A Hybrid Context Aware Recommender System with Combined Pre and Post-Filter Approach

Introduction

There is tremendous information available on the net today, and that’s why it is not convenient to find the exact information that is required. There are various personalization techniques in the market to overcome this problem, one aspect of personalization is the recommendation system which is explained by Bobadilla, Ortega, Hernando and Gutiérrez (2013). Recommendation system gained popularity in the 1990’s because of its most popular algorithm collaborative filtering, described in detail by Ekstrand, Riedl, and Konstan (2011).

A new hybrid approach has been proposed in this paper that combines the pre- filter and a variant post-filter approach. The key logic behind this approach is that - for a particular user, different contextual information holds different importance. Let’s consider an example of selecting a movie. One of the important factors for a user could be his/ her companion with whom user wants to watch the movie. And user may be least bothered about the weather. So, based on this information, pre- filter or post- filter approaches can be applied to get the recommendations. And thus, the proposed hybrid approach will be able to provide more appropriate recommendations for a user.

In order to achieve the goal of proposing an improved recommender system and to accomplish the above stated methodology, the following major objectives are developed for this research work:

• •

  To evaluate the potential of contextual information, pre-filtering and post-filtering approaches in recommender systems.

• •

  To explore the existing recommender systems that uses only single filtering approach which are stated above.

• •

  To derive a new and efficient approach to post-filtering and utilize it into the proposed hybrid approach.

• •

  To come up with an algorithm that is capable of recommending relevant movies to the user.

• •

  To use the well-known Movielens dataset to check the working of our proposed approach.

• •

  To interpret and analyze the performance of the proposed approach.

• •

  At last, to compare the efficiency of the proposed approach with existing one, based on the outcomes of the research.

The rest of the paper is organized as follows: Next section describes the related work done in the domain of context aware recommendation system (CARS). After that, the architecture of the proposed approach is explained in detail. The implementation and experimental evaluation of the proposed recommender system have been presented after that. And last section concludes the outcomes and results of the research work.

Background

A recommender system has become such an important aspect in almost all the e-commerce applications. They are being used in applications such as for tourism, mobile, music, news, social networking sites, etc. as mentioned by Lu, Wu, Mao, Wang and Zhang (2015). They typically work on user’s preferences regarding various items such as movies, music, tourism, taxi, etc. L. Chen, G. Chen, and Wang (2015) and Isinkaye, Folajimi, and Ojokoh (2015) explained that the information can be collected by two ways, either implicitly or explicitly. In implicit approach, behavior of user is observed such as movies watched, purchased items, taste in applications. On the other hand, in explicit way, information is collected on the basis of previous ratings provided by the user or past history of the user.

Recommendation system employs various techniques that include- content based: in this the system recommends items similar to the items that the users have liked in the past (Hyung, Keejun, Mun, Joonmyun, & Jinwoo, 2012), collaborative filtering: recommend items that similar taste users have liked in past (Zhang, Lin, Lin, & Liu, 2016), demographic: recommend items based on demographics (language, country) of user (Chen & He, 2009), knowledge based: recommend item based on the knowledge that how a particular item meets user demand (S. Aciar, G. Aciar, Collazos, & Gonzalez, 2016), hybrid recommender system: combination of one or more techniques described (Cheng, He & Fang, 2016; Nilashi, Ibrahim & Ithnin, 2014).