Research On Time Based Microblog Search And Filtering

Short texts,such as microblogs,are prevailing with the development of mobile Internet and social media.In face of the massive microblogs and the users’ diverse information requirements,microblog search and microblog filtering have become the indispensable components of the microblog service.In recent years,the time profile of microblog attracts the attention of researchers.Recent works reveal that the time profile is promising in improving the performance of microblog retrieval.This paper focuses on boosting the performance of microblog search and microblog filtering by using the time profile of microblog in four research issues:query modeling,document modeling,relevance estimation between query and documents,and filtering model.We attempt to leverage the time profile to reduce the impacts of short texts on content-based microblog search.Meanwhile,the historical microblogs are integrated into the online microblog filtering for an improved performance.The main contents of this paper are structured as follows.(1)A query model based on term time distribution is proposed to deal with the short query in microblog search.First,we define the term time distribution and analyze the temporal distribution of the queries and the relevant documents.Then a similarity measure for term time distribution is presented to properly decide the relevant terms for an expanded query model.The suggested method avoids the drawbacks of the classical content based query expansion approaches caused by the length limit in microblog by attempting to employ only time profile to establish the relevance between query terms and expanded terms.(2)A time-based document model is presented to enrich the short texts of microblogs.First,candidate terms are jointly weighted by their temporal closeness to the microblog and their distribution over different periods.Then,terms not in the original microblog are refined by a machine learning mechanism trained on pseudo-supervised data.Finally,a document expansion model is designed,together with two approximate solutions to optimize the time complexity to reduce the time cost.(3)The temporal relevance is investigated to enhance the relevance estimation of microblog search,as an additional information besides the well-recognized content relevance.First,three aspects of time relevance are explored in the framework of language model.Then,in the current framework of learning to rank,a loss function based on time-sensitive learning to rank is defined,which aims at a ranking more consistent with the time profile of microblogs.(4)A real time microblog filtering method is proposed by integrating the historical information into the online learning framework.In this method,the microblog to be filtered is evaluated in the past microblogs ranking list,which is employed as a prior knowledge to adjust the classical online classification algorithm dynamically.An implementation in language model based retrieval and the online logistic regression model is examined to valid its performance.