Awesome

RuSentiFrames 2.0

📓 Update 01 October 2023: this lexicon is now available in arekit-ss in text processing for most collections of the sentiment attitude extraction task (in Russian) with just single script into JSONL/CSV/SqLite` including (optional) language transferring 🔥 [Learn more ...]

Release notes: Statistics update.

Enlarged with 12% new frame entries;

cleared Other lexical units;

author→A0 and author→A1 labelings mostly removed;

Represents a lexicon which describes sentiments and connotations conveyed with a predicate in a verbal or nominal form. Checkout reference section for details in related paper.

Statistics

Lexical Unit Type	v-2.0	Diff
Frame Entries	311	+12%
Verbs	3239	+16%
Nouns	986	+20%
Phrases	2551	+6%
Other	12	removed
Unique Entries	6788	+12%
Total Entries	7034	+10%

Polarity	Polarity	v-2.0	Diff
A0→A1	POS	2558	+14%
A0→A1	NEG	3289	+17%
author→A0	POS	170	removed
author→A0	NEG	1578	-
author→A1	POS	92	removed
author→A1	NEG	249	removed

NOTE: Diff corresponds to comparison with the v-1.0

Structure

The structure of the frames includes:

Role designation: A0 is an argument exhibiting features of a Prototypical Agent, and A1 is a Theme (as in PropBank)
Dimentions:

the attitude of the author of the text towards mentioned participants;
positive or negative sentiment between participants;
positive or negative effects to participants;
positive or negative mental states of participants related to the described situation.

Format Description

The lexicon presented in JSON format.

Below is an example of the frame выйграть (to win):

"0_4": {
    "title": [
        "выиграть",
        "получить приз" ],
    "variants": [
        "выиграть",
        "выигрывать" ],
    "comment": "comment",
    "roles": {
        "a0": "победитель",
        "a1": "побежденный",
        "a2": "приз",
        "a3": "область, сфера, в которой одержана победа" },
    "frames": {
        "polarity": [
            [ "a1", "a0", "neg", 1.0 ],
            [ "a0", "a2", "pos", 1.0 ],
            [ "a1", "a2", "neg", 0.7 ],
            [ "a0", "a3", "pos", 1.0 ],
            [ "a1", "a3", "pos", 1.0 ] ],
        "value": [
            [ "a0", "a2", 1.0 ],
            [ "a1", "a2", 1.0 ] ],
        "effect": [
            [ "a0", "+", 1.0 ],
            [ "a1", "-", 1.0 ] ],
        "state": [
            [ "a0", "pos", 1.0 ],
            [ "a1", "neg", 1.0 ] ]
    }
}

Where keys denotes as follows:

title -- list of possible frame titles.

variants -- list of possible variants appeared in text.

roles -- is a dictonary of participants (keys) with the related description.

frames -- is a dictionary of parameters, which describes a frame in following directions:

polarity -- positive or negative sentiment between participants;
value -- the attitude of the author of the text towards mentioned participants;
effect -- positive or negative effects to participants;
state -- positive or negative mental states of participants related to the described situation.

Collection Reader

Folder reader contains a collection reader (source file parsers), written in Python-3.6.

Please refer to read.py, as it provides an example of how this collection could be parsed/readed.

Prior Releases

RuSentiFrames-1.0

References

@article{loukachevitch2020sentiment,
  title={Sentiment Frames for Attitude Extraction in Russian},
  author={Loukachevitch, Natalia and Rusnachenko, Nicolay},
  booktitle={Proceedings of International Conference on 
             Computational Linguistics and Intellectual 
             Technologies Dialogue-2020},
  year={2019}
}