Towards a knowledge-based probabilistic and context-aware social recommender system

Abstract

In this article, we propose (1) a knowledge-based probabilistic collaborative filtering (CF) recommendation approach using both an ontology-based semantic similarity metric and a latent Dirichlet allocation (LDA) model-based recommendation technique and (2) a context-aware software architecture and system with the objective of validating the recommendation approach in the eating domain (foodservice places). The ontology on which the similarity metric is based is additionally leveraged to model and reason about users’ contexts; the proposed LDA model also guides the users’ context modelling to some extent. An evaluation method in the form of a comparative analysis based on traditional information retrieval (IR) metrics and a reference ranking-based evaluation metric (correctly ranked places) is presented towards the end of this article to reliably assess the efficacy and effectiveness of our recommendation approach, along with its utility from the user’s perspective. Our recommendation approach achieves higher average precision and recall values (8% and 7.40%, respectively) in the best-case scenario when compared with a CF approach that employs a baseline similarity metric. In addition, when compared with a partial implementation that does not consider users’ preferences for topics, the comprehensive implementation of our recommendation approach achieves higher average values of correctly ranked places (2.5 of 5 versus 1.5 of 5).

Process Knowledge Recommendation System for Mechanical Product Design

“…Especially, under the perspective of product life-cycle, many researchers focus on how to use the knowledge service methods in collaborative product development teams [8,9,10,11] and new product development [12,13,14,15]. For the rest stages of product life-cycle, the research concern of knowledge sharing framework and method are focused on socialization mechanisms product development teams [16,17,18], product-service system development (PSS) [19,20], product supply chain management [21,22,23] and outsourcing projects of software product [24].…”

Sense-Based Topic Word Embedding Model for Item Recommendation

“…They assumed that the tastes from all child subjects of subject s should be acquired when a user is interested in s. However, this assumption is not pervasive since the child subjects are usually various and one may dislike some of the subcategories. Colombo-Mendoza et al [37] jointly combined the LDA model and the ontology-based semantic similarity metric to aid in context modeling. The knowledge-based models can address the data sparsity problem but they need knowledge acquisition, which is time consuming when the text data are massive.…”

CredSaT: Credibility ranking of users in big social data incorporating semantic analysis and temporal factor

“…Furthermore, users may deviate from their domain of knowledge to discuss general, unrelated or trending topics. Entries of W are normalised into the range of [0]- [1] by dividing each entry by the maximum weight values of the corresponding domain. For example, all users' weight values in the domain d are normalised as follows…”

“…Researchers have attempted to capture the value of BD in various contexts. For example, in online social networks (OSNs), one of the central considerations is the understanding of users' behaviour [1]. This involves the measurement of users' trustworthiness and an understanding of their influence in a particular domain.…”