luiszugasti/tagme-doc-web

Document web using TAGME API

Tagme-doc-web

Part of an ongoing project for Ryerson Capstone project EB02.

Introduction

This repo contains a flask application which allows a user to obtain the document entities of a set of documents in our Clueweb09 corpus.
The user only needs to know of the specific document name that they wish to obtain entities for. Tagme-doc-web caches each specified documentID : entity set inside a Sqlite database server, for faster document entity retrieval.

More information about the API is below.

API

The application is structured as a flask application. Hence, there are flask command line commands available as well as remote endpoint commands.

Flask CLI

db_create

Initializes the SQLite database with the default models. These models are User, Entity and Document.

db_seed

Initializes the SQLite database with dummy data.

db_drop

Drops all the tables created, if they exist.

Remote API

<server>/get_doc_entity/<params: document_names> [top:n]

The bread and butter of the API. By calling with any amount of document names, the server will check if each of them exist within the cached database. For any documents not in the local database, a separate call to TAGME will be made. Once all documents' entities are available, the server will return the requested entities.
If top is specified, this will change the behavior of the API to return only the top n hits of entities per document.
Hint: the TAGME server itself is somewhat slow, hence, it is advisable that for hundreds of documents, a high timeout is configured. As more and more documents are cached into the database, this delay is expected to decrease. The format that the api returns documents is a wrapped dictionary of {documentname : entitie(s) : occurrence of entities}, as below:

{
    "clueweb09-en0000-00-15766": {
        "Frost/Nixon (film)": 9,
        "Nixon, Pennsylvania": 1,
        "Frank Langella": 4,
        "Michael Sheen": 4,
        ...
    },
    "clueweb09-en0000-00-18760": {
        "Frost/Nixon (film)": 3,
        "Nixon, Pennsylvania": 2,
        "Frank Langella": 89,
        "Michael Sheen": 4,
        ...
    },
    ...
}

<server>/get_all_doc_entities

Returns a list of all the cached document entities found in the corpus.

<server>/get_all_docs

Returns a list of all cached document titles found in the corpus.

Installation

To get started with the API, first clone this github repo to an appropriate location on your computer. You will need to run the requirements.txt file with pip install -r requirements. Make sure you're in a virtual environment! To run the project, go to the base directory where the app.py script is located, and on a command line, type flask run. The project is tested with Python 3.7.