=Paper=
{{Paper
|id=Vol-2960/paper5
|storemode=property
|title=On the Need for a Body of Knowledge on Recommender Systems (Short paper)
|pdfUrl=https://ceur-ws.org/Vol-2960/paper5.pdf
|volume=Vol-2960
|authors=Juri Di Rocco,Davide Di Ruscio,Claudio Di Sipio,Phuong T. Nguyen,Claudio Pomo
|dblpUrl=https://dblp.org/rec/conf/recsys/RoccoRSNP21
}}
==On the Need for a Body of Knowledge on Recommender Systems (Short paper)==
https://ceur-ws.org/Vol-2960/paper5.pdf
On the Need for a Body of Knowledge on Recommender
Systems
Juri Di Rocco1 , Davide Di Ruscio1 , Claudio Di Sipio1 , Phuong T. Nguyen1 and Claudio Pomo2
1
DISIM, Università degli studi dell’Aquila, 67100 L’Aquila, Italy
2
SisInf Lab, Politecnico di Bari, 70125 Bari, Italy
Abstract
Recommender systems (RSs) are becoming widespread in different application domains to provide personalized items to given
service users. Because of such an increasing adoption of RSs, it is becoming urgent to define a precisely curated and organized
core set of concepts and practices, i.e., a Body of Knowledge (BOK), as already done in other disciplines, including software
engineering and model-driven engineering. The opportunities related to the availability of an RSBOK are manifold, and
different stakeholders would benefit from it including, developers, teachers, and newcomers to the RS community. Further
than motivating a BOK for recommender systems and discussing corresponding envisioned opportunities and challenges, we
also propose a methodology that can be employed to support the definition of an RSBOK.
1. Introduction was defined with different goals, including “1) promot-
ing a consistent view of software engineering worldwide;
Recommender systems (RSs) are complex software sys- 2) specifying the scope of, and clarify the place of soft-
tems that can provide users with relevant items of inter- ware engineering with respect to other disciplines [. . . ]; 3)
est for the particular application domains and contexts characterizing the contents of the software engineering dis-
[1, 2]. Over the last decade, different types of recom- cipline; 4) providing a foundation for curriculum develop-
mendation technologies have been conceived by both ment and for individual certification and licensing material”
industry and academia to improve the relevance of the [3]. Similarly, a Body of Knowledge for Software Lan-
items being recommended. RSs have become pervasive, guage Engineering has been recently promoted with the
and almost in any application domain, there is the avail- aim of assembling and organizing “artifacts, definitions,
ability of software systems in charge of supporting users methods, techniques, best practices, open challenges, case
in undertaking the particular tasks at hand (whether it be studies, teaching material, and other components that will
software developers who are working on some software afterwards help students, researchers, teachers, and practi-
components, or users who want to select the next movie tioners to learn from, to better leverage, to better contribute
to watch). to, and to better disseminate the intellectual contributions
While RSs are becoming ubiquitous, we believe that it and practical tools and techniques coming from the SLE
is necessary to ensure that the next generation of engi- field” [4]. A Body of Knowledge for Model-Based Soft-
neers will have a clear understanding of the fundamental ware Engineering (MBSE) has been recently promoted
techniques and tools underpinning the development and [5] as an extension of SWEBOK to characterize the MBSE
usage of RSs. To this end, it is necessary to agree on the discipline in the context of Software Engineering. Many
core concepts, mechanisms, and practices related to the other BoKs have been defined over the last decade in dif-
development and use of RSs. Therefore, as done in other ferent application domains including data management,
software disciplines, we foster an RS Body of Knowl- enterprise architecture, business analysis, project man-
edge (RSBOK) definition to formalize and characterize agement, and data management. All of them share the
the Recommender System discipline. goals of characterizing the contents of a particular disci-
To the best of our knowledge, SWEBOK is the first pline and to support various activities including training
body of knowledge that was conceived for characteriz- and the development of new technologies and tools.
ing the software engineering discipline [3]. SWEBOK In this paper, we introduce some preliminary thoughts
of an RS Body of Knowledge (RSBOK) definition for the
3rd Edition of Knowledge-aware and Conversational Recommender Software Engineering domain. We promote a comprehen-
Systems (KaRS) & 5th Edition of Recommendation in Complex sive perception of the core concepts, mechanisms, and
Environments (ComplexRec) Joint Workshop @ RecSys 2021,
September 27–October 1, 2021, Amsterdam, Netherlands practices related to the development and deployment of
" juri.dirocco@univaq.it (J. Di Rocco); davide.diruscio@univaq.it RSs. The ultimate aim is to understand the fundamental
(D. Di Ruscio); claudio.disipio@graduate.univaq.it (C. Di Sipio); techniques and tools pertinent to the development and
phuong.nguyen@univaq.it (P. T. Nguyen); claudio.pomo@poliba.it usage of recommender systems in software engineering.
(C. Pomo) Altogether, this is expected to come in handy for those
© 2021 Copyright for this paper by its authors. Use permitted under Creative
CEUR
Workshop
http://ceur-ws.org
ISSN 1613-0073
Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org) who work as recommender systems designers.
Proceedings
2. Knowledge-aware and tools to widen and simplify the adoption of
RSs in any complex software systems. The knowl-
Recommender Systems edge encoded in the envisioned RSBOK frame-
Despite their enormous popularity and the remarkable work can help developers, among others, to distill
performance that recommender systems have achieved in the algorithms that should be made available to
recent years, one of the long-standing problems affecting final users.
the performance of these systems concerns the sparsity of – Teachers: They are researchers, practitioners, and
interactions between users and items. Over the past years, educators in general who are in charge of train-
recommender system designers have relied on additional ing new generation of professionals in the design,
sources of information to overcome this issue. Modern development and operation of new RSs. To this
RSs combine collaborative information with metadata end, RSBOK can be used as a reference point, e.g.,
(e.g., tags, reviews), social connections, image and audio to link code examples, textual explanations, expe-
signal-derived features, and contextual data [6] to build riences, RS usage and development best practices,
domain-dependent, cross-domain, or context-aware rec- and code examples that demonstrate the usage of
ommendation models. Among the various sources, one RS technologies.
of the most relevant is Knowledge Graphs (𝒦𝒢𝑠). Thanks – Contributors: These users correspond to RSs de-
to the heterogeneous fields covered by 𝒦𝒢𝑠 and the velopers, and adopters who are willing to extend
myriad of specific techniques that have been developed, the knowledge formalized in RSBOK.
knowledge-based recommendation systems emerged as
a novel research field in the RecSys community. The field As previously mentioned, the wanted RSBOK
is generally known as knowledge-aware recommenda- paradigm should consist of a precise formalization of
tion systems (KaRS [7]) and blends the most advanced concepts and tools underpinning the development, usage
machine learning algorithms with cutting-edge knowl- and enhancement activities of any RSs. It is expected to
edge representation paradigms. This collective effort has provide an effective means to characterize and to support
resulted in several improvements in recommendation [8], different related activities including training and the de-
knowledge completion [9], preference elicitation, and velopment of new technologies and tools. Recently, we
user modeling research, thus producing a vast literature. had already the need to investigate and formalize the RS
field in Software Engineering, and we came up with a
model representing all the features typically supported
3. Opportunities and Challenges by RSs [10]. Figure 1 represents only the top-level fea-
Conceiving a Body of Knowledge for Recommender Sys- tures or recommendation systems, i.e., Data Preprocess-
tems (RSBOK) would disclose several opportunities based ing, Capturing Context, Producing Recommendations, and
on the availability of a common and formally defined vo- Presenting Recommendations which are the main func-
cabulary that would prescribe the usage and development tionalities typically implemented by recommendation
of recommender systems on clearly defined foundations, systems [11, 12].
instead of relying on some uncommon understanding. We performed such a conceptualization to underpin
The RSBOK definition is indeed a community effort to the design and development of the different RSs devel-
formalize and share knowledge from different stakehold- oped in the context of the EU CROSSMINER project [10].
ers on conceptual and practical RS aspects. However, the We extracted all the shown components mainly from
investment would pay off because we foresee at least the existing studies [11] as well as from our development
following kinds of users that can take advantage of the experience under the needs of the CROSSMINER project.
RSBOK paradigm [4]: A similar conceptualization work has been done to de-
sign and develop the Elliot framework [13], which aims
– Newcomers: They are perspective users and re- at supporting reproducible recommender systems eval-
searchers who do not have any knowledge about uation. The authors had to conceptualize different rec-
RSs, and are willing to understand them. They ommendation algorithms, splitting strategies, evaluation
can benefit from the results of the conceptualiza- protocols, metrics, and tasks to simplify the specification
tion efforts, e.g., to get an overview of the typ- and execution of experimental pipelines by processing
ical algorithms employed to develop RSs, or to simple configuration files as the one shown in Fig. 2
explore available linked textual explanations or Even though the opportunities related to the avail-
examples about some typically used evaluation ability of an RSBOK are immense in our opinion, its
methodologies. realization can be hampered by a number of challenges
– Developers: These include advanced RSs develop- including the following ones:
ers, who are interested in conceiving techniques – Languages to be adopted: To make the definition
Figure 1: Main design features of recommendation systems in software engineering (refinement of Di Rocco et al. [10]).
ble notations that might be employed to formalize
the results of the conceptualization efforts.
– Realization process: By looking at the ways other
BOKs have been developed, we believe that we
need a community effort, which has to be per-
formed by following precise protocols, modera-
tion mechanisms, quality check procedures, to
name a few. In this respect, it is of great impor-
tance to support contributions that may come
from different stakeholders. Consequently, to
keep the quality of the resources under control,
it is necessary to define processes and setup tools
for moderating the different contributions and to
make sure that they are all homogenized.
Figure 2: Simple Elliot configuration file. – Engagement: Even though the RSs opportunities
can be convincing, they might not be enough to
engage people in concretely contributing with
and the usage of RSBOK homogenous, it is neces- the RSBOK definition and managing its whole
sary to decide the languages and tools that need lifecycle.
to be adopted. Feature diagrams, Ecore models,
and OWL ontologies are only examples of possi-
4. Conclusion and Future Work [7] V. W. Anelli, P. Basile, D. G. Bridge, T. D. Noia,
P. Lops, C. Musto, F. Narducci, M. Zanker,
To facilitate a clear understanding of the fundamental Knowledge-aware and conversational recom-
techniques and tools underpinning the development and mender systems, in: S. Pera, M. D. Ekstrand,
usage of RSs, in this paper we envisaged the core con- X. Amatriain, J. O’Donovan (Eds.), Proceedings
cepts, mechanisms, and practices related to the develop- of the 12th ACM Conference on Recommender
ment and use of RSs. We aim to foster RSBOK, an RS Body Systems, RecSys 2018, Vancouver, BC, Canada,
of Knowledge definition to formalize and characterize October 2-7, 2018, ACM, 2018, pp. 521–522.
the Recommender System domain. URL: https://doi.org/10.1145/3240323.3240338.
For future work, we plan to tackle the challenges men- doi:10.1145/3240323.3240338.
tioned in Section 3. In particular, it is necessary to put [8] V. W. Anelli, T. D. Noia, E. D. Sciascio, A. Ragone,
into effect the conceived paradigm by realizing its con- J. Trotta, How to make latent factors inter-
stituent components. Among others, we will investigate pretable by feeding factorization machines with
and select suitable tools and languages, attempting to knowledge graphs, in: C. Ghidini, O. Hartig,
make the definition and usage of RSBOK homogeneous. M. Maleshkova, V. Svátek, I. F. Cruz, A. Hogan,
J. Song, M. Lefrançois, F. Gandon (Eds.), The Se-
mantic Web - ISWC 2019, Proceedings, Part I,
References volume 11778 of Lecture Notes in Computer Sci-
[1] M. P. Robillard, W. Maalej, R. J. Walker, T. Zim- ence, Springer, 2019, pp. 38–56. URL: https://doi.
mermann (Eds.), Recommendation Systems in org/10.1007/978-3-030-30793-6_3. doi:10.1007/
Software Engineering, Springer Berlin Heidel- 978-3-030-30793-6\_3.
berg, Berlin, Heidelberg, 2014. doi:10.1007/ [9] G. He, J. Li, W. X. Zhao, P. Liu, J. Wen, Mining
978-3-642-45135-5. implicit entity preference from user-item interac-
[2] F. Ricci, L. Rokach, B. Shapira, Introduction to tion data for knowledge graph completion via ad-
recommender systems handbook, in: F. Ricci, versarial learning, in: Y. Huang, I. King, T. Liu,
L. Rokach, B. Shapira, P. B. Kantor (Eds.), Recom- M. van Steen (Eds.), WWW ’20: The Web Confer-
mender systems handbook, Springer US, Boston, ence 2020, Taipei, Taiwan, April 20-24, 2020, ACM
MA, 2011, pp. 1–35. / IW3C2, 2020, pp. 740–751. URL: https://doi.org/
[3] P. Bourque, R. E. Fairley, IEEE Computer Society, 10.1145/3366423.3380155. doi:10.1145/3366423.
Guide to the Software Engineering Body of Knowl- 3380155.
edge, 2014. [10] J. Di Rocco, D. Di Ruscio, C. Di Sipio, P. T.
[4] B. Combemale, R. Lämmel, E. Van Wyk, SLE- Nguyen, R. Rubei, Development of recommen-
BOK: The Software Language Engineering Body dation systems for software engineering: the
of Knowledge (Dagstuhl Seminar 17342) (2018) 10 CROSSMINER experience 26 (2021) 69. URL:
pages. doi:10.4230/DAGREP.7.8.45. https://doi.org/10.1007/s10664-021-09963-7. doi:10.
[5] F. Ciccozzi, M. Famelis, G. Kappel, L. Lambers, 1007/s10664-021-09963-7.
S. Mosser, R. F. Paige, A. Pierantonio, A. Rensink, [11] J. Bobadilla, F. Ortega, A. Hernando, A. Gutiér-
R. Salay, G. Taentzer, A. Vallecillo, M. Wim- rez, Recommender systems survey, Knowl.
mer, Towards a body of knowledge for model- Based Syst. 46 (2013) 109–132. URL: https://doi.
based software engineering, in: Proceedings of org/10.1016/j.knosys.2013.03.012. doi:10.1016/j.
the 21st ACM/IEEE International Conference on knosys.2013.03.012.
Model Driven Engineering Languages and Sys- [12] A. Bellogín, A. Said, Improving accountability
tems: Companion Proceedings, ACM, Copenhagen in recommender systems research through repro-
Denmark, 2018, pp. 82–89. doi:10.1145/3270112. ducibility, CoRR abs/2102.00482 (2021). URL: https:
3270121. //arxiv.org/abs/2102.00482. arXiv:2102.00482.
[6] V. W. Anelli, V. Bellini, T. D. Noia, W. L. Bruna, [13] V. W. Anelli, A. Bellogín, A. Ferrara, D. Malitesta,
P. Tomeo, E. D. Sciascio, On the role of time and F. A. Merra, C. Pomo, F. M. Donini, T. D. Noia, El-
sessions in diversifying recommendation results, liot: A comprehensive and rigorous framework for
in: F. Crestani, T. D. Noia, R. Perego (Eds.), Pro- reproducible recommender systems evaluation, in:
ceedings of the 8th Italian Information Retrieval F. Diaz, C. Shah, T. Suel, P. Castells, R. Jones, T. Sakai
Workshop, Lugano, Switzerland, June 05-07, 2017, (Eds.), SIGIR ’21: The 44th International ACM SIGIR
volume 1911 of CEUR Workshop Proceedings, CEUR- Conference on Research and Development in Infor-
WS.org, 2017, pp. 92–96. URL: http://ceur-ws.org/ mation Retrieval, July 11-15, 2021, ACM, 2021, pp.
Vol-1911/16.pdf. 2405–2414. URL: https://doi.org/10.1145/3404835.
3463245. doi:10.1145/3404835.3463245.