摘要：AbstractIn this paper, we investigate how to recommend the meanings of newly coined words, such as newly coined named entities and Internet jargon. Our approach automatically chooses a document explaining a given newly coined word among candidate documents from multiple web references using Probabilistic Latent Semantic Analysis [1]. Briefly, it involves finding the topic of a document containing the newly coined word and computing the conditional probability of the topic given each candidate document. We validate our methodology with two real datasets from MySpace forums and Twitter by referencing three web services, Google, Urbandictionary, and Wikipedia, and we show that we properly recommend the meanings of a set of given newly coined words with 69.5% and 80.5% accuracies based on our three recommendations, respectively. Moreover, we compare our approach against three baselines where one references the result from each web service and our approach outperforms them.
