دانلود رایگان ترجمه مقاله بررسی سامانه توصیه گر مبتنی بر پالایش مشارکتی (نشریه IEEE 2016)

بررسی سامانه توصیه گر مبتنی بر پالایش مشارکتی برای اپلیکیشن های اینترنت موبایل

A Survey of Collaborative Filtering Based Recommender Systems for Mobile Internet Applications

Abstract

 With the rapid development and application of the mobile Internet, huge amounts of user data are generated and collected every day. How to take full advantages of these ubiquitous data is becoming the essential aspect of a recommender system. Collaborative filtering (CF) has been widely studied and utilized to predict the interests of mobile users and to make proper recommendations. In this paper, we first propose a framework of the CF recommender system based on various user data including user ratings and user behaviors. Key features of these two kinds of data are discussed. Moreover, several typical CF algorithms are classified as memory-based approaches and model-based approaches and compared. Two case studies are presented in an effort to validate the proposed framework.

1 Introduction

cloud computing, massive amounts of data are produced every day by both people and machines. Our society has already entered the era of Big Data [1]. Thanks to the various smart devices and mobile applications, Internet users can acquire all sorts of information about education, shopping, social activity, etc. [2] [3] [4] [5]. However, as the volume of data increases, individuals have to face the problem of excessive information, which makes it more difficult to make the right decisions. This phenomenon is known as information overload [6]. Moreover, limited by the input ability of mobile devices, users are usually unwilling to type in lots of words to describe what they want. Recommender system can alleviate these problems by effectively finding users’ potential requirements and selecting desirable items from a huge amount of candidate information. Recommender systems are usually classified into two categories, i.e., content-based and collaborative filtering (CF) [7].

Content-based recommender system utilizes the contents of items and finds the similarities among them. After analyzing sufficient numbers of items that one user has already shown favor to, the user interests profile is established. Then the recommender system could search the database and choose proper items according to this profile. The difficulty of these algorithms lies in how to find user preferences based on the contents of items. Many approaches have been developed to solve this problem in the areas of data mining or machine learning. For example, in order to recommend some articles to a specific reader, a recommender system firstly obtains all the books this reader has already read and then analyzes their contents. Key words can be extracted from the text with the help of text mining methods, such as the well-known TF-IDF [8]. After integrating all the key words with their respective weights, a book can be represented by a multi-dimensional vector. Specific clustering algorithms can be implemented to find the centers of these vectors which represent the interests of this reader.

On the other hand, collaborative filtering (CF) has become one of the most influential recommendation algorithms [9]. Unlike the content-based approaches, CF only relies on the item ratings from each user. It is based on the assumption that users who have rated the same items with similar ratings are likely to have similar preferences. CF is specifically designed to provide recommendations when detailed information about the users and items is inaccessible. Furthermore, it successfully mitigates the problem of over-specialization [10], which is quite common in content-based systems. Over-specialization is the phenomenon that recommended items are always much the same and the diversity of recommendations is neglected. As CF makes recommendations according to the neighborhood (people with similar preferences), the item one user has consumed may be something new to his neighbors. The above features are particularly attractive which make CF algorithms extensively employed in recommender systems.

However, to the best of our knowledge, very few studies have revealed the common features of the various CF algorithms for mobile Internet applications. In addition, most of the existing surveys merely introduce the principles of CF algorithms, ignoring the importance of case study, which can demonstrate the performances of typical algorithms visually and specifically. Therefore, this paper focuses on collaborative filtering based recommender systems for mobile Internet applications. In particular, main contributions of this paper are highlighted as follows:

• We introduce a general framework of CF recommender system. This framework assists recommender developers to utilize the gathered data and to generate proper recommendations. The features of data collected from both user behaviors and user ratings are also discussed and compared.

• CF algorithms are classified. Main procedures of CF are briefly summarized and introduced.

• Two case studies are presented to validate the proposed framework. Evaluations on representative CF algorithms are conducted based on real-world datasets with detailed analysis and comparison.

The rest of this paper is organized as follows. Section II presents the framework of CF. Both classification and main procedures of typical CF algorithms are introduced in Section III. In Section IV, we conduct two case studies based on realworld datasets in order to analyze the performances of CF algorithms. Finally, Section V concludes this paper.