Main Content

Seminars

Seminar PhD program Data Science

This seminar is mainly intended for the members of the structured PhD program Data Science. However, guests are always welcome.

The seminar takes place every two weeks during the lecture period and usually lasts about one hour. The next dates can be found in the list below.

Past dates

January 12, 2023, 2:15 p.m., Room 04A23 (HS V)

lecture: Allie Lahnala (PhD student, Group Prof. Flek)

Titel: A Critical Reflection and Forward Perspective on Empathy and Natural Language Processing

Summary: We review the state of research on empathy in Natural Language Processing and identify the following problems: (1) The lack or abstractness of empathy definitions, leading to (2) low construct validity%CC%88t and reproducibility. Furthermore, (3) emotional empathy is u%CC%88overemphasized, which directs our focus to a limited number of simplified tasks. We believe that these problems hinder research progress and argue that current approaches would benefit from a clear conceptualization that includes the operationalization of cognitive empathy components%C3%9Ft. Our main goals are to provide insight and guidance on the conceptualization of empathy for%CC%88r NLP research goals and to encourage researchers to pursue the neglected opportunities in this area that are highly relevant for%CC%88the clinical and educational sectors, for example.

January 26, 2023, 2:15 p.m., Room 04A23 (HS V)

lecture: Anne Kopsch (PhD student, Group Prof. Dahlke)

Titel: Construction of multiwavelets for general dilation matrices

Summary: This talk is concerned with the construction of wavelets and multiwavelets. In particular, we%CC%88would like to identify minimal requirements so that construction is mo%CC%88feasible. In addition, we%CC%88would like to minimize the number of parent wavelets. This can be achieved by including general dilation matrices in our construction procedure, whose determinant is related to the number of mother wavelets.

February 9, 2023, 2:15 p.m., Room 04A23 (HS V)

lecture: Joan Plepi (PhD student, Group Prof. Flek)

Titel: Unifying Data Perspectivism and Personalization: An Application to Social Norms

Summary: Rather than using a single ground truth for%CC%88language processing tasks, several recent studies have investigated how the labels of the set of annotators can be represented and predicted. However, often little or no information is known about the annotators, or the set of annotators is small. In this work, we study a corpus of social media contributions u%CC%88ber conflicts from a set of 13,000 annotators and 210,000 social norm ratings. We present a novel experimental setup that applies personalization methods to annotator modeling, and compare their effectiveness%CC%88t in predicting perceptions of social norms. Furthermore, we provide an analysis of performance in subsets of social situations that differ according to the na%CC%88he of the relationship between the conflicting parties, and evaluate where personalization helps the most.