Solving the Cold-Start Problem in Recommender Systems Using Contextual Information in Arabic from Calendars
- PDF / 529,964 Bytes
- 9 Pages / 595.276 x 790.866 pts Page_size
- 63 Downloads / 196 Views
RESEARCH ARTICLE-COMPUTER ENGINEERING AND COMPUTER SCIENCE
Solving the Cold-Start Problem in Recommender Systems Using Contextual Information in Arabic from Calendars Nuha Alghamdi1
· Fatmah Assiri2
Received: 12 December 2019 / Accepted: 18 August 2020 © King Fahd University of Petroleum & Minerals 2020
Abstract Cold-start problem, which is the inability to make accurate recommendations due to the unavailability of enough information about user’s preferences, is one of the challenges of recommender systems. Contextual information was widely used to solve this problem in English language. However, Arabic language is the sixth-most-spoken language in the world, Arabic text in calendars has not been used to find user’s interests in recommender systems. Our work utilizes events from users’ calendars that are written in Arabic. We first build a multi-class text classifier to classify calendar events. The classifier is trained on Wikipedia data and validated using 10-fold cross-validation. Our classifier reached an accuracy of 76.72%. We investigated the reasons for our results and we identified factors that have high impacts on them. These factors including, but not limited to, English events written in Arabic letters and Arabic names. Finally, we highlighted some research directions to tackle the existing limitations in order to improve the presented work. Keywords Recommender systems · Cold-start · Word embeddings · FastText · Multi-class classification · Arabic text classification
1 Introduction Recommender systems (RSs) are crucial parts of existing commercial systems. They address information overload by recommending items to users based on their interests, which can be places, articles, or products. Three traditional approaches are used by RSs to make recommendations to users: content-based filtering (CBF) approach, which uses products liked by the user in the past to recommend new ones, a collaborative filtering (CF) approach, which uses products liked by other users in the past with the same interest and then recommends these products to a specified user, and a hybrid approach,which combines the two previous approaches in various ways [1].
B
Nuha Alghamdi [email protected] Fatmah Assiri [email protected]
1
Faculty of Computing and Information Technology, King Abdulaziz University, Jeddah, Saudi Arabia
2
Department of Software Engineering, College of Computer Science and Engineering, University of Jeddah, Jeddah, Saudi Arabia
One of the existing problems in RSs is the cold-start problem, which entails the RS being unable to build accurate recommendations due to the unavailability of enough preference [1]. Cold-start problem comes in two forms: a new user or a new item. The former occurs when a new user registers to the system and lacks a preference history in their profile [2]. The latter occurs when a new item is added to the RS and has not been rated at all or it has been rated by few users [1]. Many research has been conducted to alleviate the coldstart problem in RSs and each prop
Data Loading...
