testbed/embeddings-benchmark__mteb/mteb/tasks/Classification/ind/IndonesianMongabayConservationClassification.py

from __future__ import annotations

import ast

import datasets
import numpy as np

from mteb.abstasks import AbsTaskClassification
from mteb.abstasks.TaskMetadata import TaskMetadata


class IndonesianMongabayConservationClassification(AbsTaskClassification):
    metadata = TaskMetadata(
        name="IndonesianMongabayConservationClassification",
        description="Conservation dataset that was collected from mongabay.co.id contains topic-classification task (multi-label format) and sentiment classification. This task only covers sentiment analysis (positive, neutral negative)",
        reference="https://aclanthology.org/2023.sealp-1.4/",
        dataset={
            "path": "Datasaur/mongabay-experiment",
            "revision": "c9e9f2c09836bfec57c543ab65983f3398e9657a",
        },
        type="Classification",
        category="s2s",
        date=("2012-01-01", "2023-12-31"),
        eval_splits=["validation", "test"],
        eval_langs=["ind-Latn"],
        main_score="f1",
        form=["written"],
        domains=["Web"],
        task_subtypes=["Sentiment/Hate speech"],
        license="Not specified",
        socioeconomic_status="medium",
        annotations_creators="derived",
        dialect=[],
        text_creation="found",
        bibtex_citation="""
        @inproceedings{fransiska-etal-2023-utilizing,
            title = "Utilizing Weak Supervision to Generate {I}ndonesian Conservation Datasets",
            author = "Fransiska, Mega  and
            Pitaloka, Diah  and
            Saripudin, Saripudin  and
            Putra, Satrio  and
            Sutawika*, Lintang",
            editor = "Wijaya, Derry  and
            Aji, Alham Fikri  and
            Vania, Clara  and
            Winata, Genta Indra  and
            Purwarianti, Ayu",
            booktitle = "Proceedings of the First Workshop in South East Asian Language Processing",
            month = nov,
            year = "2023",
            address = "Nusa Dua, Bali, Indonesia",
            publisher = "Association for Computational Linguistics",
            url = "https://aclanthology.org/2023.sealp-1.4",
            doi = "10.18653/v1/2023.sealp-1.4",
            pages = "30--54",
        }
        """,
        n_samples={"validation": 984, "test": 970},
        avg_character_length={"validation": 1675.8, "test": 1675.5},
    )

    def dataset_transform(self):
        splits = self.metadata_dict["eval_splits"]
        class_labels = ["positif", "netral", "negatif"]

        ds = {}

        # Include training because the classification task requires it
        train_split = self.dataset["train"]
        train_docs: list = []
        train_labels: list = []
        for text, label in zip(train_split["text"], train_split["softlabel"]):
            soft_label = ast.literal_eval(label)
            if len(soft_label) == len(class_labels):
                train_docs.append(text)
                hard_label = np.argmax(soft_label)
                train_labels.append(hard_label)

        ds["train"] = datasets.Dataset.from_dict(
            {
                "text": train_docs,
                "label": train_labels,
            }
        )

        documents: list = []
        labels: list = []
        # For evaluation
        for split in splits:
            ds_split = self.dataset[split]
            for text, label in zip(ds_split["text"], ds_split["softlabel"]):
                if label in class_labels:
                    documents.append(text)
                    labels.append(class_labels.index(label))

            assert len(documents) == len(labels)

            ds[split] = datasets.Dataset.from_dict(
                {
                    "text": documents,
                    "label": labels,
                }
            )

        self.dataset = datasets.DatasetDict(ds)