Conference PaperPDF Available

SemCoTrip: A Variety-Seeking Model for Recommending Travel Activities in a Composite Trip

  • LARODEC Research Laboratory
  • Prince Mohammad Bin Fahd University, Khobar, Saudi Arabia

Abstract and Figures

Selecting appropriate activities, especially in multi-destinations trips, is a hard task that many travellers face each time they want to plan for a trip.With the budget and time limitations, travellers will try to select activities that best fit their personal interests. Most of existing travel recommender systems don’t focus on activities that a traveller might be interested in. In this paper, we go beyond the specific problem of combining regions in a composite trip to propose a varietyseeking model which is capable of providing travelllers with recommendations on what activities they can engage in when visiting different regions. A semantical hierarchical clustering-based model is proposed to guarantee diversity within the set of recommended activities. Experimental results on a real dataset have shown that the proposed approach helps the traveller to avoid doing the same or similar activities in a composite trip, thus, promoting less popular activities to be selected.
Content may be subject to copyright.
SemCoTrip: A variety-seeking model for recommending
travel activities in a composite trip
Montassar Ben Messaoud1, Ilyes Jenhani2, Eya Garci3, and Toon De Pessemier4
1LARODEC, Institut Sup´
erieur de Gestion de Tunis
41, Avenue de la libert´
e, 2000 Le Bardo, Tunisia.
2Prince Mohammad bin Fahd University, Al Khobar, KSA
3Institut Sup´
erieur de Gestion de Sousse
Rue Abdelaziz Il Behi, 4000 Sousse, Tunisia
4iMinds - Ghent University, G. Crommenlaan 8/201, 9050 Ghent, Belgium
Abstract. Selecting appropriate activities, especially in multi-destinations trips,
is a hard task that many travellers face each time they want to plan for a trip. With
the budget and time limitations, travellers will try to select activities that best fit
their personal interests. Most of existing travel recommender systems don’t focus
on activities that a traveller might be interested in. In this paper, we go beyond the
specific problem of combining regions in a composite trip to propose a variety-
seeking model which is capable of providing travelllers with recommendations
on what activities they can engage in when visiting different regions. A semanti-
cal hierarchical clustering-based model is proposed to guarantee diversity within
the set of recommended activities. Experimental results on a real dataset have
shown that the proposed approach helps the traveller to avoid doing the same or
similar activities in a composite trip, thus, promoting less popular activities to be
Keywords: Multi-destination trips, leisure activities, diversity, hierarchical clus-
tering, ontology
Recommendation systems have made a significant difference in people’s lives. Being
one of the early adoption areas, the tourism industry has taken advantage of the recent
advances in recommender systems (RS) to enhance the quality of services offered to
travellers and to enrich their travel experiences [12]. One of the potential applications
of RS in tourism that has not yet been explored in details is the recommendation of
composite trips. Most of the existing approaches dealing with multi-destination trips
focus on developing ways to combine single travel items like regions and routes in
order to maximize the benefit for the traveller. Nevertheless, none of these works has
directly tackled the issue of managing activities during the stay at each destination.
This paper provides a substantially extended version of a previous work [8], in
which authors proposed an efficient algorithm for the recommendation of composite
trips. Further attempts in this direction will be initiated in order to design complemen-
tary strategies to utilize semantic prior knowledge, to improve the diversity of the rec-
ommended activities and, most importantly, to enrich visitors’ travel experience. The
remainder of this paper is organized as follows: Section 2 discusses some related works
where the diversity level is considered as a trip constraint. Section 3 gives the necessary
background for both hierarchical clustering and ontologies. We then describe in details
the SemCoTrip strategy in Section 4. In Section 5, we report experimental results that
show the effectiveness of the proposed algorithm. Concluding remarks and future works
will be given in Section 6.
In recent years, there has been a continuous line of research focusing on diversifying
the recommended lists of activities and destinations to meet tourists’ satisfaction. Diver-
sity is commonly defined as the average pairwise distance between recommendations
to users [4]. Authors in [14] proposed to use the K-means algorithm to assign users
to clusters that have similar characteristics. When executing their clustering procedure,
activities were weighted to ensure that their SigTur recommender system provides di-
verse recommendations. In [19], the diversity level was considered as a trip constraint.
To ensure diversity, the authors used a measure of balance between the attractions’ cat-
egories and the acceptable rating threshold. The work in [17] presents a search result
clustering algorithm based on semantic data representation which chooses a set of ob-
jects from each cluster to increase the diversity of the proposal made to the visitor of a
museum. [18] came up with a surprising result stating that the global error of k-Nearest
Neighbours-based recommender systems decreases when a higher diversity is associ-
ated to the recommendations.
In this section, we provide basic concepts related to semantic knowledge-based systems
and hierarchical clustering which are essential in understanding the rest of the paper.
3.1 Knowledge-based systems
Knowledge-based systems (KBs) provide domain reasoning frameworks combined with
inference engines that usually reason over logical languages. Ontology, which is one of
those popular semantic driven knowledge based systems, has received numerous defi-
nitions in the literature. The most commonly cited definition was given in [6]. It defines
the ontology as an explicit specification of a conceptualization. The ”conceptualiza-
tion”, refers to a simplified view of the world by identifying its relevant concepts.The
word ”explicit” means that all concepts (resp. their specific properties and constraints)
must be explicitly defined.
Definition 1. An ontology can be formally expressed as:
A set of concepts C={C1, ..., Cn}, which are mainly interrelated by means of taxo-
nomic (is-a) relations in the form of a hierarchy H,
A set of properties for each concept,
Semantic (i.e. non-taxonomic) relations between concepts,
A set of instances I(i.e. occurrences of concepts and semantic relations), and
A set of assertions and formal axioms (i.e. constraint-relationships like should,
should not, must, must not, etc).
3.2 Hierarchical clustering
Clustering is a typical unsupervised learning task which aims at grouping together sim-
ilar objects (with respect to their attribute values) into subsets called clusters. A cluster
is therefore a collection of objects which are similar to each others and dissimilar to
objects belonging to other clusters.
We can distinguish four main categories of clustering methods: (1) Centroid-based
clustering such as K-means [13], (2) Hierarchical clustering [9] such as single-linkage
and complete-linkage clustering methods, (3) Distribution-based clustering such as
Expectation-Maximization (EM) algorithm [5] and (4) Density-based clustering such
as DBSCAN algorithm [11].
Among these categories, we are interested in the hierarchical clustering one which
could be either agglomerative or divisive. Agglomerative methods are ”bottom up” ap-
proaches which start by assigning each element to a separate cluster then a merging of
the two least distant (most similar) clusters is successively performed leading to larger
clusters. However, Divisive methods are ”top down” approaches in which all objects
start in one cluster, and splits are performed recursively as one moves down. In prac-
tice, agglomerative techniques were more commonly used.
Distance (or similarity) between two clusters is determined by a linkage criterion,
which is a function of the pairwise distances between instances one from each cluster.
Most popular linkage criteria are: (1) Single-linkage: the distance between two clusters
is the minimum pairwise distance between elements, one from each cluster (i.e. the
shortest link between clusters). (2) Complete-linkage: the distance between two clusters
is the maximum pairwise distance between elements, one from each cluster (i.e. the
longest link between clusters). (3) Average-linkage: the distance between two clusters
is the average pairwise distance between elements, one from each cluster. Other linkage
criteria exist such as the Average group linkage (the sum of all intra-cluster variance),
Ward’s linkage (the increase in variance for the cluster being merged), V-linkage (the
probability that candidate clusters spawn from the same distribution function). A good
survey on hierarchical clustering algorithms could be found in [15].
The agglomerative clustering continues until a stopping criterion is met. We can
apply a distance-based stopping criterion to stop clustering when the clusters are too
far apart to be merged (i.e., distance between the closest clusters to be merged is greater
to a user-predefined or computed threshold). A number of clusters-based criterion can
also be used to stop clustering when there is a sufficiently predefined small number of
Fig. 1. SemCoTrip: Extending the composite trips’ RS of [8] to consider a variety of leisure
Hierarchical clustering has been mainly used in conjunction with recommender sys-
tems to deal with the problem of scalability. In fact, incremental hierarchical agglom-
erative clustering has been used in [7] to handle the large number of user profiles in
e-commerce recommender systems. Moreover, in order to better personalize naviga-
tional recommendations in social tagging systems, authors in [20] applied hierarchical
clustering to cluster the wide variety of tags. In [22], an ensemble hierarchical cluster-
ing approach has been applied to group users with similar reading profiles and get news
hierarchies which are then used in recommending news articles. More recently, hierar-
chical clustering has been used in [21] to enhance the relevance of papers to recommend
for researchers among a huge number of published papers.
The general overview of the SemCoTrip (Semantical Composite Trip) algorithm is de-
picted in Figure 1. SemCoTrip inputs are: a travel region dataset and a tourism-activity
ontology. We follow approximately the same methodology proposed by authors in [8] to
reproduce the same performance when combining regions and determining the optimal
duration of stay per region. A blue shaded area was added in Figure 1 to highlight the
differences between the two algorithms and to visualize the recommendation process
proposed by [8].
Fig. 2. (a) 2-D rating matrix as proposed in [8] and (b) SemCoTrip multidimensional model for
the User×Region×Activity recommendation space.
4.1 Search-space reduction
First, we start by reducing the search space by excluding irrelevant regions to the user
query. Using the region tree hierarchy, if a region is removed, all its sub-regions and
related activities will be removed as well.
4.2 Rating
The remaining travel activities of the pruned region tree will be then rated. At this level,
[8] used a 5-point Likert scale to rate regions’ features depending on the month (season),
which could potentially exclude many relevant destinations from the recommendations
returned to users. Alternatively, we will simply assume that activities offered in each
region are subject to change from season to season. Concepts’ attributes in the input
ontology will indicate how well the tourism activities match each traveling type group.
By doing so, travel region ratings in our scenario will thus depend on their correspond-
ing activities ratings.
Here, the standard rating schema of [8] that involves user and region dimensions is
extended to three-dimensional schema involving activities (Refer to figure 2). Such mul-
tidimensional approach is usually used to deal with context in RSs [1]. For this case, we
will further define a rating function R on the recommendation space User×Region×Activity
specifying how much user u User liked activity a Activity in (sub-)region s Re-
gion, R(u,a,s).
At the end of this step, regions with low ratings will be removed and the remain-
ing ones will be combined in a way to maximize their values for the user while still
respecting the budget and the duration constraints.
4.3 Recommendation strategy
The problem, as defined at that level, can still be considered as a variant of the knapsack
problem [3] which can be efficiently solved by means of dynamic programming [10].
Two objectives are considered here: (1) The value of the composite trip is proportional
to the distance between regions and (2) the best combination of regions is obtained
A1 A2 A3 A4 A5 A6 A7 A8 A9
A12 A13
A1 A2 A3 A4 A5
A7 A8
A9 A10 A11
Sub-Region1 Sub-Region2 Sub-Region3
Intra - regions
A1 A2 A6 A12 A13
A6 A7
A10 A11 A12 A13
A6 A7 A8
Inter - regions
Final composite trip
Fig. 3. Two-levels hierarchical clustering to optimize the diversity of the activities lists.
based on the optimal duration of the stay per region. The application of the Dynamic
programming approach to our dataset provided a candidate solution which consists of
a subset of regions along with the duration time to spend in each of these regions.
The complementary component that we propose in our approach is to select the set
of activities to recommend based on the recommended regions. An important criterion
that we introduce in selecting the activities is diversity: we want our system to recom-
mend activities which are as dissimilar as possible. To ensure that diversity, we will first
use a semantic hierarchical clustering approach which will try to select heterogeneous
clusters of activities. Then, a selection algorithm will be applied to find the optimal
combination of clusters of activities found in the previous step. Throughout the hierar-
chical clustering step, we will consider Rada’s distance [16] as the specific distance for
calculating the semantic gain intra-(resp. inter) clusters.
Definition 2. Let Ciand Cjbe two concepts in an ontology restricted to taxonomic
hierarchy. A measure of the conceptual Rada’s distance is expressed as the minimum
number of links separating the two concepts.
We choose this distance because of its simplicity and its broad adoption. Note that
Rada’s distance can be replaced by any other semantic distance (refer to [2] for a com-
parative analysis between semantic distances).
As a first step of our approach, the clustering is performed on the total set of ac-
tivities of each (sub-) region separately. As shown in Figure 3, for each recommended
(sub-)region, the hierarchical clustering will result in a set of one or more clusters.
Activities within each cluster are selected based on the maximization of the semantic
distance between the activities.
The second step of the approach is based on an inter-regions selection algorithm
that be used to find out the optimal combination of clusters of activities (no more than
one cluster will be selected from each (sub-)region). Clusters that maximize an average
semantic distance between clusters of all remaining (sub-)regions will be selected and
merged to form the final group of activities.
5.1 Data Description
Dataset Our variety-seeking model was tested on an extended version of the dataset
used in [8]. The dataset (a region tree-like structure) is composed of a total of 152
regions with 124 leaves. The main difference with the original dataset is that we have
assigned a range of seasonal activities to each (sub-)region. These activities are then
mapped to their corresponding concepts in the used ontology and a 5-point Likert scale
was used to indicate how well the proposed activity matches travellers types that we
have categorized in four families according to the Canadian Tourism Commission 5
(See table 1). All other input data (e.g. duration, budget, routing, crime level, etc) is
kept the same.
Table 1. Traveler’s types classification.
Category Traveler’s types
Learners Cultural explorer
Authentic Experiencers
Cultural History Buffs
Personal History Explorers
Indulgers Free Spirits
Familiarity Seekers Gentle Explorers
No Hassle Travellers
Virtual Travellers
Escapists Rejuvenators
Ontology The approach that we are proposing builds up on the use of a tourism ac-
tivities ontology, which is presented as a hierarchy composed of a set of more than
200 concepts taxonomically related by subsumptions. The concepts are formalized into
three related (sub-)ontologies, referred to as sport attractions, natural attractions and
cultural attractions. Figure 4 shows a small excerpt of our tourism ontology.
5.2 Experimental design
A Java-based prototype was implemented in order to evaluate our approach. We used
the prototype to handle a sample of 100 users queries. For each query, we changed
input parameters (e.g. traveller’s type, budget, total duration, etc.). The prototype exe-
cutes each query separately and provides the top-rated recommendations; based on the
recommendation procedures described in Section IV.
Fig. 4. Excerpt of the tourism activities ontology.
5.3 Results &Interpretations
The main objective of our approach is to ensure diversity when recommending a set of
activities. The diversity degree of a set A={A1, ..., An}of nrecommended activities
is measured by:
Diversity =Pn
i=1 Pn
j=i+1 Rada(Ai, Aj)
where Rada(Ai, Aj)is the normalized Rada’s distance between two activities Aiand
Ajwhich lies on the unit interval.
Figure 5 shows, for the top-10 recommendations, the diversity level within the set
of activities recommended by SemCoTrip as well as three other concurrent algorithms.
The grey line refers to a variant of SemCoTrip which is only using the Intra-regions
activities clustering algorithm. The red (resp. blue) curve shows the diversity relative to
the baseline method that recommends random (resp. most popular) activities. We stress
the fact that all four algorithms are based on the same basic regions recommendation
idea proposed in [8]. Only activities’ recommendation strategies have been changed.
For all four algorithms, a list of top-k recommendations is kept and sorted in de-
scending order of the diversity within the activities. As we can observe from the figure
5, the popularity-based strategy gave the worst results, as it drastically reduces the ac-
tivities’ search space for each sub-region.
Surprisingly, the random approach produced competitive results when compared
to the variant ”Intra”. This could be explained by the fact that many sub-regions offer
seasonal activities for several periods in the year. Those activities that depend on the
weather are, usually, susceptible to be joined together on the corresponding ontology.
This is the particular case where the random strategy provides nearly identical results
to those provided by SemCoTrip.
The difference in performance between SemCoTrip and its variant ”Intra” justifies
that the whole clustering process is required to illustrate diversity in recommendations.
Fig. 5. Diversity for the top-K recommendations.
In this paper, we have proposed the SemCoTrip recommender system which seeks to
recommend a set of diverse activities for a composite trip. A semantic-based hierarchi-
cal clustering approach has been used along with a tourism ontology to ensure diversity.
Experimental results on a real dataset have shown that activities recommended by Sem-
CoTrip are better, in terms of diversity, than activities recommended by a variant of
SemCoTrip and two baseline approaches.
For future works, we intend to manage the sequence in which the activities need
to occur and propose a complementary approach to improve novelty and serendipity
whilst maintaining high accuracy of recommendations.
1. G. Adomavicius, R. Sankaranarayanan, S. Sen, and A. Tuzhilin. Incorporating contextual
information in recommender systems using a multidimensional approach. ACM Transactions
on Information Systems, 23(1):103–145, 2005.
2. E. Blanchard, M. Harzallah, H. Briand, and P. Kuntz. A typology of ontology-based semantic
measures. In M. Missikoff and A. D. Nicola, editors, EMOI-INTEROP, volume 160 of CEUR
Workshop Proceedings., 2005.
3. J. J. Burg, J. Ainsworth, B. Casto, and S.-D. Lang. Experiments with the oregon trail knap-
sack problem. Electronic Notes in Discrete Mathematics, 1(B):26–35, 1999.
4. P. Castells, N. J. Hurley, and S. Vargas. Novelty and diversity in recommender systems.
In F. Ricci, L. Rokach, and B. Shapira, editors, Recommender Systems Handbook, pages
881–918. Springer, 2015.
5. A. Dempster, N. Laird, and D. Rubin. Maximum likelihood from incomplete data via the em
algorithm. Journal of the Royal Statistcal Society, 39(B):1–38, 1977.
6. T. R. Gruber. Towards Principles for the Design of Ontologies Used for Knowledge Sharing.
In N. Guarino and R. Poli, editors, Formal Ontology in Conceptual Analysis and Knowledge
Representation, Deventer, The Netherlands, 1993.
7. C. Haruechaiyasak, C. Tipnoe, S. Kongyoung, C. Damrongrat, and N. Angkawattanawit. A
dynamic framework for maintaining customer profiles in e-commerce recommender systems.
In EEE, pages 768–771. IEEE Computer Society, 2005.
8. D. Herzog and W. W¨
orndl. A travel recommender system for combining multiple travel
regions to a composite trip. In T. Bogers, M. Koolen, and I. Cantador, editors, CBRec-
Sys@RecSys, volume 1245 of CEUR Workshop Proceedings, pages 42–48.,
9. A. K. Jain and R. Dubes. Algorithms for Clustering Data. Prentice Hall, 1988.
10. H. Kellerer, U. Pferschy, and D. Pisinger. Knapsack Problems. Springer, Berlin, Germany,
11. H.-P. Kriegel, P. Kr¨
oger, J. Sander, and A. Zimek. Density-based clustering. Wiley Interdis-
ciplinary Reviews: Data Mining and Knowledge Discovery, 1(3):231–240, 2011.
12. C. Lu, P. Laublet, and M. Stankovic. Travel attractions recommendation with knowledge
graphs. In E. Blomqvist, P. Ciancarini, F. Poggi, and F. Vitali, editors, EKAW, volume 10024
of Lecture Notes in Computer Science, pages 416–431, 2016.
13. J. MacQueen. Some methods for classification and analysis of multivariate observations. In
Berkeley Symposium on Mathematical Statistics and Probability, 1966.
14. A. Moreno, A. Valls, D. Isern, L. Marin, and J. Borr`
as. Sigtur/e-destination: Ontology-based
personalized recommendation of tourism and leisure activities. Eng. Appl. of AI, 26(1):633–
651, 2013.
15. F. Murtagh and P. Contreras. Algorithms for hierarchical clustering: an overview. Wiley
Interdisc. Rew.: Data Mining and Knowledge Discovery, 2(1):86–97, 2012.
16. R. Rada, H. Mili, E. Bicknell, and M. Blettner. Development and application of a metric on
semantic nets. In IEEE Transactions on Systems, Man and Cybernetics, pages 17–30, 1989.
17. T. Ruotsalo, K. Haav, A. Stoyanov, S. Roche, E. Fani, R. Deliai, E. M¨
a, T. Kauppinen,
and E. Hyv¨
onen. Smartmuseum: A mobile recommender system for the web of data. J. Web
Sem., 20:50–67, 2013.
18. F. Sanchez-Vilas, J. Ismailov, and E. Sanchez. The importance of diversity in profile-based
recommendations: A case study in tourism. In TouRS, Workshop on Tourism Recommender
Systems, 9th ACM Conf. on Recommender Systems, pages 43–50. Vienna, Austria, 2015.
19. A. Savir, R. I. Brafman, and G. Shani. Recommending improved configurations for com-
plex objects with an application in travel planning. In Q. Yang, I. King, Q. Li, P. Pu, and
G. Karypis, editors, RecSys, pages 391–394. ACM, 2013.
20. A. Shepitsen, J. Gemmell, B. Mobasher, and R. Burke. Personalized recommendation in
social tagging systems using hierarchical clustering. In P. Pu, D. G. Bridge, B. Mobasher,
and F. Ricci, editors, RecSys, pages 259–266. ACM, 2008.
21. J. D. West, I. Wesley-Smith, and C. T. Bergstrom. A recommendation system based on
hierarchical clustering of an article-level citation network. IEEE Trans. Big Data, 2(2):113–
123, 2016.
22. L. Zheng, L. Li, W. Hong, and T. Li. Penetrate: Personalized news recommendation using
ensemble hierarchical clustering. Expert Syst. Appl., 40(6):2127–2136, 2013.

Supplementary resource (1)

... Messaoud et al. [16] proposed a variety-seeking model using semantic hierarchical clustering to establish diversity in a set of recommended activities. Similarly, Savir et al. [21] introduced an additional diversity constraints based on attraction types to ensure that the trip's diversity level is above a threshold. ...
Conference Paper
Given vast number of possible global travel destinations, choosing a destination has become challenging. We argue that traditional media are insufficient to make informed travel decisions, due to a lack of objectivity, a lack of comparability and because information becomes out of date quickly. Thus, travel planning is an interesting field for data-driven recommender systems that support users to master information explosion. We present unresolved research questions with working packages for a doctoral project that combines the fields of recommender systems and user modeling with data mining. The core contributions will be a framework that integrates heterogeneous data sources from the travel domain, novel user modeling techniques and constraint-based recommender algorithms to master the complexities of global travel planning.
... Messaoud et al. [15] extend [10] by focusing on the diversity of activities [6] within a composite trip. Unlike Savir et al. [17], who use a simple mechanism to uphold a certain level for diver- sity of attractions, they use hierarchical clustering to improve the heterogeneity of activities. ...
Conference Paper
Recommendations in complex scenarios require additional knowledge of the domain. Planning a composite travel spanning several countries is a challenging, but encouraging domain for rec-ommender systems, since users are in dire need for assistance: Information in typical publications, such as printed travel guides or personal blogs is often imprecise, biased or outdated. In this paper we motivate a data-mining approach to improve destination recommender systems with learned travel patterns. Specifically, we propose a methodology to mine trips from location-based social networks to improve recommendations for the duration of stay at a destination. For this we propose a model for combining data from different sources and identify several metrics that are useful to ensure sufficient data quality, i.e., whether a traveler's check-in behavior is adequate to derive patterns from it. We demonstrate the utility of our approach using a Foursquare data set from which we extract 23,418 trips in 77 countries. Analyzing these trips, we determine the travel durations per country, how many countries are typically visited in a given time span and which countries are often visited together in a composite trip. Also, we discuss how this method can be generalized to other recommender systems domains.
Full-text available
This examination targets investigating the impact of HR practices on turnover intention with regards to Bangladesh. An aggregate of 60 reactions from 10 Bangladeshi ICT organization were gathered and dissected dispassionately. It was discovered that HR practices have huge relationship with turnover intention (TI). Furthermore, training, development, job satisfaction and compensation were found to have positive effect on turnover intention (TI). Academicians, specialists, strategy producers, professionals, understudies, neighbourhood and remote business people of Bangladesh and other comparative nations could profit by this paper by investigating the relationship between HR practices and turnover intention. Keywords: Human resource management, Human resource management practices, turnover intention, ICT firms, Bangladeshi private industry I. INTRODUCTION The impact of human resource management (HRM) practices generally known as HR practices on organizational execution and employee turnover intention has been a central domain of research in the improved world for an extensive period of time (Boselie, P., Van Harten, J. what's more, Veld, M., 2019). Bangladesh is a front-positioning nation right now free progression of the information technology. A noteworthy upheaval has occurred in the improvement of the country's information technology in the course of the most recent seven years due to our administration's timely initiatives. The information and communication technology sector have seen a monstrous infrastructural improvement inferable from our earnest endeavour's (Parvez, 2018).
Full-text available
Semantic and context knowledge has been envisioned as an appropriate solution for addressing the content heterogeneity and information overload in mobile Web information access, but few have explored their full potential in mobile scenarios, where information objects refer to their physical counterparts, and retrieval is context-aware and personalized for users. We present SMARTMUSEUM, a mobile ubiquitous recommender system for the Web of Data, and its application to information needs of tourists in context-aware on-site access to cultural heritage. The SMARTMUSEUM system utilizes Semantic Web languages as the form of data representation. Ontologies are used to bridge the semantic gap between heterogeneous content descriptions, sensor inputs, and user profiles. The system makes use of an information retrieval framework wherein context data and search result clustering are used in recommendation of suitable content for mobile users. Results from laboratory experiments demonstrate that ontology-based reasoning, query expansion, search result clustering, and context knowledge lead to significant improvement in recommendation performance. The results from field trials show that the usability of the system meets users’ expectations in real-world use. The results indicate that semantic content representation and retrieval can significantly improve the performance of mobile recommender systems in knowledge-rich domains.
Conference Paper
Selecting relevant travel attractions for a given user is a real and important problem from both a traveller’s and a travel supplier’s perspectives. Knowledge graphs have been used to conduct recommendations of music artists, movies and books. In this paper, we identify how knowledge graphs might be efficiently leveraged to recommend travel attractions. We improve two main drawbacks in existing systems where semantic information is exploited: semantic poorness and city-agnostic user profiling strategy. Accordingly, we constructed a rich world scale travel knowledge graph from existing large knowledge graphs namely Geonames, DBpedia and Wikidata. The underlying ontology contains more than 1200 classes to describe attractions. We applied a city-dependent user profiling strategy that makes use of the fine semantics encoded in the constructed graph. Our evaluation on YFCC100M dataset showed that our approach achieves a 5.3 % improvement in terms of F1-score, a 4.3 % improvement in terms of nDCG compared with the state-of-the-art approach.
The scholarly literature is expanding at a rate that necessitates intelligent algorithms for search and navigation.For the most part, the problem of delivering scholarly articles has been solved. If one knows the title of an article, locating it requires little effort and, paywalls permitting, acquiring a digital copy has become trivial. However, the navigational aspect of scientific search - finding relevant, influential articles that one does not know exist - is in its early development. In this paper, we introduce EigenfactorRecommends - a citation-based method for improving scholarly navigation. The algorithm uses the hierarchical structure of scientific knowledge, making possible multiple scales of relevance for different users. We implement the method and generate more than 300 million recommendations from more than 35 million articles from various bibliographic databases including the AMiner dataset. We find little overlap with co-citation, another well-known citation recommender, which indicates potential complementarity. In an online A-B comparison using SSRN, we find that our approach performs as well as co-citation, but this new approach offers much larger recommendation coverage. We make the code and recommendations freely available at babel.eigenfactor.organd provide an API for others to use for implementing and comparing the recommendations on their own platforms.
Novelty and diversity have been identified, along with accuracy, as foremost properties of useful recommendations. Considerable progress has been made in the field in terms of the definition of methods to enhance such properties, as well as methodologies and metrics to assess how well such methods work. In this chapter we give an overview of the main contributions to this area in the field of recommender systems, and seek to relate them together in a unified view, analyzing the common elements underneath the different forms under which novelty and diversity have been addressed, and identifying connections to closely related work on diversity in other fields.
Internet-based services are available to recommend destinations and activities for organized trips. Only few systems support travelers when creating composite trips consisting of multiple destinations or activities. The idea in this work is to select travel regions that maximize the value of the composite trip for the user while still respecting her limitations in time and money. The value of a travel region can be determined by the similarity between a specified user query and the cases in a travel region database. The recommendation algorithm needs to find a decent routing between the regions while still satisfying diversity of the whole trip. We developed an algorithm based on an approximation for the knapsack problem and extended it to recognize dependencies between the regions while calculating best combinations. It is able to determine the optimal duration of stay per region and its performance improves when benefiting from the hierarchical structure of our travel database. In an expert study, we verified the results of our approach. The study proves that our algorithm for composite trips delivers good recommendations which satisfied an expert user more than baseline algorithms. Regions in the composite trip fit together better and a decent routing between the regions can be ensured. Nevertheless, the algorithm leaves room for improvement by combining less similar regions in a composite trip, thus leading to a higher diversity of the recommendation.
Approximation algorithms and in particular approximation schemes like PTAS and FPTAS were already introduced in Section 2.5 and 2.6, respectively. The main motivation in these sections was to illustrate the basic concept of constructing simple approximation schemes. The focus was put on algorithms where both the correctness and the required complexities were easy to understand without having to go deeply into the details of complicated technical constructions. Hence, an intuitive understanding about the basic features of approximation should have been brought to the reader which is a necessary prerequisite to tackle the more sophisticated methods required to improve upon the performance of these simple algorithms.
Conference Paper
Users often configure complex objects with many possible internal choices. Recommendation engines that automatically configure such objects given user preferences and constraints, may provide much value in such cases. These applications generate appropriate recommendations based on user preferences. It is likely, though, that the user will not be able to fully express her preferences and constraints, requiring a phase of manual tuning of the recommended configuration. We suggest that following this manual revision, additional constraints and preferences can be automatically collected, and the recommended configuration can be automatically improved. Specifically, we suggest a recommender component that takes as input an initial manual configuration of a complex object, deduces certain user preferences and constraints from this configuration, and constructs an alternative configuration. We show an appealing application for our method in complex trip planning, and demonstrate its usability in a user study.
Recommending online news articles has become a promising research direction as the Internet provides fast access to real-time information from multiple sources around the world. Many online readers have their own reading preference on news articles; however, a group of users might be interested in similar fascinating topics. It would be helpful to take into consideration the individual and group reading behavior simultaneously when recommending news items to online users. In this paper, we propose PENETRATE, a novel PErsonalized NEws recommendaTion framework using ensemble hieRArchical clusTEring to provide attractive recommendation results. Specifically, given a set of online readers, our approach initially separates readers into different groups based on their reading histories, where each user might be designated to several groups. Once a collection of newly-published news items is provided, we can easily construct a news hierarchy for each user group. When recommending news articles to a given user, the hierarchies of multiple user groups that the user belongs to are merged into an optimal one. Finally a list of news articles are selected from this optimal hierarchy based on the user’s personalized information, as the recommendation result. Extensive empirical experiments on a set of news articles collected from various popular news websites demonstrate the efficacy of our proposed approach.
SigTur/E-Destination is a Web-based system that provides personalized recommendations of touristic activities in the region of Tarragona. The activities are properly classified and labeled according to a specific ontology, which guides the reasoning process. The recommender takes into account many different kinds of data: demographic information, travel motivations, the actions of the user on the system, the ratings provided by the user, the opinions of users with similar demographic characteristics or similar tastes, etc. The system has been fully designed and implemented in the Science and Technology Park of Tourism and Leisure. The paper presents a numerical evaluation of the correlation between the recommendations and the user's motivations, and a qualitative evaluation performed by end users.