update README and process script

README.md

@@ -71,7 +71,21 @@ The original UIT-ViQuAD contains over 23,000 QA pairs based on 174 Vietnamese Wi
 
 The dataset has been processed to remove a few duplicated questions and answers.
 
-Questions about the private test set or the dataset should be directed to the authors.
+Version 2.0 contains the fields `is_impossible` and `plausible`, which the authors [explained](https://vlsp.org.vn/vlsp2021/eval/mrc) in the shared task announcement:
+
+```
+Context: Khác với nhiều ngôn ngữ Ấn-Âu khác, tiếng Anh đã gần như loại bỏ hệ thống biến tố dựa trên cách để thay bằng cấu trúc phân tích. Đại từ nhân xưng duy trì hệ thống cách hoàn chỉnh hơn những lớp từ khác. Tiếng Anh có bảy lớp từ chính: động từ, danh từ, tính từ, trạng từ, hạn định từ (tức mạo từ), giới từ, và liên từ. Có thể tách đại từ khỏi danh từ, và thêm vào thán từ.
+
+question: Tiếng Anh có bao nhiêu loại từ?
+is_impossible: False. // There exists an answer to the question.
+answer: bảy.
+
+question: Ngôn ngữ Ấn-Âu có bao nhiêu loại từ?
+is_impossible: True. // There are no correct answers extracted from the Context.
+plausible_answer: bảy. // A plausible but incorrect answer extracted from the Context has the same type which the question aims to.
+```
+
+Specific questions about the test set or the dataset should be directed to the [authors](https://nlp.uit.edu.vn/datasets).
 
 ### Languages
 
@@ -131,4 +145,4 @@ Shared task where version 2.0 was published:
 
 ### Acknowledgements
 
-We thank the authors of ViQuAD for releasing this dataset to the community.
+We thank the authors of ViQuAD and VLSP for releasing this dataset to the community.

process_viquad.py

@@ -6,7 +6,7 @@ import os
 import json
 import pandas as pd
 from itertools import groupby
-from datasets import Dataset, DatasetDict
+from datasets import Dataset, DatasetDict, Features, Sequence, Value
 
 def deduplicate_answers(answers):
     answers_sorted = sorted(answers, key=lambda x: (x['text'], x['answer_start']))
@@ -15,35 +15,46 @@ def deduplicate_answers(answers):
 
 data_dir = "UIT-ViQuAD 2.0"
 dataset_dict = {}
+features = Features({
+    'id': Value('string'),
+    'uit_id': Value('string'),
+    'title': Value('string'),
+    'context': Value('string'),
+    'question': Value('string'),
+    'answers': Sequence(feature={'text': Value('string'), 'answer_start': Value('int32')}),
+    'is_impossible': Value('bool'),
+    'plausible_answers': Sequence(feature={'text': Value('string'), 'answer_start': Value('int32')})
+})
 
 for split in ["train", "dev", "test"]:
     fname = os.path.join(data_dir, f"{split}.json")
     data = json.load(open(fname))
-    rows = []
-    title_i = 0
 
+    ids, uit_ids, titles, contexts, questions, all_answers, impossibles, all_plausible_answers = [], [], [], [], [], [], [], []
+    title_i = 0
+    print("-"*20, split, len(data["data"]), "-"*20)
     for title_data in data["data"]:
         title = title_data["title"]
         ctx_i = 0
         title_i += 1
 
         for ctx_and_qs in title_data["paragraphs"]:
-            questions = ctx_and_qs["qas"]
+            qas = ctx_and_qs["qas"]
             context = ctx_and_qs["context"]
             q_i = 0
             ctx_i += 1
             question_set = set()
-            # define default wherever answer is empty
-            answer_default: list = [{'answer_start': -1, 'text': ''}]
-            for q in questions:
+            for q in qas:
                 question = q["question"]
-                answers = q["answers"] if "answers" in q else answer_default
-                plausible_answers = q["plausible_answers"] if "plausible_answers" in q else answer_default
+                answers = q.get("answers", [])
+                plausible_answers = q.get("plausible_answers", [])
                 # Dedup answers
-                answers = deduplicate_answers(answers)
-                plausible_answers = deduplicate_answers(plausible_answers)
+                if answers:
+                    answers = deduplicate_answers(answers)
+                if plausible_answers:
+                    plausible_answers = deduplicate_answers(plausible_answers)
                 uit_id = q["id"]
-                is_impossible = q["is_impossible"] if "is_impossible" in q else False
+                is_impossible = q.get("is_impossible", False)
 
                 # Check duplicate questions
                 if question in question_set:
@@ -55,20 +66,29 @@ for split in ["train", "dev", "test"]:
 
                 q_i += 1
                 overall_id = f"{title_i:04d}-{ctx_i:04d}-{q_i:04d}"
-                rows.append({
-                    "id": overall_id,
-                    "uit_id": uit_id,
-                    "title": title,
-                    "context": context,
-                    "question": question,
-                    "answers": answers,
-                    "is_impossible": is_impossible,
-                    "plausible_answers": plausible_answers
-                })
+                # Append data to lists
+                ids.append(overall_id)
+                uit_ids.append(uit_id)
+                titles.append(title)
+                contexts.append(context)
+                questions.append(question)
+                all_answers.append(answers)
+                impossibles.append(is_impossible)
+                all_plausible_answers.append(plausible_answers)
                 question_set.add(question)
+
     # Convert to Dataset
-    df = pd.DataFrame(rows)
-    dataset_dict[split if split!="dev" else "validation"] = Dataset.from_pandas(df)
+    dataset = Dataset.from_dict({
+        'id': ids,
+        'uit_id': uit_ids,
+        'title': titles,
+        'context': contexts,
+        'question': questions,
+        'answers': all_answers,
+        'is_impossible': impossibles,
+        'plausible_answers': all_plausible_answers
+    }, features=features)
+    dataset_dict[split if split!="dev" else "validation"] = dataset
 
 print(dataset_dict)
 hf_dataset = DatasetDict(dataset_dict)