Classifying library resources in Library Recommender Agent using PU learning approach

Abstract

Agent based Library Recommender System is proposed with the objective to provide effective and intelligent use of library resources such as finding right book/s, relevant research journal papers and articles. The architecture consists of profile agent and library recommender agent. The main task of Library recommender agent is filtering and providing recommendations. Library resources include book records having table of contents, journal articles including abstract, keywords. Rich set of keywords are obtained to compute similarity via table of contents and abstracts. The library resources are classified into fourteen categories specified in ACM computing classification system 2012 (ACM CCS). The identified category provides a way to obtain semantically related keywords for the library resources. This paper provides the task of library resources classification using PU (Positive Unlabeled) learning approach implemented using NB (Naive Bayes) Classifier. Recommendation accuracy of the system is improved by library resources classification. The novel features of the recommender system are use of ACM CCS 2012 as ontology, semantic similarity computation, implicit auto update of user profiles, and variety of users in evaluation.