Introduction
Together with the rapid growing of information we daily produce, we have
assisted to the flourishing of new tools and techniques whose aim is to help users
in accessing such information in a personalized way. In the Information
Overload era we are living in, the amount of information exceeds the users capability
of processing and using it [
39
]. Huge and fast growing number of possibilities
overwhelms users, leading them to make poor decisions and feel anxiety and
unsatisfaction [
37
]. Recommender Systems (RSs) [
32
] are a family of information
filtering tools which have proven to be valuable means in assisting users to find,
in a personalized manner, what is relevant for them in such overflowing
complex information spaces. They provide users with personalized access to large
collections of resources.
The main task of a recommendation engine is typically to estimate the
relevance of unknown items for a target user and recommend the Top-N items by
considering for each user the best N items with highest utility [
32
]. Typically,
the utility of an item is represented by a numerical rating, which indicates how
much a particular user liked such item. However, the utility is not defined for
each pair of users and items and usually it is available only for a small subset of
them. Such subset represents the input of a generic recommender system, whose
objective is to estimate the utility of all the remaining pairs. A formal definition
of the recommendation problem has been given in [
1
] and defined as follows.
Definition 1. Let U represent the set of users and I the set of items in the
system. Potentially, both sets can be very large. Let f : U ⇥ I ! R, where R is
a totally ordered set, be a utility function measuring the relevance of item i 2 I
for user u 2 U . Then, the recommendation problem consists in finding for each
user u such item imax,u 2 I maximizing the utility function f . More formally,
this corresponds to the following:
Depending on the the way the utility function is estimated and the availability of
additional data about the characteristics of items for example, there are di↵erent
types of recommendation techniques. The main two are: collaborative filtering
and content-based. A complete list of collaborative filtering and content-based
techniques together with their hybrid versions is given in [
5
] and in [
32
].
Collaborative Filtering Recommendation. Collaborative Filtering is the process
of filtering or evaluating items using the opinions of other people [
36
]. According
to [
4
] there are two main types of collaborative filtering methods: memory-based
and model-based. Memory-based CF uses a particular type of Machine
Learning methods: nearest neighborhood (k-NN) algorithm. In particular, it does not
require any preliminary model building phase, since predictions are made by
aggregating the ratings of the closest neighbours. Conversely, model-based
techniques first learn a predictive model which is eventually used to make
predictions. Memory-based approaches can be classified either in user-based [
15
] or
item-based [
35
].
Content-based Recommendation. Content-based RSs recommend an item to a
user based upon a description of the item and a profile of the user’s interests [
30
].
Briefly, the basic process performed by a content-based recommender consists
is matching up the attributes of a user profile with the attributes of a content
object (item) [
8
]. Di↵erently from collaborative filtering, such recommendation
approach relies on the availability of content features describing the items. A
high level architecture of a content-based RS is presented in [
8
]. There are two
main content-based recommendation approaches: heuristic-based or model-based.
Approaches using heuristic functions have their roots in Information Retrieval
and Information Filtering. Items are recommended based on a comparison
between their content and a user profile. The idea is to represent both items and
users using typical IR techniques [
3
], e.g. vectors of terms, and compute a match
between their representations. The user profile consists in a vector of terms built
from the analysis of the items liked by the user. Model-based approaches [29] use
Machine Learning techniques to learn a model of the user’s preferences by
analyzing the content characteristics of items the user rated. Content-based methods
can have several limitations. For a complete and detailed description of
contentbased recommendation techniques refer to [
8,30
]. The main one is the content
overspecialization which consists in the incapability of the system to recommend
relevant items which are di↵erent to the ones the user already knows.
Knowledge-enabled Recommender Systems.
In content-based approaches, the information exploited by recommendation
engines is very often encoded by bag of words or, more recently, by word
embeddings [20]. In all these cases, no explicit semantics is associated to the contextual
data. Nowadays Linked Open Data datasets represent a huge repository of
different kinds of knowledge spanning from sedimentary-one such as encyclopedic,
linguistic, common-sense and so on, to real-time one such as data streams, events,
etc. If we consider encyclopedic datasets such as DBpedia [18] or Wikidata [
41
],
we have access to a huge amount of factual knowledge referring to a variety of
topics. In order to e↵ectively incorporate Linked Open Data in
recommendation applications there are several aspects to consider. Ultimately, the goal is to
provide the system with background knowledge about the domain of interest in
the form of a knowledge graph. In a high level architecture of a component in
charge of retrieving portions of the LOD graph regarding the items in the system
which are used to form the knowledge graph there are two main modules: the
Item Linker and the Item Graph Analyzer [28].
Item Linker. The Item Linker addresses the task of linking the items in the
system with the corresponding resources in the LOD knowledge bases. The aim
of such component is bridging the gap between the items in the catalog and LOD.
We have hypothesized two main ways for performing the linking task: Direct
Item Linking and Item Description Linking. These modules take as input
any dataset in the Linked Open Data cloud and the list of items in the catalog
with associated side information, if available, and return either the mapping
between items and URIs or the set of URIs found in each item description,
depending on the selected module.
Direct Item Linking This approach is the more straightforward way for accessing
LOD datasets. However, it requires that items have to be Linked Open Data
resources, otherwise it cannot be used. Using movie title and year information
it is possible to find the relative DBpedia resource. However, it is important to
solve possible cases of ambiguity. The simplest solution is to exploit the class
of the ontology that the item belongs to. For instance, in the movie domain we
may select the resources with class dbo:Film.
Item Description Linking This approach bases on the exploitation of side
information about the items such as textual descriptions or attributes. Such
information can be used as input for entity linking tools in order to have access to
LOD resources and link them to the item. Specifically, Entity Linking is the task
of linking the entity mentioned in the text with the corresponding real world
entity in the existing knowledge base [
38
]. Many Entity Linking tools have been
proposed in the literature and made available on the Web. Some of them are:
Babelfy [21], Dexter [
7
], DBpedia Spotlight [19], TAGME [
14
], NERD [
33
].
Item Graph Analyzer. This module is responsible of the extraction from the
knowledge base of a descriptive and informative subgraph for each item, that
is a set of RDF triples somehow related to the item resource. Eventually, all the
extracted portions of the original graph can be merged to obtain a specific
knowledge graph representative of the domain of interest covered by the recommender.
It takes as input the list of items URI returned by the Item Linker and returns a
set of RDF triples for each item. Potentially, each item resource may be connected
to a big portion of the LOD graph. However, not all entities and relations may
be informative and descriptive of the item content. Several strategies to select
a relevant subset of RDF triples for each item may be considered and adopted
[
22,31
]. One strategy can be to manually define a set of properties or sequences
of properties by using some domain knowledge. SPARQL queries are a powerful
tool to pre-filter subgraphs relevant for the recommendation task.
Evaluating LOD-based RSs
There are many datasets available for the evaluation of recommender systems.
However, such datasets are not appropriate for evaluating LOD-based
recommendation algorithms because they do not contains links to URIs. In order to evaluate
LOD-based RSs, three datasets belonging to di↵erent domains (movies, music and
book) have been processed to compute a mapping between items (movies, artists,
books) and their corresponding DBpedia URIs. The mappings for the datasets
is available at https://github.com/sisinflab/LODrecsys-datasets.
Movielens. This dataset is based on the MovieLens 1M dataset (http://www.
grouplens.org/node/73) released by the GroupLens research group. The
original dataset contains 1,000,209 1-5 stars ratings given by 6,040 users to 3,883
movies. We found a valid mapping for 3,300 out of the all movies.
LibraryThing. Derived from the LibraryThing (http://www.librarything.
com) dataset (http://www.macle.nl/tud/LT/). This dataset is related to the
book domain and contains 7,112 users, 37,231 books and 626,000 ratings ranging
from 1 to 10. In this case we found a match for 11,694 books.
LastFM. Di↵erently from the previous ones this third dataset is based on
implicit feedback consisting of user-artist listening data. Its data come from
recent initiatives on information heterogeneity and fusion in recommender systems
(http://ir.ii.uam.es/hetrec2011/datasets.html) [
6
] and has been built
on top of the Last.fm music system (http://www.lastfm.com). The original
dataset contains 1,892 users, 17,632 artists and 92,834 relations between a user
and a listened artist together with their corresponding listening counts. For this
dataset we found a match for 11,180 out of all artists.
Open Challenges
Recommender systems can be considered as a killer application for the
exploitation of the huge knowledge encoded in LOD datasets freely available on the
Web. Although, many solutions have been proposed and implemented to
deliver this new generation of knowledge enabled recommendation engines (see
[
31,12,11,28,24,23,10,40,25,9,27,26,16,2,17,34,13
]) some important issues remain
still open to a deeper investigation. Among them we cite the most important
ones: feature selection, distributed computing, cross-domain recommendation,
computing explanation, the role of formal reasoning in the recommendation
process.
feature selection The richness of Linked Open Data datasets may result in
a pitfall for data-intensive tasks (as computing recommendations) as they
potentially introduce noise in the data. Selecting the right features in
LODenabled recommender systems results not just in getting the minimal
meaningful subset of properties which are directly connected to an item but, given
the graph-based nature of the underling data, the minimal meaningful set of
semantic paths, of arbitrary length, which result representative of the item
itself.
distributed computing Over the last years solutions to horizontally distribute
graph-based data manipulation have been proposed also boosted by the
increasing production of data coming from social networks. All the methods,
algorithms and frameworks work quite well with multi-relational graphs where
the number of possible relations are just a few compared to that of Linked
Open Data. New approaches need to be proposed and developed to easily
deal with all the semantics encoded in LOD datasets.
cross-domain recommendation The highly interconnected nature of datasets
such as DBpedia or Wikidata represents an opportunity to develop
crossdomain recommender systems. That is, systems able to recommend items
in a knowledge domain which is not the same of the user profile. As an
example, we may be able to recommend books given the user profile collects
information on movies.
computing explanation Sometimes, receiving recommendations may result
frustrating as we do not know the reason why the system suggested such
items to us. Computing explanation for recommendations has been identified
as a must have feature for the new generation of recommender systems. In
this direction, all the knowledge available as Linked Open Data may surely
play a key role.
formal reasoning LOD datasets are not just a mere collection of data
represented in a graph-based way. They usually refer to a rich ontology which
in turns can be represented by means of expressive logical languages as
Description Logics. The adoption of such languages enable the application of
formal logical reasoning over the underling data. As of today, due to its
high computational complexity, such reasoning has not been exploited to
its full potential but it can surely add new value to real knowledge-enabled
recommender systems.
17. Aleksandra Karpus, Tommaso Di Noia, Paolo Tomeo, and Krzysztof Goczyla.
Rating prediction with contextual conditional preferences. In Proceedings of the 8th
International Joint Conference on Knowledge Discovery, Knowledge Engineering
and Knowledge Management (IC3K 2016) - Volume 1: KDIR, Porto - Portugal,
November 9 - 11, 2016., pages 419–424, 2016.
18. Jens Lehmann, Robert Isele, Max Jakob, Anja Jentzsch, Dimitris Kontokostas,
Pablo N. Mendes, Sebastian Hellmann, Mohamed Morsey, Patrick van Kleef, So¨ren
Auer, and Christian Bizer. DBpedia - a large-scale, multilingual knowledge base
extracted from wikipedia. Semantic Web Journal, 6(2):167–195, 2015.
19. Pablo N. Mendes, Max Jakob, Andr´es Garc´ıa-Silva, and Christian Bizer. Dbpedia
spotlight: Shedding light on the web of documents. In Proceedings of the 7th
International Conference on Semantic Systems, I-Semantics ’11, pages 1–8, New
York, NY, USA, 2011. ACM.
20. Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg Corrado, and Je↵rey Dean.
Distributed representations of words and phrases and their compositionality. In
Proceedings of the 26th International Conference on Neural Information Processing
Systems - Volume 2, NIPS’13, pages 3111–3119, USA, 2013. Curran Associates
Inc.
21. Andrea Moro, Alessandro Raganato, and Roberto Navigli. Entity Linking meets
Word Sense Disambiguation : a Unified Approach. Transactions of the Association
for Computational Linguistics (TACL), 2014.
22. Cataldo Musto, Pasquale Lops, Pierpaolo Basile, Marco de Gemmis, and Giovanni
Semeraro. Semantics-aware graph-based recommender systems exploiting linked
open data. In Proceedings of the 2016 Conference on User Modeling Adaptation
and Personalization, UMAP ’16, pages 229–237, New York, NY, USA, 2016. ACM.
23. Phuong Nguyen, Paolo Tomeo, Tommaso Di Noia, and Eugenio Di Sciascio. An
evaluation of simrank and personalized pagerank to build a recommender system
for the web of data. In Proceedings of the 24th International Conference on World
Wide Web, WWW ’15 Companion, pages 1477–1482, New York, NY, USA, 2015.
ACM.
24. Phuong T. Nguyen, Paolo Tomeo, Tommaso Di Noia, and Eugenio Di Sciascio.
Content-Based Recommendations via DBpedia and Freebase: A Case Study in the
Music Domain, pages 605–621. Springer International Publishing, Cham, 2015.
25. Vito Claudio Ostuni, Tommaso Di Noia, Eugenio Di Sciascio, and Roberto Mirizzi.
Top-n recommendations from implicit feedback leveraging linked open data. In
Proceedings of the 7th ACM Conference on Recommender Systems, RecSys ’13,
pages 85–92, New York, NY, USA, 2013. ACM.
26. Vito Claudio Ostuni, Tommaso Di Noia, Roberto Mirizzi, and Eugenio Di Sciascio.
A linked data recommender system using a neighborhood-based graph kernel. In
The 15th International Conference on Electronic Commerce and Web Technologies,
Lecture Notes in Business Information Processing. Springer-Verlag, 2014.
27. Vito Claudio Ostuni, Giosia Gentile, Tommaso Di Noia, Roberto Mirizzi, Davide
Romito, and Eugenio Di Sciascio. Mobile movie recommendations with linked data.
In Human-Computer Interaction & Knowledge Discovery @ CD-ARES’13, IFIP
International Cross Domain Conference and Workshop on Availability, Reliability
and Security, CD-ARES 2013. Springer, 2013.
28. Vito Claudio Ostuni, Sergio Oramas, Tommaso Di Noia, Xavier Serra, and Eugenio
Di Sciascio. Sound and music recommendation with knowledge graphs. ACM
Transactions on Intelligent Systems and Technology (TIST), 8(2):21:1–21:21, 2016.
29. Michael Pazzani and Daniel Billsus. Learning and revising user profiles: The
identification ofinteresting web sites. Mach. Learn., 27(3):313–331, June 1997.