data-for-change / anyway

ANYWAY - Car accidents map
http://www.anyway.co.il
MIT License
77 stars 243 forks source link

Epic: News Flashes improvements #1640

Open atalyaalon opened 3 years ago

atalyaalon commented 3 years ago

We have a few improvements we can add to location extraction from news flashes. First step will be to read and understand flow of news flashes insertion (and their infographics) in the system Relevant python command for this processing: python3 main.py process news-flash

Second will be reading suggested tasks below and related issues in algorithm tasks here Third will be to create and prioritize tasks.

The goal is: increasing the number of news flashes we show in FE and increasing their location accuracy.

  1. Improving location accuracy for רחוב resolution For example see these news flashes We were very wrong in the street. This whole flow should be improved

  2. GPS in news flash table can be not precise (not in segment at all) - we can improve this using either waze data or CBS data of accidents in segments).

  3. We have waze data - @warik21 started working on an algorithm that can improve interurban accidents location using waze data - let's push this forward (we will soon have United Hatzala data - @carmelp16 is working on it).

  4. Nowadays we only show resolution of כביש בינעירוני - and soon צומת בינעירוני - (will be added in this issue) and street (in this pr) However there are news flashes in which have non of these resolutions, or have these resolutions but we didn't recognize the road_segment / street accordingly.

  5. Adding arabic news flashes - and extraction of "CBS" location - so all news flashes "speak" in the same language.

  6. Extract relevant keywords from NewsFlash - some work was done here but not merged. This can be useful for some widgets that are based on news flash context.

  7. Create an improved classification algorithm for accident/non-accident.

  8. Combining multiple news flashes that refer to the same accident - create an algorithm that detects and relates them to the same accident - and by that perhaps create an improved location for all.

atalyaalon commented 2 years ago

After a discussion, we're focusing on no.1 and no.5 https://github.com/hasadna/anyway/issues/1740 https://github.com/hasadna/anyway/issues/2091 The rest of the tasks will wait for now