llm-research-daily
收藏资源简介:
Research Collector Dataset是一个通过Research-Collector工具从多个来源聚合的研究成果数据集,专注于大型语言模型(LLM)或语言模型的研究,时间范围为2026年4月12日至2026年4月26日。数据集包含39个条目,每个条目都附有丰富的元数据,包括机器学习子领域分类、质量评分和时间特征。数据来源包括学术平台(如PubMed、arXiv、Semantic Scholar)、专业平台(如GitHub、Stack Overflow、Kaggle)、社交平台(如Reddit、Hacker News)以及新闻和博客平台(如Medium、Towards Data Science)。数据集的核心字段包括唯一标识符、标题、来源平台、URL、作者、发布日期、引用次数、内容摘要等。此外,还包含丰富的元数据字段,如出版年份、月份、星期、季度、ML子领域分类、关键词提取、情感分析、自动摘要、数据质量评分等。该数据集适用于文本检索和文本分类任务,特别适合学术研究和机器学习应用。数据集采用MIT许可证,使用时需引用提供的仓库URL。
The Research Collector Dataset is a collection of research outputs aggregated from multiple sources using the Research-Collector tool. It focuses on research related to large language models (LLM) or language models, covering the period from April 12, 2026, to April 26, 2026. The dataset contains 39 entries, each accompanied by rich metadata, including machine learning subfield classifications, quality scores, and temporal features. Data sources include academic platforms (e.g., PubMed, arXiv, Semantic Scholar), professional platforms (e.g., GitHub, Stack Overflow, Kaggle), social platforms (e.g., Reddit, Hacker News), and news and blog platforms (e.g., Medium, Towards Data Science). Core fields of the dataset include unique identifiers, titles, source platforms, URLs, authors, publication dates, citation counts, content summaries, etc. Additionally, it includes extensive metadata fields such as publication year, month, day of the week, quarter, ML subfield classifications, keyword extraction, sentiment analysis, automatic summarization, and data quality scores. This dataset is suitable for text retrieval and text classification tasks, particularly for academic research and machine learning applications. The dataset is licensed under the MIT License and requires citation of the provided repository URL when used.
数据集总览
数据集名称:Research Collector Dataset
来源地址:https://huggingface.co/datasets/nellaivijay/llm-research-daily
许可协议:MIT License
核心信息
- 主题:大型语言模型(large language models / LLM / language models)
- 时间范围:2026-04-12 至 2026-04-26
- 数据项总数:39 条
- 导出时间:2026-04-26 15:41:22
数据来源
数据集从以下 12 个平台聚合研究内容:
| 类别 | 平台 |
|---|---|
| 学术 | PubMed, arXiv, Semantic Scholar, Crossref, Papers with Code |
| 专业 | GitHub, Stack Overflow, Kaggle |
| 社交 | Reddit, Hacker News |
| 新闻 | GDELT |
| 博客 | Medium, Towards Data Science |
数据结构
核心字段
id:唯一标识符title:研究项标题source:来源平台(如 pubmed, arxiv, github 等)url:原始内容链接author:作者published_date:发布日期(ISO 8601 格式)citations:引用数(如有)upvotes:点赞数(如有)downloads:下载数(如有)comments:评论数(如有)content:内容/摘要/描述score:相关性分数
丰富元数据字段
- 时间特征:
metadata_year,metadata_month,metadata_day,metadata_week,metadata_quarter,metadata_days_since - ML子领域:
metadata_ml_subfields(JSON 数组),metadata_subfield_count - 关键词:
metadata_keywords(JSON 数组),metadata_keyword_count - 质量评分:
metadata_quality_scores(JSON 字典),含整体质量评分 - 内容类型:
metadata_content_type(论文、预印本、仓库、讨论、问答、新闻) - 代码/DOI 标记:
metadata_has_code,metadata_has_doi - 情感分析:
metadata_sentiment_polarity(-1 至 1),metadata_sentiment_subjectivity(0 至 1),metadata_sentiment_category(positive/negative/neutral) - 自动摘要:
metadata_summary,metadata_summary_length - 数据质量指标:
metadata_data_quality(JSON 字典),含完整性、一致性、有效性评分 - 趋势指标:
metadata_trending_score,metadata_trending_category(hot/warm/cool/cold),metadata_engagement_score - 相关项:
metadata_related_items(JSON 数组),metadata_related_count
来源特定元数据
| 来源 | 特定字段 |
|---|---|
| PubMed | journal, doi, mesh_terms, publication_types, abstract_length |
| arXiv | arxiv_id, primary_category, categories, journal_ref |
| GitHub | stars, forks, language, license, topics, has_readme |
| subreddit, link_flair_text, upvote_ratio, total_awards, is_gilded | |
| Stack Overflow | tags, answer_count, has_accepted_answer, view_count, owner_reputation |
| Semantic Scholar | citation_count, influential_citation_count, fields_of_study, has_open_access |
| Medium | author, publication, read_time, claps |
| Kaggle | votes, usability_rating, file_count |
数据质量特性
- 标准化日期(ISO 8601 格式)
- ML 子领域自动分类(15+ 类别)
- 多维度质量评估(摘要长度、代码可用性、DOI、参与度、时效性)
- 时间特征提取(年、月、周、季度、发布天数)
- 自动关键词提取
- 内容类型自动检测
- 情感分析(极性、主观性、类别)
- 自动摘要生成(抽取式)
- 数据质量指标(完整性、一致性、有效性评分)
- 趋势分析(参与度速率与趋势类别)
- 交叉引用(基于共享子领域、关键词和标签的相似项检测)
- 模糊去重(智能重复检测与元数据合并)
- 元数据补全(基于回退逻辑推断缺失字段)
使用示例
python from datasets import load_dataset
加载数据集
dataset = load_dataset("nellaivijay/llm-research-daily") train_data = dataset["train"]
按来源筛选
pubmed_items = train_data.filter(lambda x: x["source"] == "pubmed")
按内容类型筛选
papers = train_data.filter(lambda x: x.get("metadata_content_type") == "paper")
按ML子领域筛选
cv_papers = train_data.filter(lambda x: "computer-vision" in x.get("metadata_ml_subfields", []))
按质量筛选
high_quality = train_data.filter(lambda x: x.get("metadata_quality_scores", {}).get("overall_quality_score", 0) > 0.7)
按分数排序
sorted_items = train_data.sort("score", reverse=True)
按日期筛选
recent_items = train_data.filter(lambda x: x.get("metadata_days_since", 999) < 30)
按趋势类别筛选
trending_items = train_data.filter(lambda x: x.get("metadata_trending_category") == "hot")
按情感分类筛选
positive_items = train_data.filter(lambda x: x.get("metadata_sentiment_category") == "positive")
获取相关项
item_with_related = train_data[0] related_items = item_with_related.get("metadata_related_items", [])
引用方式
如果使用该数据集,请引用仓库 URL:
https://huggingface.co/datasets/nellaivijay/llm-research-daily




