Individual text corpora for the language modeling of personality, knowledge and intelligence

Markus Hofmann, BUW

9.1.2025, 16:15, lecture hall 18, O.06.22, colloquium of the institute of psychology

The CNM is a special interest group on neurocognitive models hosted at the chair of General and Biological Psychology. It is in part funded by the long-term infrastructure priority program New Data Spaces for the Social Sciences. The goal of this program is to develop standards and best practices for data protection, when digital big data are integrated with survey data. Our sub-project relies on web search and web history data to train language models for each participant. Then we compute the similarity of the web information to terms describing personality, for instance. This individual similarity is then used for the predictive modeling of survey data. The speaker will present the first pilot study successfully predicting openness to experience. The predicted openness will be compared with the survey-based openness with respect to the prediction of intellectual interests, knowledge in humanities and level of education in terms of Ackerman’s (1996) theory of intellectual development. We like to discuss the planned studies in this project with the audience. As these are quite sensitive data, Markus Hofmann spent a lot of time with data protection issues. To share this knowledge with other researchers, he became data protection coordinator for third-party funded projects in the School of Human and Social Sciences. The final part will introduce into the very basic concepts of data protection and the workflow for data protection documentation for these researchers. After the talk, we will offer some drinks.

The speaker will also announce a one-day workshop on the predictive modeling on clinical trajectories (10am-16pm). There will be an introductory lecture on standard predictive modeling methods (penalized regressions, support vector machines, random forests), which are exemplified by the prediction of successful depression and psychosis therapy, as well as the prediction of cognitive decline during aging. The lecture will be followed by two hands-on sessions with JMP Student , which is freely available for all students and researchers with a university email address. Interested researchers and students can select preferred dates here no later than February, 23. Then the final date, program and room will be announced on the present homepage. 

Beyond performance: using large language models to understand human cognition during reading

Lavinia Salicchi, Hong Kong Polytechnic University

9.1.2025, 10:15, Z.02.05, colloquium of General and Biological Psychology

From the project, we also invited an external speaker for the inauguration day: Lavinia Salicchi is a computational linguist who won a shared task challenge on the prediction of eye-movements during reading. To this end, she used large language models along with other psycholinguistic predictors in a predictive modeling approach. In her talk, she will present recent research on the prediction of psycholinguistic data by such machine-learning approaches. Other interested researchers and students are cordially welcome to join her talk.