Review Aware Recommender System: Using Reviews for Context Aware Recommendation

Review Aware Recommender System: Using Reviews for Context Aware Recommendation

Fatima Zahra Lahlou, Houda Benbrahim, Ismail Kassou
Copyright: © 2018 |Pages: 23
DOI: 10.4018/IJDAI.2018070102
(Individual Articles)
No Current Special Offers


Context aware recommender systems (CARS) are recommender systems (RS) that provide recommendations according to user contexts. The first challenge for building such a system is to get the contextual information. Some works tried to get this information from reviews provided by users in addition to their ratings. However, all of these works perform important feature engineering in order to infer the context. In this article, the authors present a new CARS architecture that allows to automatically use contextual information from reviews without requiring any feature engineering. Moreover, they develop a new CARS algorithm that is tailored to textual contexts, that they call Textual Context Aware Factorization Machines (TCAFM). An empirical evaluation shows that the proposed architecture allows to significantly improve recommendation accuracy using state of the art RS and CARS algorithms, whereas TCAFM leads to additional improvements.
Article Preview

1. Introduction

The increasing number of items featured by online websites led to the necessity to assist users and help them reach relevant resources. Recommender Systems (RS) address this issue by automatically suggesting to users, all along their navigations, the most relevant items according to their needs and preferences. RS are, as defined in (Burke, 2002), systems that can provide custom recommendations and guide the user to find interesting resources in a large data space. In other words, their purpose is to filter information according to the interest of users.

While traditional RS focus only on users and items to compute predictions, another trend suggests considering additional information relative to the context within the user experienced the items, such as time, weather, or accompanying persons. These systems are called Context Aware Recommender Systems (CARS). Indeed, research proved that in situation where context matters, including contextual information when computing recommendation improves its accuracy (Herlocker & Konstan, 2001). Thus, a user may be interested in completely different items according to its situation. For example, a user may choose a restaurant for a business lunch but would prefer another one when he is with his family.

A common RS task is to predict ratings for unobserved interactions between users and items. Thus, the goal is to define a target function IJDAI.2018070102.m01, called utility function, from the set IJDAI.2018070102.m02 to IJDAI.2018070102.m03 where IJDAI.2018070102.m04 the set of users, IJDAI.2018070102.m05 the set of items, and IJDAI.2018070102.m06 and IJDAI.2018070102.m07 are the number of users and items respectively, so that: IJDAI.2018070102.m08 and IJDAI.2018070102.m09 is the predicted rating of user IJDAI.2018070102.m10 on item IJDAI.2018070102.m11. While only users and items are involved, this is called a two-dimensional recommendation (Adomavicius, Sankaranarayanan, Sen, & Tuzhilin, 2005).

In contrast, CARS assume that additional information affect ratings like Mood or Accompanying Person. Thus, the goal in CARS is to define a utility function from the set IJDAI.2018070102.m12 to IJDAI.2018070102.m13, where IJDAI.2018070102.m14 are the different sets of context that influence rating behavior. For example, IJDAI.2018070102.m15 could be the Mood, IJDAI.2018070102.m16 could be the Accompanying Person and so on. Because additional dimensions are involved, CARS are called multi-dimensional recommendation (Adomavicius et al., 2005).

In the literature (Adomavicius, Mobasher, Ricci, & Tuzhilin, 2011), CARS algorithms are classified in either:

Complete Article List

Search this Journal:
Volume 15: 1 Issue (2024): Forthcoming, Available for Pre-Order
Volume 14: 2 Issues (2022)
Volume 13: 2 Issues (2021)
Volume 12: 2 Issues (2020)
Volume 11: 2 Issues (2019)
Volume 10: 2 Issues (2018)
View Complete Journal Contents Listing