fix: Added Sentiment Analysis Bengali Dataset (#536)

* Added Hindi sentiment analysis dataset
* Made changes based on comments and added points on points table
* linted correctly
* bengali sentiment analysis dataset
* Added Sentiment Analysis Bengali Dataset
* deleted old hindi sentiment files
* removed hindi points
* updated all references, rerun tests, and added points

docs/mmteb/points/536.jsonl

@@ -0,0 +1,3 @@
+{"GitHub": "Akash190104", "New dataset": 2}
+{"GitHub": "asparius", "Review PR": 2}
+{"GitHub": "isaac-chung", "Review PR": 2}

mteb/tasks/Classification/__init__.py

@@ -6,6 +6,7 @@
 from .ara.TweetSarcasmClassification import *
 from .bam.BambaraSentimentClassification import *
 from .ben.BengaliHateSpeechClassification import *
+from .ben.BengaliSentimentAnalysis import *
 from .bul.BulgarianSentimentClassification import *
 from .bul.BulgarianStoreReviewSentimentClassfication import *
 from .ces.CzechSubjectivityClassification import *

mteb/tasks/Classification/ben/BengaliSentimentAnalysis.py

@@ -0,0 +1,44 @@
+from __future__ import annotations
+
+from mteb.abstasks.AbsTaskClassification import AbsTaskClassification
+from mteb.abstasks.TaskMetadata import TaskMetadata
+
+
+class BengaliSentimentAnalysis(AbsTaskClassification):
+    metadata = TaskMetadata(
+        name="BengaliSentimentAnalysis",
+        description="dataset contains 3307 Negative reviews and 8500 Positive reviews collected and manually annotated from Youtube Bengali drama.",
+        reference="https://data.mendeley.com/datasets/p6zc7krs37/4",
+        dataset={
+            "path": "Akash190104/bengali_sentiment_analysis",
+            "revision": "a4b3685b1854cc26c554dda4c7cb918a36a6fb6c",
+        },
+        type="Classification",
+        category="s2s",
+        eval_splits=["train"],
+        eval_langs=["ben-Beng"],
+        main_score="f1",
+        date=("2020-06-24", "2020-11-26"),
+        form=["written"],
+        dialect=[],
+        domains=["Reviews"],
+        task_subtypes=["Sentiment/Hate speech"],
+        license="CC BY 4.0",
+        socioeconomic_status="mixed",
+        annotations_creators="human-annotated",
+        text_creation="found",
+        bibtex_citation="""@inproceedings{sazzed2020cross,
+        title={Cross-lingual sentiment classification in low-resource Bengali language},
+        author={Sazzed, Salim},
+        booktitle={Proceedings of the Sixth Workshop on Noisy User-generated Text (W-NUT 2020)},
+        pages={50--60},
+        year={2020}
+        }""",
+        n_samples={"train": 11807},
+        avg_character_length={"train": 69.66},
+    )
+
+    def dataset_transform(self):
+        self.dataset = self.stratified_subsampling(
+            self.dataset, seed=self.seed, splits=["train"]
+        )

results/intfloat__multilingual-e5-small/BengaliSentimentAnalysis.json

@@ -0,0 +1,15 @@
+{
+  "dataset_revision": "a4b3685b1854cc26c554dda4c7cb918a36a6fb6c",
+  "mteb_dataset_name": "BengaliSentimentAnalysis",
+  "mteb_version": "1.7.17",
+  "train": {
+    "accuracy": 0.853759765625,
+    "accuracy_stderr": 0.029660222421585114,
+    "ap": 0.9118492318156679,
+    "ap_stderr": 0.018562450430510975,
+    "evaluation_time": 66.98,
+    "f1": 0.8318878279557277,
+    "f1_stderr": 0.03000343346915389,
+    "main_score": 0.8318878279557277
+  }
+}

results/sentence-transformers__paraphrase-multilingual-MiniLM-L12-v2/BengaliSentimentAnalysis.json

@@ -0,0 +1,15 @@
+{
+  "dataset_revision": "a4b3685b1854cc26c554dda4c7cb918a36a6fb6c",
+  "mteb_dataset_name": "BengaliSentimentAnalysis",
+  "mteb_version": "1.7.17",
+  "train": {
+    "accuracy": 0.6267578125,
+    "accuracy_stderr": 0.03611908523026426,
+    "ap": 0.7938445145047481,
+    "ap_stderr": 0.016033199516799757,
+    "evaluation_time": 52.43,
+    "f1": 0.6061352150370449,
+    "f1_stderr": 0.030922844922113873,
+    "main_score": 0.6061352150370449
+  }
+}