Update: loading script TODO test

README.md

@@ -1,3 +1,36 @@
 ---
 license: mit
 ---
+## Universal Text Classification Dataset (UTCD)
+UTCD is a curated compilation of 18 datasets revised for Zero-shot Text Classification spanning 3 aspect categories of Sentiment, Intent/Dialogue, and Topic classification. UTCD focuses on the task of zero-shot text classification where the candidate labels are descriptive of the text being classified. TUTCD consists of ~ 6M/800K train/test examples.
+
+UTCD was introduced in the Findings of ACL'23 Paper **Label Agnostic Pre-training for Zero-shot Text Classification** by ***Christopher Clarke, Yuzhao Heng, Yiping Kang, Krisztian Flautner, Lingjia Tang and Jason Mars***. 
+
+UTCD Datasets & Principles:
+
+- Sentiment
+    - GoEmotions introduced in [GoEmotions: A Dataset of Fine-Grained Emotions](https://arxiv.org/pdf/2005.00547v2.pdf)
+    - TweetEval introduced in [TWEETEVAL: Unified Benchmark and Comparative Evaluation for Tweet Classification](https://arxiv.org/pdf/2010.12421v2.pdf) (Sentiment subset)
+    - Emotion introduced in [CARER: Contextualized Affect Representations for Emotion Recognition](https://aclanthology.org/D18-1404.pdf)
+    - Amazon Polarity introduced in [Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626.pdf)
+    - Finance Phrasebank introduced in [Good debt or bad debt: Detecting semantic orientations in economic texts](https://arxiv.org/pdf/1307.5336.pdf)
+    - Yelp introduced in [Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626.pdf)
+- Intent/Dialogue
+    - Schema-Guided Dialogue introduced in [Towards Scalable Multi-Domain Conversational Agents: The Schema-Guided Dialogue Dataset](https://arxiv.org/pdf/1909.05855v2.pdf)
+    - Clinc-150 introduced in [An Evaluation Dataset for Intent Classification and Out-of-Scope Prediction](https://arxiv.org/pdf/1909.02027v1.pdf)
+    - SLURP SLU introduced in [SLURP: A Spoken Language Understanding Resource Package](https://arxiv.org/pdf/2011.13205.pdf)
+    - Banking77 introduced in [Efficient Intent Detection with Dual Sentence Encoders](https://arxiv.org/abs/2003.04807](https://arxiv.org/pdf/2003.04807.pdf)
+    - Snips introduced in [Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces](https://arxiv.org/pdf/1805.10190.pdf)
+    - NLU Evaluation introduced in [Benchmarking Natural Language Understanding Services for building Conversational Agents](https://arxiv.org/pdf/1903.05566.pdf)
+- Topic
+    - AG News introduced in [Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626.pdf)
+    - DBpedia 14 introduced in [DBpedia: A Nucleus for a Web of Open Data](https://link.springer.com/chapter/10.1007/978-3-540-76298-0_52)
+    - Yahoo Answer Topics introduced in [Character-level Convolutional Networks for Text Classification](https://arxiv.org/pdf/1509.01626.pdf)
+    - MultiEurlex introduced in [MultiEURLEX -- A multi-lingual and multi-label legal document classification dataset for zero-shot cross-lingual transfer](https://aclanthology.org/2021.emnlp-main.559v2.pdf)
+    - BigPatent introduced in [BIGPATENT: A Large-Scale Dataset for Abstractive and Coherent Summarization](https://aclanthology.org/P19-1212.pdf)
+    - Consumer Finance introduced in [Consumer Complaint Database](https://www.consumerfinance.gov/data-research/consumer-complaints/)
+
+In order to make NLP models more broadly useful, zero-shot techniques need to be capable of label, domain \& aspect transfer. As such, in the construction of UTCD we enforce the following principles: 
+
+- **Textual labels**: In UTCD, we mandate the use of textual labels. While numerical label values are often used in classification tasks, descriptive textual labels such as those present in the datasets across UTCD enable the development of techniques that can leverage the class name which is instrumental in providing zero-shot support. As such, for each of the compiled datasets, labels are standardized such that the labels are descriptive of the text in natural language. 
+- **Diverse domains and Sequence lengths**: In addition to broad coverage of aspects, UTCD compiles diverse data across several domains such as Banking, Finance, Legal, etc each comprising varied length sequences (long and short). The datasets are listed above.

utcd.py

@@ -53,6 +53,7 @@ class UtcdConfig(datasets.BuilderConfig):
 
 config = StefConfig('config.json')
 # mic(config('go_emotion'))
+_split2hf_split = dict(train=datasets.Split.TRAIN, eval=datasets.Split.VALIDATION, test=datasets.Split.TEST)
 
 
 class Utcd(datasets.GeneratorBasedBuilder):
@@ -116,10 +117,14 @@ class Utcd(datasets.GeneratorBasedBuilder):
         dnms = self._get_dataset_names()
         labels = [config(f'{dnm}.splits.{split}.labels') for dnm in dnms for split in ['train', 'test']]
         mic(dnms, labels)
+
+        labels = sorted(set().union(*labels))  # drop duplicate labels across datasets
         return datasets.DatasetInfo(
             description=_DESCRIPTION,
             features=datasets.Features(
-                text=datasets.Value(dtype='string'), labels=labels, dataset_name=datasets.ClassLabel(names=dnms)
+                text=datasets.Value(dtype='string'),
+                labels=datasets.Sequence(feature=datasets.ClassLabel(names=labels), length=-1),  # for multi-label
+                dataset_name=datasets.ClassLabel(names=dnms)
             ),
             homepage=_URL
             # TODO: citation
@@ -135,9 +140,22 @@ class Utcd(datasets.GeneratorBasedBuilder):
 
         downloaded_files = dl_manager.download_and_extract('datasets.zip')
         mic(downloaded_files)
-        raise NotImplementedError
+        # raise NotImplementedError
 
+        # return [
+        #     datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
+        #     datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
+        # ]
         return [
-            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
-            datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
+            datasets.SplitGenerator(name=_split2hf_split[s], gen_kwargs={"filepath": split2paths[s]}) for s in splits
         ]
+
+    def _generate_examples(self, filepath: str):
+        # each call is for one split of one dataset
+        dnm = filepath.split(os.sep)[-2]
+        id_ = 0
+        with open(filepath, encoding='utf-8') as f:
+            dset = json.load(f)
+            for txt, labels in dset.items():
+                yield id_, dict(text=txt, labels=labels, dataset_name=dnm)
+                id_ += 1