The Emerging News task at MediaEval 2021 involves finding emerging events in a real-time stream of news and Twitter messages, and providing relevant insights for journalists like identifying the tweets related to the news stories. Both news texts and tweets are represented as RDF-graphs following the Event Description Ontology. The RDF-graphs describe the named entities identified in the text using Linked Open Data resources from Wikiedata and DBpedia. For news organisations it is critical to identify events and emerging situations as soon as they appear, delays can cause economic and audience losses [13]. Keeping journalists and readers up-to-date is a highly demanding task. News agencies spend a lot of time and human power on continuously monitoring social media, TV programs, radio shows and blogs looking for new events. Artificial Intelligence (AI) and Big Data can assist news agencies and alleviate journalists on this tedious task by distilling events from the diferent media channels and assess their newsworthiness, while keeping journalists in the loop for human judgement. The vast amount of information that is continuously broadcasted on the internet makes it significantly challenging for journalists to distil daily events [8]. For example, Twitter publishes more than 500 million tweets a day (i.e., an average of 5700 tweets per second) [10] and more than 10000 English news articles are published online every day worldwide [9]. Some news agencies already employ software solutions to support events detection and assessing their newsworthiness [6]. Automating the detection of emerging events from social media and news feeds can help news agencies discover new events when they are not the first ones to cover them.
The Emerging News task aims to explore novel ways to detect
emerging events from a stream of social media messages and news feeds.
We define emerging events as those newsworthy events [
Participants are expected to develop a real-time solution that
identifies emerging events and relates social media messages to
them. The solution must read from a stream of news-related items
and output those stories that could be considered emerging events.
These news-related items are semantically represented using RDF
graphs following the Event Description Ontology [
This task is proposed in context with the News Angler project [
As part of the News Angler project, we developed an evolving
big-data platform that harvests potentially news-related
information in real time from textual sources, such as social media,
websites, commercial news aggregators, and open reference sources [
Participants at the Emerging News task have access to the RDFgraph stream of social media messages and news feeds produced by the News Angler platform. The data stream can be accessed through an API from where participants can get JSON-LD objects that contain semantic metadata along with strings representing the RDF graphs serialised in TURTLE. We expect participants to use the JSON-LD stream as input of their proposed solutions. Participants can choose to use either a continuous stream or time-windowed batches of, for example, 5, 10, 15, 20 minutes. We expect participants to discuss the most suitable set-up for their solutions. As an output, participants’ solutions must provide a group of JSON-LD items that belongs to the emerging event or a single JSON-LD item that is an emerging event (we leave it to the participants’ decision too). Optionally, participants can provide a user interface to better interpret and evaluate the results. 3
Data is accessed trough an API that returns one JSON-LD object from the current real-time stream on every call. The real-time stream provides tweets and news as soon as they are published. It outputs tweets from more than 70 accounts comprising diferent news agencies and journalists accounts, and a sample news articles (i.e., between 100 and 500 articles every 15 minutes) from news sources and blogs over the web.
Each returned JSON-LD objects has an URI for the news-related
item and a string representing one RDF graph in TURTLE notation.
The URI contains an MD5 hash of the original news or tweet URL.
Because the News Angler platform harvests news from news
aggregators as well as from original sources, it may contain duplicates,
hence, the URI can be employed to remove those duplicates. The
RDF graph representing the news-related item is described
following the Event Description Ontology [
In order to improve and facilitate the replication and dissemination of the results, we ask participants to (a) publish the proposed solution, associated files, AI/ML models and documentation files using the MIT copyright licence; (b) release the proposed solution as a dockerized API with its associated Dockerfile, and instructions on how to run it (optionally, it can be accompanied by another code or docker image that simulates the data ingestion); and (c) design a solution that does not require more than 16GB RAM and a quad-core CPU to run. We believe these constraints promote the accessibility of the proposed solutions increasing the research dissemination and contribution to a large audience, as well as facilitating the evaluation of the proposed solutions.
The proposed solutions will be evaluated based on their relevance by a panel of experts with relevant background in journalism and media. The experts will use the developed solution to judge if the information provided by the solutions can be considered as an emerging event or not and how useful the information is. The panel will focus on (a) the newsworthiness and completeness of the reported event, if the information related to the event provides enough insights to conform a news story; and (b) the relevance, if the emerging event reports a new aspect of an existing story or a new story. During the evaluation, all participants will use the same data set. Participants must not assume experts know RDF, therefore experts will base their decisions on the source text. 5
This was the first edition of News Emerging task and it lacked a well-established standard data-set for evaluation. Because of this, we designed an evaluation method based on experts to judge the quality of the outputs. At the same time, it forced participants to explore solutions that work on real-time data, instead of being tuned to particular data-set characteristics.
The Emerging News task is focused on RDF-graph representations instead of using text-based approaches. While this decision reduces information from text, it also reduces ambiguities, facilitates data enrichment from external sources and enables reasoning and structural marching. The resulting solutions can perhaps be applicable to other types of news-related items like images, where there is no implicit text.
Supported by the News Angler project funded by the Norwegian Research Council’s IKTPLUSS programme as project 275872.