OpenIE Demo
From Knowitall
Contents
Large Projects
- Learning algorithm for search result order.
- Grouping for queries with multiple free query slots. Might conflict with previous item.
- Classify negative and positive extractions.
- Create a reduced index by moving some filtering logic from the frontend to the index.
- Creating an in-link ratio for removing over-linked entities.
- Adding resolver or other synonym database (e.g., Freebase also-known-as or Google crosswikis)
- Creating a simple pipeline for adding new data to the demo (e.g. from current news text)
- Add OLLIE extractions
Small Projects
- Freebase autocompletion for types and entities, preferably not using old hacky method.
- Make popovers persistant.
- Progress bar on query.
- Better logic for filtering out near subset types.
- New icon.
- Remove bad FreeBase types.
Ideas
- Show all unique sentences if any sentence contains an extraction instance with confidences > threshold. Presently we don't show any sentences with extractions < threshold.
Goals
- Presentable on the smart phone
- Queries complete in less than 10 s
Queries
Good
- (type:Film, based on, true story)
- (*, symbol of, type:Country)
- What companies sell which cars?
- What actors starred in which movies?
- domesticated in
- type:Person study
Spam
Restarting the Web Demo on Reliable
- If the web page is up at http://openie.cs.washington.edu but queries are failing then the SOLR instance is likely down
- Login as knowall to reliable.cs.washington.edu
- cd to /scratch3/common
- execute stop-all.sh
- execute start-all.sh, this will create screens for SOLR Cloud instance of the OpenIE data
Restarting the Servlet Process
Stop the monitoring application
- Log in to recycle as knowall via SSH
- Run the command 'screen -ls' and check for a detached screen called 'monitor'. If it is running:
- Switch to it using the command 'screen -r monitor'
- Press Control-c to kill any process that might already be running.
Update the server
- Log in to reliable as knowall via SSH
- Run the command 'screen -ls' and check for a detached screen called 'openie-frontend'. If it is already running:
- Switch to it using the command 'screen -r openie-frontend'
- Press Control-c to kill any process that might already be running.
- If the 'openie-frontend' screen was not already running, start a new one with the command 'screen -S openie-frontend'.
- Run the command 'sbt "start 8088"'. This will start the extraction browser servlet listening on reliable.cs:8088, to which openie.cs forwards requests.
- If everything works normally, stdout should display a small amount of output ending with the text "STARTING", without any error messages or stack traces. At this point, the Servlet is ready to serve demo requests.
- Verify that the demo is accessible at http://openie.cs.washington.edu and http://reliable.cs.washington.edu:8088, and check that a few example queries behave correctly. (If only the latter URL works, support should be able to help, refer them to wreq #173026)
- Detach the screen session by pressing Control-a d.
Start the monitoring application
- Log in to recycle as knowall via SSH
- Run the command 'screen -ls' and check for a detached screen called 'monitor'. If it is running:
- Switch to it using the command 'screen -r monitor'
- Run the monitor, using the previous command in BASH history.
Updating the source code
- Kill the existing process
- 'git pull' or 'git stash', 'git pull', 'git stash pop' if there are changes
- Start the demo as normal
Running with Solr
See Solr#Open_IE_demo.