WDAqua team is looking back at the successful ESWC summer school dedicated to Data Science that took place 5-10 September 2016 in Dubrovnik, Croatia (see http://summerschool2016.eswc-conferences.org/). During the summer school week, WDAqua ESRs enjoyed tutorials on a variety of Data Science topics, actively participated in interactive hands on sessions, presented their work during poster sessions and worked in teams to conduct exciting research projects. We are very pleased to see WDAqua ESRs bringing home a lot of awards for their poster presentations and projects. Congratulations!
Let us listen to the voices of the award winners:
The poster was an incentive to try the first version of the QA system developed by WDAqua. Many people visited the poster and asked their questions during a live demo. The poster obtained the first prize for its novel way to approach researchers. By directly asking a QA system and looking what is happening behind the scenes it allowed to better understand how QA works. If you are also interested visit: http://www.wdaqua.eu/qa.
This poster outlines my PhD project and presents a mixed methods study, which aimed at better understanding how professionals search for, and work with, structured data. The results of this research can be used to inform the design of data search or data exploration tools and their user interfaces. The poster session prompted many interesting discussions and I received feedback on our study, which was very useful and inspiring.
The poster presented a response generation system that assumes that participants in a conversation base their response not only on the previous dialog utterances but also on their individual background knowledge. Throughout the poster session, I was provided with the opportunity to have fruitful discussions with fellow students and tutors about not only the content of my poster but also the broader research field of the dialogue systems, which is of great interest to me. Their valuable feedback is greatly appreciated.
The project was inspired by the Kaggle challenge: Detecting Insults in Social Commentary (https://www.kaggle.com/c/detecting-insults-in-social-commentary). It involved the design of a system that would be able to recognise hateful comments before or even after they are manifested on social media. Together with the other members of my team, we had the opportunity to explore a challenging problem in terms of both its technological aspects and its social application. History dictates that people tend to be very creative in terms of the ways with which they would express hate in a conversation. Consequently, training an accurate hate-speech detector is not a trivial procedure. As an outcome of the excellent division of labour and synergy between us, in 2 days’ time, we managed to experiment with a variety of classification models, one of which outperformed the winner of that particular Κaggle challenge. I wish to address a special “thank you” to Anke, Denis, Hady and José for sharing this summer school experience with me, and, with whom I hope to collaborate on future projects.
In this project we developed the ‘Recipator’, a tool to generate new cooking recipes based on a set of existing recipes crawled from the BBC good food recipe collection. In order to produce “cookable” recipes we extracted sequences of ingredients and their corresponding actions from the crawled dataset by applying basic entity detection and nearest verb extraction, e.g, (cook-potato). To assure the novelty of the recipes, we classified the ingredients using DBpedia Spotlight. Those sequences were used as training data for a Markov model with an additional independence assumption. The probability matrix was used to boost the ‘cookability’ of newly generated recipes. The outputs were randomly generated recipes based on the existing recipe collection.
This project was inspired by recent news that recipes section on BBC online portal is going to be closed later this year. Our group performed analysis over recipes data crawled from the BBC food subportal.