Thank you so much to everyone who submitted their executive summary drafts by the midpoint deadline. We were very excited to see so many high quality submissions, and we loved that people are using the challenge as an opportunity to grow their NLP skills.
It was difficult to choose just 3 most promising midpoint submissions to receive bonus prizes. The finalists for the bonus prizes were notified on September 13, 2023.
Below, we provide feedback and guidance on building strong submissions based on the midpoint submissions. As you prepare your submissions, be sure to follow these tips and to leverage the resources and guidance provided in the problem description. (And don’t forget about the bonus prizes!)
Methods
- Focus on the narrative field! The strongest submissions in this challenge will primarily explore the narrative field, using other variables to enrich insights that come from the narrative field.
- Describe your methodology in detail (even when it didn’t result in a concrete finding). Overall, submissions were strong in their descriptions of the methodological approach and reasoning. Keep it up! Be sure to include the approaches you took and your reasoning, including both the approaches that didn’t pan out as well as those that resulted in a key finding (… you could even win a bonus prize for the most interesting null result).
- Consider your downsampling strategy. One approach to working with constraints on computing resources is downsampling. Conclusions from properly drawn random subsamples can still generalize to larger populations. An alternative approach is to focus on a particular demographic or ‘fall type’ and analyze the entire sample relevant to that grouping. This would reduce computational needs but also provide findings for the entire (sub)sample, and solvers could discuss or explore other groupings or additional research questions the established methodology could address.
Executive summary write-ups
- Make your claims as detailed and specific as possible. Many of you found patterns and fit interesting models relating to fall outcomes. Most reports described planned next steps methodologically, which makes sense for these midpoint drafts. The most competitive submissions will explore various possible methodological approaches and dig into preliminary patterns to look for detailed, specific insights. What features (native to the dataset or engineered) did your classifier use to predict your outcome, and what can we learn from that? Can you characterize clusters or types of falls in clinically meaningful ways?
- Provide rationales for your analytic choices. Communicate clearly about your rationales and hypotheses, e.g., why did you choose this method, what did you expect to happen, and what did you learn from the results. Keep in mind that preprocessing, including removing or standardizing information, is a critical part of analysis.
Submission requirements and formatting
- Mind the submission requirements. In your final submissions, adhere to requirements like the executive summary page limits for text and figures. Use your notebooks to provide full detail and your executive summary to summarize your observations into a tidy, single page of text (with up to two additional pages of figures, tables, and references). Please do not include datasets or other files you’ve used in your exploration; the submission page was not designed to handle large files.
- Ensure that text in figures is legible. If your data visualizations include text, ensure that the text is readable.