pruizf/elco

Entity Linking system combination using weighted voting (*SEM 2015)

The application consists of clients to call Entity Linking services (EL) in English, and modules to operate on the results. Implements the Entity Linking System Combination described in our *SEM 2015 paper.

The EL services currently supported are:

• TagMe
• DBpedia Spotlight
• Wikipedia Miner (public instance no longer accessible)
• AIDA: both installed locally and in the public web service
• Babelfy

Requirements

• Python 2.7
• lxml
• MySQL-python (aka MySQLdb)
• nltk
• pyspotlight
• requests

To call TagMe and Babelfy, you need to request a key: Tagme, Babelfy. The application's config module has variables to enter the keys.

Modules

• analysis: Parses client responses. Computes entity-cooccurrence tables.
• clients: Clients to call the services
• config: Configuration
• main: Example how to use. Creates runners and calls them for each service
• model: Data types and some methods for them
• readers: To preprocess input before calling a client
• runners: Classes here use a reader, client and writer to create an annotation workflow
• utils: General tools useful for several modules
• writers: To postprocess the annotations and output them (to a file etc)

Usage

• activate the services to call in config.py

• call main.py

   usage: App to work with Entity Linking [-h] [-i MYINPUT] [-o MYOUT]
                                      [-s MYSKIPLIST] [-c CORPUS_NAME]
  
   optional arguments:
     -h, --help            show this help message and exit
     -i MYINPUT, --input MYINPUT
                           Input file, directory or text. A default can be set in
                           config.py (default: /path/to/some/default/input)
     -o MYOUT, --output MYOUT
                           Output file or files. Default names are created 
                           dynamically by code in writers.py module (default: None)
     -r MYOUTRESPS, --resp_output MYOUTRESPS
                           Output directory for client responses. A default is
                           created dynamically by code in writers.py module
                           (default: None)
     -s MYSKIPLIST, --skip_list MYSKIPLIST
                           File with filenames to skip (default:
                           /path/to/some/default/list)
     -c CORPUS_NAME, --corpus CORPUS_NAME
                           Name of the corpus (for output files etc.). A default
                           can be set in config.py (default: SOME_DEFAULT_NAME)