search

ropensci / textworkshop18

9 stars 6 forks source link

Interactive interfaces - do we need Javascript as a platform? #21

Open brendano opened 6 years ago

brendano commented 6 years ago

Interactivity poses a major challenge for text analysis applications, since you need not only to do statistical-style query-based analysis (e.g. with R data frames), but you should also be reading and/or annotating/tagging text too. The traditional numerical- and tabular-oriented R and Python interfaces are good at flexible SQL and statistical-style queries, but they're bad at viewing text data.

I'm curious to hear whether people find Shiny useful. I don't use it myself much. I've seen a few nice application-specific corpus analysis interfaces with it.

I'd be happy to talk a little about Javascript (and/or Typescript and other related technologies) as a potential platform for future text analytics work. For better or worse, it's the only way to get decent user interfaces that are widely usable. I keep checking back every few years and I think it's finally gotten to the point where you can be reasonably productive in it. It's much faster than either R or Python and can run either from a commandline or inside a web browser. It would even be reasonable to write statistical and NLP libraries in it, though unfortunately quite a bit would have to be written from scratch. I'll demo a little topic model viewer I made in it last week too.

bmschmidt commented 6 years ago

I'm interested in this session.

brendano commented 6 years ago

Cool! Another idea to consider is embedding interfaces inside Jupyter -- for example, certain libraries can embed scrollable tables now (Pandas, I think? Others?). Text viewers or term list viewers could be nice to embed also. I'm curious if anyone in this community (which I understand to be more plugged-in to R/Jupyter type stuff than I am) has experimented with this.