e-Business

An Online Shop in the Area of Technical and Scientific Publications

José António S. Pereira, Paulo Pita, Élsio Santos and Joaquim Filipe

Superior School of Technology, Polithecnical Institute of Setúbal, Rua Vale de Chaves, Setúbal, Portugal

Keywords: e-Commerce, User Profiling, Context-based Software, Machine Learning, Data Mining, Recommender

Systems, Intelligent Systems.

Abstract: The explosive growth of the world-wide-web and the emergence of e-commerce enabled the development

of recommender systems that became to an independent emerged research area in the mid-1990s. The

recommender systems are used to solve the prediction problem or the top-N recommendation problem.

However, recommendation systems feel ever more the pressure related to a change on users habits. In order

to capture users interests it is necessary a representation of information about an individual user. Our Online

Shop in the Area of Technical and Scientific Publications intends to add the best of the user-based

collaborative filtering and content-based collaborative filtering methodologies into a single hybrid

methodology in order to answer some issues raised about new users and new items added to the

recommender system. And also try to combine inference and prediction to assist the user in finding content

that is of personal interest or even combine data mining techniques to provide recommendations.

1 INTRODUCTION

The RS (Recommender systems) became to an

independent emerged research area in the mid-

1990s. And their foundations are based on work

done in the fields of cognitive science, information

retrieval, approximation theory, forecasting theories,

consumer choice modelling in marketing and also

have links to management science (Adomavicius,

G., Tuzhilin, A., 2005).

According to (Deshpande and Karypis, 2004) ,

the recommender systems are personalized

information filtering technology used to either

predict whether a particular user will like a

particular item (prediction problem) or to identify a

set of N items that will be of interest to a certain user

(top-N recommendation problem).

The problem of prediction and recommendation

is increased, because users have their own culture,

expectations, commitments and beliefs, behaving

differently when they act alone or in an organized

way.

In order to capture user’s interests, knowledge,

background, skills, goals, behaviour interaction

preferences, individual characteristics and the user’s

context it is necessary a representation of

information about an individual user (i.e. user

profile) that is essential for the (intelligent)

application we are considering (Schiaffino and

Amandi, 2009).

Another pressure that recommendation systems

feel is related to a change in habits of users, now

more than ever they use e-commerce to make their

purchases, express their views (e.g. commenting,

rating) on collaboration environments and

maintaining links with friends and family; or simply

users with the same preferences.

2 MAIN RS APPROACHES

Most of the researchers agree in the existence of two

main approaches for recommendation systems –

item-based analysis and user-based analysis – the

first approach determines items that are related to a

specific item (e.g. when a user likes a particular

item, all of which are related are recommended), the

second approach uses personal user information to

suggest the best recommendations (e.g. based on

profile information, user actions, and lists of the

contacts or user's friends) (Patel and Balakrishnan,

2009).

In order to overcome some disadvantages of

approaches exclusively item-based or content-based

289

António S. Pereira J., Pita P., Santos É. and Filipe J..

e-Business - An Online Shop in the Area of Technical and Scientiﬁc Publications.

DOI: 10.5220/0004163402890293

In Proceedings of the International Conference on Data Communication Networking, e-Business and Optical Communication Systems (ICE-B-2012),

pages 289-293

ISBN: 978-989-8565-23-5

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

and pure user-based methods of recommendation,

hybrid methods have been created based upon

collaborative filtering and content-based, which

maintains user profiles based on content analysis,

and directly compares these profiles to determine

similar users for collaborative recommendation

(Balabanovic and Shoham, Y., 1997).

To capture the context in which

recommendations are made the multidimensional

approach to recommendations extended to support

additional dimensions capturing the context in which

recommendations are made (Adomavicius and

Tuzhilin, 2001).

All of the work is not only focused on

recommending items to users and users to items, but

takes into consideration additional contextual

information such as time, place, the company of

other people, and other factors affecting

recommendation experiences (Adomavicius et al.,

2005).

Although these different approaches there are

three main recommendation methods, as can be seen

in Figure 1, for finding similar items and similar

users, and can be applied to each of them, techniques

based on heuristics or on models for the rating

estimation (Adomavicius and Tuzhilin, 2005).

Figure 1: Methods used in recommendation systems -

Adapted from: (Adomavicius and Tuzhilin, 2005).

The RS now have the capability to capture

context, through several methods of

recommendation and use of techniques based on

models or heuristics.

To unify user-based and item-based collaborative

filtering approaches (Wang et al., 2006) uses the

similarity fusion. This unification allows the

estimation of final rating by fusing predictions from

three sources: i) predictions based on ratings of the

same item by other users; ii) predictions based on

different item ratings made by the same user; iii)

ratings predicted based on data from other but

similar users rating other but similar items.

The user-based collaborative filtering has been

proven to be the most successful technology for

building recommender systems so far, and is

extensively used in many commercial recommender

systems, although content-based collaborative

recommendation system solves many of the

problems (Patel and Balakrishnan, 2009).

3 OBJECTIVES

The main objective is to develop a website for

electronic commerce oriented to the publication of:

(Scientific Magazines, Books, E-books and

Articles).

Although the existence of several online e-

commerce systems, the purpose is to develop a

flexible website, better than most of the already

existing systems giving each user a different

experience with the online store based on his profile.

The focus is on RS, and should be considered the

several approaches previously mentioned by

focusing on user-based collaborative approach by

applying filtering techniques based on models and

heuristics, putting an emphasis on the user profile.

The differentiating aspect of this system is that it

must be sufficiently intelligent to take the initiative

and proactively recommending the user, content that

we believe is of interest to him.

The user profile must contain essential

information about an individual user and the

motivation of building user profiles is that users

differ in their interest, preferences, backgrounds and

goals when using software applications, discovering

these differences is essential to provide customized

services (Schiaffino and Amandi, 2009).

For the content of the user profile this project

will not only consider the content provided by the

user explicitly, but must infer unobservable

information about users from observable information

about them (Zukerman and Albrecht, 2001).

Figure 2: Predicted observations strategy - Source: (Oard

and Kim, 1998).

It should be combined inference and prediction

to assist the user finding content that is of personal

ICE-B 2012 - International Conference on e-Business

290

interest. This combination helps using implicit

feedback, because it is based on previous

observations, which are used to predict user

behaviour in response to new information, and then

the inference phase seeks in order to estimate the

value of information based on the predicted

behaviour, as Figure 2, shows (Oard and Kim,

1998).

4 DISCOVERY OF THE

FUNCTIONALITIES

To achieve the desired objectives it is necessary to

study the e-commerce competitors and see what kind

of recommendation technology they use.

A comparative study was conducted to analyze

the market and see what other e-commerce sites

have to offer. This comparative study was based on

the direct experience of use.

High-level functionalities have been defined in

order to compare these functionalities between some

of the most popular e-commerce sites. The sites

selected for this study were Amazon.com, FNAC,

Pixmania and Barnes & Noble. Figure 3 shows the

results of the comparison of the functionalities.

Figure 3: Functionalities comparative study.

The results show that Amazon.com has all the

features that our website is intended to have.

In addition to the results, and due to the study by

(Schafer et al., 1999), when he established a

comparison of six e-commerce sites for business,

has allowed a nice overview on the types of

applications used, the interface of recommendation,

recommendation technology and how to find the

recommendations.

As our focus is on Amazon.com as a model to

follow, we show in Figure 4 some of the

applications used to interface recommendation,

recommendation technology and how to find

recommendations.

Figure 4: Amazon.com Recommender System - adapted

from: (Schafer et al., 1999).

This research shows some of the applications

used by Amazon.com.

These applications support the recommendation

of books frequently purchased by customers who

purchased the selected book, recommend authors

whose books are frequently purchased by customers

who buy books by the author of the book selected.

Such applications also do the notification of new

books added to the catalog requests and provide

recommendations based on research performed and

persistent data.

They also offer the possibility of registered users

receive text with recommendations based on the

opinion of other registered users.

Beyond the implementation of these

functionalities mentioned, our website will apply the

implicit feedback techniques to avoid the effort of

cognitive load in assigning precise ratings to large

user populations and thus contributing to avoid the

dispersion of data within these populations. These

techniques seek to avoid this bottleneck inferred

from observations that are available to the system,

something similar to the ratings assigned by a user.

According to (Oard and Kim, 1998) in addition

to explicit ratings were identified three major

categories of potentially useful behavior

observations: examination, retention and reference.

As Figure 5 shows, the category Examination

extends beyond a single interaction between the user

and the system it is characterized by repetition of the

previous user behavior. The category of Retention

aims to group these behaviors that suggest some

future intention to use an object. Finally, the

e-Business - An Online Shop in the Area of Technical and Scientific Publications

291

Reference category is distinguished by the

opportunity to direct observation of negative

evaluations.

Figure 5: Observable behaviour for implicit feedback-

source: (Oard and Kim, 1998).

For the contents of the user profile our website

should have all the information cited above in

Section 1.

Finally our website should not just allow the user

to provide the content about their profile, but also

should be able to infer the information which is not

observable about their profile by using techniques

based on Machine Learning. The user or customer

profile is used to make personalized offers and to

suggest or recommend products the user is supposed

to like (Schiaffino and Amandi, 2009).

5 ISSUES

Taking into account the several approaches and the

main recommendation methods as well as based

techniques, the objectives defined in Section 3 and

the functionalities mentioned in Section 4, many

issues arise.

The companies that use Recommender Systems

in their e-commerce are facing some issues such as:

i) the need to have large amounts of data; ii) based

on the preferences and previous behavior of the user;

iii) the need for a large number of variables.

The first issue is related to the amount of data

and the time required to provide effective

recommendations by RS. It is necessary to save the

data about the items, as well as all user behavior and

profile. The lack of data could become a problem.

This problem is related to the occurrence of a

new user, because it has to rate a sufficient number

of items before a content-based RS can really

understand the user’s preferences and present the

user with reliable recommendations. Therefore, a

new user, having very few ratings, would not be able

to get accurate recommendations (Adomavicius and

Tuzhilin, 2005).

The lack of data is also related to a new item

added to RS, because it relies solely on users

preferences to make recommendations. Therefore,

until the new item is rated by a substantial number

of users, the recommender system would not be able

to recommend it (Adomavicius et al, 2005).

The second issue takes into account trends and

user intentions. Trends are based on past behavior,

and the user can change their tendencies; about the

intentions of the user, it does not always have the

same intentions when browsing a site. The problem

is related to changes in data and user preferences.

The last issue is related to the need of adding

contextual information, and will raise the problem of

complexity because there are a larger number of

variables.

This complexity problem can also be related to a

large retailer that might have huge amounts of data,

tens of millions of customers and millions of distinct

catalog items, and can also be related to old

customers who may have an excess of information

based on thousands of purchases and ratings (Linden

et al., 2003)

6 CONCLUSIONS AND FUTURE

WORK

Although some major issues that have been

mentioned, obviously our web site is by its nature

limited by the peculiarities of the data and the

recommendation domain.

We hope to achieve with this web site an

improvement over existing approaches from the use

of user profiles that contain information about user’s

tastes, preferences, actions and needs.

The profiling information can be elicited from

users explicitly, e.g., through questionnaires, or

implicitly (e.g. learned from their transactional

behavior over time).

We intend to use techniques for content-based

recommendation such as Bayesian classifiers, and

various machine learning techniques, including

clustering, decision trees, and artificial neural

networks.

The methods that we intend to use for searching

similar items and similar users, should apply,

techniques based on heuristics or on models for the

rating estimation.

Even though is no more than an idea, our future

work shall be to find some way of aggregating the

best of the user-based collaborative filtering and

ICE-B 2012 - International Conference on e-Business

292

content-based collaborative filtering methodologies

into a single hybrid methodology.

If the peculiarities of the site allow, this could be

considered an approach based on multidimensional

data model used for data warehousing and OLAP

(Online Analytical Processing) applications in

databases on hierarchical aggregation capabilities,

and on user, item and other profiles defined for each

of these dimensions (Adomavicius et al, 2005).

ACKNOWLEDGEMENTS

We would like to thank the Polytechnic Institute of

Setubal, School of Technology of Setubal, for

supporting the research work reflected in this paper,

presented at ICE-B 2012.

REFERENCES

Adomavicius, G., Tuzhilin, A.: Using Data Mining

Methods to Build Customer Profiles. IEEE Computer

34(2) (2001)

Adomavicius, A., Sankaranarayanan, R., Sen, S., Tuzhilin,

A.: “Incorporating Contextual Information in

Recommender Systems Using a Multidimensional

Approach,” ACM Trans. Information Systems, vol. 23,

no. 1, Jan. (2005)

Adomavicius, G., Tuzhilin, A.: “Toward the Next

Generation of Recommender Systems: A Survey of

the State-of-the-Art and Possible Extensions”, vol. 17,

no. 6, pp. 734-749, (2005)

Balabanovic, M., Shoham, Y.:‘Fab: Content-Based,

Collaborative Recommendation,’ Comm. ACM, vol.

40, no. 3, pp. 66-72, (1997)

Deshpande, M., Karypis, G.: “Item-Based Top-N

Recommendation Algorithms”, ACM Trans. Info.

Systems, vol. 22, no. 1, pp. 143-177, (2004)

Linden, G., Smith, B., York, J.:“Amazon.com

Recommendations: Item-to-Item Collaborative

Filtering,” IEEE Internet Computing, Jan./Feb. 2003.

Oard, D. W., Kim, J.: Implicit feedback for recommender

systems. In Recommender Systems. Papers from 1998

Workshop. Technical Report WS-98-08. AAAI Press

(1998)

Patel, A., Balakrishnan, A.: Generic framework for

recommendation system using collective intelligence -

International Conference for Internet Technology and

Secured Transactions, (2009)

Schafer, J. B., Konstan, J. and Riedl, J.: ‘Recommender

Systems in E-Commerce’. In: EC ’99: Proceedings of

the First ACM Conference on Electronic Commerce,

Denver, CO, pp. 158-166. (1999)

Schiaffino, S., Amandi, A.: M. Bramer (Ed.): Artificial

Springer-Verlag Berlin Heidelberg (2009)

Wang, J., de Vries, A. P. M., Reinders, J. T., “Unifying

user-based and item-based collaborative filtering

approaches by similarity fusion”, In SIGIR '06: Proc.

of the 29th annual int’l ACM SIGIR Conf. on R&D in

information retrieval , pp. 501-508 (2006)

Zukerman, I., Albrecht, D.: Predictive Statistical Models

for User Modeling. User Modeling and User-Adapted

Interaction 11(1-2), 5–18 (2001)

e-Business - An Online Shop in the Area of Technical and Scientific Publications

293