Following the United Kingdom’s decision to leave the European Union on 23 June 2016, the EU-funded SENSEI project has revealed that it accurately predicted the result of the referendum by analysing 6 million social media conversations in the weeks preceding the vote.

The SENSEI (Making Sense of Human-Human Conversation Data) project aims to make sense of the millions of blog posts and social media conversations that take place every day…

>> Read the entire article

On 19/09/2016 a dissemination event called “Language Technology Meets Social Media Conversation in the News” (see: www.meetup.com/textanalytics/events/233885011/) was held in central London at the British Computer Society. Invited were media professionals from various media organisations/media units across the UK and, in addition, members of the London Text Analytics meet up group. The event was organised by USFD and UESSEX (Kruschwitz). Presentations were made by project members from USFD (Gaizauskas, Hepple, Foster) and Websys (Poch) covering SENSEI work on summarisation and on the Brexit poll predictions. In addition, a lively round table discussion was held in which participants at the event as well as the presenters were able to share views on the current state of play with respect to user-generated content in the news as well as future directions. Around 35 people attended, including representatives from the traditional content-generating media organisations (Bloomberg, Telegraph Media Group), media monitoring, personalisation and analysis organizations (Signal Media, Watching America, Lumi News, eRevalue, etc.), management consultancy companies (A.T. Kearney), government (UK Government Digital Service) and universities (King's College London, University of Regina (Canada)).

The Sensei team has participated  in the international “E Commerce Customer Experience” in September in Paris, an event with over 500 companies, 4 specialized exhibition areas and a Start-Up Village.

The Sensei industrial partner Teleperformance had a booth dedicated to the Sensei Project, and Hugo Zaragoza (Websays) and Vincenzo Giliberti (Teleperformance) held a very successful workshop on  “Social Media + Machine Learning = A new Customer Care Experience. The Sensei Project.”.

>> Event Websites

Monitoring Brexit – The Story

In the month preceding the referendum date, SENSEI’s system monitored millions of social media conversations to predict the outcome of the referendum.
Every day, more than 300 000 posts across multilingual media sources on the topic of the UK EU Referendum are captured and automatically analysed by the SENSEI technology. 
Most exit polls were showing confidence the REMAIN side would prevail. In contrast, the SENSEI system hit with very high accuracy the final outcome.

Monitoring BREXIT – What Happened
 

Press

Press Release (English)

Press Release (Italian)

Interview (Italian) at Radio 24 (Italian Radio) about SENSEI monitoring Brexit Referendum

Go to the Radio webpage

or listen it here

SENSEI started to monitor the Brexit Campaign.

Find it at www.sense-eu.info

Press Release here

SENSEI-UNITN team ranked 3rd at the CoNLL 2016 Shared Task on Shallow Discourse Parsing in the end-to-end parsing and 1st in the argument extraction sub-task.
 
The SENSEI team participated to the CoNLL 2016 Shared Task on Shallow Discourse Parsing on end-to-end parsing of English news. Discourse parsing has utility for many other NLP tasks such as summarization and opinion mining. Penn Discourse Treebank style discourse parsing is a composite task of detecting explicit and non-explicit discourse relations, their connective and argument spans, and assigning a sense to these relations. Due to the composite nature of the task, end-to-end discourse parsing is very challenging.

The system developed by SENSEI emphasized argument extraction, as it is one of the most difficult subtasks (and ranked 1st). The approach to discourse parsing consists of token-level sequence labeling with Conditional Random Fields for connective and argument span extraction tasks, and supervised classification for sense assignment tasks. 
The SENSEI system ranked third on the end-to-end parsing task out of 14 participants. The detailed description of the system will be available in the CoNLL 2016 proceedings.

Read the full paper here

The SENSEI team participated to the Semeval scientific evaluation campaign, under the sentiment analysis track 4.A (polarity detection
in tweets). Polarity detection consists in detecting whether the author expressed a positive or negative sentiment in a text. This task
is a basic building block for analysing social media conversations.

The system developed by SENSEI consists in a family of convolutional neural networks (CNNs) trained from multiple views of the input, and combined with a deep neural network. The different views are created by training various flavors of word embeddings, from lexical, syntactic and semantic evidence, on large datasets collected through the project. They are combined by extracting hidden layers of the CNNs, concatenating them as input of the final classifier. This type of embedding fusion helps characterizing correlations between hidden representations of the data which leads to more robust decisions

The SENSEI system ranked second out of 34 participants, with non significant difference with the first participant. Its detailed description will be available in the Semeval proceedings.

Following papers have been accepted at LREC 2016:

• Celli F., Riccardi G. and Alam F., “Multilevel Annotation of Agreement and Disagreement in Italian News Blogs”,  Proc. LREC, Portroz, 2016.
• Chowdhury S.,  Stepanov A. E. and Riccardi G., “Transfer of Corpus-Specific Dialogue Act Annotation to ISO Standard: Is it worth it ?”, Proc. LREC, Portroz, 2016. 
• Danieli M., Balamurali A. R., Stepanov A. E., Favre B., Bechet F. and Riccardi G., “Summarizing Behaviors: An Experiment on the Annotation of Call-Centre Conversations”, Proc. LREC, Portroz, 2016. 
• Barker, E., Funk, A., Paramita, M., Kurtic, E., Aker A., Foster, J., Hepple M. and  Gaizauskas, R., " What’s the Issue Here?: Task-based Evaluation of Reader Comment Summarization Systems", Proc. LREC, Portroz, 2016.
• Funk, A., Gaizauskas, R. and  Favre, B., "A Document Repository for Social Media and Speech Conversations", Proc. LREC, Portroz, 2016.
• Kabadjov, M., Kruschwitz, U., Poesio, M., Steinberger, J., Poch, M., Zaragoza, H., "The OnForumS corpus from the Shared Task on Online Forum Summarisation at MultiLing 2015", Proc. LREC, Portroz, 2016.