Snowball Algorithm

(Note: This program requires non-code resources that can be locally accessed. Preprocessing/building the database takes a while. This README will be updated with what resources are needed and where to get them, plus a recommended way to organize the resource directory.)

Based on this paper.

How to run

1. Go to ~/resources/snowball/

2. Make a directory <some_name>

3. Go to <some_name>/

4. Create a file called 'seeds'. Contents are as follows:

   • 1st line: A name for your relation (e.g. located_in). No whitespace.
   • 2nd line: Subject tag and object tag. Must be separated by at least one whitespace character. See stanford corenlp to know which tags are allowed. Make sure they are capitalized.
   • 3rd line onwards: Pair of tokens where whitespace is replaced by one underscore and tokens are separated by whitespace.

5. Open config/config.py in project directory. Replace <some_name> in SNOWBALL_SEEDS_FILE with the name you chose.

6. The other SNOWBALL_* parameters can be changed, but be careful. Increasing iterations will make the program run more slowly per iteration.

7. Start elasticstart. Go to ~/resources/elasticsearch-1.7.0/bin/ and run:

   ./elasticsearch

   It is good to do this in a screen session.

8. Return to project directory and run the following to start Snowball: python main/main.py

9. You can monitor the system by running: tailf ~/resources/logs/snowball.log

10. Tuples will be in ~/resources/snowball/<some_name>/tuples. Each line is a Tuple object that can be evaluated in python. For a cleaner read of the results, run:

     python classes/read.py > out
    

    The results will be in out. Each line is of the form SUBJ OBJ.