CLiPS, University of Antwerp

Mentors: Student: Pierre Voué

<p>In the wake of the research I have undertaken for my Master's degree's thesis on automatic detection and assessment of right-wing extremists' online speech, I intend to leverage the insights gained thereby to help CLiPS refine and expand their resources in the field of hate speech. It will include, among other possible sub-tasks, the automatic collection of hate speech textual data, the definition of formal annotation guidelines for such data building on existing literature such as "Hate Speech Dataset from a White Supremacy Forum" (De Gibert et al., 2018), the fine-grained annotation of the data for ulterior automatic processing, investigate the best automatic assessment methods for the data (both in terms of performance and explainability), study the opportunities and limitations (from a technical, ethical and practical perspective) of said automatic processing, transpose the results to develop hate speech assessment tools for other languages such as French or German.</p>

Mentors: Student: Fabricio I Layedra

<p>viNLaP is an interactive and data-driven web dashboard with three modules, each one based on one of these three main analysis: Spatial, Temporal and Statistical/Traditional. Each module will include traditional visualizations related to the respective analysis but also novel visualizations based on the proposed ones in the literature. The proposed viNLaP is to visualize in this first scenario: polarized data; but it is built to be useful for new types of dataset that would come in the future.</p>

Mentors: Student: Anna A. Soboleva

<h4>Cross-language analysis of U.S.-Russian relations (Task 3)</h4> <p>Tweets from US democrats (or just anyone actively opposing Trump's presidency) and Russian conservators (or anyone supporting government) are collected and brought to a common denominator via translation-based and non-translation cross-lingual methods. Resulting corpus would be fascinating to research through several perspectives: for example get the political views through speech analysis, comparison of similarities in the vocabulary between people of similar and different political views, etc. Information can be retrieved through the Twitter API (Search or Stream, using Tweepy library).</p> <h4>Early Alzheimer syndrome detection in speech analysis (Task 9)</h4> <p>Research showed that it is possible to detect early signs of Alzheimer Syndrome in speech. I propose to train a model, which will learn to do it (and later make an app, assisting people with the high risk of Alzheimer). Data can be extracted from the Talk Bank (<a href="https://dementia.talkbank.org/" target="_blank">DementiaBank</a>) or, for example, <a href="https://dementiadiaries.org/" target="_blank">Dementia Diaries</a> website. The app could prompt their users to talk about their day and using speech recognition module could analyze the possibility of Alzheimer and changes overtime.</p>

Mentors: Student: Panagiotis Lantavos

<p><strong>Meaningful Adversarial Examples for Natural Language Models</strong></p> <p>A project to create adversarial examples and resulting counterfactuals for text classifiers using the relations in the word embedding vector space. This allows for meaningful alterations to be made in the input documents to test a model for biases. What would happen if the subject of this document was female instead of male? If it was a person of color instead of being white? How would a state of the art model change its results when such changes are made?</p> <p>This project means to address these questions by creating a framework that allows for the testing against such biases as well as the creation of augmented datasets to dissuade their development.</p>

Mentors: Student: Alexander Rossa

<p>This project explores the topics of context-free generation, topic analysis and sentiment analysis. It focuses on work with Seed (an open-source procedural content generator) and extending its capabilities with, for example, conditional generation. Moreover, a real-world application of Seed's capabilities, combined with NLP machine learning backend for topic and sentiment analysis, is used for creation of a Twitter bot which is be able to participate in a debate and calm it down if it gets too heated.</p>