class: center, middle, inverse, title-slide .title[ # Projects in Computational Modelling of Social Systems ] .author[ ### David Garcia, Petar Jerčić, Jana Lasser <br><br> <em>TU Graz</em> ] .date[ ### Computational Modelling of Social Systems ] --- layout: true <div class="my-footer"><span>David Garcia - Computational Modelling of Social Systems</span></div> --- # Project basics - Participation in group presentation sessions (both as presenter and discussant): 25% of final grade - Project report and codes: 25% of final grade - Timeline: - 2.06: First project feedback exercise session - 9.06: Second project feedback exercise session and general Q&A session - 10.06: last chance to get any feedback from lecturers online via discord - 11.06: Deadline to register your group in Teach Center. - 23.06: First presentations session - 30.06: Second presentations session - 10.07: Project report deadline --- # Projects objectives With the projects, we want to evaluate the following skills and learning: - Planning a computational project as in the exercises and course examples: model, simulation, analysis, interpretation - Describing a model in detail, referring to other relevant models (especially if covered by the course) - Ability to implement and code the model, showing it works as designed with visualizations and other outputs - Assessing the project question with systematic analysis of simulations --- # Project routes A) Reproducing the analysis of a research paper - Example: Reproducing the results of Axelrod's culture model paper - Reflect on similarities or differences in your results - If paper is short, explore a small extra question or analysis B) Extending a model from the course - Example: Neighbor overlap in the Barabasi-Albert model - Motivation: online we don't only interact with our friends in a direct manner, but also through common friends. Users with larger and more clustered neighborhoods will be less likely to leave - Analysis: measuring average neighbor overlap as network grows - Question: does overlap change with node age and network size? --- # Project routes (II) C) Designing and analyzing a new model - Example: privacy concerns in social resilience - Model probability to leave a social network increasing with too many friends - Study % of users leaving with value of parameter of that increase - Check previous work! Is this really a new model? - Can be the most challenging approach ** Data is not always necessary but desirable in some cases** - Data-driven simulations (e.g. using network data) - Comparing outcomes with empirical data (hard) - Quantitatively validating outcomes (harder) - Calibrating dynamics (hardest) --- # Modeling project pitfalls - **Not understanding your question** - Bad: *I want to model how people play Pokemon Go* - Good: *I want to understand the role of city size in Pokemon Go players using only two teams* - Risk: not knowing when you are done and if you are doing it right - **Not searching for previous work** - Bad: *Here is the first model ever for fashion* - Good: *My model is similar to (cite,cite) in this and that aspect* - Risk: specifying the problem wrong or missing interesting parts --- # Modeling project pitfalls (II) - **Not being systematic** - Bad: *I made some simulations and they look like this* - Good: *Outcome measure X grows with parameter Y as shown in the boxplot* - Risks: Not having any substantial analysis and just anecdotes. Going back and forth between model and simulations in an endless loop. - **Being too ambitious** - Bad: *I want to explain polarization* - Good: *I want to understand how the strength of social influence affects polarization of opinions in the social network of German politicians on Twitter* - Risks: Too much work and open paths, large models lead to complicated analyses. Might have too much previous work behind.