recommender helping viewers in their choice for educational programs in digital tv context

6

Click here to load reader

Upload: elaine-cecilia-gatto

Post on 25-May-2015

156 views

Category:

Education


0 download

TRANSCRIPT

Page 1: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-1

Recommender: Helping Viewers in their Choice for Educational Programs in Digital TV Context

Paulo Muniz de Ávila, Elaine Cecília Gatto, Sergio Donizetti Zorzo pauloavila @pucpocos.com.br,[email protected],[email protected]

Abstract - Currently in Brazil, a fundamental change is taking place in TV: the migration from analogue to digital TV system. This change has two main implications: an increase in transmission capacity for new channels with the same bandwidth and the ability to send applications with multiplexed audio-visual content. Brazilian government aims to exploit the transmission capacity for new channels offering programming created to distance learning and thereby promoting social inclusion in the vast majority of the population. This information overload demands mechanisms to help students to browse and select what education programs are best suited to their current level. Personalized recommendation systems emerge as a solution to this problem, providing the viewer with educational programs relevant to his profile. In this paper we present a personalized recommendation system, the Recommender consistent with the reference implementation of the Brazilian digital TV system. Finally, we present the results obtained after using the proposed system. Key-words - Personalization, Multimedia, Recommendation System, Digital TV, Middleware Ginga.

INTRODUCTION

Digital television has created new services, products, contents, channels and business models. The Brazilian Digital TV System allows high quality audio and video, as well as interactivity, creating different contents for users. There are two main implications with Brazil Digital TV System: the increase of the number of channels being broadcasted with the same bandwidth and the possibility of sending multiplexed applications with the audio-visual content. As new channels emerge due to the transmission increase, it is necessary to create ways that allow the TV viewers to search among these channels.

The Electronic Program Guide (EPG) helps the TV viewers. However, as new channels are available, an information overload is unavoidable making the EPG system inappropriate. In Shangai [1], a big city in China, the TV operators provide different services (in the analogical system, channels), and this number has been increasing at a 20% rate per year. Thus, the traditional EPG system became unattractive because it takes too long for the viewers to search among hundreds of options available to find their

favorite program. In face of this situation, personalized recommendation systems are necessary.

Different from EPG functions which allow basic search, a personalized TV system can create a profile for each TV viewer and recommend programs that best match this profile, avoiding the search in many EPG options to find the favorite program. Elementary and secondary education schools and universities generally seek to explore this new model offering personalized content to their students. In this context, a recommendation system is able to analyze the profile of a group of students, suggesting the educational content that best suits the needs of the group.

To make the benefits (new channels, interactive applications) offered by the digital system possible, the TV viewers with analogical system need new equipment called set-top box (STB). STB is a device which works connected to the TV and converts the digital sign received from the provider to audio/video that the analogical TV can exhibit. To have the advantages offered by the digital TV, the STB needs a software layer which connects the hardware to the interactive applications called middleware. The DTV Brazilian System middleware is Ginga [2,3]. It allows declarative and procedural applications through its components Ginga-NCL [2] and Ginga-J [3]. Ginga-NCL performs declarative application written in Nested Context Language (NCL) while Ginga-J can perform procedural application based on JavaTM known as Xlets [4].

This paper proposes an extension to Ginga middleware through implementation of a new module incorporated to Ginga Common Core called Recommender. The Recommender module is responsible for gathering, storing, processing and recommending TV education programs. To develop the Recommender module, Ginga-NCL middleware developed by PUC-RIO (Pontifical Catholic University of Rio de Janeiro) was used, implemented in C/C++ language with source code available under GPLv2 license and according with the patterns defined by the Brazilian system digital television [4].

TVDI IN BRAZIL AND EDUCATION

One of the reasons to implement TVDi in the national territory is its potential to social inclusion. In Brazil, in many cases, the open TV is the only source of information for people who do not frequently read newspaper, magazine or any other kind of printed media. If we consider that the access to written information is low and that the information

Page 2: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-2

transmitted through TV newscasts is the biggest link between the world and the daily routine of Brazilian people, we have many reasons not to ignore the reach power of this technology. If it is correctly and consciously explored, with the help of interactive resources, TVDi can represent a powerful tool to have access to differentiated educational knowledge at the same time it can include Brazilian citizens digitally excluded nowadays. Thus, it can be said that in Brazil, the access to the Internet is low and high-class people are those who have more access to it and participate somehow in the educational scenario. The low number of personal computers and the high number of TV sets in Brazilian houses defend the efforts to use all TVDi potential in issues in the educational extent. If public policies are well structured, TVDi can reinforce a new educational paradigm, allowing the entire population to have access to Internet resources, video, images, sounds, interactivity to introduce new knowledge, entertainment, education, leisure, services. It can allow the unlimited access to written and audiovisual information. As the great part of Brazilian population has a limited access to information and Internet, and considering the fact that the TV is the durable good which is in almost all Brazilian houses, we can consider the TVDi a way to significantly change the perspective of Brazilian distance learning. Even knowing that the TVDi inclusion in Brazil will not solve the social inclusion problem, it is certain that all its power can improve the digital inclusion, for it will ensure the information access, services and education to people with low purchasing power. [5]

EDUCATIVE BROADCASTING IN BRAZIL

According to the Communication Department, educative broadcasting is the Sound Broadcasting Service (radio) or Sounds and Images Services (TV) intended for the transmission of educative-cultural programs which, besides performing together with teaching systems of any level or modality, aims the basic and higher education, the permanent education and the professional education, besides comprehending educational, cultural, pedagogical and professional orientation activities. The execution of broadcasting services with exclusively educative purposes is granted to legal entities with internal public right, including universities, which will be given the preference to obtain the grant, and foundations privately established and others Brazilian universities. The first educative broadcasting station, the University TV of Pernambuco pertaining to the Education Department, was on TV in 1967. Until 1980´s, educative TV broadcasting in Brazil gave priority to essentially educative programs and in 1997, the Brazilian Association of Public, Educative and Cultural Broadcasting (ABEPEC) was created. In 1999, the participant broadcastings created the RPTV (Public TV Network) which aims at establishing a common and mandatory programming guide to the associated broadcastings. Today, the programming is different from that one in the beginning of educative broadcasting transmissions, that is, it does not have the strict educative

features anymore due particularly to the financial survival of theses broadcastings. It is possible to note, according to legislation, that the programming only admits transmission of programs with educative-cultural purposes. However, there is the option to recreational, informative or sport programs considered educative-cultural since they present instructive elements or educative-cultural focus identified in its presentation. Digital TV implantation in Brazil has been advancing. Some obstacles – among them the situation of commercial broadcastings, political interests, influences (and models) of digital television international systems, legislation ruling the radio broadcasting – still prevent its complete operation, but when it is defined, a social participation never seen before in other historical moments can take place in Brazil, ensuring access to information and culture. [6]

RELATED WORKS

There are several recommendation systems for DTV (Digital Television) designed to offer a distinct personalization service and to help TV viewers to deal with the great quantity of TV programs. Some systems related to the current work are presented here. The AIMED system proposed by [7], presents a recommendation mechanism that considers some TV viewer characteristics as activities, interests, mood, TV use background and demographic information. These data are inserted in a neural network model that infers the viewers’ preferences about the programs. Unlike the work proposed in this paper, which uses the implicit data collection, in the AIMED system, the data are collected and the system is set trough questionnaires. This approach is doubtful, mainly when limitations imposed to data input in a DTV system are considered. In [8] a method to discover models of multiuser environment in intelligent houses based on users’ implicit interactions is presented. This method stores information in logs. So, the logs can be used by a recommendation system in order to decrease effort and adapt the content for each TV viewer as well as for multiuser situations. Evaluating the TV viewers’ background of 20 families, it was possible to see that the accuracy of the proposed model was similar to an explicit system. This shows that collecting the data in an implicit way is as efficient as the explicit approach. In this system, the user has to identify himself in an explicit way, using the remote control. Unlike this system, the proposal in this paper aims at promoting services to the recommendation systems for a totally implicit multiuser environment. In [9], a program recommendation strategy for multiple TV viewers is proposed based on the combination of the viewer’s profile. The research analyzed three strategies to perform the content recommendation and provided the choice of the strategy based on the profile combination. The results proved that the TV viewers’ profile combination can reflect properly in the preferences of the majority of the members in a group. The proposal in this paper uses an approach similar to a multiuser environment, however,

Page 3: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-3

besides the profile combination, the time and day of the week are also considered. In [1] a personalized TV system is proposed loaded in the STB compatible with the Multimidia Home Plataform (MHP) model of the digital television European pattern. According to the authors, the system was implemented in a commercial solution of the MHP middleware, and for that, implemented alterations and inclusions of new modules in this middleware. Offering recommendation in this system requires two important information that must be available: programs description and the viewer visualization behavior. The description of the programs is obtained by demultiplexing and decoding the information in the EIT (Event Information Table) table. EIT is the table used to transport specific information about programs, such as: start time, duration and description of programs in digital television environments. The viewing behavior is collected monitoring the user action with the STB and the later persistence of this information in the STB. The work of [1] is similar to the work proposed in this paper. The implicit collection of data, along with the inclusion of a new module in the middleware architecture, is an example of this similarity. In [10], the Personalized Electronic Program Guide is considered a possible solution to the information overload problem, mentioned in the beginning of this work. The authors compared the use of explicit and implicit profile and proved that the indicators of implicit interests are similar to the indicators of explicit interests. The approach to find out the user’s profile in an implicit way is adopted in this work and it is about an efficient mechanism in the context of television environment, where the information input is performed through remote control, a device that was not designed to this purpose. In [11], the AVATAR recommendation system is presented, compatible to the European MHP middleware. The authors propose a new approach, where the recommendation system is distributed by broadcast service providers, as well as an interactive application. According to the authors, this approach allows the user to choose among different recommendation systems, what is not possible when we have an STB with a recommendation system installed in plant. The AVATAR system uses the approach of implicit collection of user profile and proposes modifications in the MHP middleware to include the monitoring method. The Naïve Bayes [12] is used as a classification algorithm and one of the main reasons for that is the low use of STB resources.

SYSTEM OVERVIEW

The recommendation system proposed in this paper is based on Ginga middleware. As mentioned before, the version used was the open source version of Ginga-NCL middleware. Figure 1 presents its architecture consisting of three layers: Resident applications responsible for the exhibition (frequently called presentation layer); Ginga Common Core,

a set of modules responsible for the data processing, information filtering in the transport stream. It is the architecture core; Stack protocol layer responsible for supporting many communication protocols like HTTP, RTP and TS.

FIGURE 1 – GINGA MIDDLEWARE ARCHITECTURE (ADJUSTED

WITH THE RECOMMENDATION SYSTEM) The proposed system extends the Ginga middleware functionalities including new services in the Ginga Common Core layer. The Recommender module is the main part of the recommendation system and it is inserted in the Common Core layer of Ginga-NCL architecture. The Recommender module is divided in two parts. The first one describes the components integrated to the source code of the middleware such as Local Agent, Schedule Agent, Filter Agent and Data Agent. The second part describes the new component added to the STB: Sqlite [13], a C library which implements an attached relational database. Figure 2 presents the Recommender module architecture.

I. Implemented Modules

This subsection describes the modules added to the Ginga-NCL middleware source code and the extensions implemented to provide a better connection between middleware and the recommendation system. Local Agent is the module responsible for constant monitoring of the remote control. Any interaction between the viewer and the control is detected and stored in the database. The Local Agent is essential for the recommendation system that uses implicit approach to perform the profile. Scheduler Agent is the module responsible for periodically request the data mining. Data mining is a process that demands time and processing, making its execution impracticable every time the viewer requests a recommendation. Scheduler Agent module guarantees a new processing every 24 hours preferably at night, when the STB is in standby.

Page 4: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-4

FIGURE 2 – RECOMMENDER MODULE ARCHITECTURE

Mining Agent is the module that accesses the information in the viewer’s behavior background and the programming data from the EIT and SDT tables stored in cache to perform the data mining. In order to process the data mining, the Mining module has direct access to the database and recovers the TV viewer’s behavior background. From the point of view of the system performance, this communication between mining module and user database is important. Without this communication, it would be necessary to implement a new module responsible for recover the database information and then make such data available to the mining algorithm. The second data set necessary to make possible the data mining is the program guide. The program guide is composed by information sent by providers through EIT and SDT tables. These tables are stored in cache and are available to be recovered and processed by the Mining module. Ginga-NCL Middleware does not implement storage mechanism in cache of EIT and SDT tables. This functionality was implemented by the Recommender system. Filter Agent & Data Agent The raw data returned by the Mining Agent module need to be filtered and later stored in the viewer’s database. The Filter Agent and Data Agent modules are responsible for this function. The Filter Agent module receives the data from the mining provided by the Mining Agent and eliminates any information that is not important keeping only those which are relevant to the recommendation system such as the name of the program, time, date, service provider and the name of the service. The Data Agent module receives the recommendations and stores them in the viewer’s database. If there were many educative programs on open TV, it would be very useful to recommend other educative programs. However, “educative” is one of the many TV program categories. The system can be used inside a distance learning system to recommend several types of

items. For example, the system can be used to create a top-10 question topic; the students would classify extra material with a grade and the best extra materials would be recommended. It would be also possible to have a top-10 favorite and a top-10 best students. Moreover, the system could also provide a way to look for old content interesting for the user to improve what is being studied at that moment.

METHODOLOGY AND TESTS User history and EPG data are necessary to perform the

tests. These data were provided by IBOPE (Brazilian Institute of Public Opinion and Statistics) [14] through a treatment process almost entirely manual in order to be in accordance to the standard format which must be used in the Brazilian digital TV system and also in the tests.

Many technologies have been arising with the aim at identifying behavior standards and its application in the personalization. The recommendation systems operation is found on these techniques and the most used are the Collaborative Filtering and Content-Based Filtering which includes several algorithms for each one. A recommendation system can use only one technique or two together, becoming a hybrid system.

In order to study, analyze and choose an algorithm to be used in Technical module, some information filtering algorithms were tested. The tests were performed in three steps. In the first step, tests were performed with Apriori algorithm. In the second step, the forecast method was used, applying Cosine as measure of similarity. The third step was to compare the results and the operation with both algorithms, analyzing the facilities and difficulties, especially for the implementation.

The association techniques algorithms identify associations between the data registers which are related in some way. The basic premise finds elements which imply the presence of others in a same operation aiming at determining which are related. The association rules interconnect objects trying to show characteristics and tendencies. The association discoveries present trivial and non trivial association. The data was adapted in order to be used in Apriori algorithm, that is, it was submitted to a pre-processing phase. The user history was created from IBOPE data. For the implementation, it is not necessary that the data go through adjustments, as it will be collected in the correct format to be used. The results were satisfactory verifying that Apriori can be applied to the system for it can be adapted to the system needs. [15, 16]

The Cosine is a similarity measure, a forecast method which calculates the similarity between items and users, consults similar items to a given item and matches item content and user profile. The data also had to be adjusted to be used with Cosine. Database in sqlite was used with the EPG and the user history. From these two tables, it was possible to derive two more, one with the profile of the program watched by the user and other with the profile of genres. It was necessary that the EPG passed through a modification which should also occur in the implementation.

Page 5: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-5

A new table was created, identical to the EPG table, but added with fields containing the genres names. According to the adjustment of the program in the genres, these fields were populated with 0 or 1, becoming a matrix. From these tables it was possible to find the Cosine for the programs and genres, the profile and what could be recommended to the user. The results from the Cosine were also satisfactory confirming that this technique can be applied to the system for it can be adjusted to the system needs. [17, 18]

ANALYSIS During the tests, it was possible to note some

particularities. Our system recommends contents based on the programs genres and our analyses were performed according to this standard. With Apriori algorithm, the data are collected in the correct format to be used. For the Cosine, the EPG needs to be changed to a matrix before starting the process of discovering profiles and recommendations.

In a desktop, the feedback of the Cosine calculation is faster in relation to the feedback of Apriori association rules. However, further studies about these algorithms processing in these devices are still being performed. Apriori is able to discover the profile from the standards, but to select the programs to be recommended, another technique must be used and the Cosine can find both the profile and the recommendations.

The Cosine cannot discover these characteristics, but reaches our goal. In order to discover behaviors similar to the association rules, it is necessary to consult the databank. Apriori output must be operated in order to give the correct user profile, that is, the rules must be understood, and that is very hard concerning implementation. The Cosine output is clearer; the result straightly reaches intended goal, allowing the output to be used without the need of a post-treatment.

Regarding the input, there is no need of treatment for Apriori, since all data will be used as they are collected. However, for the Cosine, whenever the EPG is updated, the table containing the EPG matrix must be changed according to the new EPG, becoming something hard to work. The profile of the genres founded by both algorithms is similar.

RESULTS In order to measure the evolution of the recommendation offered to the students viewer, the following formula was applied:

Ef = (α / β) 100 (1) 

Where Ef is the efficacy of the recommendation system, ranging from 0 to 100, α is the recommendation number accepted by the students viewers and β is the number of recommendation presented. In order to monitor these data provided by IBOPE were used. The validation adopted an accuracy formula presented in (1).

FIGURE 3. ACCURACY OF THE RECOMMENDATION SYSTEM

Figure 3 presents the results obtained after 4 weeks of

monitoring considering the best value obtained among the 8 schools analyzed. It is clear that on the first weeks, as the collected data were few, Apriori algorithm did not extract relevant information from the preferences of the group. With the data increase in the visualization background on the third and fourth week, the algorithm obtained better results and the index of recommendation acceptance increased.

FIGURE 4 ACCURACY OF THE RECOMMENDATION SYSTEM PER

SCHOOL Figure 4 presents the accuracy per school. The main

characteristic of the schools is the socioeconomics difference among them. The conclusion is that Apriori algorithm had a good performance unrestricted to the students’ ´socioeconomic profile.

Page 6: Recommender helping viewers in their choice for educational programs in digital tv context

Session S1E

978-1-4244-6262-9/10/$26.00 ©2010 IEEE October 27 - 30, 2010, Washington, DC 40th ASEE/IEEE Frontiers in Education Conference S1E-6

FIGURE 5 – RECOMMENDERTV SYSTEM

Figure 5 shows Recommender system. The application

used as front-end is written in NCL and allows the students to search the recommendation list selecting the education program.

CONCLUSION With the appearance of digital TV, a variety of new

services (in the analogical system, channels) will be available. This information overload requires the implementation of new mechanisms to offer facilities to the students looking for their education programs. These new mechanisms suggesting the viewers programs are known as recommendation systems. A recommendation system compatible with Ginga middleware is presented in this paper and it is implemented according to the standards of the digital television Brazilian system. The recommendation system was modeled considering the current characteristics of the television, and this model can be adjusted to other standards and also to new portable devices which will be on the market. At last, future works can include algorithms of collaborative filtering and also a new architecture using client-server, providing and offering other kinds of personalization services for the users.

ACKNOWLEDGMENT We thank IBOPE for providing real data about the

electronic program guide and also the viewer’s behavior data from March, 05, 2009 to March, 19, 2009.

REFERENCES

[1] H. Zhang, S. Zheng and J. Yuan. 2005, “A Personalized TV Guide System Compliant with MHP”, IEEE Transactions on Consumer Electronics, pages 731-737, Vol. 51, No. 2, MAY 2005.

[2] L.F.G. Soares, R.F. Rodrigues, M.F. Moreno. 2007, “Ginga-NCL: The declarative Environment of the Brazilian Digital TV System”, Journal of the Brazilian Computer Society. V.12, n.4, p.37-46, March 2007.

[3] Souza Filho, G. L., Leite, L. E. C., Batista, C. E. C. F. 2007, “Ginga-J: The Procedural Middleware for the Brazilian Digital TV System.”, Journal of the Brazilian Computer Society, v. 12, n. 4, p. 47-56, March 2007.

[4] Ginga-NCL Virtual STB, (March 2009), Available at: http://www.ncl.org.br/ferramentas/index_30.html.

[5] Silva, Dirceu et al. Possibilidades educativas e de inclusão social e digital com a TVDi: uma breve análise do cenário brasileiro. Universidade Estadual de Campinas, Brasil. Available in: http://www.google.com.br/url?sa=t&source=web&ct=res&cd=1&ved=0CAYQFjAA&url=http%3A%2F%2Fwww.rieoei.org%2Fdeloslectores%2F2907Veraszto.pdf&ei=4ar5S-eeM8imuAeA1PG9Dg&usg=AFQjCNHaebMawhyd-xbrBsw0JjSHbTr7LQ&sig2=VGSGkTroJZu8t3NYiHA9_A. Acess in 2010-05-20.

[6] Fort, Mônica Cristine. Televisão + Educação = Televisão Educativa. Available in: http://www.google.com.br/url?sa=t&source=web&ct=res&cd=4&ved=0CBUQFjAD&url=http%3A%2F%2Fencipecom.metodista.br%2Fmediawiki%2Fimages%2Fb%2Fbf%2FGT10_-_008.pdf&ei=N8n5S5TaDYmHuAfp5Py9Dg&usg=AFQjCNFLgg9ng2elo7UJAjo9dpf8-3I4hg&sig2=RrpI9k-iotN31hCBrc6yNw. Acess in 2010-05-20.

[7] S. H. Hsu, M. H. Wen, H. C. Lin, C. C. Lee, C. H. Lee. 2007. “AIMED – A personalized TV Recommendation System” in Proc 2007 Interactive TV: A Shared Experience. 5th European Conference, EuroITV 2007, Amsterdam, the Netherlands.

[8] Vildjiounaite, E., Kyllonen, V., Hannula, T. and Alahuhta, P. 2008. Unobtrusive Dynamic Modelling of TV Program Preferences. In Proceedings of the Changing Television Environments, 6th European Conference, EuroITV 2008, pages 82-91.

[9] Zhiwen, Y., Xingshe, Z., Yanbin, H. and Jianhua, G. 2006. TV program recommendation for multiple viewers based on user profile merging. In Proceedings of the User Modeling and User-Adapted Interaction, pages 63-82. Publishing Springer Netherlands.

[10] O’Sullivan, D., Smyth, B., Wilson, D. C., McDonald, K. and Smeaton, A. 2004. Interactive Television Personalization: From Guides to Programs. Personalized Digital Television: Targeting Programs to Individual Viewers. L. Ardissono, A. .Kobsa and M. Maybury editors, pages 73-91, Kluwer Academic Publishers

[11] Blanco-Fernandez, Y., Pazos-Arias, J., Gil-Solla, A., Ramos-Cabrer, M.,Lopes-Nores, M., Barragans-Martinez, B. 2005. AVATAR: a Multi-agent TV Recommender System Using MHP Applications. In: IEEE International Conference on E-Technology, E-Commerce and E-Service (EEE '05), pp. 660-665.

[12] Wu, X., Kumar, V., Ross Quinlan, J., Ghosh, J., Yang, Q., Motoda, H., McLachlan, G. J., Ng, A., Liu, B., Yu, P. S., Zhou, Z., Steinbach, M., Hand, D. J., and Steinberg, D. 2007. Top 10 algorithms in data mining. Knowl. Inf. Syst. 14, 1 (Dec. 2007), 1-37. DOI= http://dx.doi.org/10.1007/s10115-007-0114-2

[13] Sqlite, (March 2010) Available at: http://www.sqlite.org/. [14] “IBOPE”. Available in: http://www.ibope.com.br. Access in

December 2009. [15] Witten, I. H.; Frank, E. Data Mining: Practical Machine Learning

Tools and Techniques, 2nd Edition, Morgan Kaufmann, 525 pages, June 2005.

[16] Gatto, Elaine C.; Zorzo, Sergio D. “Sistema de Recomendação para TVDPI,” in 8th International Information and Telecommunication Technologies Symposium. Florianópolis, Santa Catarina, Brasil. 09-11/12/2009.

[17] Torres, Roberto. “Personalização na Internet.” Novatec Editora. 2004. 158p.

[18] Gatto, Elaine C.; Zorzo, Sergio D. Application of recommendation techniques for Brazilian Portable Interactive Digital TV. In: IWSSIP 2010 - 17th International Conference on Systems, Signals and Image Processing. June 17-19, 2010, Rio de Janeiro, Brazil.