A Knowledge Graph Approach for Exploratory Search in Research

Institutions

Tim Schopf

, Nektarios Machner

and Florian Matthes

Department of Computer Science, Technical University of Munich, Germany

Keywords:

Knowledge Graphs, Exploratory Search.

Abstract:

Over the past decades, research institutions have grown increasingly and consequently also their research

output. This poses a signiﬁcant challenge for researchers seeking to understand the research landscape of

an institution. The process of exploring the research landscape of institutions has a vague information need,

no precise goal, and is open-ended. Current applications are not designed to fulﬁll the requirements for

exploratory search in research institutions. In this paper, we analyze exploratory search in research institutions

and propose a knowledge graph-based approach to enhance this process.

1 INTRODUCTION

Scientiﬁc literature has grown exponentially over the

past centuries, with a two-fold increase every 12

years (Dong et al., 2017). Concurrently, the num-

ber of research institutions as well as the number

of researchers and research areas within these insti-

tutions has also been growing. Once research in-

stitutions reach a certain size, it becomes challeng-

ing to determine which topics are being researched

at an institution and who is researching which top-

ics. The most basic approach to disclosing ongoing

research at research institutions is to post this infor-

mation as unstructured text on the institution’s or its

sub-units’ websites. Usually, these websites are de-

signed according to the organizational structures of

research institutions rather than research areas, fur-

ther complicating the process of understanding what

research is being conducted. More advanced solu-

tions attempt to consolidate the entire research output

of researchers and research units in an institution us-

ing speciﬁc Research Information Management Sys-

tems (RIMS). These RIMS can identify and visualize

the domain of expertise of researchers and research

units based on research topic tags automatically ex-

tracted from publications. This may accelerate the

process of a targeted search, such as ﬁnding an ex-

pert in a speciﬁc domain. However, these systems are

https://orcid.org/0000-0003-3849-0394

https://orcid.org/0009-0001-8359-6668

https://orcid.org/0000-0002-6667-5452

not capable of representing the relationships between

individual research areas and assess the similarity of

researchers based on organizational afﬁliations. As a

result, the search for related research areas and fur-

ther potentially relevant experts remains challenging.

In addition, RIMS often lack comprehensible statis-

tics and analyses about the speciﬁc research ﬁelds

of researchers and research units. Therefore, despite

RIMS, it is still very time-consuming for researchers

to obtain an overview of the research landscape of a

research institution.

To understand the current process of researchers

seeking insights into the research landscape of insti-

tutions, we conducted several interviews. Our analy-

sis is based primarily on one-on-one interviews with

researchers, ranging from early to late career stages.

From this, we conclude that the use of RIMS is still

not widely adopted. Consequently, researchers cur-

rently heavily rely on search engines to ﬁnd relevant

institution websites and then use these as a starting

point for both browsing and formulating new search

queries. The process of searching and browsing con-

tinues iteratively until researchers have a satisfac-

tory overview of an institution’s research landscape.

However, simply using returned lists of relevant el-

ements that need to be analyzed manually limits the

knowledge search of researchers (Jaradeh et al., 2019;

Brainard, 2020). Two main reasons can be identiﬁed

for this. First, it is very tedious and time consuming

to search and browse all relevant websites. Second,

important information that is potentially relevant to

researchers but was not searched for because the ap-

Schopf, T., Machner, N. and Matthes, F.

A Knowledge Graph Approach for Exploratory Search in Research Institutions.

DOI: 10.5220/0012223800003598

In Proceedings of the 15th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2023) - Volume 3: KMIS, pages 265-270

ISBN: 978-989-758-671-2; ISSN: 2184-3228

265

propriate search queries are unknown can be missed.

We argue that the exploratory knowledge search

process for research activities in research institutions

can be signiﬁcantly enhanced by the semantic linking

of research topics and their subsequent association

with further relevant entities.

Our aim is to advocate for more clarity and con-

ciseness in presenting the areas of competence of re-

search institutions. In addition, we propose an ap-

proach to enhance the exploratory knowledge search

process for research activities in research institutions.

Researchers seeking an overview of current research

at an institution should be able to identify potential

collaborators more easily and be less likely to miss

important information. Enhancing this knowledge

search process has the potential to encourage and fa-

cilitate more research collaborations within research

institutions. Eventually, enhanced knowledge search

and the resulting potential for more research collabo-

rations may also have a positive impact on an institu-

tion’s overall research output.

2 REQUIREMENTS ANALYSIS

2.1 Exploratory Search in Research

Institutions

The information need for obtaining an overview of an

institution’s research landscape is complex, evolves

during the search, and is open-ended. In this case,

the user objective is rather vague and can be di-

vided into several smaller sub-goals during the search

(e.g., which research areas exist at the institution, who

works in certain research areas, or what exactly is

being studied in the research areas). Furthermore,

users can not rely on a single search result to sat-

isfy their information need. They need to assemble

the relevant information from multiple search results

and institution websites to get an overview of the re-

search landscape. Users perform several reformula-

tions and reﬁnements of their search queries based on

the new information obtained from institution web-

sites. The search process continues until users con-

sider they have obtained enough information to reach

their vague goal. However, users may never know if

their acquired knowledge is complete or if they have

missed important aspects of the research landscape at

the institution. Therefore, the search does not have a

deﬁned end, but can be continued after receiving ad-

ditional information, if users feel that there are still

unanswered questions to be investigated.

Existing search engines are optimized for sim-

ple lookup searches, which are characterized by low

complexity and a precise objective (Athukorala et al.,

2016). However, the described knowledge search for

research activities in research institutions has a rather

vague, complex, and open-ended goal, showing many

characteristics typical of an exploratory search (Palagi

et al., 2017; Grether and Witschel, 2022). In our

interviews, users were able to use search engines to

quickly ﬁnd answers to speciﬁc questions, such as

whether there exists a particular research area at a

research institution. However, in the case of rather

vague objectives, e.g., searching for key research ar-

eas of a particular department in an institution, the

search process turned out to be quite inefﬁcient and

took a considerable amount of time. Moreover, users

were not able to properly assess whether their results

were exhaustive or incomplete at the end of the search

process. We conclude that it is not sufﬁcient to rely

only on existing search engines to gain insights into

the research landscape of institutions. Moreover, we

argue that a new approach is required to sufﬁciently

address the need for exploratory search in research in-

stitutions.

2.2 Requirements

To investigate the exploratory search process of re-

searchers seeking insights into the research landscape

of institutions, we conducted semi-structured inter-

views. Our participants included graduate students,

PhD students, and professors. We gave each partic-

ipant an exploratory task with a rather vague objec-

tive, such as e.g., identifying the most relevant re-

search areas at an unknown research institution. They

were allowed to complete this task using any re-

sources at their disposal. In many cases, search en-

gines were the tool of choice. We observed the search

and asked participants to express their thought pro-

cesses aloud. Throughout the search, we kept asking

questions about the challenges participants encoun-

tered and what might help them to better deal with

them.

From our interviews, we derive the following

key requirements for an application designed to en-

hance the exploratory knowledge search process for

research activities in research institutions:

• Requirement 1: Hierarchical Navigation

Structure Based on Research Topics.

Users are used to navigating through applications

and websites in a hierarchical manner. Thereby,

more general concepts are presented in the upper

levels and more speciﬁc concepts are presented in

the lower levels of the navigation hierarchy. An

interface using hierarchical navigation simulta-

neously shows previews of where to go next and

KMIS 2023 - 15th International Conference on Knowledge Management and Information Systems

266

how to return to previous states in the exploration

(Hearst and Stoica, 2009). In the scientiﬁc

domain, the most important concepts are the

speciﬁc research topics and areas. Therefore, to

enable a streamlined and satisfactory semantic

exploration of scientiﬁc knowledge, we need to

use a well-organized hierarchical structure of

research topics and areas that comprehensively

covers the broad spectrum of academic disci-

plines (Shen et al., 2018).

• Requirement 2: Semantic Linking of Scientiﬁc

Entities.

In addition to the hierarchically structured

research topics, the application has to include ac-

curate representations of the research institution

and its research units as well as the associated

researchers and their publications. The respective

entities have to be semantically linked to ensure

an extensive exploratory search that offers many

opportunities for discovering new information.

Most importantly, the relationships between the

hierarchically structured research topics and the

other entities must be modeled precisely. This

allows for a comprehensive exploratory search

within the research institution based on research

topics. In addition, other relationships such as the

afﬁliation of researchers with research units or

the similarity of researchers need to be modeled

in order to recommend other relevant entities

during the search process.

• Requirement 3: Single Source of Truth.

As a countermeasure to the complex and

open-ended nature of exploratory search, the

application has to act as a single source of truth

containing all relevant data of one research

institution. With all relevant data being available

from one source, the user should feel conﬁdent in

receiving a comprehensive overview and it should

not be necessary to involve further external tools

and search engines to ﬁnd the desired information.

• Requirement 4: Support for Lookup Search as

Well as Recommendations.

Exploratory search involves a signiﬁcant amount

of lookup activities in addition to a wide range

of other goals and tasks (White and Roth, 2009).

Therefore, the application has to implement a

semantic search that supports basic lookup tasks.

Furthermore, based on the semantic relationships,

the application must be able to recommend

further relevant entities for the user to explore.

This facilitates the discovery of new and relevant

knowledge that was potentially unknown to users

prior to conducting the exploratory search.

• Requirement 5: Aggregation of Data Points

Into Insights.

The application should provide both a static

overview of all relevant data, as well as dynamic

aggregations of important data points displayed

in meaningful ways, in order to facilitate drawing

impactful conclusions for the respective research

ﬁelds.

3 CONSTRUCTING THE

RESEARCH INSTITUTION

KNOWLEDGE GRAPH

In recent years, knowledge graphs (KGs) have es-

tablished as an approach for semantically represent-

ing knowledge about real-world entities in a machine-

readable format (Schneider et al., 2022). In contrast to

relational databases or document databases, KGs can

explicitly capture all kinds of semantic relationships

between different entities. Since our data is highly in-

terconnected and the primary focus of this approach

is on data retrieval and analysis, we propose the us-

age of a KG as the database for exploratory search in

research institutions.

To construct the KG, ﬁrst a hierarchy of research

topics is needed that can be used to semantically

link other relevant entities. Manually deﬁned hierar-

chies although very precise, are usually very domain-

speciﬁc or generic and cannot fully capture the whole

range of existing research topics in a research institu-

tion. Therefore, we propose to use the automatically

constructed ﬁeld-of-study (FoS) hierarchy of the Mi-

crosoft Academic Graph (MAG) (Sinha et al., 2015;

Shen et al., 2018; Wang et al., 2020). The hierarchy

consists of over 200K hierarchically structured FoS

concepts, covering a broad spectrum of academic dis-

ciplines. Thereby, each FoS concept represents one

distinct academic discipline. After obtaining the FoS

hierarchy, the other entities need to be semantically

linked to speciﬁc research topics. To this end, the

publications of researchers can be classiﬁed accord-

ing to the existing FoS concepts. Classiﬁcation of

articles according to their related FoS concepts can

be performed by using semantic similarity scores be-

tween the respective text representations of concepts

and publications (Shen et al., 2018; Toney and Dun-

ham, 2022). Subsequently, the ontology shown in

Figure 1 can be used as a data model to semantically

A Knowledge Graph Approach for Exploratory Search in Research Institutions

267

Research Institution

Researcher

has_associate

University

subclass_of

School

subclass_of

Professorship

Department

subclass_of

Research Group

subclass_of

Chair

subclass_of

Other Research Institution

subclass_of

subclass_of subclass_of

is_associated_to

Publication

is_author_of has_author

references is_referenced_by

Field of Study

is_studied_in has_ﬁeld_of_study

subﬁeld_of superﬁeld_of

Figure 1: Simpliﬁed ontology of the research institution knowledge graph. Here, only the University entity, representing

German universities, with its respective sub-units is expanded. Other types of research entities are grouped in the node Other

Research Institution. All entities are transitively connected to the Field of Study entity. This means that each researcher and

each sub-unit of the research institute can be linked to their respective ﬁeld-of-study by traversing the knowledge graph.

link the remaining entities. Finally, the classiﬁcation

of publications together with transitive relations can

be used to infer the FoS concepts of researchers and

their associated sub-units.

4 PROOF OF CONCEPT

4.1 Architecture

We propose a proof of concept based on a classic

client-server architecture. Thereby, the backend con-

sists of a NodeJS app which operates as a server and

is connected to a Neo4j database containing all se-

mantically linked data. The frontend is a client-side

single-page application implemented in React with a

thin server architecture where most business logic is

moved from the server to the client that requests data

only as needed, thereby allowing for a seamless user

experience.

4.2 Data Model

As data model of the application, we propose the on-

tology shown in Figure 1. For a prototypical imple-

mentation Research Institution should the central en-

tity. This entity concept, which can be substituted

by several sub-classes, is linked to important entities

such as researchers and transitively also to publica-

tions, as well as to ﬁelds-of-study. The data model

can be discretionarily extended to support further en-

tities and their associated data.

4.3 Features

• Search & Browse.

Users can start looking for the desired information

either by using the full-text search to search for

speciﬁc data entities from the underlying model,

or they can browse a sortable list of research ﬁelds

that serves as a starting point to navigate deeper

into the data hierarchy and explore the related data

by following the embedded links. As shown in

Figure 2, users can use the input ﬁeld to search in

the application and browse through research ﬁelds

using the FoS tiles.

• Statistics & Analytics.

All data entities are enriched with meaningful

statistics and analytics supported by graph visual-

izations for a better understanding. Among other

things, it is possible to compare research ﬁelds

regarding amount of citations or check which re-

search topics are currently trending. Figure 2 il-

lustrates how the data can be used to display cur-

rent research trends within the institution.

• Semantically Linked Data.

Since all data is semantically linked, it is possible

to navigate to related content from any point in

the data hierarchy. Furthermore, the application is

able to identify and recommend additional related

content by means of a similarity search that is au-

tomatically conducted over the data set, thereby

complementing the identiﬁcation of related con-

tent through metadata.

KMIS 2023 - 15th International Conference on Knowledge Management and Information Systems

268

Figure 2: Mock-up of the proof of concept homepage. Users can either use the search to retrieve speciﬁc information

or explore the existing research landscape at the institution by using the hierarchically structured FoS concepts. On the

homepage, the FoS concept of each high-level research domain is shown in a convenient layout. Additionally, users can

immediately gain insights into current research trends within the institution. Here, users can examine research trends within

different depth levels of the FoS hierarchy.

5 LIMITATIONS

Since our proposed solution requires a single source

of truth, it is vital that relevant data is available and

of high quality. Complex data structures are prone

to becoming stale quickly, which reduces the overall

expressiveness of the presented information. The ne-

cessity to acquire all relevant data as well as to keep it

up-to-date and consistent requires the maintainers of

the data source to continually check and if necessary

update the data.

By design, the solution is limited to all relevant

data of one research institution. Further enriching the

data hierarchy with semantic links to external sources

is out of scope. Additionally, since all FoS concepts

are inevitably tied to publications, only those research

areas for which publications exist can be represented.

Due to the inherent complexity of displaying large

amounts of data in a clear and concise manner, our

proposed approach still requires users to invest time

before they can gain an extensive overview of more

complex topics.

A Knowledge Graph Approach for Exploratory Search in Research Institutions

269

6 CONCLUSION

Understanding the landscape of research institutions

is a challenging task for researchers. Current solu-

tions do not fulﬁll the requirements imposed by the

resulting exploratory search process. Through several

interviews we derived key requirements for a possi-

ble solution that we prototypically sketched out as a

minimal web application serving as proof of concept.

Our proposed application is able to provide extensive

information for a rather vague exploratory objective.

Further, it enables users to gain insights into the de-

sired search space by semantically linking related en-

tities and aggregating relevant data points into insight-

ful analytical views. However, the approach is limited

by the quality and availability of relevant data and is

restricted to data of one research institution. Since

accumulating large amounts of data and presenting

them in an exhaustive yet concise manner is inher-

ently complex, the degree to which such a task can

be simpliﬁed is limited and still requires the user to

invest a certain amount of time and effort in order to

reach the desired goal. That being said, we do believe

that having relevant data semantically linked and eas-

ily accessible within one application helps alleviate

the necessary time and effort investment to achieve

an exhaustive overview of a particular research land-

scape. In future work, we aim to implement the pro-

posed application as a minimum viable product that

we can then systematically evaluate. To further en-

hance the proposed approach, we subsequently plan

to extend the initial implementation based on the user

feedback.

REFERENCES

Athukorala, K., Głowacka, D., Jacucci, G., Oulasvirta, A.,

and Vreeken, J. (2016). Is exploratory search dif-

ferent? a comparison of information search behav-

ior for exploratory and lookup tasks. Journal of the

Association for Information Science and Technology,

67(11):2635–2651.

Brainard, J. (2020). Scientists are drowning in covid-19

papers. can new tools keep them aﬂoat. Science,

13(10):1126.

Dong, Y., Ma, H., Shen, Z., and Wang, K. (2017). A cen-

tury of science: Globalization of scientiﬁc collabora-

tions, citations, and innovations. In Proceedings of

the 23rd ACM SIGKDD International Conference on

Knowledge Discovery and Data Mining, KDD ’17,

page 1437–1446, New York, NY, USA. Association

for Computing Machinery.

Grether, L. and Witschel, H. F. (2022). Studying inter-

action patterns for knowledge graph exploration. In

Proceedings of the 14th International Joint Confer-

ence on Knowledge Discovery, Knowledge Engineer-

ing and Knowledge Management - KDIR, pages 257–

264. INSTICC, SciTePress.

Hearst, M. A. and Stoica, E. (2009). NLP support for

faceted navigation in scholarly collection. In Pro-

ceedings of the 2009 Workshop on Text and Citation

Analysis for Scholarly Digital Libraries (NLPIR4DL),

pages 62–70, Suntec City, Singapore. Association for

Computational Linguistics.

Jaradeh, M. Y., Oelen, A., Farfar, K. E., Prinz, M., D’Souza,

J., Kismih

ok, G., Stocker, M., and Auer, S. (2019).

Open research knowledge graph: Next generation in-

frastructure for semantic scholarly knowledge. In

Proceedings of the 10th International Conference on

Knowledge Capture, K-CAP ’19, page 243–246, New

York, NY, USA. Association for Computing Machin-

ery.

Palagi, E., Gandon, F., Giboin, A., and Troncy, R. (2017).

A survey of deﬁnitions and models of exploratory

search. In Proceedings of the 2017 ACM Workshop

on Exploratory Search and Interactive Data Analyt-

ics, ESIDA ’17, page 3–8, New York, NY, USA. As-

sociation for Computing Machinery.

Schneider, P., Schopf, T., Vladika, J., Galkin, M., Simperl,

E., and Matthes, F. (2022). A decade of knowledge

graphs in natural language processing: A survey. In

Proceedings of the 2nd Conference of the Asia-Paciﬁc

Chapter of the Association for Computational Lin-

guistics and the 12th International Joint Conference

on Natural Language Processing (Volume 1: Long

Papers), pages 601–614, Online only. Association for

Computational Linguistics.

Shen, Z., Ma, H., and Wang, K. (2018). A web-scale system

for scientiﬁc knowledge exploration. In Proceedings

of ACL 2018, System Demonstrations, pages 87–92,

Melbourne, Australia. Association for Computational

Linguistics.

Sinha, A., Shen, Z., Song, Y., Ma, H., Eide, D., Hsu, B.-

J. P., and Wang, K. (2015). An overview of microsoft

academic service (mas) and applications. In Proceed-

ings of the 24th International Conference on World

Wide Web, WWW ’15 Companion, page 243–246,

New York, NY, USA. Association for Computing Ma-

chinery.

Toney, A. and Dunham, J. (2022). Multi-label classiﬁca-

tion of scientiﬁc research documents across domains

and languages. In Proceedings of the Third Workshop

on Scholarly Document Processing, pages 105–114,

Gyeongju, Republic of Korea. Association for Com-

putational Linguistics.

Wang, K., Shen, Z., Huang, C., Wu, C.-H., Dong, Y.,

and Kanakia, A. (2020). Microsoft Academic Graph:

When experts are not enough. Quantitative Science

Studies, 1(1):396–413.

White, R. W. and Roth, R. A. (2009). Exploratory Search:

Beyond the Query-Response Paradigm, pages 1–98.

Springer International Publishing, Cham.

KMIS 2023 - 15th International Conference on Knowledge Management and Information Systems

270