A Look into the Project Results

Although SocialSensor is still in progress (approaching the end of its second year), there are already practical outcomes that could be of interest to the research community. Here, we will briefly summarize these outcomes providing pointers to further resources and giving an outlook into the future. At the moment, there are two basic types of outcomes: datasets and software.

In terms of datasets, there are at the moment two publicly datasets available. The first is the Social Event Detection 2012 (SED2012) dataset, consisting of 167,332 CC-licensed Flickr images and event annotations for 149 target events, as well as an evaluation script for deriving performance metrics (measuring event detection accuracy). The dataset was used for the Social Event Detection task of MediaEval 2012. The second publicly available dataset actually consists of multiple datasets collected from Twitter in 2012 around specific news events: the Super Tuesday in the US, the FA Cup final, and the US Elections. Apart from the collected tweet IDs, the dataset also contains ground truth information for a list of manually created topics, which should be detected by topic detection methods. In addition, an evaluation script is provided for automatically assessing the topic detection accuracy of different methods. Apart from the aforementioned datasets, SocialSensor will soon make available the data and ground truth for the Social Event Detection 2013 task, as well as other pertinent datasets, e.g. concept detection in multimedia, etc.

In terms of reference implementations, interested research can already download and use several topic detection methods implemented by the project (described in the Featured Article of this e-letter), a framework and experimental testbed for large-scale similarity image search (also available as an open-source project), a reference implementation of the event detection approach used for the Project participation in SED2012, as well as a framework for detecting and characterizing evolving communities in graphs created from Twitter interactions (also available as an open-source project). Another interesting outcome of the project is the development and availability of the ThessFest API, which is used to power the mobile apps and online dashboard of the Thessaloniki International Film Festival.

Soon, we plan to release more software, for instance an open-source implementation of the social crawler and indexer described in the respective e-letter paper, and an efficient implementation of concept detection in images. The Project maintains a SocialSensor GitHub organization, where the open-source results will be made available.

For a more up-to-date list of the results, check out the project Dataset and Software pages!