FGCS

repo for paper 《FGCS: A Fine-grained Scientific Information Extraction Dataset in Computer Science》

Data format

Each line represents a instance and is organized in the following format:

{
  "tokens": ["We", "start", "from", "analyzing", "the", "procedures", ...],
  "ners": [
    [16, 16, "Metric"], 
    [13, 13, 16, 16, "Metric"],
    ...
  ],
  "relations": [
    [0, 1, "Hyponym-of"],
    ...
  ],
  "discontinuous": true
}

Entries in ners are organized as [token_start, token_end, type] for general entities and [first_span_start, first_span_end, second_span_start, second_span_end, type] for discontinuous entities.

Entries in relations are organized as [head_entity_index, tail_entity_index, type].

discontinuous indicated whether there is a discontinuous entity in this sentence.

Convert to PURE format

PURE is a simple but effective entity and relation extraction method. You can run scripts/pure.py to convert the data to the PURE format:

python scripts/pure.py --input_file={input_file_path} --output_file={output_file_path}