Recommender Systems (RSs) can be described as any system that can generate personalized recommendations for the user, guiding them towards more useful and interesting items in a vast space of options. Over the years they have become integral parts of many e-commerce and television industries. They effectively filter large amounts of information based on what items the user is likely going to interact with best. The number of research papers that have been published in this field has increased significantly since recommender systems came into being in the 1990s. There are various applications of this system in todays world such as images, books, movies, music, online shopping, and others. There are two domains that this paper is focused on; Television content and e-commerce websites. With the rapid growth of TV content and e-commerce, now more than ever before, RSs play an important role keeping users tied into an ecosystem for long periods of time and also encourage them to come back for more. The data gathered by such companies, help them get deeper understanding about the user and enable them to get an edge over their rivals. For this reason, Recommender Systems need to be analyzed and implemented accurately. This paper presents an overview of recommender systems, algorithms, user profiling and evaluation.
Originally defined as systems that “people provide recommendations as inputs, which the system then aggregates and directs to appropriate recipients" cite{res:paper}, Recommender Systems (RSs) now have a broader definition, describing them as any system that can generate personalized recommendations for the user, guiding them towards more useful and interesting items in a vast space of options. The point that separates, information retrieval systems from RSs is the concept of personalization of options available to users. General search engines, simply match the inputs of the user against its database and then displays the results based on the degree of match cite{hybrid:paper}. “ In general, recommender systems directly help users to find content, products, or services (such as books, digital products, movies, music, TV programs, and web sites) by aggregating and analyzing suggestions from other users, which mean reviews from various authorities, and users" cite{park:paper}. There are two filtering categories into which RSs falls into. Collaborative Filtering (CF) and Content-based Filtering (CB). CF works by filtering information based on using a user's purchase or evaluation of items history while CB “uses the contents of the documents, as well as the provided ratings, to infer a user profile that can be used to recommend additional items of interest"cite{park:paper}.
newlinehspace*{10pt} In this paper, articles on recommender systems were used that have been published from 1997 to 2015 in order to better understand how they work. This paper is organized as follows: Classifications methods, Algorithms, User Profiling, Results, and Final conclusions
section{Classification Methods}
These recommender system classifiers are based on Ricci et al.,2012. cite{ricci:book}
subsection{Non-Personalized}
These classifiers are used to present the user with a predefined items These recommendations remain the same for all users.
subsection{Content-based Filtering (CB)}
These recommendation systems are used to recommend similar items that the user has purchased or liked in the past. “the basic process performed by a content-based recommender consists in matching up the attributes of a user profile, in which preferences and interests are stored, with the attributes of a content item, in order to recommend to that user new interesting items" cite{veras:paper}. For example, e-commerce website would recommend items similar to the ones a user placed in their wishlist and/or shopping list.
subsection{Collaborative Filtering (CF)}
This system generates different recommendations based off several users' ratings and suggests items that users with similar rating profiles liked previously. This is again based on the degree of similarity between users' rating profiles. This classifier is most commonly used in RSs. cite{veras:paper}
subsection{Data Mining}
These techniques are generally used to extract large amounts of data which allows RS to discover patterns. Data mining is an important tool to predict decisions as well as aid in decision making processes cite{park:paper}.
subsection{Context-awareness}
In this system the personalization results is based on the user's context and resources. This could be location, time, friends and others. Proactive services generally use a user's location to provide “context aware" information. For example, an RS could recommend specific shopping deals to user if they are in a mall.
subsection{Semantic-based}
This method deals with meta-data. “Semantic-based strategy discovers semantic relationships between the users? preferences and the items available in the domain ontology through semantic similarity metrics"cite{veras:paper}.
subsection{Community-based}
Community based filtering involves recommending items to a user that is based off the user's friends preference cite{veras:paper}. This RS is related to a user's social interactions and preferences of their friends.
subsection{Group-filtering}
According to Marilly et al., this filtering is done by producing recommendations for single users and then aggregating all of them and generate a group recommendations. Where as according to Brusilovsk et al., this RS can be used to generate recommendations for a single group profile (a pseudo user) which consists of individual users. This system is highly applicable to the TV content industry where content is consumed by both single users as well as families/friends.