Multidimensional User Proﬁle Model to Support System

Recommendations in Complex Social Networks: Application to Hashtag

Recommendations

Abir Gorrab, Wala Rebhi, Narjes Bellamine Ben Saoud and Henda Hajjami Ben Ghezala

RIADI Laboratory, National School of Computer Sciences, University of Manouba, 2010, Tunisia

Keywords:

Recommendation Systems, Multidimensional Model, Generic User Proﬁle, Complex Social Network,

Context, Topic Recommendations.

Abstract:

Recommendation systems play a crucial role in providing relevant information through data analysis. One

of the pivotal challenges in the recommendation process is modeling user proﬁles. However, many existing

models focus on a single aspect to describe users, overlooking other valuable data. In response to this lim-

itation, this paper introduces a comprehensive multidimensional model that captures various dimensions of

a user within their complex social network. This model encompasses demographic, social, behavioral and

homophilic dimensions, with the goal of offering more holistic recommendations tailored to different con-

texts. Towards the end of this article, we introduce a focused application of the multidimensional model. This

speciﬁc application revolves around providing hashtag recommendations within the X platform (Twitter plat-

form). This serves as a tangible demonstration of how the proposed model can be applied in a practical context

within a real social network. The main goal is to comprehensively assess the model’s efﬁcacy in generating

recommendations by utilizing a varied set of user-related information. To accomplish this, we introduce and

evaluate a recommendation approach driven by our proposed user proﬁle model, showcasing relevant and no-

table results.

1 INTRODUCTION

The use of recommender systems (RS) has expanded

dramatically over the last decade, mostly due to their

enormous business value (Deldjoo et al., 2021). In

fact, RS are considered as a helpful tool for helping

the user in cutting the time needs to ﬁnd personal-

ized products, documents, friends, places and services

(Alhijawi and Kilani, 2020). Thus, RS have become

an important part of the web sites (SKazienko et al.,

2011). In particular, e-commerce sites since they help

people to make decision, what items to buy (Kazienko

and Kolodziejski, 2006), or which movie to watch

(Degemmis et al., 2007). For example, According to

the statistics revealed by Netﬂix, 75% of the down-

loads and rentals come from their recommendation

service (Deldjoo et al., 2021).

One of the requirements to create a successful and

usable recommender system is to build a detailed user

model (Martijn and Verbert, 2022). This model serves

as a foundation for the system to suggest items effec-

tively or adjust the interface accordingly (Graus and

Ferwerda, 2019).

In this context, many user models have been pro-

posed. For example, in many RS, a user has been rep-

resented by his preferences or his interests. For other

recommendation systems, only his behaviors have

been considered. Furthermore, with the widespread

use of social networks such as Facebook, X (Twitter),

LinkedIn, an enormous amount of information could

be analyzed and considered in the recommendation

process. This is why the user proﬁle model needs

to be further enriched to take into account a greater

amount of information.

Thus, in this paper, we propose a new hybrid mul-

tidimensional user proﬁle model that provides a com-

prehensive representation of a user within his social

network, including all relevant information that can

be used for generating recommendations. The model

considers the social, demographic, behavioral and

homophilic dimensions to ensure a thorough under-

standing of the user’s preferences and interests. Simi-

larly, a model-driven graph-based hashtag recommen-

dation approach is proposed in order to demonstrate

Gorrab, A., Rebhi, W., Ben Saoud, N. B. and Ben Ghezala, H. H.

Multidimensional User Proﬁle Model to Support System Recommendations in Complex Social Networks: Application to Hashtag Recommendations.

DOI: 10.5220/0013118200003890

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2025) - Volume 3, pages 241-248

ISBN: 978-989-758-737-5; ISSN: 2184-433X

241

how the proposed model could be utilized in a real-

world context.

The remainder of this article is structured as fol-

lows. Section 2 investigates related works to user pro-

ﬁle models devoted to recommendation systems. The

new proposed user proﬁle model for recommendation

support is depicted in section 3. In section 4, we

evaluate the performances of our model by integrat-

ing it in a hashtag recommendation system, proving

the practical usefulness of the contribution via a real

data based example. In the last section, we conclude

and present an outlook on future works.

2 RELATED WORKS

Various techniques and algorithms have been devel-

oped to build and use user proﬁle models for recom-

mendation systems (Ko et al., 2022).

In this context, the proposed user proﬁle mod-

els could be classiﬁed into four categories, notably:

content-based, demographic, collaborative ﬁltering

and contextual models.

To start with, content-based models or behaviour-

based models are widely used in recommender sys-

tems (Middleton and De Roure, 2004; Zhang et al.,

2023). These models rely on the user’s historical be-

havior and preferences, often utilizing a binary classi-

ﬁcation model to represent what users ﬁnd interesting

and uninteresting (Middleton and De Roure, 2004).

By analyzing the content of items that the user has in-

teracted with, these models recommend similar items

that align with their preferences (Jerry et al., 2024).

As for demographic models in recommendation

systems, they rely on user demographic information

such as age, gender, and location. These models

leverage this information to recommend items that are

popular among users with similar demographic char-

acteristics (Tahmasebi et al., 2021). An important as-

pect of demographic models is their role in mitigating

the cold-start problem, a common challenge in rec-

ommendation systems (Safoury and Salah, 2013; Al-

hijawi and Kilani, 2020; Tahmasebi et al., 2021). By

leveraging demographic data, these models improve

the accuracy of user proﬁle modeling, enabling more

precise predictions for cold-start users, even in plat-

forms like Reddit (Sharma et al., 2021).

When it comes to collaborative ﬁltering models or

social models, they are based on the user’s interac-

tions with other users. They recommend items based

on the preferences of users with similar tastes. The

key to the success of personalized recommendation

lies in the correct use of “collective intelligence,” and

one user behaves similarly to some other users (Qian

et al., 2023).

Finally, contextual models take into account the

user’s current context, such as time, location, and

mood, and recommend items that are relevant to the

user’s current situation (Rattanajitbanjong and Ma-

neeroj, 2009; Minsung et al., 2024).

These existing proﬁle models in recommendation

systems, while offering beneﬁts, are not without crit-

icisms. One major criticism is their potential over-

simpliﬁcation of user preferences, as they often rely

on limited feedback sources that may fail to cap-

ture the intricacies of individual tastes. Furthermore,

the representation of user preferences within pro-

ﬁles can be incomplete, overlooking certain dimen-

sions and resulting in less comprehensive recommen-

dations. These models may also neglect the impor-

tance of serendipity and exploration, focusing solely

on personalized recommendations based on past be-

havior. Privacy concerns arise due to the collection

and utilization of user data for proﬁling purposes. Ad-

ditionally, proﬁle models face challenges in handling

the cold-start problem for new users or items with

limited data. Their limited adaptability to evolving

user preferences further raises concerns. Recogniz-

ing these limitations, researchers should explore hy-

brid or alternative approaches to enhance recommen-

dation accuracy and ensure a more satisfying user ex-

perience.

Furthermore, with the emergence of social net-

works today, a vast amount of data can be analyzed

to improve recommendations. In this regard, an indis-

pensable component is a multidimensional proﬁle that

represents users across various dimensions such as so-

cial, demographic, behavioral and homophilic within

a speciﬁc context.

This proﬁle will be described in the next section.

3 PROPOSED

MULTIDIMENSIONAL USER

PROFILE MODEL

In this paper, we aim to propose a new generic mul-

tidimensional user proﬁle model to support recom-

mendations in complex social networks. This model,

given in Figure 1, is:

• Generic. Our proposed model is designed to be

generic and applicable across multiple recommen-

dation systems. It can effectively support various

domains, including movies, products, and even

social connections.

• Multidimensional. Our model is characterized

by its multidimensional nature as it incorporates

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

242

various user dimensions, including social, behav-

ioral, and demographic factors. By consider-

ing these diverse aspects, our model can capture

a more comprehensive understanding of users’

preferences and characteristics.

• Dynamic. Our model is designed to be dynamic

and adaptable, capable of evolving based on the

speciﬁc context and domain it operates in. It

recognizes that user data is not static but rather

evolves over time. As users interact with the

system, their preferences, behaviors, and demo-

graphic information may change or be updated.

• Contextual-Based. Due to its generic nature, our

model has the ﬂexibility to support a wide range

of contexts. It can seamlessly adapt and provide

recommendations in various domains and scenar-

ios. Whether it’s recommending movies, books,

music, restaurants, or any other type of item, our

model can handle different contexts effectively.

Figure 1: Proposed multidimensional user proﬁle model.

As illustrated in Figure 1, user is at the core of

the proposed model, and it is crucial to represent

him accurately by considering all relevant informa-

tion which are: Dynamic multilayer social networks,

user’s Proﬁle and recommendation’s Context. These

three classes are outlined below.

3.1 Dynamic Multilayer Social

Networks

The ﬁrst essential aspect to consider in system rec-

ommendations is the users’ environment, which in-

cludes their participation in complex social networks.

Indeed, users are not isolated entities but are intercon-

nected within complex social structures that inﬂuence

their preferences and behaviors. Moreover, as the use

of social media has become much more widespread

than before (Roozbahani et al., 2022), a user has been

involved in multiple social networks simultaneously

including popular platforms such as Facebook, Twit-

ter, Instagram, and many others. This phenomenon

has given rise to what is commonly referred to as a

multilayer social network (Li et al., 2023).

Furthermore, a key characteristic of these multi-

layer social networks is their evolution and their con-

tinual change over time (Ceria et al., 2022). This is

why, in our model, we are considering dynamic mul-

tilayer social networks.

Illustratively, Figure 2 captures a segment of a dy-

namic multilayer social network, akin to the structure

found in platforms like Twitter. Within this network,

users cultivate two discernible relationship types: one

layer encapsulates connections formed through the

act of following, and another layer delineates interac-

tions encompassing tweets and retweets. Moreover,

this intricate structure is depicted at two speciﬁc in-

stances, labeled as t1 and t2, emphasizing the dy-

namic character of the network and revealing the pro-

gression of relationships and interactions over time.

Figure 2: Twitter as a dynamic multilayer social network.

Within the dynamic multilayered social network,

each user is characterized by a proﬁle. This will be

the subject of the next section.

3.2 User’s Proﬁle

User’s proﬁle includes all necessary knowledge for

effective query evaluation and production of relevant

information tailored to each user (Rebhi et al., 2017a).

It is usually integrated into the system to impart the

user knowledge to the system to enable personalized

adaptations and avoid ”unnecessary” dialogues be-

tween the system and the user (Liu et al., 2009). As

shown in Fig.1 and based on what we have recurrently

found in the literature, a user’s proﬁle could be con-

sidered as a combination of two dimensions: generic

and speciﬁc.

Multidimensional User Proﬁle Model to Support System Recommendations in Complex Social Networks: Application to Hashtag

Recommendations

243

Generic Dimension. Regarding personal informa-

tion, including details such as name, birthday, ad-

dress, nationality, education, profession, and physi-

cal factors like weight, height, and eye color, these

demographic elements collectively contribute to un-

derstanding a user’s context (Beel et al., 2013; Al-

Shamri, 2016; Rebhi et al., 2017a). They aid recom-

mender systems in comprehending individual prefer-

ences, facilitating the delivery of culturally relevant

and suitable recommendations. By integrating such

information, the recommendations become more per-

sonalized, thereby enhancing the likelihood of user

satisfaction and engagement with the platform.

Speciﬁc Dimension. Concerns data that is speciﬁc

to each layer of the social network. This data could

be social data or behavioral data. As for social data

(Li et al., 2019), it involves structural interactions

within each layer. Much more, the ways in which

users interact with each other within layers. For ex-

ample, friendships on Facebook or links of tweets and

retweets on Twitter. Additionally, these connections

or links between users might exhibit a homophilic di-

mension (Aiello et al., 2012). This means that users

with similar interaction patterns or behaviors tend to

connect more. For instance, the presence of a high

number of common friends between two users indi-

cates a similarity in their social networks, reﬂecting a

homophilic aspect in their interactions.

For the behavioral dimension (Dhelim et al.,

2022), it pertains to the patterns of actions and ac-

tivities undertaken by users within each layer of the

network. This dimension involves the analysis of user

behavior and activities to comprehend and model how

individuals navigate, communicate, and engage with

others on the social network. Furthermore, the key

aspect of the behavioral dimension is content. The

content encompasses two signiﬁcant aspects:

• Posting Content. Understanding the type and fre-

quency of content that users post, such as status

updates, photos, videos, or links, offers insights

into their interests and preferences.

• Reactions to Content. Assessing how users re-

spond to various types of content, including their

emotional reactions, aids in gauging the impact of

content on the community.

Moreover, the behavioral dimension can incorpo-

rate homophilic elements, with the number of com-

mon posts between two users serving as an exemplary

illustration of such a dimension. Homophily denotes

the inclination of individuals to connect with others

who share similarities with them in certain aspects. In

the context of social networks, this similarity is man-

ifested in their behavioral patterns. For instance:

• Number of Common Posts. Users frequently post-

ing about similar topics or sharing common in-

terests may exhibit a higher number of common

posts, indicating a homophilic dimension. This

suggests that these users engage with and ex-

press similar content, reinforcing the idea that

they share commonalities.

• Shared Interests: Analyzing the types of content

that users engage with or post about can unveil

shared interests. Users with a signiﬁcant overlap

in the topics they post about or the content they in-

teract with demonstrate a homophilic connection

based on shared interests.

• Similar Posting Patterns. Examining the timing,

frequency, and style of posts can unveil similar-

ities in posting behavior. Users with compara-

ble posting patterns are likely to have homophilic

connections, aligning their behaviors in the way

they contribute to the social network.

This user’s proﬁle is always linked to a context in

which the recommendation process is executed.

3.3 Recommendation’s Context

The term ”context” lacks a singular deﬁnition, given

its diverse applications across various ﬁelds (Casillo

et al., 2023). Considering the expansive nature of the

concept (Zimmermann et al., 2007), multiple deﬁni-

tions have been proposed.

From the array of context deﬁnitions, we embrace

one of the most widely accepted and formalized con-

cepts (Zainol and Nakata, 2010), as articulated by

(Dey, 2001): ”context is any information that can be

used to characterize the situation of an entity. An en-

tity, in this context, refers to a person, place, or object

considered relevant to the interaction between a user

and an application, including both the user and the ap-

plications themselves.” In the realm of recommenda-

tion systems, a situation is akin to the snapshot of the

multilayer social network at a speciﬁc moment. Con-

sequently, the recommendation context encompasses

any information pertaining to this situation that holds

relevance for the recommendation process.

To structure our multidimensional model, we draw

upon the generic context model proposed by (Zainol

and Nakata, 2010), incorporating three distinct con-

text categories: Extrinsic Context, Interface Context,

and Intrinsic Context. These categories correspond to

the triggering elements of the recommendation pro-

cess, addressing the ”who” question through user pro-

ﬁle attributes (Intrinsic Context), the ”why” question

via recommendation needs (Interface Context), and

the ”where” question concerning the environment it-

self (Extrinsic Context) (Zainol and Nakata, 2010).

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

244

4 EVALUATION: APPLICATION

TO HASHTAG

RECOMMENDATIONS

To evaluate the proposed multidimensional user pro-

ﬁle model, our objective in this section is to introduce

an application for hashtag recommendation. To ac-

complish this, we furnish information about the uti-

lized data, the recommendation approach employed,

and the resulting outcomes.

4.1 Data Set Description

To validate the proposed modeling approach, we used

a set of data collected from X social network (for-

merly Twitter) (Danisch et al., 2014). From this data

set, we extracted a signiﬁcant sample of 20000 tweets

corresponding to 1000 users. We have thus obtained a

set of tweets, including the user identiﬁer, tweet iden-

tiﬁer, tweet body (list of hashtags in the tweet), the

timestamp at which the tweet was issued, and whether

the tweet is a response to another tweet or not. We

note that these tweets and retweets are fully formed of

hashtags. We chose users with different social char-

acteristics (i.e. a different number of tweets, retweets,

followers and hashtags). We process to data cleaning

by eliminating stop words and short hashtags (com-

posed of less then 3 characters). Thus, from a start-

ing list of 135420 hahstags, we eliminated 3451 stop

words and short ones to maintain 131969 hashtags.

With this data, the ﬁrst thing to do is to build our

multidimensional model. Thus, we instantiate our

user proﬁle model taking X (Twitter) as our dynamic

social network. We differentiate 5 time periods dur-

ing which tweets are sent, notably T0, T1, T2, T3 and

T4. From X, we distinguish two layers: Following

layer, tweets and retweets layer. Figure 3 represents

an example of instantiation of the proposed model at

a moment T1. For clarity, we have just represented

one user (U1).

In the initial ’Following’ layer, we examine the so-

cial dimension, emphasizing the connections formed

through user following and the count of shared fol-

lowers. In contrast, within the second layer involving

tweets and retweets, our focus shifts to the relation-

ship established through retweets among users. Here,

in terms of the behavioral dimension, our attention

centers on the hashtags included in each tweet. Lo-

cation and age are the subject of demographic dimen-

sion. Those dimensions form the generic proﬁle of

each user.

After establishing our model, our aim is to offer

the most pertinent hashtag recommendations for each

user. To fulﬁll this objective, the subsequent section

Figure 3: Instance of the proposed multidimensional user

proﬁle model.

introduces an approach to recommendation guided by

this model.

4.2 Used Model-Driven Hashtag

Recommendation Approach

Drawing from our multidimensional model and

within the recommendation context, which is in our

case hashtag recommendations for individual users,

our objective is to identify relevant suggestions. In

line with this, as depicted in Figure 4, we advocate

for the implementation of a hybrid recommendation

approach. This approach is based on three phases.

The initial phase involves graph construction, trans-

forming the model into an exploitable structure. Sub-

sequently, a community detection algorithm is em-

ployed to identify clusters for each user. Finally, a

hashtag selection process is executed to provide per-

tinent hashtag recommendations tailored to each user.

These phases will be detailed in the following.

Phase 1: Graph Construction. To exploit the

proposed model with its different dimensions, we

choose to use graph as a powerful mathematical

abstraction, especially for user proﬁle (Daoud et al.,

2009; Caro-Mart

ınez et al., 2023). To do so, we reuse

the temporal multirelational information graph-based

model (Rebhi et al., 2017b) for representing entities

(i.e., actors in social network) and their relationships

(Rebhi et al., 2022). For the collected data, each

user is described by a set of tweets. To simplify

Multidimensional User Proﬁle Model to Support System Recommendations in Complex Social Networks: Application to Hashtag

Recommendations

245

Figure 4: Hashtag recommendation case study.

the analysis of tweets, we apply Latent Dirichlet

Allocation(LDA)(Blei et al., 2003). This step is

of major importance, as LDA allows us to analyze

the semantics of hashtags in our context and group

them into signiﬁcant topics. LDA is applied to

each user proﬁle to extract the meaningful topics

that represent it. In our data, for each proﬁle, we

tested different values of k (number of signiﬁcant

topics), and ultimately, we chose the value k=3. This

decision was based on the observation that for a

number of topics greater than 3, we did not observe

a signiﬁcant difference in the hashtags within each

topic. Consequently, there is a stabilization of the

content within each group of hashtags. As a result

one user is represented by 3 topics. For example, for

the user U1, the 3 topics are: (america, usa, news);

(science , technology , Google , tech) and (job, team).

Then, at each time t

and for each layer j, a simi-

larities matrix MS

(i)

is deﬁned as follows:

(i)

= (ms

(i)

)

1≤k,l≤Nbr

. (1)

Indeed, each element (ms

(i)

)

k,l

represents the sim-

ilarity between the node N

and the node N

within

the layer j at t

. For the tweets and retweets layer,

we consider similarities between users based on each

extracted meaningful topics. As for following layer,

we consider, similarities based on age and location.

Phase 2: Users Clustering. Once, we have

constructed the graph, our aim now is to form

users clusters. To do so, we propose to apply the

Stable Communities Detection Method for Temporal

Multiplex Graphs (Rebhi et al., 2021). We opted

for this approach due to its effectiveness and its

capability to build communities by utilizing the

diverse dimensions proposed in our model (Rebhi

et al., 2021). Applied to our data , we obtain 8

clusters grouping similar users.

Phase 3: Hashtags Selection. Now, as each

user belongs to his pertinent cluster, we use these

representative hashtags from each cluster and sort

them based on their decreasing frequency within the

cluster. We then suggest to each user, in order, the

most frequent and recent hashtags in the cluster that

they have not already used.

4.3 Results and Discussion

The proposed Hashtag recommendation approach is

applied to the collected data, considering only the ﬁrst

four time instances (T0, T1, T2, and T3). The ﬁ-

nal instance is allocated for validation, where the ob-

tained recommendations are assessed by comparing

them with the hashtags users have actually shared at

T4. Therefore, we measure the performance of our

system in retrieving the same hashtags contained in

this last tweet. Furthermore, to compare the perfor-

mance of our system with existing works, we have

chosen to position ourselves in relation to the study

by (Dovgopol and Nohelty, 2015), which proposed

two recommendation methods. The ﬁrst relies mainly

on Naive Bayes, while the second utilizes cosine sim-

ilarity with the KNN method. As evaluation metric,

we choose to calculate the Recall which is in our case

represents the system’s ability to provide all relevant

hashtags in response. For us, relevant hashtags are

those that appear in the last tweet T4. We depict re-

sults of recall after respectively 10, 20 then 30 hash-

tags recommended. Results are displayed in table 1.

Table 1: Recall results.

Recall R10 R20 R30

Our system 0.8157 0.8504 0.9175

Naive Bayes 0.6312 0.6987 0.7278

Cosine with KNN 0.7237 0.7687 0.8097

As shown in Table 1, for our recommender sys-

tem, the recall for 10 recommended hashtags is

0.8157 , 0.8504 for 20 hashtags, and 0.9175 for 30

recommended hashtags. For the Naive Bayes system,

the recall values are 0.6312, 0.6987, and 0.7278, re-

spectively. Regarding the cosine similarity measure

with KNN, the recall values are 0.7237, 0.7687, and

0.8097. Hence, our system produced recall values su-

perior to those of the two systems we benchmarked

against.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

246

We calculated the improvement rate in terms of

recall for our system by comparing it with the two

methods of (Dovgopol and Nohelty, 2015). Our sys-

tem shows a signiﬁcant improvement rate of 15.7%

compared to other comparative approaches in terms

of recall.

The results obtained afﬁrm the efﬁcacy of the

model-driven approach in retrieving a greater num-

ber of relevant hashtags compared to other meth-

ods. In contrast, the Naive Bayes and KNN ap-

proaches rely predominantly on a proﬁle composed

solely of user-shared hashtags. As a result, depend-

ing solely on hashtag similarity may not consistently

yield accurate recommendations, highlighting a lim-

itation of these two approaches. In contrast, our

proposed approach capitalizes on all dimensions of-

fered by our model, encompassing social interactions,

shared hashtags, and similarity with diverse users.

Additionally, it takes into account the inherent char-

acteristics of the social network itself, including its

multi-layered structure and dynamism. Indeed, by in-

corporating information about the user’s social net-

work connections, we can create a more comprehen-

sive representation of the user. This multilayer dy-

namic social network perspective allows us to cap-

ture the inﬂuence of social ties, community dynam-

ics, and evolving relationships on user preferences

and decision-making. Thus, taking into account the

user’s environment in this way enables us to leverage

the power of social inﬂuence and network effects for

more accurate and effective recommendations.

5 CONCLUSION

This paper has proposed a multidimensional user

proﬁle model designed to enhance system recom-

mendations within intricate social networks. The

model comprehensively captures diverse dimensions

of a user’s presence in their complex social network,

including demographic, social, behavioral, and ho-

mophilic aspects. The overarching objective is to pro-

vide more comprehensive recommendations tailored

to distinct contexts. As a practical application, we

have implemented this model in the realm of hashtag

recommendations within Twitter platform, demon-

strating its utility within a real social network rec-

ommender system. Employing a recommendation ap-

proach, we sought pertinent recommendations, show-

casing that the proposed modeling approach enables

the acquisition of relevant suggestions for each user.

In future works, we aim to test the performance

and the scalability of the proposed modeling approach

for recommender systems in other contexts using real

large-scale multilayer social networks or benchmarks.

REFERENCES

Aiello, L. M., Barrat, A., Schifanella, R., Cattuto, C.,

Markines, B., and Menczer, F. (2012). Friendship pre-

diction and homophily in social media. ACM Trans-

actions on the Web (TWEB), 6(2):1–33.

Al-Shamri, M. Y. H. (2016). User proﬁling approaches

for demographic recommender systems. Knowledge-

Based Systems, 100:175–187.

Alhijawi, B. and Kilani, Y. (2020). The recommender

system: a survey. Int. J. Adv. Intell. Paradigms.,

15(3):229–251.

Beel, J., Langer, S., N

urnberger, A., and Genzmehr, M.

(2013). The impact of demographics (age and gender)

and other user-characteristics on evaluating recom-

mender systems. In Research and Advanced Technol-

ogy for Digital Libraries: International Conference

on Theory and Practice of Digital Libraries, TPDL

2013, Valletta, Malta, September 22-26, 2013. Pro-

ceedings 3, pages 396–400. Springer.

Blei, D. M., Ng, A. Y., and Jordan, M. I. (2003). Latent

dirichlet allocation. Journal of machine Learning re-

search, 3:993–1022.

Caro-Mart

ınez, M., Jim

enez-D

ıaz, G., and Recio-Garcia,

J. A. (2023). A graph-based approach for minimis-

ing the knowledge requirement of explainable recom-

mender systems. Knowledge and Information Sys-

tems, pages 1–31.

Casillo, M., Colace, F., Conte, D., Lombardi, M., San-

taniello, D., and Valentino, C. (2023). Context-aware

recommender systems and cultural heritage: a survey.

Journal of Ambient Intelligence and Humanized Com-

puting., 14(4):3109–3127.

Ceria, A., Havlin, S., Hanjalic, A., and Wang, H. (2022).

Topological–temporal properties of evolving net-

works. Journal of Complex Networks., 10(5):cnac041.

Danisch, M., Dugu

e, N., and Perez, A. (2014). On the im-

portance of considering social capitalism when mea-

suring inﬂuence on twitter. Paper presented at the 20

International Conference on Behavioral, Economic,

and Socio - Cultural Computing, 1-7 October 2014.

Daoud, M., Lechani, L.-T., and Boughanem, M. (2009).

Towards a graph-based user proﬁle modeling for a

session-based personalized search. Knowledge and

Information Systems, 21:365–398.

Degemmis, M., Lops, P., and Semeraro, G. (2007).

A content-collaborative recommender that exploits

wordnet-based user proﬁles for neighborhood forma-

tion. User Modeling and User-Adapted Interaction.,

17(3):217–255.

Deldjoo, Y., Anelli, V. W.and Zamani, H., Bellogin, A.,

and Di Noia, T. (2021). A ﬂexible framework for

evaluating user and item fairness in recommender sys-

tems. User Modeling and User-Adapted Interaction,,

31(3):457–511.

Dey, A. K. (2001). Understanding and using context. Per-

sonal and ubiquitous computing, 5:4–7.

Multidimensional User Proﬁle Model to Support System Recommendations in Complex Social Networks: Application to Hashtag

Recommendations

247

Dhelim, S., Aung, N., Bouras, M. A., Ning, H., and Cam-

bria, E. (2022). A survey on personality-aware rec-

ommendation systems. Artiﬁcial Intelligence Review,

pages 1–46.

Dovgopol, R. and Nohelty, M. (2015). Twitter hash tag rec-

ommendation.

Graus, M. and Ferwerda, B. (2019). Theory-grounded user

modeling for personalized hci. Personalized human-

computer interaction.

Jerry, C., Shankar, K., and Kiranmai, E. S. (2024). Harm

mitigation in recommender systems under user pref-

erence dynamics. In 2024 30th ACM SIGKDD Con-

ference on Knowledge Discovery and Data Mining,

pages 255–265.

Kazienko, P. and Kolodziejski, P. (2006). Personalized inte-

gration of recommendation methods for e-commerce.

Int. J. Comput. Sci. Appl., 3(3):12–26.

Ko, H., Lee, S., Park, Y., and Choi, A. (2022). A survey

of recommendation systems: recommendation mod-

els, techniques, and application ﬁelds. Electronics.,

11(1):141.

Li, W., Li, T., and Berahmand, K. (2023). An effective link

prediction method in multiplex social networks us-

ing local random walk towards dependable pathways.

Journal of Combinatorial Optimization., 45(1):31.

Li, Y., Liu, J., and Ren, J. (2019). Social recommendation

model based on user interaction in complex social net-

works. PloS one, 14(7):e0218957.

Liu, H., Salem, B., and Rauterberg, M. (2009). A survey

on user proﬁle modeling for personalized service de-

livery systems. In Proceeding of IADIS International

Conference on Interfaces and Human Computer Inter-

action, pages 45–51.

Martijn, M.and Conati, C. and Verbert, K. (2022). Know-

ing me, knowing you”: personalized explanations for

a music recommender system. User Modeling and

User-Adapted Interaction., 32(1):215–252.

Middleton, S. E.and Shadbolt, N. R. and De Roure, D. C.

(2004). Ontological user proﬁling in recommender

systems. ACM Transactions on Information Systems

(TOIS),, 22(1):54–88.

Minsung, H., Sojung, A., Rajendra, A., D, C., and J, J. J.

(2024). Cross-cultural contextualisation for recom-

mender systems. Journal of Ambient Intelligence and

Humanized Computing, pages 1–12.

Qian, F., Qin, K., Chen, H., Chen, J., Zhao, S., Zhou, P., and

Zhang, Y. (2023). Utilizing the inﬂuence of multiple

potential factors for social recommendation. Knowl-

edge and Information Systems, pages 1–20.

Rattanajitbanjong, N. and Maneeroj, S. (2009). Multi cri-

teria pseudo rating and multidimensional user proﬁle

for movie recommender system. Paper presented at

the 32nd IEEE International Conference on Computer

Science and Information Technology, 596-601 August

2009.

Rebhi, W., Ben Yahia, N., and Bellamine, N. (2022). Life-

long and multirelational community detection to sup-

port social and collaborative e-learning. Computer

applications in engineering education, 30(5):1321–

1337.

Rebhi, W., Ben Yahia, N., and Bellamine Ben Saoud, N.

(2021). Stable communities detection method for tem-

poral multiplex graphs: Heterogeneous social network

case study. The Computer Journal, 64(3):418–431.

Rebhi, W., Yahia, N. B., and Saoud, N. B. B. (2017a).

Hybrid modeling approach for contextualized com-

munity detection in multilayer social network: Emer-

gency management case study. Procedia computer

science., 112:673–682.

Rebhi, W., Yahia, N. B., Saoud, N. B. B., and Hanachi, C.

(2017b). Towards contextualizing community detec-

tion in dynamic social networks. In Modeling and Us-

ing Context: 10th International and Interdisciplinary

Conference, CONTEXT 2017, Paris, France, June 20-

23, 2017, Proceedings 10, pages 324–336. Springer.

Roozbahani, Z., Rezaeenour, J., Katanforoush, A., and

Jalaly Bidgoly, A. (2022). Personalization of the col-

laborator recommendation system in multi-layer sci-

entiﬁc social networks: A case study of researchgate.

Expert Systems., 39(5):e12932.

Safoury, L. and Salah, A. (2013). Exploiting user demo-

graphic attributes for solving cold-start problem in

recommender system. Lecture Notes on Software En-

gineering., 1(3):303–307.

Sharma, M., Pant, B., and Singh, V. (2021). Demographic

proﬁle building for cold start in recommender sys-

tem: a social media fusion approach. Materials To-

day: Proceedings., 46:11208–11212.

SKazienko, P., Musial, K., and Kajdanowicz, T. (2011).

Multidimensional social network in the social rec-

ommender system. IEEE Transactions on Systems,

Man, and Cybernetics-Part A: Systems and Humans.,

41(4):746–759.

Tahmasebi, F., Meghdadi, M., Ahmadian, S., and Valial-

lahi, K. (2021). A hybrid recommendation system

based on proﬁle expansion technique to alleviate cold

start problem. Multimedia Tools and Applications.,

80(2):2339–2354.

Zainol, Z. and Nakata, K. (2010). Generic context ontology

modelling: A review and framework. In 2010 2nd In-

ternational Conference on Computer Technology and

Development, pages 126–130. IEEE.

Zhang, Z., Patra, B. G., Yaseen, A., Zhu, J., Sabharwal, R.,

Roberts, K., Cao, T., and Wu, H. (2023). Scholarly

recommendation systems: a literature survey. Knowl-

edge and Information Systems, pages 1–46.

Zimmermann, A., Lorenz, A., and Oppermann, R. (2007).

An operational deﬁnition of context. In Modeling

and Using Context: 6th International and Interdisci-

plinary Conference, CONTEXT 2007, Roskilde, Den-

mark, August 20-24, 2007. Proceedings 6, pages 558–

571. Springer.

ICAART 2025 - 17th International Conference on Agents and Artiﬁcial Intelligence

248