Format.md

Quick Start: Format Description

• The template of each line is as follows:

    [KEYWORD] ':' [VALUE] [OPTIONS]

• Checkout sample;

• Checkout reader example, written in Python.

Detailed Options Description

Options presented in case of the following keywords:

1. Object;
2. Attitude;
3. FrameVariant.

For Object keyword

Object: 'сша' b:(3,1) oi:[1] si:{0} d:ner <AUTH>

• b:(position, length) -- beginning position description.
  • position -- index of word, starts with 0;
  • length -- length of related object in words.
• oi:[line_index] -- Object Index.
  • line_index -- an unsigned int value, is a reference to the Object;
• si:{index} -- Synonym Index.
  • index -- index of synonym group, if synonym group exists; otherwise -1 value used.
• d:type -- a method by which named entity has been extracted, where type denotes:
  • ner -- entity has been found using NER tool (Bi-LSTM+CRF model) [paper];
  • restored -- entity was missed by NER but restored using a list of authorized objects (see NOTE);
• t:[TYPE] -- entity type:
  • TYPE -- could be one of the following: LOC, PER, ORG;
• <AUTH> -- optional; if present, denotes that related object is belong a list of authorized object (see NOTE);

NOTE: List of authorized objects list presented in a form of relations, where each relation has the following format: source->target.

Object x is belong the list of authorised objects, when there is at least a single relation with source == x or target == x.

For Attitude keyword

Attitude: 'сша'->'украина' b:(1) oi:[1, 2] si:{0,180}

• b:(score) -- Sentiment score of the related atittude.

  • score -- integer value; 1 -- positive, -1 -- negative.

• oi:[source, target] -- Object Index.

  • source -- unsigned int value, is a reference to the Object;
  • target -- unsigned int value, is a reference to the Object;

• si:{source, target} -- synonym indices of source and target.

  • source -- index of synonym group;
  • target -- index of synonym group;

For FrameVariant keyword

FrameVariant: помогать (6, 1) b:[a0->a1[pos]] id:(0_8)

• (position, length) -- position description.

  • position -- index of word, starts with 0;
  • length -- length of related object in words.

• b:[a0->a1[score]] -- sentiment score of a relation A0->A1 (according to RuSentiFrames). score parameter could have one of the following values:

  • pos -- denotes a positive sentiment
  • neg -- denotes a negative sentiment

• id:(id) -- identifier in RuSentiFrames lexicon:

  • id -- key in json dictionary of RuSentiFrames lexicon;