Social recommendation: A user profile clustering‐based approach

Abstract

The recommendation in information systems is a specific form of information filtering that aims to present the relevant information interesting the user. This technique is used in different contexts such as social networking, e-commerce and information retrieval. Generally, existing recommender system techniques implement collaborative filtering by deducing a part of user interests from the preferences of other users with similar profiles. Many techniques can be used to implement Collaborative Filtering such as Bayesian Networks, latent semantic, and clustering.We present in this work a novel clustering approach using a modified partitional algorithm. We propose a user model that integrates the relevant user information and a clustering algorithm that generates groups of similar user profiles by implementing a profile similarity function. The proposed approach is then evaluated based on a set of user profiles data corresponding to the context of an e-commerce website. KEYWORDSclustering algorithm, recommender systems, similarity measures, user profile INTRODUCTIONRecent technologies offer access to large masses of information. In order to help the user get relevant information that specifically concerns him/her, personalization is the appropriate solution. The concept of personalization, also called adaptation, is an extension of a research trend that has been actively used since the beginning of the 1980s in human-machine interaction. 1 Personalization can be as much about the content (data, information, document, etc) as about the container (IHM, interaction platform, communication mode, communication channel, etc). This concept is often used in a variety of domains, ie, Information Retrieval, e-commerce, tutoring systems, etc.In order to offer the user the personalized content adapted to his/her expectations, it is necessary to identify his/her needs and preferences. 2-4A personalized system has to satisfy preferences of the user based on his profile. 5 The user profile is built using information gathered about the user and his/her use of the system. Offering personalized content adapted to user's expectations implies identifying his/her needs and preferences. [2][3][4] The implementation of personalization can be performed through different techniques, ie, personalizing the appearance of the interface, filtering the incoming information or recommending content to the user. In this paper, we focus on the last option that is recommending the content to the user.It has been shown recently 6 that a normal user can lose interest after 60 to 90 seconds of choosing through reviewing 10 to 20 items on one or two screens. 6 That points to the risks of the user abandoning the system if he/she doesn't find what interests him/her. Recommendation aims to ensure that each user will find something corresponding to his/her preferences.In this work, a novel approach is proposed for a user profile clustering based recommendation. This approach considers a global user model represented with a keyword vector model structure usin...

Research on User Profile Combined with Collaborative Filtering Recommendation Algorithm for Intelligent Tourism

“…In recent years, the development of deep learning technology has further promoted the evolution of user portraits. Deep learning can process more complex user behavior data and text content, and extract deeper user characteristics, so as to achieve higher levels of personalized recommendation and precision marketing [6].…”

A Systematic Study on the Recommender Systems in the E-Commerce

“…Thus, even though e-commerce websites widely use recommendation techniques to improve their business, Sivapalan et al [116] expressed that they still face some research and practical challenges, including scalability, rich data, consumer-centered recommendations, anonymous users, and connecting recommenders to markets. Suitable data types for use in recommender systems in ecommerce include on-site user activity (clicks, searches, page, and item views), and off-site user activity (tracking clicks in the email and mobile applications), particular items or user profiles data [135], and contextual data (the device used, current user location, and referral URL). On the other hand, the RSs requires sufficient data to run their algorithms more carefully and provide the relevant recommendations.…”