A Cross-Domain Recommender System Based on Common-Sense Knowledge Bases

A system able to extract and recommend technical terms from various domains is proposed in this paper. The motivation is to provide keywords that users may not be familiar with in the beginning but will be interested in after studying. To acquire domain knowledge, we collect documents from various sources, and the words in the documents are then represented as semantic word vectors. Given queries from users, the system first extracts important terms from given documents and computes the semantic similarity between those terms. Next, we utilize third party common-sense knowledge bases such as ConceptNet and Wikipedia to connect the queries to those extracted keywords through the network structures. Finally, the system will collect all keywords traversed and recommend the top-n of them. We propose and compare four models for the recommendation, and the differences between using ConceptNet and Wikipedia for discovering related knowledge are also investigated in this work.