Optimizing Academic Pairings in Smart Campuses: A Recommendation
System for Academic Communities
Elvandi da Silva Junior
1 a
, Gabriel Casanova
1 b
, Daniel Youssef de Hollanda Lopes
1 c
Ana Paula Militz Dorneles
1 d
, Renan Bordignon Poy
1 e
, Jos
e Palazzo M. de Oliveira
2 f
and Vin
ıcius Maran
1 g
Laboratory of Ubiquitous, Mobile and Applied Computing (LUMAC), Polytechnic School,
Federal University of Santa Maria, Av. Roraima, 1000, Santa Maria, Brazil
Informatics Institute, Federal University of Rio Grande do Sul, Porto Alegre, Brazil
Academic Pairings, Hybrid Filtering, Recommendation System, Smart Campus.
Collaborating across disciplines can advance research fields by offering new ways of addressing complex prob-
lems and fostering integration and acknowledging similarities among different fields. Pairing preferences, such
as teaching the same or different content areas, grade spans, or buildings, are also crucial in mentorship pro-
grams, as they can significantly impact the effectiveness of the partnership. This necessary academic pairing
can be difficult to a set of factors, as cultural, geographical or personal. This study addresses the challenge
of academic pairings, emphasizing the need to publicize university resources and projects to promote inter-
connection between professionals from different areas within the same academic environment. The research
describes the ”Unified Recommendation System” - a recommendation system for academic communities. This
is a hybrid recommendation system and was developed to recommend relevant projects of interest to the user,
in addition to being easy to access through an application for students, teachers and the general community.
Thus, the developed prototype demonstrated significant potential as a relevant tool in the context of smart
campuses, with user interest recommendation rates of more than 81% in the evaluated scenario.
Smart campuses are part of a broad category of smart
solutions designed to enhance human experiences in
a variety of ways, such as fostering connectivity, ef-
ficiency, and personalization (Sneesl et al., 2022b),
while also contributing to a more sustainable planet
(Clark and Eisenberg, 2008).
This category strongly emphasizes the integra-
tion of innovative tools into education (Huang et al.,
2019), striving to equip students, teachers and the
campus community with access to resources that hold
the potential to revolutionize the educational enviro-
ment into a more digitized and contemporary setting
(Yin, 2014).
One challenge that smart campuses educational
tools are well-equipped to address is Academic Pair-
ings. This involves a collaborative process where
academics and practitioners coexist in a shared
space, jointly generating knowledge (Stahl and Hesse,
2009). This process can also foster interprofessional
collaboration (Davoli and Fine, 2004), positively im-
pacting both the projects and the individuals involved
(Chiocchio et al., 2011; Karam et al., 2018; Green and
Johnson, 2015). Furthermore, these smart resources
can be useful for promoting ongoing university initia-
tives to the academic community.
Academic communities refer to groups of indi-
viduals who are involved in academic pursuits, such
as students, faculty, researchers, and administrators,
and who share common goals and interests related to
the pursuit, exchange, and validation of knowledge
(Wakeling et al., 2019). An effective strategy involves
the use of Recommendation Systems (RS), which are
commonly divided into the types: content-based fil-
Silva Junior, E., Casanova, G., Lopes, D., Paula Militz Dorneles, A., Poy, R., M. de Oliveira, J. and Maran, V.
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities.
DOI: 10.5220/0012630200003693
In Proceedings of the 16th International Conference on Computer Supported Education (CSEDU 2024) - Volume 1, pages 124-134
ISBN: 978-989-758-697-2; ISSN: 2184-5026
Copyright © 2024 by Paper published under CC license (CC BY-NC-ND 4.0)
tering (CBF), collaborative filtering (CF), and hybrid
filtering (HF) (Adomavicius and Tuzhilin, 2005).
These systems have the capability to suggest items
to users (Aamir and Bhusry, 2015), including projects
and professors. In this context, this study aims to ad-
dress the problem of Academic Pairings at universi-
Thus, an intermediary–a recommendation sys-
tem must be established between users and poten-
tial educational items of interest, through an appli-
cation that employs a hybrid recommendation system
(Maruyama et al., 2023). To adress this, we developed
a recommendation system, considering the specifics
of educational items related to academic pairings. We
developed a prototype of this recommendation system
and evaluated it in the UFSM
The paper is structured as follows: Section 2
presents a background on the subjects of Recom-
mender Systems and Smart Campuses; Section 3 pro-
vides an overview of related work; Section 4 de-
scribes the recommendation system built and its soft-
ware architecture; Section 5 discusses the data acqui-
sition and evaluation process in testing the recommen-
dation system and Section 6 summarizes the conclu-
sionss of this paper.
This section provides a comprehensive review of rec-
ommender systems, focusing on the most frequently
employed algorithms in recommendation systems.
Additionally, it explains the concept of smart cam-
puses, describes their various types, and different ap-
plication areas.
2.1 Recommender Systems
Recommender systems, which emerged in the 90s, are
filtering tools designed to improve the quality of rec-
ommendations and are extensively utilized across the
internet (Batmaz et al., 2018). Currently, these sys-
tems are prevalent on various online platforms, such
as e-commerce, where they assist customers in lo-
cating products, and social networks,where they aid
users in identifying content of interest, as well as
other applications, such as banking.
These systems have become increasingly inte-
gral to users’ daily lives, due to their personalized
suggestions and their effectiveness surpassing that
of keyword-based search engines (Bai et al., 2020).
Whithin an academic context, recommender systems
can prove to be exceptionally useful for a variety of
reasons. They function as facilitators of learning and
teaching tasks, enable students to make more appro-
priate and informed decisions, and assist organiza-
tions in gaining a better understanding of users (Ver-
bert et al., 2012; Hagemann et al., 2019; Zhou et al.,
Recommendation systems in a smart campus can
employ various techniques to provide relevant recom-
mendations to members of the academic community.
Below, some of these techniques and some of the al-
gorithms associated with each of them will be pre-
2.1.1 Content-Based Filtering
Content-Based Filtering utilizes detailed information
about academic items and user preferences to provide
relevant recommendations. ”Items” in the academic
context include courses, lectures, learning resources,
academic articles, events, and other elements related
to the academic environment. Content-Based Filter-
ing algorithms analyze the characteristics of the items
and create user profiles based on previously demon-
strated preferences (Banik, 2018). Among these algo-
rithms, we can mention:
TF-IDF (Term Frequency-Inverse Document
Frequency). TF-IDF is a widely used algorithm
for text analysis. It assesses the importance of
keywords in documents based on how frequently
they appear relative to a set of documents. In the
academic context, TF-IDF can be applied to rec-
ommend study materials, academic articles, and
resources based on keywords and topics related
to the user’s interests. The algorithm ranks items
based on how well their characteristics (e.g., key-
words) match the user’s profile (Karabiber, 2020).
Word Embedding. Word embedding algorithms,
such as Word2Vec and GloVe, map words to real-
number vectors. These vectors represent the se-
mantic meaning of words and, by extension, aca-
demic items. In the academic context, these vec-
tors can be used to calculate semantic similarity
between items and suggest those with similar se-
mantic content to the user’s interests (Kenyon-
Dean et al., 2020).
2.1.2 Collaborative Filtering
Collaborative filtering is based on the interactions and
behaviors of members of the academic community to
make recommendations. It considers how students,
professors, researchers, and other users interact with
each other, participate in academic activities, and col-
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities
Figure 1: The framework of collaborative filtering-based
RS (Chen et al., 2018).
laborate on research projects (Banik, 2018) (Figure
This method can be user-based or item-based, as
described below:
User-Based Collaborative Filtering: In this
method, user similarity is calculated based on
their past behaviors and preferences. The k-
Nearest Neighbors (KNN) algorithm is commonly
used to find similar users. Once similar users are
identified, the system recommends items based on
the choices of these similar users that have not yet
been evaluated by the current user (Boehmke and
Greenwell, 2019).
Item-Based Collaborative Filtering: In this ap-
proach, item similarity is calculated based on
user preferences. The KNN algorithm is an ex-
ample, and it calculates the similarity between
items based on user ratings. Then, it recommends
items similar to those that a user has already liked
(Boehmke and Greenwell, 2019).
The rationale of user-based CF and item-based CF
is shown in Figure 2.
Figure 2: The rationale of user-based CF and item-based
CF (Chen et al., 2018).
2.1.3 Hybrid Filtering
Hybrid filtering combines elements of content-based
filtering and collaborative filtering to provide more
accurate and diverse recommendations. This ap-
proach can be executed in various ways, including in-
tegrating results from both approaches or switching
between them based on specific criteria. The algo-
rithms used in hybrid filtering can include any combi-
nation of content-based and collaborative filtering al-
gorithms, depending on the adopted strategy (Banik,
There are some techniques and models that can be
employed to achieve better results in hybrid filtering.
Below are some of the most common types of hybrid
recommendations (Kharsa and Al Aghbari, 2023):
Weighted Models: In hybrid systems, the results
of content-based filtering and collaborative filter-
ing techniques are combined using a weighted
model. This model assigns weights to recommen-
dations generated by each technique based on user
context and preferences. The weighted combina-
tion of results helps provide balanced and person-
alized recommendations.
Technique Switching: In this approach, the sys-
tem alternates between content-based filtering and
collaborative filtering based on user needs and
contextual characteristics. For example, it may
use collaborative filtering to recommend study
groups based on users’ past interactions and then
apply content-based filtering to suggest study
materials related to the topics discussed in the
Recommendation-Level Machine Learning:
Machine learning algorithms, such as neural net-
works, can be used to automatically learn how
to combine the results of different filtering tech-
niques in a hybrid recommendation model. This
enables dynamic adaptation to user preferences
and interaction patterns, resulting in highly per-
sonalized recommendations (Kharsa and Al Agh-
bari, 2023).
CSEDU 2024 - 16th International Conference on Computer Supported Education
Considering that the proposal of this work is cat-
egorized as Recommendation-Level Machine Learn-
ing, in the next sections we will present important
concepts that were used in developing the proposal:
Cosine Similarity and the KNN algorithm.
2.1.4 Cosine Similarity
Cosine similarity plays an important role in collab-
orative filtering and content-based recommendation,
helping to identify similar users or items to create per-
sonalized recommendations (Chen et al., 2018). It can
be applied both to measure similarity between users
and between items in recommendation systems:
Cosine Similarity for User: In this case, each
user is represented as a vector of item ratings.
Cosine similarity is calculated between the rating
vectors of two users to measure how similar their
likes and preferences are. High cosine similarity
between two users indicates that they have sim-
ilar preferences, which can be used to make rec-
ommendations based on the interactions of similar
Cosine Similarity for Item: In this case, each
item is represented as a vector of user ratings.
Cosine similarity is calculated between the rating
vectors of two items to measure how similar the
items are in terms of user preferences. High co-
sine similarity between two items indicates that
they are similar in terms of how they are rated by
users, which can be used to suggest similar items
when a user interacts with a specific item.
2.1.5 KNN Algorithm
The k-Nearest Neighbors (KNN) algorithm is a super-
vised machine learning method used for classification
and regression (Khoa et al., 2013). It operates based
on the idea that similar objects tend to be close to each
other in a feature space (Zhang et al., 2018). KNN is
a simple and intuitive approach (Boehmke and Green-
well, 2019):
Training. During the training phase, the algo-
rithm stores all examples from the training dataset
in a multi-dimensional space, where each example
is represented by a feature vector. Each example
is also associated with a class or target value.
Classification/Regression. When making predic-
tions for a new example, KNN searches for the k
nearest examples in the feature space. Proxim-
ity is typically measured using a metric like Eu-
clidean distance. The k nearest examples either
vote for the class of the example or contribute to
calculating a weighted average for regression.
Choosing k. The choice of the k value is crucial
in KNN. A small k value (e.g., 1) makes the algo-
rithm sensitive to noise and outliers, while a large
k value smoothens the decision boundary, making
it less sensitive to local data variations.
Weight Function. To address different weights
for nearest neighbors, you can assign weights
based on distance. Closer neighbors may have
larger weights than farther ones.
Classification vs. Regression. KNN can be used
for both classification and regression tasks. In
classification, the example is assigned to the most
frequent class among the k nearest neighbors. In
regression, a weighted average of the target values
of the k nearest neighbors is calculated.
Performance. KNN’s performance can be influ-
enced by the appropriate choice of distance met-
ric, data preprocessing, and selecting an optimal k
value. It’s also important to have a representative
training dataset.
2.1.6 Cold Start
The Cold start in recommendation systems refers to
situations in which the system needs to make rec-
ommendations without relevant historical data. This
challenge is particularly common when dealing with
new users or items and requires the application of cre-
ative strategies to provide a useful experience to users,
even when information is limited.
User-related cold start occurs when a new user
registers in the recommendation system and hasn’t
had significant interactions yet. Given that the system
lacks information about this user’s preferences, mak-
ing personalized recommendations becomes a chal-
lenge. On the other hand, item-related cold start hap-
pens when a new item is introduced into the recom-
mendation system and lacks a history of interactions
with users. In this scenario, the system has no data on
how users have reacted to that specific item (Joy et al.,
2.2 Smart Campuses
In the realm of technological advancements, a con-
cept that has gained significant prominence is that of
a Smart Campus. It can be characterized as an entity
that strategically employs technology and infrastruc-
ture with the main goal of improving its processes,
thus improving their utility for individuals (S
Torres et al., 2018).
The realization of this concept can be achieved
through the deployment of a diverse array of tools.
These include IoT (Internet of Things) technology,
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities
sensors, cloud computing, user interfaces, blockchain
and a variaty of other applications (Gilman et al.,
2020; Fern
es and Fraga-Lamas, 2019).
Each of these components can facilitate the creation
and implementation of Smart Campus, contributing
unique benefits to the system.
Some of the benefits include the enhanced qual-
ity of life for inhabitants of these environments; en-
hanced user experience for students, staff, and es-
tate managers, optimized space utilization, improved
management of resources such as computer labs and
air conditioners and energy conservation, among oth-
ers (Villegas-Ch. et al., 2019; Sutjarittham et al.,
2018; Wang, 2014).
In spite of these benefits, smart campuses also face
significant challenges. These include limitations of
perspectives that rely entirely on data, the integration
of individuals’ practices and the importance of con-
sidering ethics and domestication (Bates and Friday,
2017). Additional challenges involve the absence of a
technology adoption model, perceived fees, perceived
trust and perceived value (Sneesl et al., 2022a).
This section provides an overview of related research
in recommender systems and smart campuses, high-
lighting recommended item types and filtering tech-
niques employed in each work.
(Ibrahim et al., 2019) introduced a framework tar-
geting the academic domain, recommending various
courses for students. Their hybrid system combines
collaborative and content-based filtering with ontol-
ogy for information extraction.
(Bai et al., 2019) explored a collaborative recom-
mendation system for researchers with similar inter-
ests. The model incorporates collaborative filtering,
content-based filtering, and social network-based fil-
tering, forming a hybrid approach.
(Mrhar and Abik, 2019) proposed a recommenda-
tion system for online course platforms, offering per-
sonalized suggestions based on user profiles. The au-
thor utilizes content-based filtering and deep learning
to enhance algorithm accuracy.
(Xiao et al., 2018) presents a personalized recom-
mendation system that tailors suggestions to individ-
ual users based on their interests and learning history.
The system employs both content-based and collabo-
rative filtering techniques to recommend didactic ma-
terials and essential learning resources.
This work aims to develop a platform that recom-
mends a wide variety of items based on previous stud-
ies. Table 1 summarizes the types of recommended
resources and the methods used for processing data.
The proposed system adapts precisely to individ-
ual user preferences. It offers a personalized and
profile-based recommendation experience that accu-
rately aligns with user-defined preferences during
item acquisition. This is achieved through the inte-
gration of advanced algorithms and recommendation
elements, named as ”Unified Recommendation Sys-
The “Unified Recommendation System” serves as a
software architecture for smart campuses, offering a
variety of recommender systems and techniques tai-
lored to different item types.
4.1 Data Aquisition
Obtaining data is a crucial step in optimizing recom-
mendation systems, providing the essential founda-
tion for filtering algorithms to analyze a wide range
of information. In the world of recommendation sys-
tems, data is the lifeblood that yields valuable insights
into user preferences, behaviors, and trends. This pro-
cess entails not only the collection of diverse data but
also ensuring that recommendation algorithms have a
robust basis for generating accurate and pertinent sug-
A Python script was used to extract data from the
UFSM projects and professors system, throught a web
scrapping process. In this way, a total of 56865 items
were obtained, divided between projects, professors
and events.
From the data obtained, it was necessary to to an-
alyze the collected resources, separating them into
different categories and topics, so that the filter-
ing algorithm could obtain greater precision in the
search and recommendation. To accomplish this task,
we used the ChatGPT 3.5 Large Language Model
(LMM)(OpenAI, 2023) to categorize items based on
a series of frequently used generic categories. This
step resulted in 9540 categories and subcategories.
4.2 Software Architecture Definition
The software architecture to support recommenda-
tion processes is segmented into three main modules,
which is as shown in Figure 3: The User Access Mod-
ule (UAM), the Recommendation Management mod-
CSEDU 2024 - 16th International Conference on Computer Supported Education
Table 1: Studies and Characteristics of Recommendation Systems.
Work Type of Recommended Item Filtering Strategy
(Ibrahim et al., 2019) Undergraduate and post-
graduate courses
Ontology-based filtering, Content-based filter-
ing, Collaborative filtering
(Bai et al., 2019) Collaborating researchers Collaborative filtering, Content-based filtering,
Social network-based filtering
(Mrhar and
Abik, 2019)
Online courses Content-based filtering, Deep Learning
(Xiao et al., 2018) Didactic materials for learning Collaborative filtering, Content-based filtering
(Hoang et al., 2022) Courses in specialized fields Ontology-based filtering, Content-based filter-
ing, Collaborative filtering
This research Educational resources (courses,
mini-courses, video lessons,
teaching materials, e-books,
lectures, events, scientific articles,
thesis, similar user profiles,
other educational platforms)
Collaborative filtering, Content-based filtering
ule (RMM), and the Administrative and Persistence
Module (APM).
The modules of the software architecture are pre-
sented below:
4.2.1 User Access Module (UAM)
In the user access module, users engage with the ma-
terial provided through recommendations, by using
their personal devices such as cell phones, comput-
ers, and alternative devices. This section allows users
to interact with resources, direct themselves to other
university portals, and communicate their interests.
The module is composed by three submodules:
(i) User Interface Module, responsible for presenting
recommendations appropriately to the user, (ii) Con-
trol Module: Responsible for capturing user usage in-
formation and sending usage statistics to the recom-
mendation and administration modules, and (iii) Per-
sistence and Communication module, responsible for
storing information collected in the recommendation
process and make requests to other modules of the ar-
chitecture (using the REST standard).
UAM was prototyped using the Ionic framework
(Yusuf, 2016), which allows the generation of web
and mobile applications.
4.2.2 Administrative and Persistence Module
The Administrative and Persistence Module is the re-
sponsible for the communication with university ex-
ternal systems. It is also responsible to manage and
store user profile and navigation information.
The module is composed by four submodules:
(i) Communication module, responsible for make re-
quests to other modules of the architecture (using the
REST standard), (ii) Persistence module, responsible
for the storage of the data collected in the recom-
mendation process, and to retrieve items to present
to users in a recommendation, (iii) User Navigation
Module, responsible to store and retrieve user data
given by the UAM module, and (iv) User Profile Pro-
cessig Module, responsible for the management of
user profile data and his/her topics of interest.
APM was prototyped using Java SpringBoot
framework (Yusuf, 2016) using JDBC driver to the
communication with PostgreSQL database. APM
stores resource descriptions, user interests, interacted
items, and recommendation history. All recommen-
dations are saved in this section and, when requested,
sent to the recommendation management module to
initiate the filtering and recommendation process.
4.2.3 Recommendation Management Module
The recommendation management module contains
all the code blocks and functions of the recommen-
dation system. Here, new codes are added, edited and
corrected, building the core of the personalized rec-
ommendation process for each user. This environ-
ment processes user information, sends requests of
the database, and operates as an indirect approach, re-
flecting the user’s actions and interactions.
One of the key strengths of the platform’s recom-
mendation system lies in its utilization of a hybrid ap-
proach, seamlessly blending both content-based and
collaborative filtering techniques.
By incorporating these two distinct filtering meth-
ods, the platform optimizes the diversity and fresh-
ness of recommendations. Content-based filtering al-
lows the system to understand the intrinsic character-
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities
Figure 3: “Unified Recommendation System” software architecture definition.
istics of educational resources and match them with
the user’s preferences. On the other hand, collabora-
tive filtering leverages collective user behavior to sug-
gest resources that align with the preferences of users
with similar tastes. Changing between these filters
guarantees that whenever a user asks for recommen-
dations, they receive new suggestions.
The algorithm encompassing Content-Based Fil-
tering adheres to a sequence of operations for gener-
ating recommendations. The first step begins by cat-
aloging the user’s interests into tables, which are sub-
sequently associated with the user’s topics. Following
this, educational resources that align with these topics
are then listed based on the the user-topic relationship.
Each resource undergoes a process known as ‘bag-
of-words’, which isolates keywords, quantifies their
occurrences, and compiles a list accordingly. An anal-
ogous process is conducted for user-favorite items.
Cosine similarity is then employed to calculate the
similarity between text features. Ultimately, the fea-
tures bearing the highest similarity are recommended
to the user and and can be seen later displayed on the
Having users previously declared their interests,
the algorithm employs Collaborative Filtering by
seeking out other users with comparable interests,
creating a list for each. The interests of both users
are combined, with a numerical increment for each
shared interest. The algorithm then selects the recom-
mended resources for the users with the highest sim-
ilarity and compiles them into a list. This list is then
shuffled and ultimately presented to the user.
RMM was prototyped in Python language, with
Scikit (Kramer and Kramer, 2016) and Surprise (Hug,
2020) libraries.
5.1 Evaluation Scenario
In order to assess the accuracy of resource recom-
mendations, a profile representing a student from the
Polytechnic School of UFSM who has a broad inter-
est in the field of technology has been created, which
is as shown in Figure 4. This profile was created with
the intention of delving deeper into the analysis of the
effectiveness of recommendations provided to users,
considering their preferences and affinities with the
technology field as a whole.
In this situation, topics of interest such as Tech-
nology, Applied Technology, Assistive Technology,
Automotive Technology, Banking Technology, BIM
Technology, and Bluetooth Technology (as shown in
Figure 5a) were selected. This approach enabled the
system to search for relevant data and information re-
lated to these specific subjects and to identify other
users who shared the same interests. Thus, a dy-
namic platform was created that promotes intercon-
nection among individuals with diverse technology-
related affinities, enriching the user experience and
fostering knowledge exchange and collaboration.
The evaluation of each recommended resource oc-
curs by sliding it to the side of the screen, which is as
shown in Figure 5b, where the left represents disin-
terest and the right signifies that the user wants more
recommendations of that type. This simple and intu-
itive interaction provides users with effective control
over the suggested content, allowing them to further
customize their experience according to their prefer-
ences. As users interact with the system, they provide
CSEDU 2024 - 16th International Conference on Computer Supported Education
Figure 4: New user registration screen.
valuable feedback that continually enhances the ac-
curacy of recommendations, making the learning and
discovery environment even more tailored to their in-
dividual needs.
5.2 Results and Discussion
In the proposed test scenario, the recommendation
system returned a sample consisting of 58 recom-
mended items in about 2 seconds, as shown in Figure
6. Among these items, we noticed that 47 of them
piqued the user’s interest, while 11 did not generate
interest, resulting in an accuracy rate of 81% in rec-
ommendations related to this specific profile.
The accuracy of the algorithm tends to improve as
users interact more with the system and accumulate a
more substantial data history. This enhancement be-
comes even more noticeable as we add more users
and data to our database. This growth results in an
increase in the number of users with similar interests,
which, in turn, leads to more refined recommenda-
tions aligned with each user’s individual preferences.
The recommendation process, based on user pref-
erences and interests, has proven to be an effective
tool for presenting projects aligned with their selected
topics of interest. The recommendations have high-
lighted projects directly related to specific areas of
interest, as well as professors in those fields or who
were directly involved in a project.
The presence of projects that piqued the interest of
users with similar profiles added an additional layer of
relevance, suggesting that the recommendations not
only met individual criteria but also reflected com-
mon trends and preferences within the community of
users with similar interests. Furthermore, the identi-
fication of professors with expertise and involvement
in the areas of interest enriched the recommendations
by providing a direct connection to the academic and
practical knowledge of these professionals.
However, it is important to note that projects with
similar keywords were recommended, even if they
did not align with the user’s specific interest. This
occurred, for example, when projects related to Lan-
guage Technologies or Methodologies for Social Sci-
ences were suggested due to the broad search for
”Technology, even if the user had not explicitly se-
lected these categories.
The similarity in keywords between the recom-
mended resources and the selected categories can also
lead to confusion during the evaluation process. Mis-
interpretation of an item due to shared project titles
or similar elements can result in inaccurate assess-
ments. This inaccuracy in evaluation can, in turn, lead
to errors in recommendations, underscoring the im-
portance of enhancing recommendation accuracy to
avoid misunderstandings and optimize the user expe-
As users continue to use the system and provide
more feedback, the algorithm becomes more profi-
cient in understanding and anticipating their needs
and preferences. This process initiates an ongoing
cycle of improvement, making the system progres-
sively more effective in suggesting relevant and perti-
nent content.
Furthermore, as our user community grows, and
more people share similar opinions and interests, the
algorithm also benefits from collective wisdom. This
further contributes to the increased accuracy of rec-
ommendations. Therefore, the continuous enhance-
ment of the algorithm’s accuracy is a dynamic process
that benefits all users, enriching their experience and
making it more personalized.
The implementation of innovative technologies has
brought significant modernization to the field of e-
learning, especially in smart academic environments.
This has resulted in the creation of a new scenario
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities
(a). Interface to select topics of inter-
est by the user (in Portuguese).
(b). Recommendation evaluation screen – the user can swipe to right or left,
see more details about the item or mark it as favorite.
Figure 5: Example of interfaces provided by the mobile application.
Figure 6: Interest statistics screen.
where communication among various mobile devices
plays a fundamental role.
This study highlights the effectiveness of a rec-
ommendation system that is based on individual user
interests to address the challenges that teachers, stu-
dents, and the academic community as a whole face
in finding compatible academic peers and facilitating
connections within the academic community. How-
ever, it also recognizes the importance of improving
the accuracy of recommendations in order to avoid
potential inaccuracies, enhance the user experience,
and increase the efficiency of the proposed solution.
One innovative feature of this study is its approach
to connecting the academic community with projects
aligned with their interests, revealing opportunities
that might otherwise have gone unnoticed. The most
notable aspect is that the recommendation process
uses multiple techniques to recommend items to aca-
demic communities.
However, it is important to note that this study has
limitations, such as the evaluation process being con-
ducted at a single university, the absence of testing
with a large number of users, and the lack of longi-
tudinal data to track changes in user interests over
time. Possible areas for future research include ex-
panding testing to other universities, enhancing sys-
tem accuracy, incorporating Deep Learning and Neu-
ral Network techniques, conducting performance tests
CSEDU 2024 - 16th International Conference on Computer Supported Education
on different hardware and platforms, adding georef-
erencing features for points of interest in the applica-
tion, and utilizing ontologies for even more personal-
ized recommendations.
This research is supported by CNPq/MCTI/FNDCT
n. 18/2021 grant n. 405973/2021-7 and CNPq/MCTI
10/2023 - UNIVERSAL grant n. 402086/2023-6.
The research by Jos
e Palazzo M. de Oliveira is par-
tially supported by CNPq grant 306695/2022-7 PQ-
SR. The research by Vin
ıcius Maran is partially sup-
ported by CNPq grant 306356/2020-1 DT-2, CNPq
Aamir, M. and Bhusry, M. (2015). Recommendation sys-
tem: State of the art approach. International Journal
of Computer Applications, 120:25–32.
Adomavicius, G. and Tuzhilin, A. (2005). Toward the
next generation of recommender systems: a survey
of the state-of-the-art and possible extensions. IEEE
Transactions on Knowledge and Data Engineering,
Bai, X., Wang, M., Lee, I., Yang, Z., Kong, X., and Xia, F.
(2019). Scientific paper recommendation: A survey.
IEEE Access, 7:9324–9339.
Bai, X., Wang, M., Lee, I., Yang, Z., Kong, X., and Xia, F.
(2020). Scientific paper recommendation: A survey.
IEEE Access, 7:9324–9339.
Banik, R. (2018). Hands-On Recommendation Systems
with Python: Start building powerful and personal-
ized, recommendation engines with Pytbhon. Packt
Bates, O. and Friday, A. (2017). Beyond data in the smart
city: Repurposing existing campus iot. IEEE Perva-
sive Computing, 16:54–60.
Batmaz, Z., Yurekli, A., Bilge, A., and Kaleli, C. (2018).
A review on deep learning for recommender systems:
challenges and remedies. Artificial Intelligence Re-
view, 52:1–37.
Boehmke, B. and Greenwell, B. (2019). Hands-On Ma-
chine Learning with R. Chapman & Hall/CRC The R
Series. CRC Press.
Chen, R., Hua, Q., Chang, Y.-S., Wang, B., Zhang, L., and
Kong, X. (2018). A survey of collaborative filtering-
based recommender systems: From traditional meth-
ods to hybrid methods based on social networks. IEEE
Access, 6:64301–64320.
Chiocchio, F., Forgues, D., Paradis, D., and Iordanova, I.
(2011). Teamwork in integrated design projects: Un-
derstanding the effects of trust, conflict, and collabo-
ration on performance. Project Management Journal,
42:78 – 91.
Clark, W. and Eisenberg, L. (2008). Agile sustainable com-
munities: On-site renewable energy generation. Utili-
ties Policy, 16:262–274.
Davoli, G. W. and Fine, L.-J. (2004). Stacking the deck
for success in interprofessional collaboration. Health
Promotion Practice, 5:266 – 270.
es, T. and Fraga-Lamas, P. (2019). To-
wards next generation teaching, learning, and context-
aware applications for higher education: A review
on blockchain, iot, fog and edge computing enabled
smart campuses and universities. Applied Sciences.
Gilman, E., Tamminen, S., Yasmin, R., Ristimella, E.,
Peltonen, E., Harju, M., Lov
en, L., Riekki, J., and
Pirttikangas, S. (2020). Internet of things for smart
spaces: A university campus case study. Sensors
(Basel, Switzerland), 20.
Green, B. and Johnson, C. D. (2015). Interprofessional col-
laboration in research, education, and clinical prac-
tice: working together for a better future. The Journal
of chiropractic education, 29 1:1–10.
Hagemann, N., O’Mahony, M. P., and Smyth, B. (2019).
Visualising module dependencies in academic recom-
mendations. Proceedings of the 24th International
Conference on Intelligent User Interfaces: Compan-
Hoang, V. N., Le Thanh, P., My, L. O. T., Vinh, L. C.,
and Xuan, V. T. (2022). Towards a novel architecture
of smart campuses based on spatial data infrastruc-
ture and distributed ontology. In Intelligent Systems
and Applications: Proceedings of the 2021 Intelligent
Systems Conference (IntelliSys) Volume 3, pages 662–
673. Springer.
Huang, L.-S., Su, J., and Pao, T. (2019). A context aware
smart classroom architecture for smart campuses. Ap-
plied Sciences.
Hug, N. (2020). Surprise: A python library for recom-
mender systems. Journal of Open Source Software,
Ibrahim, M. E., Yang, Y., Ndzi, D. L., Yang, G., and
Al-Maliki, M. (2019). Ontology-based personalized
course recommendation framework. IEEE Access,
Joy, J., Raj, N. S., and Renumol, V. G. (2021). Ontology-
based e-learning content recommender system for ad-
dressing the pure cold-start problem. Journal of Data
and Information Quality, 13.
Karabiber, F. (2020). Tf-idf—term frequency-inverse doc-
ument frequency.
Karam, M., Brault, I., Durme, T. V., and Macq, J. (2018).
Comparing interprofessional and interorganizational
collaboration in healthcare: A systematic review of
the qualitative research. International journal of nurs-
ing studies, 79:70–83.
Kenyon-Dean, K., Newell, E., and Cheung, J. C. K. (2020).
Deconstructing word embedding algorithms. arXiv
preprint arXiv:2011.07013.
Kharsa, R. and Al Aghbari, Z. (2023). Leveraging associa-
Optimizing Academic Pairings in Smart Campuses: A Recommendation System for Academic Communities
tion rules in feature selection for deep learning classi-
fication. SN Computer Science, 5(1):112.
Khoa, N. M., Viet, D. T., and Hieu, N. H. (2013). Clas-
sification of power quality disturbances using wavelet
transform and k-nearest neighbor classifier. In 2013
IEEE International Symposium on Industrial Elec-
tronics, pages 1–4.
Kramer, O. and Kramer, O. (2016). Scikit-learn. Machine
learning for evolution strategies, pages 45–53.
Maruyama, M. H. M., Silveira, L. W., de Oliveira, J. P. M.,
Gasparini, I., and Maran, V. (2023). Hybrid recom-
mender system for educational resources to the smart
university campus domain.
Mrhar, K. and Abik, M. (2019). Toward a deep recom-
mender system for moocs platforms. Proceedings of
the 3rd International Conference on Advances in Ar-
tificial Intelligence.
OpenAI (2023). Chatgpt 3.5. https://chat.openai.com. Ac-
cessed: 2023-12-20.
anchez-Torres, B., Rodr
ıguez, J. A., Rico-
Bautista, D. W., and Guerrero, C. D. (2018). Smart
campus: Trends in cybersecurity and future develop-
ment. Revista Facultad de Ingenier
ıa, 27(47):104–
Sneesl, R., Jusoh, Y. Y., Jabar, M. A., and Abdullah, S.
(2022a). Conceptualizing iot-based smart campus
adoption model for higher education institutions: A
systematic literature review. 2022 Applied Informat-
ics International Conference (AiIC), pages 7–12.
Sneesl, R., Jusoh, Y. Y., Jabar, M. A., and Abdullah, S.
(2022b). Revising technology adoption factors for iot-
based smart campuses: A systematic review. Sustain-
Stahl, G. and Hesse, F. (2009). Paradigms of shared knowl-
edge. International Journal of Computer-Supported
Collaborative Learning, 4:365–369.
Sutjarittham, T., Gharakheili, H., Kanhere, S., and Sivara-
man, V. (2018). Realizing a smart university campus:
Vision, architecture, and implementation. 2018 IEEE
International Conference on Advanced Networks and
Telecommunications Systems (ANTS), pages 1–6.
Verbert, K., Manouselis, N., Ochoa, X., Wolpers, M.,
Drachsler, H., Bosni
c, I., and Duval, E. (2012).
Context-aware recommender systems for learning: A
survey and future challenges. IEEE Transactions on
Learning Technologies, 5:318–335.
Villegas-Ch., W., Molina-Enriquez, J., Chicaiza-Tamayo,
C., Ortiz-Garc
es, I., and Luj
an-Mora, S. (2019). Ap-
plication of a big data framework for data monitoring
on a smart campus. Sustainability.
Wakeling, S., Spezi, V., Fry, J., Creaser, C., Pinfield, S.,
and Willett, P. (2019). Academic communities: The
role of journals and open-access mega-journals in
scholarly communication. Journal of Documentation,
Wang, H.-I. (2014). Constructing the green campus within
the internet of things architecture. International Jour-
nal of Distributed Sensor Networks, 10.
Xiao, J., Wang, M., Jiang, B., and Li, J. (2018). A person-
alized recommendation system with combinational al-
gorithm for online learning. Journal of ambient intel-
ligence and humanized computing, 9:667–677.
Yin, F. (2014). Construction and application of wisdom
campus. Computer Programming Skills & Mainte-
Yusuf, S. (2016). Ionic framework by example. Packt Pub-
lishing Ltd.
Zhang, S., Li, X., Zong, M., Zhu, X., and Wang, R. (2018).
Efficient knn classification with different numbers of
nearest neighbors. IEEE Transactions on Neural Net-
works and Learning Systems, 29(5):1774–1785.
Zhou, X., Zhang, J., and Zhang, Y. (2019). The 1st inter-
national workshop on context-aware recommendation
systems with big data analytics (cars-bda). Proceed-
ings of the Twelfth ACM International Conference on
Web Search and Data Mining.
CSEDU 2024 - 16th International Conference on Computer Supported Education