Skip to content

Latest commit

 

History

History
41 lines (34 loc) · 2.43 KB

README.md

File metadata and controls

41 lines (34 loc) · 2.43 KB
Raw

Interactive QA (IQA)

IQA is an interactive framework to construct SPARQL Query from a natural language question.

While IQA interactive schema is generally able be implemented using an arbitrary semantic question answering pipeline, this repository provides a concrete realization of IQA using following external components:

  1. MDP-Parser: that acts as the shallow parser to extract the mentions of entity and relations from the input question.
  2. EARL is a entity/relation linker that provides a list of candidate linked item for each entity/relation mention from the previous step. A demo version is available at http://earldemo.sda.tech/ .
  3. SQG is a query building component that produces a list of candidate queries given the candidate linked items and the input question. It alo has a ranking mechanism to arrange the candidate queries in respect to their structural similarity to the input question.

Getting Started

  • Please install Python 3.7 and the required packages as listed in here.
  • Please follow the documentation of the aforementioned components to get them running.
  • Update the config.py to point the web service of the components.
  • run python scripts/service/backend.py --base_path . .
  • run python UI/app.py --base_path . --port 5001 .
  • Now you can navigate to http://localhost:5001/ and check out the web based user interface.

Data

Please find all required data at: http://tiny.cc/IQA-Data

data
└LC-QuAD (Benchmarkin dataset)
│   │   linked.json (Original)
│   │   linked_IOB.pk (IOB format to train the parser)
│   └───EARL (cached the output of EARL using various parser)
│       │   output_nltk.json (nltk based parser)
│       │   output_gold.json (gold annotations)
│       │   output_gold_EARL_ER_PRED.json (default)
└output
│   │  wd_perf.json
│   │  stats-general.json (detailed results for oracle evaluation)
│   │  gold_answer.pk (Gold answers for LC-QuAD dataset)
│   └───pipeline (Full interpretation space for all LC-QuAD questions)
│       │   fbc52b963d5141a6b89216bae276bf21.pickle ({questionid}.pickle)
│       │   ...

License

This program is subject to the terms of the General Public License (GPL), Version 3.0.