditional macro-perspective is insufficient for tracking the evolution of knowledge and expertise. Hence, Expertise Profiling is presented with major challenges in the context of dynamic and evolving knowledge. In our previous study, we proposed a comprehensive, domain-independent model for expertise profiling in the context of evolving knowledge. In this paper, we incorporate Language Modeling into our methodology to enhance the accuracy of resulting profiles. Evaluation results indicate a significant improvement in the accuracy of profiles generated by this approach. In addition, we present our profile visualization tool, Profile Explorer, which serves as a paradigm for exploring and analyzing time-dependent expertise profiles in knowledge-bases where content evolves overtime. Profile Explorer facilitates comparative analysis of evolving expertise, independent of the domain and the methodology used in creating profiles.

Modelling expertise at different levels of granularity using semantic similarity measures in the context of collaborative knowledge-curation platforms

“…We contrast these results to those discussed in our previous studies (Ziaimatin et al 2012, 2013), where at weight thresholds of 0, 0.1 and 0.2, we achieved F-score values of 18.91 %, 21.03 % and 20.31 % respectively. While we cannot compare them directly (since the previous results were generated on unstructured contributions and achieved via exact matching), we can draw the conclusion that comparing profiles using semantic similarity methods and the structure of ontologies results in more accurate comparisons than detecting exact matches between the content of profiles.…”

“…In the second group of models (Model 2), it first identifies important documents for a given topic, and then determines which entities are most closely associated with these documents. In previous studies, we conducted experiments with EARS using our biomedical use cases (see Ziaimatin et al 2012, 2013); however, this system also relies on a given set of queries. Furthermore, as with other studies that target expert finding, EARS relies on large corpus of static publications, while we aim at building expert profiles from micro-contributions, without relying on any queries.…”

User-driven efforts in creating knowledge graph information system

“…Being strongly grounded in a particular place (geolocation, implying an organization, even a unit within), entails having local norms and requirements related to the research topic and practices, while keeping the focus on the needs of the student population. Furthermore, discussed technologies as well as other relevant research regarding expert recommender systems in academia (for example, STEP methodology by Ziaimatin et al, (2013)) are designed to suggest experts in a given field based on their research (usually faculty or researchers). In the case where PhD students are the user population, peers might serve as equally valuable information sources (Catalano, 2013).…”

“…An additional relevant research domain is related to the development of graph-based models and systems that tackle the issue of information discovery in academia. Considering User-driven efforts that "science can be described as a complex, self-organizing, and evolving network of scholars, projects, papers and ideas" (Fortunato et al, 2018, p. 2), there is a plethora of studies using graph technology and social network analysis (SNA) approaches, for example, those based on co-authorship and/or publishing venues graphs (Cabanac, 2011;Makarov et al, 2018;Tchuente et al, 2013) Ziaimatin et al (2013) are designed to suggest experts in a given field based on their research (usually faculty or researchers). In the case where PhD students are the user population, peers might serve as equally valuable information sources (Catalano, 2013).…”