ARDIAS: AI-Enhanced Research Management, Discovery, and Advisory System

Abstract

In this work, we present ARDIAS, a web-based application that aims to provide researchers with a full suite of discovery and collaboration tools. ARDIAS currently allows searching for authors and articles by name and gaining insights into the research topics of a particular researcher. With the aid of AI-based tools, ARDIAS aims to recommend potential collaborators and topics to researchers. In the near future, we aim to add tools that allow researchers to communicate with each other and start new projects.

Keywords

• eol>scholarly data
• scholarly community
• research recommendations

1. Introduction

In spite of the existing tools available for the discovery of research work, a major gap remains in tools that allow collaboration. ARDIAS aims to enable the following workflow:

Historically, scientific research has been performed in closely-knit groups inside individual institutes or corporations, with little to no collaboration across diferent geographies [ 1 ]. The usual mode of exchange of scien- • User visits ARDIAS (either logged in or as a guest) tific findings used to be publishing articles in journals, • User discovers relevant research and authors and more rarely, researchers used to meet each other in • User communicates with the relevant author of person during scientific conferences. interest and starts a new research project on the

With the advent of the Internet, it is finally possible platform to follow current research topics and researchers more • The project members track experimental progress closely online. Institutions host on their websites a col- and log results on the platform lection of publications by their researchers, along with • The research article is drafted on ARDIAS and some contact details. Moreover, conferences and journals submitted to a conference or journal often publish their proceedings online for the public to access. In the case of open-access publications, people pmuabyliaccactieosnsst,huesdeorscummuesnttesitfhreeer opfaycoastf,eheoowrebveelro,nfogrtopaaind 2. Related Work afiliated institute to access these articles. For the purpose of scholarly communication and its meth

To make such research even more accessible, in the ods, most solutions so far have emphasized on techniques past decade, a number of aggregators of research arti- for linking metadata about articles, people, data and other cles have appeared online. Some notable products and relevant concepts [ 2, 3 ]. The common mode today is projects in this area are DBLP1, OpenAlex2, Research to extend the representation to a document structure Gate3, Microsoft Academic Network, Semantic Scholar4, [ 4 ]. Some other solutions propose more comprehensive Google Scholar5 and ORKG6. conceptual models for scholarly knowledge that capture The Third AAAI Workshop on Scientific Document Understanding 2023 problems, methods, theories, statements, concepts, and $ [e-mail] (D. Banerjee); their relations [ 5, 6 ]. Our focus is not to reinvent any of [e-mail] (S. M. Yimam); these methods, but to use them as a baseline system for (sCu.shBiile.amwaanlen@)tib.eu (S. Awale); [e-mail] our initial discovery phase within ARDIAS.

© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License The next step in ARDIAS is a recommendation com1 hCPWrEooUrctkReshtdoinpgpssIhStpN:/c1e:6u1r3-w/-0s.o7r3g/dbAClttpEri.bUoutRriognW/4.0oInrtekrnsahtioonpal (PCCroBYce4.0e).dings (CEUR-WS.org) cphoinneenlte,awrnhiinchg rheasseasrecehn isniggneinficaenratl,acatnivditayls[o i7n] tihnemfoa-2https://openalex.org/ cused area of scholarly research [ 8, 9 ]. For the purpose of 34hhttttppss::////wwwwww..rseesmeaanrcthicgsactheo. nlaert./org/ recommending relevant future research topics, potential 5https://scholar.google.com/ co-authors, and relevant citations, one of the required 6https://orkg.org/ steps is to extract relevant information from a scientific

3. Architecture of the System

manuscript and then process it. In this regard, there are plenty of works, particularly in the area of information extraction from scholarly documents [ 10 ]. Figure 2 shows the diferent components of ARDIAS. In

The task of finding relevant people with a given ex- the back end, ARDIAS currently depends on the Opepertise is called Expert Finding, and several works have nAlex API to fetch and update the latest research artifocused on this problem [ 11, 12, 13, 14 ]. For this task cles and corresponding metadata. A local copy of the certain datasets have been developed, for example, the OpenAlex data is stored in an Elasticsearch instance. Adenriched version of DBLP 7 provided by the ArnetMiner ditionally, the OpenAlex data dump is converted into a project [ 15 ], or the W3C Corpus 8 of TREC used by Balog Knowledge Graph (KG) and indexed into a graph database et al. [ 11 ]. hosted in Neo4J locally. The front end currently inter

From a communication perspective, the internet has faces with 1) OpenAlex API 2) Elasticsearch and 3) Neo4J provided several new avenues for interaction among re- to display relevant information to the user. A batch of searchers. It has been shown that social media is an helper scripts downloads the latest OpenAlex updates efective means of collaborative learning in new domains and updates the local Neo4J graph. [ 16 ]. Researchers are active on privately owned plat- At the moment we do not have a local database for forms like Linkedin and Twitter, however, very recently storing user data, settings, or preferences. Hence the there has been a migration of researchers from Twit- user experience is largely built for a user who has not ter to Mastodon9. Mastodon is an open-source publish- registered or logged in. However, a login feature does subscribe protocol-based broadcast platform that is dis- exist that stores login information in memory per session. tributed and self-hosted in nature. Among the current In the near future, we plan to implement a more comtop Mastodon servers for the specific category of AI re- prehensive login and registration system which verifies searchers, sigmoid.social seems to be the most popular. users thoroughly before onboarding them to ARDIAS. Hosting a Mastodon server, and an XMPP server to en- In the future, additional infrastructure for machine able real-time collaboration is a possible future direction learning and big data ingestion shall be added to ARDIAS. that ARDIAS may take to enable communication among researchers. 4. Features of ARDIAS

For collaboration, several products and projects exist that allow tracking, collection, analysis, and visualization 4.1. Discovery of scientific results. Some such platforms include Overleaf, TensorBoard, Wandb, Trello, GitHub, and others. In ARDIAS, it is our endeavor to select the best features from existing products and implement an open-source version for them.

7https://aminer.org/lab-datasets/expertfinding/

8https://tides.umiacs.umd.edu/webtrec/trecent/parsed_w3c_ corpus.html 9https://mastodon.social/explore

The first stage of the workflow of ARDIAS is discovery.

This begins with user login, where a first-time user is presented with the view shown in Figure 4. The OpenAlex data we consume may contain several authors of the same name and several orphan nodes for the same author. Hence, it is important to ask the authors to identify themselves on the first login, and to solve this problem, our login page produces a list of authors by text match over name and afiliation. The user can then select the correct author from the list and self-identify. Additionally, it is also possible for a user to explore most features of ARDIAS without logging in. Scholarly work discovery begins with the discovery of one’s own work on the platform. If the user is logged in, ARDIAS fetches the works via an OpenAlex API call and populates the Works page. If the user is not logged in, the user can search for himself or herself using the prominently located search bar in the UI. The search box allows a user to fetch relevant results based on a handful of diferent criteria, namely 1) Works 2) Author 3) Institutions 4) Venues. 4.1.1. Works Searching by works criterion displays a paginated results page showing the relevant publications based on a text search of the title. The page contains the title of the paper, the list of authors, the publication year, and the venue. A short abstract follows, and at the bottom, the citation count is displayed. The author names and venue names are hyperlinks that may be clicked, and they open to the respective author and venue pages. On clicking the title of the work itself, a dedicated publication page is loaded which additionally displays DOI information, OpenAlex and Microsoft Academic Graph (MAG) IDs, and keywords for this work. A tab is also available to view the citations of this work from where the citing papers may also be visited. The next section of the page displays similar papers to the current work. The last section of the Work page contains a discussion section, where logged-in users may add comments and start a discussion on the paper topic. 4.1.2. Authors As depicted in Figure 1, searching by author fetches a list of authors matched by text search, with a short display of their institute of afiliation, the number of publications, citation counts, and OpenAlex and MAG IDs. Several authors may share the same name, hence the institution name is helpful in disambiguating the authors. On clicking on an author’s name on this page, a more detailed author page opens, which displays the published works of the author in sorted order of citation counts. If a publication in this list is Open Access, it is marked as so. The results on this page may be further sorted based on title, date, and citations. In a separate tab, one can see the co-author network for the author, arranged in a hot-spotstyled graphic. The subsequent tab displays the research focus areas of the author in a similar style. 4.1.3. Institution Searching by institution displays a paginated list of institutes based on text search similarity of the institute name. The title for each institute displays the name and location, while the subsequent lines display the website address, the sector of establishment, and the acronym popularly used for the institute if any. This short selection of information allows disambiguation of similarly named institutes. On clicking on an institute, a dedicated institute page opens, which on the right side displays the institute logo, institute name, location, homepage, Wikipedia links, publication, and citation counts. In the central section of the page, a list of authors belonging to this institute sorted by citation counts is displayed. in the scholarly KG hosted in the graph database. This allows author A to find common connections to author B. This makes communication and collaboration stages possible for author A.

4.3. Communication

4.2. Recommendation

4.1.4. Venue

Searching by venue displays a list of similarly named

venues with corresponding historical publication and citation counts. Clicking on a venue opens a new page that lists the publications for the given venue in sorted order based on citations. This sorting order can be changed to be based on title and date.

After the phases of discovery and recommendation, an

author may want to get in touch with another author on our platform. At the moment we have not implemented any communication method explicitly, however, the vision is to enable this kind of communication through an open-source and well-established messaging platform. Specific methods need to be developed to keep spam at bay, and verification of users on our platform remains vital. Some methods for logged-in user verification may include making an institution or work e-mail mandatory for sign-up, and also having to supply an ORCID or a similar academic network ID.

Recently there has been an exodus of academics shifting from the social media site of Twitter to Mastodon. Twitter is a privately held social media company that allows micro-blogging, while Mastodon is a free and opensource software that allows the hosting of independent and federated nodes, each with its own set of policies. ARDIAS may host a dedicated Mastodon node for researchers to join and also allow communication across other Mastodon nodes on our platform. From a communications perspective, we additionally plan to explore the XMPP protocol to allow chat between researchers who already accept each other’s request to communicate.

Users of ARDIAS also receive recommendations from the

system on a number of diferent categories, both with and without the aid of Machine Learning methods.

Some basic non-ML recommendations are as follows: 1) ARDIAS produces a list of researchers who work at the same institute as the user 2) ARDIAS is able to show the least number of hops between a user and another researcher on the platform through work and author nodes. 3) ARDIAS produces a list of related sub-topics of interest that are currently popular in the research community.

Some ML-based recommendations we plan for ARDIAS in the future include 1) a list of similar researchers based on common interests. This would be achieved via graphbased node embedding algorithms and nearest neighbor computations of the embedding space 2) a list of similar 4.4. Collaboration works based on common publications using similar node After researchers have discovered each other and decided embedding methods 3) a list of potential and high-impact to collaborate on a project, ARDIAS aims to provide a baresearch topics based on node prediction approaches in sic platform to start and manage a project. At the moment graph-based ML. the components are still in the planning phase, however, Apart from recommendations in the discovery process of some basic requirements for project collaboration are research and authors, we see scope for recommendations clear. On the creation of a project on the platform, the algorithms in the manuscript authoring phase as well. For collaborating researchers should gain access to an online example, a beginner scholar may be making elementary document editing and storage platform, which allows not formatting, grammar, spelling, or structural errors in just plain text documents, but also binary files, models, the paper being authored. In some cases, a significant and archived logs. The project space should have an issection like Related Work may be missing. In such cases, sue tracker so planning on individual components of the AI-based technologies will be able to help the scholar project is possible and progress is trackable. It would also with helpful suggestions on how to improve the draft. be desirable to provide an online platform that allows Another area where AI-based recommendations may help various forms of visualization of results, and logging of in the authoring process is the automatic suggestion of experimental data. Finally, it would be desirable to allow citations for the given topic of the manuscript, providing editing of the research manuscript itself on the platform, an easy selection between diferent formats of citations perhaps through a LaTeX editor. within the Latex editor window, eg: BibTex, Crossref, etc.

A special aspect of non-ML recommendation is depicted in Figure 3. If author A wishes to work with author B, ARDIAS fetches the shortest path between the two nodes

5. Use cases and applications

Searching for research projects and collaborators: ARDIAS shall store information about users and their expertise, lists of tools and datasets, machine learning models and their applications, detailed documentation of projects, concept notes, as well as research publications. Traditionally, one should look for a research collaborator either by searching online profiles, seeking or contacting the responsible ofice in the research institute, or advertising a call for collaboration. ARDIAS can help by automatically searching or recommending collaborators using the knowledge stored in the repository. The system ranks and recommends collaborators based on their expertise and relevance.

Research approach suggestion: Research on a multidisciplinary project is challenging. Medical researchers are aware of problems in their discipline but they might lack the proper approaches to build an automated system with a machine learning component. Similarly, in a computational linguistic research project, experts from diferent fields such as linguistics, mathematics, statistics, and computer science work together. Employing computational linguistics approaches for social science projects needs a further understanding of diferent research problems. Using an AI-enhanced system, where prior multidisciplinary projects are indexed, can help in suggesting research approaches for new research problems.

Virtual help-desk: In large research institutes or companies, where there are several research projects, tools, and research collaborators, it is dificult to provide accurate information. A research knowledge management system with interactive UI can help in finding requested services that can be enhanced with faceted search functionality to support help desk professionals.

6. Conclusion

In this work, we described ARDIAS, which is a web appli

cation with a vision for researchers to discover, communicate and collaborate on new research topics. The demo is made available via a public URL10. The development is currently at an initial phase, with basic features having been implemented. We present our long-term goal for ARDIAS as being a common platform for researchers from all backgrounds and fields to discover each other and collaborate.

7. Acknowledgements

ARDIAS is funded by the "Idea and Venture Fund" re

search grant by Universität Hamburg, which is part of 10https://ardias.ltdemos.informatik.uni-hamburg.de/ the Excellence Strategy of the Federal and State Governments.

References

[1] G. M. Olson , A. Zimmerman , N. Bos , Scientific Collaboration on the Internet , The MIT Press, 2008 . URL: https://dl.acm.org/doi/10.5555/1522423.

[2] A. Aryani , J. Wang , Research Graph: Building a Distributed Graph of Scholarly Works using Research Data Switchboard ( 2017 ). URL: https://bridges.monash.edu/articles/preprint/ Research_Graph_ Building_a_Distributed_Graph_ of_Scholarly_Works_using_Research_Data_ Switchboard/4742413.

[3] A. Burton , H. Koers , P. Manghi , M. Stocker , M. Fenner , A. Aryani , S. La Bruzzo , M. Diepenbroek , U. Schindler, The scholix framework for interoperability in data-literature information exchange, D-Lib Magazine 23 ( 2017 ). URL: http://mirror.dlib. org/dlib/january17/burton/01burton.html.

[4] S. Peroni , Law, Governance and Technology Series, 15 , 2014 . URL: https://link.springer.com/book/10. 1007/978-3- 319 -04777-5.

[5] B. Brodaric , F. Reitsma , Y. Qiang, SKIing with DOLCE: Toward an e-Science Knowledge Infrastructure , in: Proceedings of the 2008 Conference on Formal Ontology in Information Systems: Proceedings of the Fifth International Conference (FOIS 2008 ), 2008 , p. 208 - 219 . URL: https://dl.acm.org/ doi/10.5555/1563953.1563975.

[6] M. Y. Jaradeh , A. Oelen , K. E. Farfar , M. Prinz , J. D'Souza , G. Kismihók , M. Stocker , S. Auer , Open Research Knowledge Graph: Next Generation Infrastructure for Semantic Scholarly Knowledge , in: Proceedings of the 10th International Conference on Knowledge Capture, K-CAP '19 , New York, NY, USA, 2019 , p. 243 - 246 . URL: https://doi.org/10.1145/ 3360901.3364435.

[7] F. Isinkaye , Y. Folajimi , B. Ojokoh , Recommendation systems: Principles, methods and evaluation , Egyptian Informatics Journal 16 ( 2015 ) 261 - 273 . URL: https://www.sciencedirect.com/science/ article/pii/S1110866515000341.

[8] F. Xia , H. Liu, I. Lee , L. Cao , Scientific Article Recommendation: Exploiting Common Author Relations and Historical Preferences , IEEE Transactions on Big Data 2 ( 2016 ) 101 - 112 . URL: https: //doi.org/10.1109% 2Ftbdata . 2016 . 2555318 .

[9] M. Färber , A. Jatowt , Citation Recommendation: Approaches and Datasets, International Journal on Digital Libraries 21 ( 2020 ) 375 - 405 . URL: https: //doi.org/10.1007% 2Fs00799 - 020 -00288-2.

[10] Z. Nasar , S. W. Jafry , M. K. Malik , Information Extraction from Scientific Articles: A Survey, Scientometrics 117 ( 2018 ) 1931 - 1990 . URL: https://doi.org/10.1007/s11192-018-2921-5.

[11] K. Balog , L. Azzopardi , M. de Rijke, Formal Models for Expert Finding in Enterprise Corpora , in: Proceedings of the 29th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR '06 , 2006 , p. 43 - 50 . URL: https://doi.org/ 10.1145/1148170.1148181. doi: 10 .1145/1148170. 1148181.

[12] H. Fang , C. Zhai , Probabilistic Models for Expert Finding , in: G. Amati, C. Carpineto , G. Romano (Eds.), Advances in Information Retrieval , 2007 , pp. 418 - 430 . URL: https://link.springer.com/chapter/10. 1007/978-3- 540 -71496-5_ 38 .

[13] P. Serdyukov , H. Rode , D. Hiemstra , Modeling multi-step relevance propagation for expert finding , in: Proceedings of the 17th ACM Conference on Information and Knowledge Management , CIKM '08 , 2008 , p. 1133 - 1142 . URL: https://doi.org/10.1145/ 1458082.1458232.

[14] T. Fischer , S. Remus , C. Biemann, LT expertfinder: An evaluation framework for expert finding methods, in: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations) , Association for Computational Linguistics , 2019 , pp. 98 - 104 . URL: https://aclanthology.org/N19-4017.

[15] J. Tang , J. Zhang , L. Yao, J. Li , L. Zhang , Z. Su , ArnetMiner: Extraction and Mining of Academic Social Networks, in: Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD '08, Association for Computing Machinery, 2008 , p. 990 - 998 . URL: https://doi.org/10.1145/1401890.1402008.

[16] J. A. N. Ansari , N. A. Khan , Exploring the role of social media in collaborative learning the new domain of learning , Smart Learning Environments 7 ( 2020 ) 9 . URL: https://doi.org/10.1186/s40561-020-00118-7.