Targeted Topic Modeling for Focused Analysis

Wang, Shuai; Chen, Zhiyuan; Fei, Geli; Liu, Bing; Emery, Sherry

doi:10.1145/2939672.2939743

Cited by 38 publications

(33 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…TTM : The method requires input documents from one domain but of different aspects (Wang, Chen, Fei et al, ). Because we cannot assign all the tweets into a specific domain, we consider all tweets in TC from one domain.…”

Section: Methodsmentioning

confidence: 99%

“…Yang, Kiang, and Shang (2015) collected messages from social media related to adverse drug reaction through Latent Dirichlet allocation (LDA) and a partially supervised classification approach. Wang, Chen, Fei, Liu, and Emery (2016) proposed a targeted topic model that leverages a latent variable to identify documents belonging to some specific aspects. Although Kasiviswanathan, Melville, Banerjee, and Sindhwani (2011) detected novel topics, the authors claim their dictionary learning-based method could collect topic-related data with good precision and recall on test data.…”

Section: Methods For Data Collectionmentioning

confidence: 99%

“…When selecting top-ranked words, we rank all the words belonging to the same topic according to their values in the corresponding column in D T . TTM: The method requires input documents from one domain but of different aspects (Wang, Chen, Fei et al, 2016). Because we cannot assign all the tweets into a specific domain, we consider all tweets in TC from one domain.…”

Section: Baseline Methodsmentioning

confidence: 99%

“…The term correlation matrix could further be decomposed to term‐topic matrix, and topic representative keywords could be obtained. Except for extracting keywords for all the topics, Wang et al () proposed a targeted topic model that could identify keywords of specific aspects. Given a few words that define the aspect, their model, which is a variation of LDA, could return keywords related to the specified aspect.…”

Section: Related Workmentioning

confidence: 99%

See 3 more Smart Citations

Collecting event‐related tweets from twitter stream

Zheng

Sun

2018

Asso for Info Science & Tech

View full text Add to dashboard Cite

Twitter provides a channel of collecting and publishing instant information on major events like natural disasters. However, information flow on Twitter is of great volume. For a specific event, messages collected from the Twitter Stream based on either location constraint or predefined keywords would contain a lot of noise. In this article, we propose a method to achieve both high‐precision and high‐recall in collecting event‐related tweets. Our method involves an automatic keyword generation component, and an event‐related tweet identification component. For keyword generation, we consider three properties of candidate keywords, namely relevance, coverage, and evolvement. The keyword updating mechanism enables our method to track the main topics of tweets along event development. To minimize annotation effort in identifying event‐related tweets, we adopt active learning and incorporate multiple‐instance learning which assigns labels to bags instead of instances (that is, individual tweets). Through experiments on two real‐world events, we demonstrate the superiority of our method against state‐of‐the‐art alternatives.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methods For Data Collectionmentioning

confidence: 99%

Section: Baseline Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Collecting event‐related tweets from twitter stream

Zheng

Sun

2018

Asso for Info Science & Tech

View full text Add to dashboard Cite

show abstract

“…A cluster has been defined for news stories occurring at a snapshot interval of time which describe the same sub goal or sub topic of the goal topic [105] or whole story [106]. A clustering of events has been defined on the commonality of attributes describing events, where exact match of attributes causes events to be positioned in the same cluster, whereas a partial match causes a link to be described between events across their corresponding clusters as has been described in [106].…”

Section: Related Work In Clustering Of Nodes In the Provenance Graphmentioning

confidence: 99%

Learning topic description from clustering of trusted user roles and event models characterizing distributed provenance networks: a reinforcement learning approach

Mukherjee

Bandyopadhyay

2017

J Big Data

View full text Add to dashboard Cite

This paper proposes a reinforcement learning based message transfer model for transferring news report messages through a selected path in a trusted provenance network with the objective of maximizing the reward values based on trust or importance based and network congestion or utility based cost measures. The reward values are calculated along a dynamically defined policy path connecting start topic or event node to a goal topic or event or issue nodes for incrementally defined time windows for a given network congestion situation. A hierarchy of agents of trusted roles is used to accomplish the sub-goals associated with sub-story or subtopic in the provenance structure where an agent role may assume the semantic role of the associated subtopic. The twitted news story thread or plan of events is defined in this work from the starting topic or event node to the goal topic or event node for incrementally defined intervals of time. The graphs are clustered into subtopic and these sub-goals or sub topic nodes of a topic node at every level of granularity are associated with cluster of news reports which describe activities associated with sub-goal or sub-topic events. Such cluster of nodes may also represent drilled down sequence of sub-events describing a sub-topic or sub-goal node. The policy path in a topic or story graph model is defined by applying reinforcement learning principles on dynamically defined event models associated with evolution of topic definition observed from incrementally acquired samples of input training data spanning multiple time windows. We provide a methodology for unifying similar provenance graph models for adapting and averaging the policy path classifiers associated with individual models to produce a reduced set of unified models derived during training. A minimum set cover of classifiers is identified for the models and a clustering procedure of the models is suggested based on these classifiers. Other database clustering methods have also been suggested as alternatives for clustering these models. A collection of unified models are identified from the models identified within a cluster and the policy path classifiers associated with these models provide the story or topic descriptions destined to goal topic or event nodes characterizing these models within a cluster.

show abstract