Topic Extraction Service

Classifies the given text within a set of concepts, retrieved using an unique service id and a language.

Installation

Python 3 is necessary to run the service. Required libraries are:

• tornado

Configuring the service

The service is provided with an example taxonomy, but it is designed for having the final user providing its own.

The taxonomy should be in a python file, such as taxonomy.py. The format there is a json map. Its keys are the phrases to tag. They will be tag with all the tags that are in the value array for that tag.

Example:

taxonomy = {
 'gerente': ['DIRECTIVOS'],
 'competidores': ['COMPETENCIA'],
 'jefe': ['DIRECTIVOS'],
 'publicidad': ['IMAGEN_DE_MARCA'],
 'opinion online': ['IMAGEN_DE_MARCA', 'COMPETENCIA'],
 'competidoras': ['COMPETENCIA'],
}

Starting and stopping the service

There is a script that starts and stops the service with the desired configuration:

./launcher.sh start
./launcher.sh stop

This same script contains the configuration for running the service with a given number of separated processes, using the port range as defined.

Calling the service

This service admits both GET and POST requests.

Calling the service via GET

An example call would be:

http://[ip_address]:[port]/?text=La tarifa de movistar es muy mala&lang=es

This call returns a JSON array with the matched concepts for the given text, language and service, or "N/A" if there is not any match. E.g.:

[
  "DIRECTIVOS"
]

Calling the service via POST

The service expects a body with a text per line.

The response is a JSON response with a line per entry in the request. E.g.:

{
  "response": [
    {
      "topics": [],
      "text": "some jefe"
    },
    {
      "topics": [],
      "text": "nada por aquí"
    },
    {
      "topics": [
        "DIRECTIVOS"
      ],
      "text": "otro jefe"
    }
  ]
}

Creating a Docker Image

For creating a docker image, just configure your own taxonomy.py and run:

docker build -t {name} .

Acknowledgement

This topic extraction module was developed by Paradigma Digital. This development has been partially funded by the European Union through the MixedEmotions Project (project number H2020 655632), as part of the RIA ICT 15 Big data and Open Data Innovation and take-up programme.

http://ec.europa.eu/research/participants/portal/desktop/en/opportunities/index.html