|dc.description.abstract||Tracking topical events in social media streams, such as Twitter, provides a means for users to keep up-to-date on topics of interest to them. This tracking may last a period of days, or even weeks. These events and topics might be provided by users explicitly, or generated for users from selected news articles. Push notification from social media provides a method to push the updates directly to the users on their mobile devices or desktops.
In this thesis, we start with a lexical comparison between carefully edited prose and social media posts, providing an improved understanding of word usage within social media. Compared with carefully edited prose, such as news articles and Wikipedia articles, the language of social media is informal in the extreme. By using word embeddings, we identify words whose usage differs greatly between a Wikipedia corpus and a Twitter corpus.
Following from this work, we explore a general method for developing succinct queries, reflecting the topic of a given news article, for the purpose of tracking the associated news event within a social media stream. A series of probe queries are generated from an initial set of candidate keywords extracted from the article. By analyzing the results of these probes, we rank and trim the candidate set to create a succinct query. The method can also be used for linking and searching among different collections.
Given a query for topical events, push notification to users directly from social media streams provides a method for them to keep up-to-date on topics of personal interest.
We determine that the key to effective notification lies in controlling of update volume, by establishing and maintaining appropriate thresholds for pushing updates. We explore and evaluate multiple threshold setting strategies. Push notifications should be relevant to the personal interest, and timely, with pushes occurring as soon as after the actual event occurrence as possible and novel for providing non-duplicate information. An analysis of existing evaluation metrics for push notification reflects different assumptions regarding user requirements. This analysis leads to a framework that places different weights and penalties on different behaviours and can guide the future development of a family of evaluation metrics that more accurately models user needs.
Throughout the thesis, rank similarity measures are applied to compare rankings generated by various experiments. As a final component, we develop a family of rank similarity metrics based on maximized effectiveness difference, each derived from a traditional information retrieval evaluation measure. Computing this maximized effectiveness difference (MED) requires the solution of an optimization problem that varies in difficulty, depending on the associated measure. We present solutions for several standard effectiveness measures, including nDCG, MAP, and ERR. Through experimental validation, we show that MED reveals meaningful differences between retrieval runs. Mathematically, MED is a metric, regardless of the associated measure. Prior work has established a number of other desiderata for rank similarity in the context of search, and we demonstrate that MED satisfies these requirements. Unlike previous proposals, MED allows us to directly translate assumptions about user behavior from any established effectiveness measure to create a corresponding rank similarity measure. In addition, MED cleanly accommodates partial relevance judgments, and if complete relevance information is available, it reduces to a simple difference between effectiveness values.||en