Citation Context

...pics range from music and sports to more serious ones like politics and religion. Much work has been done to analyze topics on Twitter (Ramage et al., 2010; Hong and Davison, 2010; Zhao et al., 2011; =-=Lau et al., 2012-=-). (2) Events. These are things that take place at a certain time and attract many people’s shortterm attention in social media. Example events include concerts, sports games, scandals and elections. ...

Citation Context

...tc. In addition, most of the content found on Twitter is not related to any event and the high volume of data on a diverse number of topics poses a big challenge in terms of scalability. Despite these challenges, it would be beneficial to develop a system that can lead to real time detection and tracking of events as Twitter does not provide a tool to see event summaries except for searching on trending topics or using hashtags. Different approaches have been taken by the researchers to deal with the event detection task. The approaches based on term interestingness [2]–[6] and topic modeling [7]–[11] suffer from high computational cost among other things. However, incremental clustering based approaches [12]–[14] usually provide a low computational cost solution. Taking this into consideration, we propose an incremental clustering based end-to-end solution to detect newsworthy events from a stream of time ordered tweets. The problem of event detection from the Twitter data stream in an incremental clustering context can be divided into two major stages. The first stage involves detecting a burst in the number of tweets discussing a topic/event and the second stage involves grouping/c...

Citation Context

... Obama’s victory speech, there are 327k tweets per second posted/reposted to discuss Obama’s re-election6. One of the typical tweets like “Thank The Lord for that!! Well done Obama in Chicago, Illinois!! That was a Great victory speech!”. In this tweet: Time is “November 7, 2012”, “Chicago” and “Illinois” representLocations, “Obama” representsEntity,Keywords are “Great”, “victory”, and “speech”, and the rest are General words. GEAM can provide a clearer description of events by separating General words from Event-related Aspects words. The most related work is the model proposed by Lau et al. [12]. They introduce an online processing variant of topic model Latent Dirichlet Allocation (LDA) [13, 14] to analyze tweets trend. Their graphic structure is the same as LDA, which models each tweet as a multinomial mixture of all topics or events. However, this assumption is obviously unreasonable due to Twitter’s short text length, most users only discuss one event in a tweet. In this paper, we improve LDA for Twitter event detection by assigning only one event to an event-related tweet and differentiating General words from Event-related Aspects words. To the best of our knowledge, we are the...

Citation Context

...escribed an application of partially observable Markov model to analyze a large-scale query log. Among the diverse types of probabilistic models, topic models are found to be an effective tool for query log analysis [6, 15, 17, 23]. For example, [17] proposed a topic-concept cube that supports online multidimensional mining of query log. [34] presented a topic model that captures latent structure of textual data and how the structure changes over time. More recently, with the popularity of microblogs, researchers are aware of the importance of analyzing text streams in real-life applications. [19] presented a topic model to track emerging events in microblog data such as tweets. [13] presented an algorithm to model diversity phenomena in tweet streams based on topical diversity and geographical diversity. [9, 8] described how to efficiently capture the statistics of stream data. However, to the best of our knowledge, none of the existing techniques are primarily proposed for processing massive web search streams. Their capabilities are limited in term of the effectiveness and the efficiency of topic discovery from web search data. Besides designing effective probabilistic models, explo...

Citation Context

...e of the wave represents the number of co-occurrences between the main word and the other words inside a topic: high amplitudes represent elevate co-occurrences frequency. In (Wang and McCallum, 2006) Wang and McCallum proposed a modified LDA model, called Topics Over Time, where topic discovery is influenced both by word co-occurrences and temporal information. In their work, the authors model the time as a continue distribution, defined by a Beta distribution over a parameter Ψ, associated with each topic which is responsible to generate both patterns and topics distribution. Lau et al. in (Lau et al., 2012) proposed a novel method to track emerging events in microblogs (e.g., Twitter). Their method defines a window of time slices, where each time slice contains several documents, and updates parameters α and β for each old word and document. Novel words and documents are initialised using two parameters, α0 and β0, that are defined a priori. Another related work, not linked to the LDA model, is (Di Caro et al., 2011). Di Caro et al. proposed a framework called TMine which defines a navigable tag-flag. A tag-flag can be thought as a topic because it contains a set of related words. 3. Research Qu...

Citation Context

...ar, often represent the basis of several works which analyze data from social networks in order to detect trending topics. A topic modeling technique based on an online variant of LDA is presented in =-=[10]-=-. The Twitter stream is processed in time slices and older tweets are discarded allowing the model to be constant in size. Moreover, differently from the on-line LDA approach [11], a dynamic vocabular...

Citation Context

...n languages used in emotional spiking tweets. Many tasks on Twitter including burst detection (Kleinberg, 2003; Diao et al., 2012), first story detection (Petrović et al., 2010), and topic tracking (=-=Lau et al., 2012-=-) failed to effectively incorporate the textual characteristics of tweets and regard it is out of their scope. Being able to characterize tweets from emotional upsurge would open a window to the ident...

Citation Context

...ive information in text and gives suggestions to sanitize sentences. Their method involves querying the Internet for detections and recommendations. There have been numerous studies on topic modeling =-=[18]-=-, named entity recognition [25], and sentiment analysis [6] on Twitter as well as normalizing micro-text [30] though not focusing on tweets in particular. 3. PROBLEM STATEMENT AND THREAT MODEL The mai...