A Hybrid User Profile Model for Personalized Recommender System with Linked Open Data

Abstract

In order to better enhance user experience on the web, varies applications such as search engines have integrated with recommender systems. Users will get some relevant items recommended when browsing the result page. However, building such recommendation services needs a large amount of item information, which makes it hard to start a new recommender service. Due to the development of Semantic Web and Linked Data, a vast amount of RDF data can be accessed via the Internet and naturally used as a knowledge base for recommender systems. In this paper, we study modeling user profiles in semantic environment and building a personalized recommender system exclusively on Linked Open Data. First, we define two concepts related to user browsing history and propose a hybrid user profile model (Hy-UPM). Second, we design a generic and personalized recommender system to utilize the semantic information between the items and user profile model to make recommendations. Finally, we conduct our experiment on the movie dataset of DBpedia and MovieLens. The result shows Hy-UPM has a better Mean Reciprocal Rank performance compared with other recommendation methods.

Sparse Feature Factorization for Recommender Systems with Knowledge Graphs

“…The adoption of KG𝑠 as a source of side-information has generated several advancements in the tasks of recommendation [16], knowledge completion [33], preference elicitation [14], user modeling [69], and thus produced a vast literature. In recent years, the Knowledgeaware Recommender Systems have been particularly impactful for several recommendation tasks: hybrid collaborative/contentbased recommendation [16,47], exploiting the KG information to suffice the lack of collaborative information and to improve the performance; knowledge-transfer, cross-domain recommendation [29,41,77], where the KG𝑠 allow to find semantic similarities between different domains; interpretable/explainablerecommendation [6,13,16,73,76], with KG being a backbone for understanding the recommendation model and providing humanlike explanations; user-modeling [39,50,54,69], since the resource descriptions can drive the construction of the user profile; graphbased recommendation [27,61,62,68,70,71], where the topologybased techniques have met the semantics of the edges/relations, and the ontological classification of nodes (classes); the cold-start problem [29,51,60,74], since the KG𝑠 can overcome the lack of collaborative information; the content-based recommendation [15,53] that solely relies on KG and still produces high-quality recommendations. KGFlex could be considered a Knowledge-aware hybrid collaborative/content-based recommendation model.…”

Personalised Information Recommender Using Framework for Ontology Alignment Among Digital Libraries

“…A digital library should not only adjust to the specific characteristics of each user profile, but also to the particular necessities and preferences of each user combining both library archives and profile level personalisation 10 . Luo 11 , et al introduced concepts related to user browsing history and proposed a hybrid user profile model and a personalised recommender system to utilise the semantic information between the items and user profile model to make recommendations. Kruk 12 , et al highlighted that a semantic personalised digital library should enhance information extraction, facilitate query refinement and also provide recommendation services using community-aware ontologies.…”