DRSR research meeting, 28 March at Schiphol

About designing the research and understand how to collaborate with stakeholders.

Deniz and Pinar are not at the same level of process. Deniz needs to attend courses and the 8-month paper is still quite ahead. Pinar has done courses already and next month (late april-early may) she submits her 8-month paper.

We follow 2 production lines:

• Space of places (physical)
• Space of flows (data)

And we explore the space in between these two with BUF and the choice of proxies

Case study on public space:

• Physical aspect
  • understand the work with afdeling reiniging
  • Interviews with city district + cleaners
  • Participatory observation with reinigers
  • Analysis of the situation and proposing interventions
• Date aspect
  • Data cleaning
  • Connect to the BUF
  • Identify proxies
  • Data analysis: understanding what has been already done and what datasets are available.
  • Getting the data
  • Intervention

We try to identify which aspects of this study can be done in collaboration between data and social sciences. The result of the social science analysis can be used for identifying proxies, and also interventions can be carried out together.

Requirements data science PhD:

• The work cannot be framed as social policy
• Contribution to MAS
• Single cases cannot be used in papers
• Relation between spatial flows and space of places has to be clear
• General understanding for publishing 

Requirements social science PhD

• Cleaning can be rephrased as a social problem
• Contribution to urban planning
• Contribution to ZO policy 

Social geography can be a good angle that integrates both disciplines. Space of places and space of flows. Negotiation between data and social is always active.
 

CRISP-DM Model

This is an industry standard method that aims to guide the data mining processes when working with non-data-expert stakeholders. During the meeting, we attempted to create an analogy of this model to be used in social sciences, to ensure that will be at similar stages of our processes. The model has 6 main steps: problem understanding, data understanding, data preparation, modelling, evaluation and deployment. To these steps, we assign tasks both for data science and social sciences.

Problem understanding:

• Data: Focuses initially on understanding the objectives and the requirements of the project with respect to the context that they exist in. The next task in this part is to translate/convert these context understanding into a data mining problem definition. After this translation, a preliminary design is created to achieve the objectives in both domains (Ideally, they should align). How the data can help the context problem is defined at this stage.
• Social: Focuses on understanding the research problem and case study understanding (general question, understand the space of places, and space of flows)

Data understanding:

• Data: Building on top of the problem identification performed on the last step, this step is initialized by developing an understanding of the existing datasets. Previous related work from both academic and private backgrounds can be investigated to supplement this. Once the relevant datasets are identified, an initial exploration of them is also performed here. This process called exploratory data analysis (EDA) may result in visualizations which help us understand the data quality problems and help us gain initial insights into the data. Results of EDA can also be used to check back at the previously defined requirements and estimate how adequate the dataset is to answer them.
• Social: explore what methodologies of analysis are possible 

+ Extra step: Evaluation of the understanding

Data preparation:

• Data: This process starts with the initial datasets and the final result should be a dataset that is ready to be an input to a model. Manipulating, transforming, cleaning and formatting of the data is performed as it is most fitting to the data at hand, without a fixed order.
• Social: Interviews, qualitative data collection

Modelling:

• Data: Selection and application of various modelling techniques. followed by calibration and optimization of model parameters. Typically, multiple modelling approaches are compared to each other regarding how well they perform in answering the initially defined problems. Because multiple models are applied, it is very common at this stag to go back to the data preparation step.
• Social: bring together results

Evaluation: 

• Data: The models that were built and ran at the last step are put to a quality test here. The steps of creation of the models are reviewed to ensure they work as intended. It is at this stage that we look at the initial requirements and as ourselves "Did we answer every objective/problem that we initially decided on?" If we missed or failed to answer any of the initial objectives, a decision on how to handle this situation should be made here.
• Social: go back to beginning to evaluate

Deployment:

• Data: This step is about the implementation of the results to address the initial goals. The knowledge attained from the model is organized and made into a format that is presentable to the initial problem owner. This step can be putting up dashboards on a website or delivering high-quality visualizations that give a message.
• Social: propose rhythm intervention

* The steps marked in bold above are the important ones to collaborate with OIS and Stadsdeel (and other stakeholders).

The figure above shows an outline of the phases of CRISP-DM methodology. Bolded text show generic tasks and the italic text indicates the outputs that are expected from said tasks.

Solution – case working:

• An alternative way to CRISP-DM model.
• A business-minded way of working that relies on a solid understanding of the initial problem.
• Delivering proof of concepts and informing on cases.
• Spending more time on problem understanding.

We try to map the steps for urban/rhythm-based working, on the four main steps of this model. Below are the data steps in correspondence to urban/rhythm steps:

1. Determine business objectives / define a goal or task
2. Assess situation / understanding the context
3. Determine data mining goals / identify rhythms
4. Produce project plan / concept development
5. Additional step: verifications, comparing with other measurements (?) 

Expected final results:  in each short case study (4 times 3 months), we follow the steps above, with the aim of achieving conclusions below:

1. State of the art (introduction case)
2. What Amsterdam has: data, social, literature reviews
3. Bring together results: identify rhythms in 2 domains
4. Possibilities for big case: identify the potential for the big case 

It is important that the conclusion of the small cases end up with a reflection on the potential for the big case. So this step will define rhythm results and discuss whether these are interesting.

Question: What experiments can we do to make the argument that rhythm analysis exists?