Exploiting Context for Rumour Detection in Social Media

Zubiaga et al.'s context-learning approach used conditional random fields as a sequential classifier to learn reporting dynamics during an event.

“Their context-learning approach relied on conditional random fields (CRF) as a sequential classifier that learns the reporting dynamics during an event, so that the classifier can determine, for each new tweet, whether it is or not a rumour based on what has been seen so far during the event.”

Zubiaga et al. avoid reliance on enquiring tweets because not all rumours trigger them, which may lead to low recall.

“Moreover, they avoid the reliance on enquiring tweets, which they argue that not all rumours may trigger and hence may lead to low recall, as rumours not provoking enquiring tweets would be missed.”

The state-of-the-art approach for rumour detection is Zubiaga et al. (2017), which leverages context from earlier posts associated with an event to determine if a tweet is a rumour.

“The state-of-the-art approach for rumour detection is that presented by Zubiaga et al. (2017), which leverages context from earlier posts associated with a particular event to determine if a tweet constitutes a rumour.”

Zubiaga et al. (2016b, 2017) proposed an alternative approach that learns context throughout a breaking news story to determine if a tweet is a rumour.

“In contrast, Zubiaga et al. (2016bZubiaga et al. ( , 2017) ) suggested an alternative approach that learns context throughout a breaking news story to determine if a tweet constitutes a rumour.”

Zubiaga et al. build on the hypothesis that a tweet alone may not suffice to know if its underlying story is a rumour due to lack of context.

“They build on the hypothesis that a tweet alone may not suffice to know if its underlying story is a rumour, due to the lack of context.”

Zubiaga et al.'s approach improved performance over Zhao et al.'s baseline and over several non-sequential classifiers, and achieved state-of-the-art recall results.

“Their approach led to improved performance over the baseline classifier by Zhao et al. (2015), improving also a number of non-sequential classifiers compared as baselines.The classifier was also evaluated in terms of recall in this case, achieving state-of-the-art results.”

Recent research shows that using context and interactions can help in determining whether a social media post reports a rumour.

“Recent research shows that the use of context (Zubiaga et al. 2017) and interactions (Zhao et al. 2015) can be of help”