Track description
forChallenges track

Four challenges have been selected for The Web Conference 2018:

• Financial Opinion Mining and Question Answering
  Authors: Andre Freitas, Manel Zarrouk and Brian Davis

  Keywords: Aspect based sentiment analysis, multilingual opinion mining, Fintech, Financial Analytics, NLP for social media, Question Answering

  Abstract:
  The increasing interest and investment around technologies which can support better financial analysis and decision making creates the demand for an increasing dialog between academia and industry. The specificity of the language use and its underlying conceptualizations in the financial and economic domains requires the creation of new fine-grained models and techniques which capture the particular semantic phenomena of this field. This challenge aims at providing an experimentation and discussion ground for novel NLP approaches targeting the interpretation of financial data using the tasks of multi-lingual aspect-based sentiment analysis

  View details: https://sites.google.com/view/fiqa/home

  Schedule:

  • Open Challenge: Financial Opinion Mining and Question Answering – André Freitas – 10mins
  • Best paper for subtask 1: Aspect-based Financial Sentiment analysis with Deep Neural – Shijia E – 15mins
  • Best paper for subtask 2: A Neural Network-based Framework for Non-factoid Question Answering – Nam Khanh Tran – 15mins
• Learning to Recognise Musical Genre from Audio
  Authors: Michaël Defferrard, Sharada Mohanty, Sean Carroll and Marcel Salathé

  Keywords: challenge, audio, musical genre

  Abstract:
  Like never before, the web has become a place for sharing creative work – such as music – among a global community of artists and art lovers. The Free Music Archive (FMA, https://freemusicarchive.org) is a web library of high-quality, legal downloads that are provided freely and openly, and that are curated by established audio curators. Recently, Defferrard et al. (Defferrard 2017) have published a dataset from this library, the FMA dataset, which is a collection of 917 GiB and 343 days of Creative Commons-licensed audio from 106,574 tracks from 16,341 artists and 14,854 albums, arranged in a hierarchical taxonomy of 161 genres. It provides full-length and high-quality audio, pre-computed features, together with track- and user-level metadata, tags, and free-form text such as biographies. The task of this challenge is to recognise the musical genre of a piece of music of which only a recording is available. Genres are broad, e.g. pop and rock, and each song only has one target genre. Other metadata, e.g. the song title or artist name, shall not be used for the prediction. The submitted algorithms shall learn to map an audio signal, i.e. a time series, to one of the 16 target genres. The training data for this challenge will consist of the “medium set”, already available at https://github.com/mdeff/fma. That subset is composed of 25,000 clips of 30 seconds, categorised in 16 genres. The categorisation is unbalanced with 21 to 7,103 clips per genre, but only one genre per track. As the data is public, we will collect test data once all participants will have submitted their solutions to prevent access to the test set.

  View details: https://www.crowdai.org/challenges/www-2018-challenge-learning-to-recognize-musical-genre

  Schedule:

  • Dataset & challenge – Michaël Defferrard and Sharada P. Mohanty – 20mins
  • Detecting Music Genre Using Extreme Gradient Boosting – Benjamin Murauer – 10mins
  • Transfer Learning of Artist Group Factors to Musical Genre Classification – Jaehun Kim – 10mins
• Knowledge Extraction for the Web of Things (KE4WoT)
  Authors: Amelie Gyrard, Mihaela Juganaru-Mathieu, Manas Gaur, Swati Padhee and Amit Sheth

  Keywords: Web of Things (WoT), Knowledge Extraction, Entity Recognition, Natural Language Processing (NLP), Ontologies, Internet of Things (IoT), Healthcare

  Abstract:
  The Web of Things (WoT) is an extension of the Internet of Things (IoT) to ease the access to data using the benefits of Web technologies. Data is generated by things/devices and then exploited by more and more web-based applications to monitor healthcare or even control home automation devices. There is a growing interest within standardization in designing models to represent devices and produced data as demonstrated by the following standards. The purpose of this challenge would be to automatically extract the knowledge (e.g. the most common concepts and properties) in already designed and available Knowledge Bases (e.g., datasets and/or models) released on the Web. We will focus on KBs from standards, and/or ontology-based WoT research projects applied to numerous domains. It will demonstrates that the complementary knowledge is constantly redesigned in different communities.The purpose of this challenge would be to automatically extract the knowledge (e.g. the most common concepts and properties) in already designed and available Knowledge Bases (e.g., datasets and/or models) released on the Web. We will focus on KBs from standards, and/or ontology-based WoT research projects applied to numerous domains. It will demonstrates that the complementary knowledge is constantly redesigned in different communities.

  View details: http://wiki.knoesis.org/index.php/KE4WoTChallengeWWW2018

  Schedule:

  • Introduction to the challenge – Amelie Gyrard, Manas Gaur, Swati Padhee, Amit Sheth, Mihaela Juganaru-Mathieu
  • Posters and demos responding to the challenge tasks (TBC)

• Question Answering Mediated by Visual Clues and Knowledge Graphs
  Authors: Fabrício Faria, Andre Freitas, Tingting Mu, Alessio Sarullo and Ricardo Usbeck

  Keywords: Visual Question Answering, Visual Knowledge Graph, Content Based Image Retrieval

  Abstract:
  This challenge focuses on the use of Web data to support semantic Visual Question Answering: given a large image collection, find a set of images matching natural language queries. The task will support advancing the state-of-the-art in Visual Question Answering by focusing on semantic representation and reasoning mechanisms. In order to address natural language queries, participating systems will need to integrate graph descriptors extracted from images to use structured and unstructured Web data sources.

  View details: https://visual-question-answering-challenge.github.io/

  Schedule:

  • Description of the challenge, including the tasks available and the resources – André Freitas
  • Discussion on the response from the community – André Freitas