RECOMMENDATION OF WEB RESOURCES FOR ACADEMICS

Architecture and Components

Pavel Cech and Vladmir Bures

Faculty of Informatics and Management, University of Hradec Kralove, Rokitanskeho 62, Hradec Kralvoe, Czech Republic

Keywords: Recommendation system, recommendation system architecture, web services, heterogeneous tools.

Abstract: Vast amount of resources in digital libraries and on the Internet makes the selection of relevant and

appropriate resources rather difficult especially for novices and less experienced academics such as

students. The article presents an architectural solution for personal recommendation system of web pages

targeted to the academic sphere. The solution aims to meet the objective of minimizing the overheads and

provide support for PDF documents. The solution is based on a set of heterogeneous components connected

via web services.

1 INTRODUCTION

Vast amount of resources in digital libraries and on

the Internet makes the selection of relevant and

appropriate resources rather difficult especially for

novices and less experienced academics such as

students. The article presents an architectural

solution for personal recommendation system of

web pages targeted to the academic sphere. For

inferring potentially interesting pages the content

filtering and collaborative filtering approaches were

used. There are many theoretical papers on how to

infer similar web pages and also many case studies

with particular implementations (Balabanovic 1997,

Balabanovic & Shoham 1997, Basu 1998, Sarwar

2000, Herlocker 2004). The goal in this article is to

show how one practical solution can be implemented

using various tools. The problem is that the tools

might often be fairly heterogeneous i.e. implemented

in different programming environments. Thus the

solution has to offer a way how to split the system.

The solution is also different in that it puts stress on

the usability of the system especially by academics.

The aim of the technological solution was to offer

user friendliness both for rating and also for

recommending (i.e. the overheads of doing the rating

has to be minimized) and to support also formats

other then simple HTML. In particular, the focus of

the project was first to integrate the recommendation

system into the internet browser so that user can rate

and be recommended while browsing the web pages;

second, to support PDF (Portable Document Format)

documents since the system is targeted for the

academic audience and PDF is de facto standard for

academic papers.

2 SYSTEM ARCHITECTURE

The architecture of the system is guided by the need

for centralized store of user preferences on one hand

and on the other hand by the necessity to track the

users’ behaviour. Therefore, the whole system is

split into components. Splitting the whole system

into several components makes the system more

flexible and easier to develop. However, the main

reason for dividing the system to components rose

from the implementation tier in the form of a

constraint of available tools, especially for PDF

parsing and user tracking. To make the system use

various tools and technologies it was necessary to

implement corresponding parts in different

programming environments. Thus, the whole system

is divided into several components. Each component

has different role in the system and can be

programmed in different programming languages.

Since the individual components are programmed in

different programming languages the appropriate

communication infrastructure, with common

communication protocol had to be selected. Thus,

components are interconnected using the web

services infrastructure that enables the cooperation

437

Cech P. and Bures V. (2007).

RECOMMENDATION OF WEB RESOURCES FOR ACADEMICS - Architecture and Components.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Web Interfaces and Applications, pages 437-440

DOI: 10.5220/0001284304370440

 SciTePress

Figure 1: Architecture of the recommendation system.

of programs developed in different programming

languages. The overall architecture and the

communication lines between components are

illustrated in the figure 1.

2.1 User Component

The user component is responsible for monitoring

user behaviour and interacting with him. The user

component provides the interface to the system and

to the functions ensured. User component offers the

interface to search the web for resources. The system

is using the web services of various search engines

to perform the first search result. Those results

obtained from the search engine (if not specified

otherwise) are then refined based on the information

in the user profile.

The comfort offered by the user component to

the user is critical for the user acceptance of the

system. To minimize the overheads necessary while

rating resources the user component is integrated

directly into the web browser window. Thus, a part

of the user interface of the system is in the form of

an explorer window bar (placed vertically on the left

hand side) and explorer toolbar (placed horizontally)

– see figure 2.

However, there are features that cannot be placed

in the toolbar and the window bar of the browser

since the user probably wants to have the maximum

space for the opened resource. Therefore, some of

Legend:

request

response

User

Component

User Profile

Component

Collect

Component

Ratings

Use

results

displayed

initiated

resources to

process

resources to

compute

similarity

specified

resources to

process

resources to

recommend

WEBIST 2007 - International Conference on Web Information Systems and Technologies

438

the details and reports that the system produces are

to be presented in the main window of the web

browser as a normal web page.

Since the most common browser is today MS

Internet Explorer, it had been selected for the user

interface layer. Hence, integrating the system user

interface directly to the MS Internet Explorer

window required the use of the COM (Component

Object Model) technology. However, the similar

user interface could be offered to Mozilla Firefox

users as well. This, however, was not covered in the

project.

The remote communication mainly between the

user component and the user profile component

required including the asynchronous processing of

requests and the display of the results. For this

reason the user component was developed as a

multithreaded application. The coordination of

individual requests as threads was provided by one

main thread that communicates with user interface

(which is usually not thread safe).

2.2 Search Component

Search component’s role is to query the search

engines and obtain result. Typically most of the

major search engines provide web services or other

interfaces to programmatically use the searching

capabilities. The search component waits for the

command of a user component to start the search.

The results from the search engine can then be

refined by computing the similarity and filtering out

those resources that do not correspond with the

preferences in the user profile. Before the similarity

can be computed the resources are processed in the

collect component.

2.3 Collect Component

Collect component is responsible for processing the

resources and for storing the resource information.

Typically the collect component obtains resource

from the search component or directly from the user

component, in case user gets the resource in other

ways than searching. The resources can also be

obtained from the digital libraries and, if appropriate

and the communication interfaces are specified, the

collect component can also supply some of the

resources into the digital library.

The obtained resource is first checked on the

type so that the particular parsing and extracting

engine can be used. Currently, only the resources in

the PDF (Portable Document Format) are supported.

The extension to Microsoft Word Documents and

the resources in XML or HTML pose only a minor

problem.

If the resources are parsed successfully, further

information is attempted to be extracted. The

information being extracted concerns title, authors,

keywords and an abstract. Optionally the publisher

and other information used for citations and

referencing can be included in the system. If

extraction does not succeed then it can be filled

manually using the user component interface in the

internet browser. The resource then is processed

against the terms identified and the normalized

frequencies are computed.

2.4 Recommendation Component

The recommendation component is the core of the

system. It has the role of the server providing

services for other components namely user

component. The main task of the recommendation

component is twofold: First, to obtain the necessary

resource information about a resource and rating

information of a particular user and store such data

in the database; Second, to infer the rank of searched

documents according to the preferences of users.

If request for recommending possibly interesting

resources is sent, then the recommendation

component computes the similarity between the user

preferences stored in the user profile and the

resources stored in the database or obtained from the

search engine. In this way the recommendation is

based on the content of the resources and the user

preferences i.e. content filtering (Herlocker 2004).

The weight of the recommended resources is given

by the similarity function.

The recommendation can also be based on the

computation of similarities between user preferences

in which the users with similar user preferences are

determined and the system then recommends the

resources that similar users have rated as interesting

i.e. collaborative filtering (Herlocker 2004).

2.5 User Interface

As stated above, the system was built with usability

in mind. The figure 2 shows window of the MS

Internet Explorer with recommendation system user

interface. First, the toolbar provides means for

documents rating. The rating can be done explicitly

or user can enable the implicit rating based on the

time spent on a page and other patterns (Herlocker

2004). The bar on a left hand side then serves for

recommendation or assisted search. The list of

recommended resources is displayed as hyperlinks

so that they can be used instantly.

RECOMMENDATION OF WEB RESOURCES FOR ACADEMICS - Architecture and Components

439

Figure 2: User interface of the recommendation system integrated to the internet browser window.

The user can select to open the resource or to see

additional information and metainformation about

the resource. In the figure 2 the abstract and rating

information is being displayed.

3 CONCLUSIONS

The future trends in the Internet tend to

personalization. The paper presented a solution of

recommendation system based on the user profile

and heterogeneous components connected via web

services. The prototype developed under the code

name Personal Recommendation (PRECO) is

located on the following website http://preco.uhk.cz.

ACKNOWLEDGEMENTS

This paper is partially supported by AMIMADES,

the GACR project No. 402/06/1325.

REFERENCES

Balabanovic, M., 1997a. An adaptive web page

recommendation service, In Proceedings of the 1st

International Conference on Autonomous Agents,

California.

Balabanovic, M., Shoham, Y., 1997b. Fab: content-based,

collaborative recommendation, Communications of

the ACM, Volume 40 Issue 3.

Basu, C., et al., 1998. Recommendation as classification:

using social and content-based information in

recommendation, In: Proceedings of the 15th National

Conference on Artificial Intelligence, California.

Herlocker, J., L., et al., 2004 Evaluating collaborative

filtering recommender systems, ACM Transactions on

Information Systems, Volume 22 , Issue 1, January

2004, pp. 5 – 53

Sarwar, B., M., et al., 2000 Analysis of recommendation

algorithms for E-commerce, In Proceedings of the 2nd

ACM Conference on Electronic Commerce, New

York.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

440