jessmatthews1
commented
6 years ago - Problem definition: It is helpful to hear your articulate the multiple views of the problem and how that is confusing. I have focused more narrowly on one problem.
- Methods: I don't want to go into too much depth on the solution but I think it will make the solution more compelling if I some details about how it would work.
Nice work on your big idea, Jess. I know very little about public companies in general, so it was interesting for me to get a better sense of a problem in that space and some ideas about solving it.
In the state-the-problem section, it seemed like you were describing two different problems: 1) the qualitative portion of companies' annual reports, once written, are too lengthy and aren't read by most consumers of the report, and 2) the reports are too costly to generate because of the many hours it takes analysts to write them. Those both sound like pretty compelling problems. It seems like for most of the middle section, you were describing using NLP to solve problem 1, but I wasn't completely sure. I would stick to just one of those problems in the first section to make it clear what you're trying to solve.
Additionally, in your NLP section, I find myself wanting to know more about how exactly NLP would be used to make the process of understanding these reports easier. It doesn't have to go into great technical detail (I struggled with this somewhat in my own idea because I don't know a ton about the particular methods I described). At the very least, though, I want to know what you imagine the output of the NLP methods to be. Would a natural language processing model analyze the sentiment of the report? Would it predict which categories out of some set the report falls into? Would it predict how much a company is likely to be worth in x number of years? Would you use NLP to cluster a bunch of reports to try to figure out industry trends? And what data would you use to train any of these potential models? There are a lot of different scenarios you could imagine, and I think being more specific about what the end goal is would make your idea stronger and wouldn't necessarily require tons more technical detail.
This idea seems like a good, high-impact choice. You're suggesting that a lot of information is already out there that could help stakeholders make better choices, and what they need is tools to be able to handle the vast quantity of information. It's a really compelling problem. I'm excited to read the final version!
Emily