Recommender Systems: Techniques and Uses Research Paper

Introduction

Recommender systems have generated various benefits to both consumers and product providers, especially now with the emergence and popularity of the internet. Technological advancement has to some extent provided the backdrop for recommender systems.

Emerging technologies (iPod, iPhones, and other Mac products) have provided the necessity of recommenders. Without this, downloads of music can have some difficulties. Recommenders have a wide range of applications that include movies, television, and music, to name a few.

First of all, what are recommenders?

Recommender systems provide recommendations on items to consumers. Consumers can have access to so many items on sale over the internet and recommender systems are there to help them make the right choice or pick according to their liking after they have inputted their data in the system.

When consumers purchase items, they provide data according to their purchases, product ratings, user profiles, and likes or dislikes of items, goods, and other consumer products. This information is then processed for comparison in the systems’ databases by software that uses a type of algorithm.

Recommender systems have become a popular technique to prune large information spaces so that users are directed toward those items that best meet their needs and preferences. (Li and Kim, 2004, p. 100)

Recommenders are intelligent systems that employ prediction techniques to determine what is and what is not interesting by learning from the user and sometimes other users. Recommenders use techniques such as information filtering, social filtering, genre-based recommendations, case-based reasoning, and item-filtering (Setten et al., 2004, p. 13).

There are hybrid recommenders too which are combinations of two or more recommender systems and which work by combining collaborative and content-based algorithms.

Recommenders help users to locate and choose items on sale. In the Amazon store for example, as users browse for books, they are guided by a recommender phrase that says: Customers who bought this item also bought… A push of the button can lead to choosing and picking other items of the customer’s liking.

Recommenders are more beneficial if organizations share their resources (i.e. products and customer database) and recommendations boundlessly (i.e. apply recommendation systems on to inter-organizational level), and more importantly, the great business value might be generated during the resource sharing process among the organizations (Weng et al., 2004, p. 32).

This means to say that organizations or business should not isolate their databases or the vast information they have acquired from customers. By sharing their databases, a whole lot more benefits can be derived from it.

Methodology

This brief paper will analyze and examine how recommenders work and their positive purposes in present-day businesses using different technologies and the internet.

First, we delve into the technological aspect and the practical uses and later touch on the fragmentation and homogenizing effects, as observed by some researchers.

Method

There are various ways of analyzing this special type of software, one of which is to dig up from the voluminous literature available, from publications of talks and conferences; the topic of recommenders is a popular one among researchers and software engineers. We have touched on the background literature, analyzed it, and later attempted to arrive at a conclusion and recommendation.

Sampling

The next task is to take a sample from among the many software available which have been applied with much success.

Tàtari (a New Zealand word meaning filter) is software that aims “to provide an open-source tool that researchers can use to develop and evaluate recommender algorithms without the need to create an entire recommender system from scratch” (Hassan and Watson, 2004, p. 47).

Tàtari provides all the background functionality leaving the researcher free to concentrate on implementing and evaluating their recommender algorithm. Thus, Tàtari will minimize the programming effort required and will maximize the time available for evaluating algorithm performance. Tàtari is recommended for researchers seeking advancement in technology.

Piloting Process

Recommenders have been with businesses for decades now. It first gained prominence in promoting consumer products than into the movie and television industries and later the music industry. Emerging technologies have enormously taken advantage of, such as Mac products, e.g. iPod, iPhones, etc.

Ethical consideration

The task now is to discuss possible ethical ramifications on recommender applications. Whilst there are always the negative sides of new applications, especially on technology, and emerging technologies for that matter, recommender systems have aided both consumers or users and the implementers which are the big businesses.

Recommender systems and corresponding software have been the subject of various talks and conferences worldwide. This is because of the benefits they generate for both the business organizations and consumers who are seeking the right products for them.

Recommender systems are tools for researchers and consumers/users. They aid users in their choice of products and also help researchers seeking technological advancements. However, improvements and innovations have still to be implemented on the technology.

There can be ethical aspects that we can discuss here. Despite the recommenders’ usage and practicality to both consumers and businesses, there are negative sides to the sharing of information. Technology has always been ethically questioned in the long run, and recommenders could be another bad side that can be cited.

Some say that recommender systems can cause fragmentation; meaning users will choose only those they like and every user will have a unique preference, or they will not have common choices anymore. The opposing view is that recommenders will have homogenizing effects, which means users are being pushed to choose on the same items or they can share information.

While fragmentation is a negative result for recommender systems, homogenization counters this and is said to produce a network wherein users are more similar or are using the same items and products.

Data Analysis

Recommenders can be taken from successful researchers. One is that which combines three prediction techniques to arrive at a recommender or prediction.

One weighted hybrid TV recommender also combines three prediction techniques: a stereotype-based technique, a technique based on explicitly provided interests from the user, and a technique that employs a Bayesian belief network that learns from implicitly gathered user behavior data. The weights used to combine the predictions are based on confidence scores provided by the individual techniques. (Setten et al, 2004, p. 14)

Switching hybridization is another technique (Setten et al, 2004). This is the prediction strategy whereby “the decision is based on the most up-to-date knowledge about the current user, other users, the information for which a prediction is requested, other information items and the system itself” (p. 13).

On the other hand, Wei (cited in Weng et al, 2006, p. 32) proposed a multi-agent-based recommender system in which a recommender system is considered as a marketplace consisting of one auctioneer agent and multiple bidder agents.

There is a proposed distributed recommender system that consists of multiple recommender systems (or recommender peers) of different organizations (Weng et al, 2006, p. 32).

It works this way. When anyone of these recommender peers receives a request from a user, not only does it generate recommendation from its resources, but it also consults (and interact) with other recommender peers for the suggestion to improve its recommendation quality to the user.

Reliability

We can always say that there are negative sides to this application. But in the long run, recommenders can be very helpful and reliable for consumers and businesses. As already said here, recommenders have been in use for decades now, even before the popularity of the internet. Now, with everything interconnected, databases are filled with so much information for users and businesses. Databases have so much use for businesses and organizations.

Validity and Limitations

Recommenders can be limited when businesses do not share their databases. The vast information can be very useful, as experts say if that information stored by various businesses can be shared for the advantage of the users. Of course, there are limitations to their sharing.

As to their validity, it all depends on the information stored therein. Recommenders still depend on the users. It’s still man against machine.

Reference

Hassan, H. and Watson, I., 2004. Tàtari: An Open Source Software Tool for the Development and Evaluation of Recommender System Algorithms. In Carbonell, J. and Siekmann, J. eds. Knowledge-Based Intelligent Information and Engineering Systems: 8^th International Conference, KES 2004, Wellington, New Zealand, September 2004 Proceedings, Part II. New York: Springer. p. 46.

Li, Q. and Kim, B. M., 2004. Constructing User Profiles for Collaborative Recommender System. Korea: Department of Computer Science. In Yu, J. X., Lin, X., and Hongjun, L., Advanced Web Technologies and Applications (6^th Asia-Pacific Web Conference, APWeb 2004, Hangzhou, China, Proceedings), p. 100.

Setten, M. van et al., 2004. Case-Based Reasoning as a Prediction Strategy for Hybrid Recomender Systems. In: Favela, J., Menasalvas, E. and Chavez, E. (Eds.). Advances in Web Intelligence: Second International Atlantic Web Intelligence Conference, AWIC 2004. New York: Springer. pp.13.

Weng, L., Xu, Y., Li, Y., and Nayak, R., 2006. A Fair Peer Selection Algorithm for an Ecommerce-Oriented Distributed Recommender System. In Li, Y., Looi, M., and Zhong, N. eds. Advances in Intelligent IT: Active Media Technology 2006. Fairfax, VA: IOS Press, Inc. p. 31.