Recommending Sources in News Recommender Systems

Ozlem

Ozg¨obek

1 2

, Jon Atle Gulla

and R. Cenk Erdur

Department of Computer and Information Science, NTNU, Trondheim, Norway

Department of Computer Engineering, Ege University, Izmir, Turkey

Keywords:

Recommender Systems, News Source, News Recommendation.

Abstract:

Recommender systems aim to deliver the most suitable item to the user without the manual effort of the

user. It is possible to see the applications of recommender systems in a lot of different domains like music,

movies, shopping and news. Recommender system development have many challenges. But the dynamic and

diverse environment of news domain makes news recommender systems a little bit more challenging than other

domains. During the recommendation process of news articles, personalization and analysis of news content

plays an important role. But beyond recommending the articles itself, we think that where the news come from

is also very important. Different news sources have their own style, view and way of expression and they may

give the user a complete, balanced and wide perspective of news stories. In this paper we explain the need

for including news sources in news recommendation and propose a news source recommendation method by

ﬁnding out the implicit relations and similarities between news sources by using semantics and association

rules.

1 INTRODUCTION

News recommendation is a challenging task which

includes many difﬁculties compared to other recom-

mendation domains. News domain has a very dy-

namic environment with usually hundreds of new ar-

ticles published every hour. While the number of on-

line articles increases, the recency and popularity of

articles change too fast which makes the recommen-

dation more challenging. In (Ozgobek et al., 2014)

challenges in recommender systems and news rec-

ommenders are explained in detail. To be able to

make suitable recommendations to the user, the rec-

ommender system needs the detailed user information

and/or the content of news items.

News recommendation is considered as the news

article recommendation. So each item that the recom-

mender system recommends is a news article. There

is a vast amount of research going on about news rec-

ommender systems and there are very successful re-

sults. But all the news recommender system research

is too much focused on the articles, analysis of the

text, ﬁnding similarities between news articles and

predicting user’s like on each article. There is no news

recommender system which takes the source of the

news into account that we could ﬁnd.

Differentnews sources may have different special-

izations like sports, science etc. Even though they

publish the same story with other sources the way

they express the news, the words they select to use

may be different. Each news source has its own style,

view and way of expression that some users like and

some does not. So it is important to consider the dif-

ferences between news sources and to recommend ar-

ticles from the sources that users like.

Serendipity problem for recommender systems

addresses the problem of recommending similar or

the same items with the already recommended ones.

A different article describing the same event should

not be recommended while keeping the diversity of

recommendations (Lops et al., 2011). For news rec-

ommendersystems it is challenging to solve this prob-

lem because it is harder to understand the differences

of the meaning of the whole article. It is really hard to

distinguish the news articles if they are the same news

written in a different way or different news stories re-

lated to the same topic. For example two news articles

may be identical which are telling the same story or

they may be the different parts of one story. By rec-

ommending news sources we mitigate this problem.

When there is more than one news article on a speciﬁc

topic coming from the same news source, it is usually

possible to say that they are different news articles on

the same topic. We can recommend very similar ar-

ticles from the same source without worrying about

serendipity problem.

526

Özgöbek Ö., Gulla J. and Erdur R..

Recommending Sources in News Recommender Systems.

DOI: 10.5220/0005489205260532

In Proceedings of the 11th International Conference on Web Information Systems and Technologies (WEBIST-2015), pages 526-532

ISBN: 978-989-758-106-9

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

In (Ofcom, 2013) it is stated that less than half

of online news users (45%) use only one source to

get the news. Similarly in (Media, 2012) it can be

seen that the average number of online news sources

consumed changes between 1.4 and 2.4 according to

the demographic differences, where the total average

is 2.0, including websites and apps.

News recommender systems usually include news

resources that are hard coded by the developers. Even

though it is possible to give the right to select the

news sources to the user, the user would select the

most trusted, liked or known news sources for her-

self. This prevents the users to discover new sources

of news articles and narrows the scope of the news

recommender system. Since many online news read-

ers usually check only a few online news sources in

their daily routine, it is hard for users to discover new

sources and gain a wider perspective by what they

read. In (Media, 2012) it is stated that “Our qualita-

tive research showed that reasons for multi-sourcing

can often be active choices, where a person seeks to

get a balanced, complete view of a story from across

a range of providers and platforms.”. So beyond mak-

ing personalized news article recommendation, it is

also important to make news source recommenda-

tions and consider the news sources in article rec-

ommendations. Among many online news sources it

is also a challenge to ﬁnd the source that the users

would like. Considering the news sources including

websites, blogs and other possible sources during the

news recommendation process, helps us to increase

the quality of recommendations, provides better solu-

tions for dealing with the news recommender system

challenges and gives the chance to the users to dis-

cover different sources which they may like.

In this paper we propose a news source recom-

mendation method. To be able to recommend news

sources, we used two different methods to ﬁnd out

the implicit relations and similarities between news

sources. The ﬁrst method is a semantic analysis of

news content. For this method we built a small scale

ontology which includes important terms from spe-

ciﬁc news sources. The second method is the asso-

ciation analysis which is a data mining method. We

used the association analysis to ﬁnd out the relations

between news sources according to the users’ read-

ing patterns. Then we compared and discussed these

two different approaches to ﬁnd the hidden relations

between news sources.

The rest of the paper organized as follows. In

Section 2, we give information about the background

work done within the SmartMedia project which this

work also belongs to. Section 3 describes the related

work about semantic news recommender systems and

association rules for recommender systems. The de-

tails of our approach is explained in Section 4. In Sec-

tion 5 the results and discussion is explained. Finally

in Section 6 conclusion and future work is given.

2 BACKGROUND WORK

SmartMedia project

in Norwegian University of Sci-

ence and Technology (NTNU) was started in 2011 in

close collaboration with Scandinavian media indus-

try. With this project it is aimed to present the on-

line newsinformation in an effective and personalized

way to the users while considering the point of view

of the journalists and media companies. The main

focus of the project is recommender systems and se-

mantic search.

Within the SmartMedia project it is presented a

user proﬁling approach for the mobile news recom-

mender systems (Gulla et al., 2013). In this work

the users’ actions on the mobile device is observed

and by using this information an approach for learn-

ing the user proﬁles is proposed. The implemented

mobile news recommender of SmartMedia project is

proposed in (Tavakolifard et al., 2013). There is also

ongoing work about the multi-platform implementa-

tion of our news recommender system. The progress

of building a complete and publicly available dataset

in Norwegian news domain is continuing (Ozg¨obek

et al., ) within the SmartMedia project.

The proposed work in this paper is also continuing

as a part of the SmartMedia project.

3 RELATED WORK

3.1 Semantic News Recommenders

Semantic approach is one of the methods for recom-

mendation. The main motivation to use semantics in

recommender systems is to be able to use the cultural

and linguistic background knowledge of the content

(Peis et al., 2008). The use of semantics reduces am-

biguity compared to keyword based systems, it allows

hierarchical representation of concepts and inference

(Cantador and Castells, 2009). In (Lops et al., 2011),

the semantic recommenders are grouped according to

their use of different semantic approaches. Since the

challenges to solve and approaches applied to solve

these challenges differs from domain to domain (Oz-

gobek et al., 2014), it is also possible to see different

semantic approaches for different domains. Although

http://research.idi.ntnu.no/SmartMedia/

RecommendingSourcesinNewsRecommenderSystems

527

there are a lot of semantic recommender system re-

search available, in this section we are going to con-

sider the semantic news recommenders which is our

main focus of interest.

In (Cantador and Castells, 2009), it is proposed a

news recommender system News@Hand, which uses

semantic technologies to provide recommendations.

Ontolgies are populated from the news contents by

extracting the noun terms. Also Wikipedia articles are

used to populate ontology classes. To overcome the

data sparsity problem in user proﬁles, it is proposed a

mechanism to expand the user preferences. The cur-

rent context is also considered and deﬁned in a way

that the importance of concepts decreases within time

for making better context aware recommendations.

Hermes framework uses a semantic approach to

build personalized news service. (IJntema et al.,

2010) It is proposed an extension to the Hermes

framework for the semantic news recommendation

which is called Athena. For the recommendations,

ﬁrst a user proﬁle is constructed by using the user’s

reading history. Then by using the different simi-

larity measures and the ontology populated by the

Hermes framework recommendations are done. The

news recommendation method which is implemented

in Athena is proposed in (Goossen et al., 2011). In

this work, the well known method TF-IDF (Term

Frequency - Inverse Document Frequency) for con-

tent based recommenders is applied to the semantic

recommenders as CF-IDF (Concept Frequency - In-

verse Document Frequency). CF-IDF considers only

the key concepts in the news articles where TF-IDF

considers all the terms. Similarly, in (Capelle et al.,

2012) it is proposed two new methods called Synset

Frequency - Inverse Document Frequency (SF-IDF)

which uses the WordNet synonym sets and Semantic

Similarity (SS) to calculate the similarities between

news items. It is used to recommend news items

based on the user behavior proﬁle and semantic sim-

ilarity measure. The proposed methods are imple-

mented as the Ceryx frameworkwhich is an extension

of Athena.

(Rao et al., 2013) proposes an ontology based sim-

ilarity model to calculate the news-user similarity in

a semantic news recommender system. The ontolo-

gies are populated by using the online encyclopedias

as DBPedia. The proposed similarity model is built

on these ontologies and the background information

is used to measure the similarities between news arti-

cles and users.

In (Laˇsek and Vojt´aˇs, 2011) it is proposed a se-

mantic information ﬁltering workﬂow in the news ﬁl-

tering use case. The workﬂow includes many steps

including entity identiﬁcation, semantic data crawl-

ing and building user proﬁle. This work aims to use

the advantages of semantic background information

used together with the user proﬁles and improve the

results of name entity recognition by the help of user

feedback.

Even though there are many works on person-

alized news recommendation, there is no research

which considers the news sources that we could ﬁnd.

We think that having a wider perspective of news sto-

ries around the world or ignoring the news sources

that the user does not want to read is important. So

considering news sources is an important aspect of

news recommender systems to work on.

3.2 Association Rules for Recommender

Systems

Association analysis is a data mining method which

is used to discover the hidden relationships of items

in large datasets (Pang-Ning et al., 2006). The re-

lations are represented as association rules which is

shown as X → Y where X and Y are disjoint item-

sets. The strength of association rules is measured

with two metrics called support and conﬁdence. Sup-

port deﬁnes the proportion of the number of transac-

tions containing X and Y together to the total number

of itemsets in the dataset. Support is shown as s(X →

Y) =

σ(X∪Y)

. Conﬁdence deﬁnes the proportion of the

transactions which includes X and Y together to the

number of transactions only contains X. Conﬁdence

is shown as c(X → Y) =

σ(X∪Y)

σ(X)

Association analysis is widely used in many areas

for different purposes. A very common and success-

ful usage of association analysis is for the supermar-

kets to ﬁnd out the hidden relations between prod-

ucts and change the sales policies accordingly. By

analyzing the users’ buying patterns it is possible to

ﬁnd a relation between two disparate products (Pang-

Ning et al., 2006). There are also some studies about

the association rule mining in recommender systems.

In (Mobasher et al., 2001) a personalization frame-

work based on association rule discovery is proposed.

Association rules are also used to develop more se-

cure recommender systems. For example in (Sand-

vig et al., 2007) an association rule based algorithm

is presented to prevent the proﬁle injection attacks in

recommender systems. In (Lemdani et al., 2010) it

is presented a collaborative ﬁltering method by using

the association rules. Similarly (Sun et al., 2005) pro-

poses a method which applies quantitativeassociation

rules in collaborative ﬁltering and compares with the

conventional Pearson method.

WEBIST2015-11thInternationalConferenceonWebInformationSystemsandTechnologies

528

4 OUR APPROACH

In our approach to the news source recommendation,

we used two different methods to ﬁnd the similari-

ties between news sources. Our aim was to ﬁnd out

some rules between news sources according to the

users’ reading patterns. We preferred to use associa-

tion rules because it givesexactly what we are looking

for. By using association rules we discovered some

rules like: ”Who reads from source X also reads from

source Y.” which can be shown as X → Y. This ap-

proach is a well known and used method in recom-

mender systems by itself. In addition to the ﬁndings

from association rules we wanted to see if the source

X and source Y are semantically related. To do that

we built a small ontology to detect the similarities be-

tween news sources. The details and results are given

below.

4.1 Dataset

To test our approach we are using the YOW dataset

(Wolfe and Zhang, 2010). YOW dataset includes two

parts. The ﬁrst part is publicly available on the web

and it is collected in a user study at Carnegie Mellon

University in 2004. This part of the dataset contains

the article information(document id, source etc.) read

by users, explicit feedback (user likes) and very de-

tailed implicit user feedback like the time spent on

each article. A small example from the dataset is

shown in Figure 1. In the dataset, each user, arti-

cle and source has a unique id. In Figure 1, we can

see which user (user id) read which article (DOC ID),

how much she liked (user like) and the source number

of the news article (RSS ID). YOW dataset contains

data collected from 25 users in total. The number of

read articles from each source by each user included

in this part of the dataset is used to ﬁnd out the possi-

ble similarities between news sources.

Figure 1: A small example of the dataset.

The second part of the dataset is crawled from

RSS feeds (Zhang, 2005). This part includes textual

http://users.soe.ucsc.edu/ yiz/papers/data/YOWStudy/

information about news articles including the head-

line, an introductory text, URL and source number

(RSS ID). So by using the RSS IDs we can group the

news articles according to their sources. Since there

are many news articles from each source, it is pos-

sible to analyse the textual information for speciﬁc

sources. By using text analysis we extracted the key

concepts for every news source and built an OWL on-

tology for the ontology based similarity detection of

news sources.

4.2 Ontology based Similarity Detection

of News Sources

Ontologies and semantic reasoning is a useful way

of revealing relations between entities. For the news

sources, each source may have different areas of fo-

cus like sports, politics, movies etc. Or they may

use different terminology for news stories. On one

hand, ﬁnding important terms used in articles from

a news source can tell us more about the proper-

ties of that source. On the other hand, ﬁnding com-

mon similar terms of different news sources gives us

the chance to discover the similarities between news

sources. The YOW dataset contains the headlines and

short introductory text of each article. So we started

our progress of revealing the similarities between dif-

ferent news sources using ontologies by ﬁnding the

important terms of articles contained in each news

source.

For ontology construction we use Prot´eg´e

and

TerMine

. Prot´eg´e is a widely used open source plat-

form to build ontologies. TerMine is a text mining

tool to extract candidate terms from a text. It can also

be used as a Prot´eg´e plug-in.

The class hierarcy in the ontology is built accord-

ing to the common news article categorization. So it

is easy to see the detailed category and topic similar-

ities between news sources. A small example of the

ontology class hierarcy is shown in Figure 2. Each

individual in the ontology has an object property ’has

source’ which shows the source(s) of entites.

The developed ontology is a very small scale on-

tology only to study the possibility of recommending

news sources by using semantic relations.

When we analyse the resulant ontology which

contains more than 350 individuals from 26 differ-

ent sources and 16 main categories, we found some

relations between news sources. In Figure 3 relation

between two news sources number 8156 and 8174 is

shown as an exmple. They both include common top-

ics about actors. So it is possible to say that these two

http://protege.stanford.edu/

http://www.nactem.ac.uk/software/termine/

RecommendingSourcesinNewsRecommenderSystems

529

Figure 2: A small example of the ontology class hierarcy.

Figure 3: Relation between two news sources as an exam-

ple.

news sources include related topics and if one user

likes one of the sources, it is possible to recommend

her articles from the other news source.

Also there is a possibility that even though the

sources are related in topic based, the user might not

like one of them because of the different language

used in that source or the different view of the writ-

ers over the topic (how they discuss on the topic).

This aspect of the news source recommendation will

be handled in future work. In this work our aim

is to show that it is possible to make recommenda-

tions based on semantic analysis and machine learn-

ing techniques together.

In our analysis of news sources like the exam-

ple shown in Figure 3, we consider a source relation

only if that source has two or more common enti-

ties with the other sources. By doing this we elim-

inate the chance factor on discovering relations of

news sources. For example, the name of actor might

be included in a news article related with sports, but

the news source can be a source related with only

sports, not with movies or actors. We eliminate the

chance factor in ﬁrst step when we extract the impor-

tant terms from the text contained in a news source.

And by choosing only the relations containing more

than one common entity we make sure that we ﬁnd

the related news sources.

In Table 1 it is provided some of the common re-

lations of news sources discovered by the semantic

relations in the ontology. In this table X-axis denotes

different news sources and y-axis shows different cat-

egories. The numbers in the table denotes the fre-

quencies of instances for each source in different cat-

egories.

4.3 Association Rule based Relation

Detection of News Sources

A user’s interest to a speciﬁc news source can be un-

derstood by looking at the number of articles read

from that source. For a user, as the number of read

articles from a speciﬁc source increases, it is more

likely that the user is interested in that news source.

If a user is interested in a few news sources, these

sources may be similar to each other or they may have

similar impact levels on readers. So ﬁnding the un-

derlying relations of news sources that is not clearly

visible becomes an important issue on news recom-

mendation. In this work we used association analysis

to ﬁnd out the implicit relations between news sources

according to the users’ interests.

To analyze the users’ reading patterns by asso-

ciation rules, we extracted the number of reads per

user from each news source for 24 users and 26 news

sources. In Table 2 a small example of users’ number

of reads from different sources is shown. The news

sources used in this experiment are the same news

sources that we used to build an ontology which is

described in previous section. For applying the asso-

ciation analysis we used Orange

which is an open

source data visualization and analysis tool.

In the analysis of the results to preventconsidering

the rules which may occurred by chance, we ignore

the association rules which has a support number be-

low 0,4 and conﬁdence number below 0,7. Especially

the support is known as an important measure for the

rules occur by chance (Pang-Ning et al., 2006).

5 RESULTS AND DISCUSSION

In this work we proposed a news source recommen-

dation method which uses two methods to ﬁnd the

http://orange.biolab.si/

WEBIST2015-11thInternationalConferenceonWebInformationSystemsandTechnologies

530

Table 1: Some relations of news sources extracted from ontology. X-axis denotes different news sources.

1265 1302 1892 3616 8152 8155 8156 8157 8172 8174 8185 8195

Business 3 8 6

Entertainment 3 5 8 3 2 2

Economy 5

Event 2 6 3 2

Health 3 5 8

City 3 2 2

Sports

Software 6 2

Web 3 8

Table 2: Users’ number of reads from different sources.

Source X Source Y Source Z ...

User A 35 2 15 ...

User B 28 73 6 ...

User C 4 12 26 ...

... ... ... ... ...

most relevant news sources. The ﬁrst method con-

tains semantic analysis of news content from differ-

ent sources. As it is explained in detail in Section 5.2

we built an ontology where each entity belongs to a

news source. So by looking at the number of entities

within a class which belongs to a speciﬁc news source

it is possible to see the topics included in that news

source. The second method uses the association anal-

ysis to ﬁnd out the relations between news sources

according to the users’ reading patterns. When we

merge the results from these two methods we get the

most relevant news sources that the system can rec-

ommend.

Our results show that it is possible to ﬁnd a cor-

relation between the related news sources and users’

reading patterns. If there is an extracted rule like X

→ Y also there is usually a semantic relation between

news sources X and Y. In Table 3 it is shown the top

5 results of this correlation between association rules

and the ontology. All the relations seen in this table

occurs both in ontology and association rules. Each

source is represented by a four digit number, also

the association rule support and conﬁdence values are

given.

We see the news source recommendation in three

different aspects:

• Possible solution to the serendipity problem.

• Discovery of news sources for users. It gives a

wider perspective of news stories.

• Better personalized news recommendations by

considering news sources. Some people like to

read from several different sources, some like to

follow only one.

Table 3: Ontology and association rule results.

Source 1 Source 2 Support Conﬁdence

8156 8195 0,619 1,000

8153 8195 0,572 1,000

8156 8155 0,461 0,769

8172 8155 0,429 0,900

8172 8152 0,333 0,700

6 CONCLUSION AND FUTURE

WORK

In this paper we presented a news source recommen-

dation method. As the number of online news sources

increases it becomes harder for users to ﬁnd the suit-

able news sources for themselves. A user may spend

many hours to discover new sources of news which

she likes. It is also possible that she may never ﬁnd

some sources that she would like. The same news

topic may be represented differently in different news

sources. So when the user likes how news topics

are represented and expressed in a speciﬁc source,

she would like to receive more news items from the

same source. For the news domain it is a challenge

not to recommend the same story from different news

sources. This approach may also be a solution for this

challenge. On one hand, reading several articles from

the same source may also give the user a coherent and

consistent view of stories. On the other hand, discov-

ering new sources and reading the stories from differ-

ent sources may give the user a complete, balanced

and wider perspective.

The news source ontology that we built is a small

scale ontology built for testing the idea of considering

news sources in news recommendation process. It is

possible to observe more correct relations in a bigger

ontology which we consider as a future work. Also

any improvements of the quality of the ontology will

help to get better results.

The importance of news recommendation which

considers the differences and similarities between

RecommendingSourcesinNewsRecommenderSystems

531

news sources looks promising to improve the recom-

mendation quality. On the other hand recommend-

ing the news sources itself is something that the news

recommenders should consider. This approach that

we used to evaluate the similarities between news

sources can also be used to ﬁnd the correlations be-

tween news categories and make recommendations

considering the news categories where the categoriza-

tion of news articles is a challenge by itself.

REFERENCES

Cantador, I. and Castells, P. (2009). Semantic contextualisa-

tion in a news recommender system. In Workshop on

Context-Aware Recommender Systems (CARS 2009).

Capelle, M., Frasincar, F., Moerland, M., and Hogenboom,

F. (2012). Semantics-based news recommendation.

In Proceedings of the 2nd International Conference

on Web Intelligence, Mining and Semantics, page 27.

ACM.

Goossen, F., IJntema, W., Frasincar, F., Hogenboom, F., and

Kaymak, U. (2011). News personalization using the

cf-idf semantic recommender. In Proceedings of the

International Conference on Web Intelligence, Mining

and Semantics, page 10. ACM.

Gulla, J. A., Ingvaldsen, J. E., Fidjestl, A. D., Nilsen, J. E.,

Haugen, K. R., and Su, X. (2013). Learning user pro-

ﬁles in mobile news recommendation. pages 183–194.

IJntema, W., Goossen, F., Frasincar, F., and Hogenboom,

F. (2010). Ontology-based news recommendation.

In Proceedings of the 2010 EDBT/ICDT Workshops,

page 16. ACM.

Laˇsek, I. and Vojt´aˇs, P. (2011). Semantic information

ﬁltering-beyond collaborative ﬁltering. In 4th Inter-

national Semantic Search Workshop.

Lemdani, R., Bennacer, N., Polaillon, G., and Bourda, Y.

(2010). A collaborative and semantic-based approach

for recommender systems. In Intelligent Systems De-

sign and Applications (ISDA), 2010 10th International

Conference on, pages 469–476. IEEE.

Lops, P., De Gemmis, M., and Semeraro, G. (2011).

Content-based recommender systems: State of the art

and trends. In Recommender systems handbook, pages

73–105. Springer.

Media, K. (2012). Measuring news consumption and atti-

tudes.

Mobasher, B., Dai, H., Luo, T., and Nakagawa, M. (2001).

Effective personalization based on association rule

discovery from web usage data. In Proceedings of the

3rd international workshop on Web information and

data management, pages 9–15. ACM.

Ofcom (2013). News consumption in the uk - 2013 report.

Ozgobek, O., Gulla, J. A., and Erdur, R. C. (2014). A sur-

vey on challenges and methods in news recommenda-

tion. In In Proceedings of the 10th International Con-

ference on Web Information System and Technologies

(WEBIST 2014).

Ozg¨obek, O., Shabib, N., and Gulla, J. A. Data sets and

news recommendation.

Pang-Ning, T., Steinbach, M., Kumar, V., et al. (2006). In-

troduction to data mining. In Library of Congress.

Peis, E., del Castillo, J. M., and Delgado-L´opez, J. (2008).

Semantic recommender systems. analysis of the state

of the topic. Hipertext. net, 6:1–5.

Rao, J., Jia, A., Feng, Y., and Zhao, D. (2013). Per-

sonalized news recommendation using ontologies har-

vested from the web. In Web-age information manage-

ment, pages 781–787. Springer.

Sandvig, J. J., Mobasher, B., and Burke, R. (2007). Robust-

ness of collaborative recommendation based on asso-

ciation rule mining. In Proceedings of the 2007 ACM

conference on Recommender systems, pages 105–112.

ACM.

Sun, X., Kong, F., and Chen, H. (2005). Using quantita-

tive association rules in collaborative ﬁltering. In Ad-

vances in Web-Age Information Management, pages

822–827. Springer.

Tavakolifard, M., Gulla, J. A., Almeroth, K. C., Ingvaldesn,

J. E., Nygreen, G., and Berg, E. (2013). Tailored news

in the palm of your hand: a multi-perspective transpar-

ent approach to news recommendation. In Proceed-

ings of the 22nd international conference on World

Wide Web companion, pages 305–308. International

World Wide Web Conferences Steering Committee.

Wolfe, S. R. and Zhang, Y. (2010). Interaction and person-

alization of criteria in recommender systems. In User

Modeling, Adaptation, and Personalization, pages

183–194. Springer.

Zhang, Y. (2005). Bayesian graphical models for adaptive

information ﬁltering - ph.d. dissertation.

WEBIST2015-11thInternationalConferenceonWebInformationSystemsandTechnologies

532