ScholarLensViz: A Visualization Framework for Transparency in Semantic User Pro les

Abstract

Personalized applications are a two-edged sword. They are convenient and assist users by keeping the focus on relevant topics, but they are often black boxes and users typically do not know why certain entries appear in their pro le. As transparency and provenance are essential for researchers, in this paper, we introduce ScholarLensViz, a visualization component for scholarly user pro les displaying a scholar's research competences including the provenance. It also provides visualizations to inspect the diversity of a pro le and to analyze the semantic similarity of the pro le entries.

Keywords

• Visualization
• Provenance
• Semantic User Pro les
• LOD

-

Scholarly user pro les are gaining increasing attention and are becoming more important for various application scenarios, such as expertise retrieval [ 10 ], search and recommendation of research articles [ 1 ] or research network analysis [ 6 ]. These intelligent systems support scholars in time-consuming daily tasks, such as data discovery and data reuse, by providing content tailored to a scholar's preferences. In most applications, scholars are permitted to inspect these collected user preferences but do not know where the information comes from. However, the justi cation of entries is an important issue in scholarly pro les. Users need explanations why a certain preference is presented. ? Copyright c 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).

In this work, we address these issues and introduce ScholarLensViz 5, a visualization framework to display scholarly pro les based on a semantic user model. In earlier work, Bakalov et al. [ 2 ] proved the feasibility of a transparent visualization of semantic user pro les. We further developed that idea and present a new graphical user interface with linkage to the LOD cloud. ScholarLensViz presents a scholar's competences and their provenance. We obtain the competences from publications and store them in an RDF graph, using a work ow being introduced in our previous work ScholarLens [ 8 ]. The system also enables the inspection of a pro le's diversity and provides visualizations to explore the semantic similarity of the pro le entries. 2

Related Work

Various scholarly applications have emerged that create user pro les out of different resources, for di erent purposes and with various visualizations. Semantic Scholar [ 1 ], a smart search over research articles, generates user pro les from publications based on paper heuristics. In contrast, AMiner [ 10 ] focuses on expertise retrieval, a research eld that aims at nding experts on a certain topic. Following a topic modeling approach, data is collected from institutional websites, documents and conferences. The visualization provided includes radar diagrams for author's statistics and stacked area charts for a scholar's automatically extracted research interests. Full semantic approaches are VIVO [ 5 ] and Scholia [ 6 ]. While VIVO provides a framework, an ontology and a graphical user interface for representing and visualizing scholars and their research context within an organization, Scholia aims at visualizing scienti c bibliographic information through Wikidata. Both applications provide visualizations for co-author networks and research topics a scholar is competent in. In addition, Scholia offers various entry points to explore not only authors and publications but also organizations, locations or projects and supports numerous display formats such as timelines, scatter charts, line chart or trees. Provenance information, e.g., explanations from which source a topic or research interest has been extracted, is less considered in current approaches for scholarly user pro les. Our system attempts to close that gap and to leverage transparency and provenance in the visualization of semantic user pro les. 3

Architecture

5 ScholarLensViz source code, https://github.com/fusion-jena/ScholarLensViz Competence as a literal. These classes are used from the Competence Management Ontology (CM) [ 4 ]. In addition, we model a user's publications with the PUBO Ontology [ 9 ] that provides properties to describe relations between documents and semantic annotations generated by the pipeline. The nal output is an RDF graph containing the competences and being visualized by ScholarLensViz (blue-colored). We separated the visualization (client-side) and the calls to the knowledge base (server-side). Thus, a server handles the SPARQL requests and provides HTTP methods to the client. This separation minimizes security risks such as sending direct SPARQL queries from the client to the knowledge base. Embedding ScholarLensViz into personalized applications, for instance as widget or standalone application, requires additional authentication mechanism to be handled by the integrating system (green-colored). 4

ScholarLensViz

ScholarLensViz provides three dialogs. (A) The start dialog aims to display a user's Top-25 ranked competences and their provenance. The elements of the pie chart (Fig. 2) represent a user's competences. The arc of the pie slices denotes the competence rank, and the color represents the category obtained from the Computer Science Ontology (CSO) [ 7 ] by matching the competence entries (which are presented as DBpedia entities) with CSO labels. The comment tab, becoming visible by clicking on a pie element, provides the competence description obtained from DBpedia. In order to justify the origin of the competences, we provide a provenance tab displaying the sentences from a user's publications and highlighting the relevant phrases that contain the competence and its rank, the paper title and, if available, the DOI.

In order to inspect the diversity of a user's pro le and the semantic similarity between the pro le entries, ScholarLensViz o ers two further charts (Fig. 3). (B) A force-directed graph and (C) a chord chart visualizing the semantic similarity and relatedness between the competences computed with Sematch [ 11 ]. While semantic similarity is hierarchy-based, the semantic relatedness considers all relations between concepts [ 11 ]. We use the average of both values as both values in uence the diversity of a pro le. The more diverse a pro le is, the less strong connections appear in the force-directed graph per competence. The chord chart allows a better analysis on the accumulated similarity values over the selected competences. The length of the outer arc re ects the number of similar connections a competence provides.

Demonstration

In this demo6, visitors can inspect the competences from known computer scientists working in the elds of Semantic Web. In addition, we added experts from other computer science domains as a basis for comparison. Users can select the competences and can change the thresholds for the semantic similarity. Acknowledgements This work was supported by the German Academic Exchange Service (DAAD) within the scope of the PPP program and the Deutsche Forschungsgemeinschaft (DFG) within the scope of the GFBio project (229241684). 6 https://dev.gfbio.org/scholar/

References

1. Ammar , W. , Groeneveld , D. , Bhagavatula , C. , Beltagy , I. , Crawford , M. , Downey , D. , Dunkelberger , J. , Elgohary , A. , Feldman , S. , Ha , V. , Kinney , R. , Kohlmeier , S. , Lo , K. , Murray , T. , Ooi , H. , Peters , M. , Power , J. , Skjonsberg , S. , Wang , L. , Wilhelm , C. , Yuan , Z. , van Zuylen, M. , Etzioni , O. : Construction of the literature graph in semantic scholar . In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics ( 2018 )

2. Bakalov , F. , Meurs , M.J. , Konig- Ries , B. , Sateli , B. , Witte , R. , Butler , G. , Tsang , A. : An approach to controlling user models and personalization e ects in recommender systems . In: IUI. ACM , New York, NY, USA ( 2013 )

3. Jovanovic , J. , Gasevic , D. , Siadaty , M. : Intelleo user model ontology . http: //intelleo.eu/ontologies/user-model/spec/ ( March 2012 )

4. Jovanovic , J. , Siadaty , M. , Gasevic , D. , Milikic , N.: Intelleo competences ontology . http://www.intelleo.eu/ontologies/competences/spec/ ( 2011 )

5. Kra t, D., Cappadona , N. , Caruso , B. , Corson-Rikert , J. , Devare , M. , Lowe , B. : Vivo: Enabling national networking of scientists . In: Proceedings of the Web Science Conference 2010 ( 2010 )

6. Nielsen , F.A. , Mietchen , D. , Willighagen , E.: Scholia, scientometrics and wikidata . In: Blomqvist, E. , Hose , K. , Paulheim , H. , Lawrynowicz , A. , Ciravegna , F. , Hartig , O . (eds.) The Semantic Web: ESWC 2017 Satellite Events . pp. 237 { 259 . Springer International Publishing, Cham ( 2017 )

7. Salatino , A.A. , Thanapalasingam , T. , Mannocci , A. , Osborne , F. , Motta , E. : The computer science ontology: A large-scale taxonomy of research areas . In: International Semantic Web Conference 2018 , Monterey (CA), USA, 2018 .

8. Sateli , B. , Lo er, F. , Konig- Ries , B. , Witte , R.: Scholarlens: extracting competences from research publications for the automatic generation of semantic user pro les . PeerJ Computer Science 3 , e121 ( Jul 2017 )

9. Sateli , B. , Witte , R.: Semantic representation of scienti c literature: bringing claims, contributions and named entities onto the linked open data cloud . PeerJ Computer Science 1 , e37 ( Dec 2015 )

10. Wan , H. , Zhang, Y. , Zhang , J., Tang , J.: Aminer: Search and mining of academic social networks . Data Intelligence ( 2019 )

11. Zhu , G. , Iglesias , C.A. : Computing semantic similarity of concepts in knowledge graphs . IEEE Transactions on Knowledge and Data Engineering 29 ( 1 ) ( 2017 )