Using a Fuzzy-Based Cluster Algorithm for Recommending Candidates in E-Elections

1 Motivation

The rapid increase of information on the Internet is currently a key issue when one is looking for relevant information. In the political sector, the amount of available information about candidates and political parties is also drastically increasing. This is becoming a significant issue for voters when they face election processes that require them to select their representatives from a big list of candidates since, in many cases, the candidates are relatively unknown to their constituents.

In this chapter, the use of recommender systems for eElections is presented as an alternative to solve the problems of information overload.

Recommender systems are computer-based techniques that attempt to present information about products that are likely to be of interest to a user. This technique is mainly used in eCommerce in order to provide suggestions on items that a customer is, assumable, going to like.

Yager (2003) distinguishes between recommender systems and targeted marketing by considering that a recommender system is a “participatory” system in which the user intentionally provides information about his preferences. In a targeted marketing effort, the recommendation is based on extensional information, which is nothing but information predicated upon the actions or past experiences with respect to specific objects.

A recommender system for eCommerce specifies two basic entities, which include the user (i.e., customer) and the item (i.e., product). The main goal of this type of recommender system that is used in eCommerce is to basically increase the sales of products. The main problems of recommender systems, according to Vozalis et al. (2003), include the following:

• •

  Quality of Recommendations: The information received from a recommender system must be reliable; for that reason, recommender systems should minimize the number of false positive results (i.e., the products that the customer does not like).

• •

  Sparsity: A recommendation system is related to the number of recommendations made by customers. The sparsity problem of recommender systems emerges when the number of rated items is small compared to the total number of items, which leads to weak recommendations since the recommender systems are based on similarities between individuals.

• •

  Scalability: Increasing the number of users and products elevates the cost in terms of computations in recommender systems.

• •

  Lost of Neighbor Transitivity: The correlations between users cannot be expressed unless they have purchased and rated common items.

• •

  Synonymy: Recommender systems generally cannot link products with different names that belong to the same category.

• •

  First Rater Problem: A product cannot be recommended unless another customer has previously rated it.

• •

  Unusual User Problem: This problem refers to users who cannot define their opinion about a product. This causes inconsistent recommendations.

The most-used techniques in recommender systems are based on collaborative filtering technologies according to Guo et al. (2007) and Sarwar et al. (2001). They include collaborative filtering algorithms that are memory-based (i.e., user-based) and model-based (i.e., item-based).

• •

  Memory-based collaborative filtering algorithms: These techniques are based on the computation of “neighborhood formation” that uses the user-item matrix R, which contains the ratings of items by users (users are not required to provide their opinion on all items). This may cause the previously mentioned problem of sparsity. The most common techniques used to reduce the effect of sparsity consist of default voting, preprocessing using averages, the use of filterbots, and the use of dimensionality reduction techniques.

• •

  Model-based collaborative filtering algorithms: The model-based (i.e., item-based) collaborative filtering algorithm uses the set of items that the active user u_a has ranked to compute the similarities between this item and a target item (i_j) and to select the n most similar items. In order to compute similarities between items i_i and i_j, model-based techniques isolate the users that have rated both items.