Die Erholunsgzone vor dem D4 Gebäude über dem Brunnen.

Abstracts

  • Camilla Damian - Detecting Gender Bias in Children's Books

    Gender stereotypes form early in the child’s development and are carried over throughout adolescence into adulthood, leaving long-lasting effects which may impact activity and career choices, as well as academic performance. Books, in particular, can have considerable influence, as their characters serve to shape role models of femininity and masculinity for young children. Thus, gender under- and misrepresentation in children’s textual literature can contribute to the internalization and reinforcement of stereotypes. To address this issue, we aim to identify and measure relevant dimensions of gender bias in children’s books with the aid of both qualitative and quantitative techniques: systematic literature review across disciplines on the one hand and natural language processing (NLP) methods on the other. By exploiting such an integrated research framework, we believe that it is possible to automate the detection of potentially biased text while enhancing the interpretability and transparency of the results.

  • Sourav Adhikari - Combining Topic Modeling and Word Embedding to Predict Match Outcomes in Football

    This project proposes an approach to predict match outcomes in football by leveraging text data to learn features of individual teams and individual matches. The dataset comprises newspaper match previews along with bookmakers' odds, and the objective is to predict outcomes of football games in terms of home team win, draw, or away team win.

    Team-specific features are extracted using Word2Vec word embedding based on the skip-gram approach, which consider the entire corpus of match previews to provide nuanced representations for each team. Second, topic modeling is employed to derive match-specific features in the form of topic proportions. This step enhances predictions by capturing the specific underlying themes for each match. The model further incorporates bookmakers' odds, reflecting the perceived outcomes. These features are fed into a random forest classifier to make predictions.

    The model's accuracy is then assessed in comparison with actual results, bookmakers' odds and existing prediction methods. The holistic approach presented aims to offer an alternate and informative approach to football match predictions, by using entity specific and match specific features as determinants of match results.

  • Daniel Winkler - B-DiD Bayesian estimation of causal treatment effects in quasi-experimental settings

    > Abstract
    > Overview

  • Jan Greve - The Use of Riordan Arrays to Characterize Some Random Combinatorial Structures

    Probability distributions on random combinatorial structures such as partitions, permutations, trees and graphs are often representable as a random walk on the corresponding Bratelli diagrams. Here, we consider a family of distributions on random combinatorial structures whose sufficient statistics can be embedded in a nonnegative 2D lattice, thereby constructing a weighted Pascal triangle. Such distributions arise in many applied works such as the exchangeable partition probability function of Pitman-Yor processes commonly referred to as Ewens-Pitman partitions. We show that elementary computations involving Riordan arrays enable us to compute factorial moments of the sufficient statistics of these distributions.