NUDTSNA at TREC 2015 Microblog Track: A Live Retrieval System Framework for Social Network based on Semantic Expansion and Quality Model

Abstract

Abstract : This paper describe our approaches to real-time filtering task in the TREC 2015 Microblog track, including push notifications on a mobile phone task and periodic email digest task. In the push notifications on a mobile phone task, we apply a recommendation framework with rank algorithm and dynamic threshold adjustment which utilizes both semantic content and quality of a tweet. External information extracted from Google search engine and word2vec model based on existing corpus are well incorporated to enhance the understanding of a tweets or a profiles interest. In the email digest task, based on the candidate tweets retrieved from the first task, we calculate the score of a tweet considering semantic features and quality features, all the tweets classified into a topic are ranked by our key word bool logistic model.