Finding keyphrases of readers' interest utilizing writers' interest in social media

Abstract

Suggesting further reading materials is an application of recommendation. Considering context, current systems usually rely on topic information and related materials to propose options for users, while users behavior is also commonly used if log information is involved. However, the users interests, which are aroused by the content of the current article they read instead of what they have had, are seldom detected from the context, and they are usually the motive that readers want to read more. This paper presents an approach to detect readers' interest from the current article they read and the users feedback of it. TED talks are utilized as the experimental materials. InterestFinder proposes interest keywords/keyphrases for each talk, where different kind of words and phrases are provided to it to find suitable candidate terms. Experiments show that the best setting proposed achieves a NDCG@50 0.6392, and the detail results are discussed. Results conclude that considering both words and phrases in a proper selection criteria benifits, and finding conceptual keyphrases as interest terms is necessary to further improve the system performance. © 2014 Springer International Publishing.