department of informatics

Fuzzy-based recommender system (FRS)

Description: 

Finding information on the World Wide Web is not an easy task. One of the main problems is the exponential growing of data available on the Internet, which can be considered as an almost infinite, non structure and evolving network.

Different techniques have been introduced in order to deal with information overload such as: search engines, semantic web, recommender systems, among others.

Recommender systems are computer-based techniques used to reduce information overload, and to provide recommendation of products likely of interest by an user given some information about the user’s profile. This technique is mainly used on eCommerce to suggest items that a customer is assumably going to buy.

The use of recommender systems in eGovernment is a new research topic, and aims to solve the problem of information overload on eGovernment services, which could help to improve the interaction between public administrations, citizens and the private sector.

Recommender Systems for eGovernment

The term eGovernment refers to the use of information technologies to improve the interaction between public administrations, citizens, and the private sector. Three types of relationships are defined for eGovernment: Administration to citizens (A2C), Administration to Business (A2B), and Administration to Administration (A2A).

 

The Information Systems (IS) research group at the University of Fribourg developed, in collaboration with other domain experts, a process oriented stage model. This model differentiates between the following tree process levels: Information and Communications, Production, and Participation.

The lowest level provides information and communication for eGovernment; it focuses on the design of communal web portals. The second level consists of the actual public services (electronic procurement, taxation, and electronic payments, among others). The third level refers to citizen participation.

This project focuses on the participation level, specifically on eDemocracy, and uses the approach proposed by Meier in his book "eDemocracy & eGovernment". Meier mentions the importance of citizen participation in eDemocracy (eElection and eVoting) and defines eDiscussion as a stage where citizens could know more about the candidates or the subject in a voting process. It uses information and communication technologies such as forums, decision aids, and subscription services, among others, to aid voters in making decisions. In the same way, once an eVoting or eElection process has been completed, another stage defined as  ePosting. This stage facilitates the publication of results, and it gives voters the possibility to open discussion channels about the process. 

 

 

Fuzzy Recommender System Architecture

This project introduces a Fuzzy Recommender System (FRS) for eElections which provides information about closest candidates to a voter, and it provides the distribution of political parties organized in fuzzy clusters.

The recommendation process is given in three steps. In the first step, voters and candidates must create their profiles using a fuzzy interface, described in more detail in the following section, to be stored in a database.

In the second step, once all necessary profiles have been created, the user selects the recommendation target and the type of output (Top-N Recommendation or Fuzzy Cluster Analysis). In the final step, once all information has been computed by the recommendation engine, the user receives the recommendation in the pre-established format. The architecture of the FRS is presented below.

 

Fuzzy Cluster Analysis

Fuzzy clustering analysis allows voters to visualize and analyze relationships to closest “neighbors”. Fuzzy clustering analysis differs from classic clustering (sharp clustering) in that the observations belong to one and only one cluster. Moreover, classic clustering makes no use of gradual membership.

Top-N Recommendation

The top N candidates similar to voter v are generated using the bi-dimensional "Fuzzy Profile". The distances of all candidates with respect to voter v are computed and the N closest candidates are displayed

Poster

 

Relevant Documents and Files: