Identifying Witness Accounts from Social Media Using Imagery

Abstract

This research investigates the use of image category classification to distinguish images posted to social media that are Witness Accounts of an event. Only images depicting observations of the event, captured by micro-bloggers at the event, are considered Witness Accounts. Identifying Witness Accounts from social media is important for services such as news, marketing and emergency response. Automated image category classification is essential due to the large number of images on social media and interest in identifying witnesses in near real time. This paper begins research of this emerging problem with an established procedure, using a bag-of-words method to create a vocabulary of visual words and classifier trained to categorize the encoded images. In order to test the procedure, a set of images were collected for case study events, Australian Football League matches, from Twitter. Evaluation shows an overall accuracy of 90% and precision and recall for both classes exceeding 83%.

Mining social media for newsgathering: A review

“…They define four types of accounts: (1) witness accounts, or accounts reporting a direct observation of the event or its effects, (2) impact accounts, or those being directly impacted or taking direct action because of the event and/or its effects, (3) relay accounts, in which a micro-blogger relays a witness or impact account, and (4) other accounts, which are users relevant to the event, but do not fall in any of the other three categories. Besides textual content, [84,85] used geotags and image features to perform supervised classification of users as witnesses or non-witnesses. They used unigrams, bigrams and part-of-speech tags as textual features.…”

“…In line with previous research, they also emphasise the difficulty of the task provided the small ratio of witness accounts with respect to non-witness accounts. One limitation of the work by [84] is that they only analyse geolocated tweets, which is a small fraction of the entire Twitter stream, and could therefore be missing out important updates from non-geolocated, witness users.…”

Testing the event witnessing status of micro-bloggers from evidence in their micro-blogs

“…Additionally, seeking evidence of witnessing to test a micro-blogger’s status from a number of micro-blog content sources distinguishes this work in comparison to previous research e.g. [ 7 – 10 ].…”

“…A secondary contribution of this paper is exploring the potential of the search micro-blogger processes. The search micro-blogger processes are distinguished from the search event processes that provide event relevant micro-blogs from hashtag or keyword searches, the on-hash datasets [ 10 ]. This research seeks to further identify evidence in targeted micro-bloggers’ off-hash micro-blogs, and establish whether this evidence improves the certainty of their witnessing status.…”

“…Twitter case studies, Australia Football League (AFL) matches held at the Melbourne Cricket Ground (MCG) are utilized to support this research [ 6 , 10 ]. Sporting events are a popular social search topic [ 13 ] and of interest to journalism, emergency management related to mass gatherings, and event detection and summarization research [ 14 – 16 ].…”