SKIML-ICL/hoh
收藏资源简介:
该数据集是一个用于时间敏感问答或信息更新研究的数据集,包含问题、答案、文档和时效性相关信息。数据集中每个样本具有唯一的问题ID(qid)、问题文本(question)和多个可能的答案(answers)。关键特征包括冲突段落(conflict_passage)和冲突时间(conflict_time),用于表示与当前信息不一致的内容及其时间戳。此外,数据集提供当前时间(current_time)、文档标题(document_title)和文档ID(document_id)。all_outdated_infos字段列出过时信息,包括答案、证据和最后修改时间。ctxs字段包含上下文信息,如段落ID(pid)、排名(rank)、相关性分数(score)、文本内容(text)、标题(title)、是否包含答案(hasanswer)和自然语言推理标签(nli)。其他字段包括问题是否可回答(answerable)、可回答前缀列表(answerable_prefix)、冲突段落有效性(is_valid_conflict_passage)、冲突答案列表(conflict_answers)和整体上下文(context)。数据集仅包含测试集(test),有829个样本,适用于评估模型在动态信息环境中的问答能力。
This dataset is designed for time-sensitive question answering or information updating research, containing questions, answers, documents, and timeliness-related information. Each sample in the dataset includes a unique question ID (qid), question text (question), and multiple possible answers (answers). Key features include a conflict passage (conflict_passage) and conflict time (conflict_time), representing content inconsistent with current information and its timestamp. Additionally, the dataset provides the current time (current_time), document title (document_title), and document ID (document_id). The all_outdated_infos field lists outdated information, including answers, evidence, and last modified time. The ctxs field contains context information, such as passage ID (pid), rank (rank), relevance score (score), text content (text), title (title), whether it contains an answer (hasanswer), and natural language inference label (nli). Other fields include answerability of the question (answerable), a list of answerable prefixes (answerable_prefix), validity of the conflict passage (is_valid_conflict_passage), a list of conflict answers (conflict_answers), and overall context (context). The dataset only includes a test split (test) with 829 examples, suitable for evaluating model performance in dynamic information environments for question answering tasks.




