WhatsApp Anonymized Privacy-focused Interactions Dataset (WAPI dataset)

1. Summary This dataset contains processed and fully anonymized metadata from WhatsApp chat histories. It is designed for researchers in fields such as Computational Linguistics, Social Network Analysis (SNA), and Human-Computer Interaction (HCI). Unlike raw chat logs, this dataset preserves user privacy by removing all message content and personally identifiable information (PII), replacing them with structural descriptors (e.g., message length ranges, emoji arrays) and cryptographic hashes. 2. Methodology The raw data was processed using a custom Python pipeline centered on privacy-by-design. The transformation includes:Temporal Anonymization: Timestamps were shifted with a random noise offset (-4 to +5 seconds) and converted into relative_time_seconds to mask actual dates and times while preserving the cadence of interaction.Identity Masking: Senders and mentioned phone numbers were transformed into 16-character SHA256 hashes using a unique salt for each file.Content Abstraction: Textual content was discarded. In its place, the dataset provides binned message lengths (ranges), punctuation markers (interrogative/exclamatory), and extracted emoji lists.Conversation Clustering: An Inter-Quartile Range (IQR) based algorithm was used to segment messages into "sessions" or conversation_ids, identifying natural breaks in communication based on temporal gaps. 3. Data Description The data is provided in CSV format (semicolon-separated). Each file represents a specific group or chat, containing the following features: Column,Description id,Sequential message identifier within the file. conversation_id,Cluster ID representing a continuous session of interaction. num_characters,"Binned message length (e.g., ""1-10"", ""11-50"", ""500+"")." relative_time_seconds,Seconds elapsed since the start of the observation period. message_type,"Categorization of the entry (e.g., text, audio, image, sticker, system)." responds_to_id,The ID of the parent message in a reply thread (if applicable). array_emojis,List of unique emojis present in the original message. interrogative,Boolean; indicates the presence of question marks. exclamatory,Boolean; indicates the presence of exclamation marks. sender_hash,Salted SHA256 hash of the message sender. mentioned_phones_hash,"List of hashes for phone numbers mentioned via ""@"" in the text." 4. Potential Research Applications Interaction Dynamics: Analyzing response times and turn-taking patterns in digital communication.Non-Verbal Communication: Studying the usage and frequency of emojis across different conversation types.Network Topology: Mapping the flow of information through "responds_to" hilos (threads) and mentions.Behavioral Modeling: Detecting session-based patterns and communication bursts.1. 摘要 本数据集包含经过处理且完全匿名的 WhatsApp 聊天记录元数据，面向计算语言学（Computational Linguistics）、社交网络分析（Social Network Analysis, SNA）、人机交互（Human-Computer Interaction, HCI）等领域的研究人员。与原始聊天日志不同，本数据集通过移除所有消息内容与个人可识别信息（PII），并以结构化描述符（如消息长度区间、表情数组）及加密哈希替代这些内容，从而保护用户隐私。 2. 研究方法 原始数据通过以隐私设计（privacy-by-design）为核心的自定义 Python 流程进行处理，具体转换步骤包括： 时间匿名化：时间戳通过随机噪声偏移（-4至+5秒）进行偏移，并转换为relative_time_seconds，以隐藏实际日期与时间，同时保留交互的节奏。 身份掩码：发件人与提及的电话号码通过为每个文件生成唯一盐值，被转换为16位SHA256哈希。 内容抽象：文本内容已被移除，取而代之的是分箱后的消息长度（区间）、标点标记（疑问/感叹）以及提取的表情列表。 会话聚类：基于四分位距（Inter-Quartile Range, IQR）的算法被用于将消息分割为"会话"或conversation_ids，通过时间间隔识别通信中的自然断点。 3. 数据说明 数据以CSV格式（分号分隔）提供，每个文件对应一个特定群组或聊天，包含以下特征： 列名 描述 id 文件内的序列消息标识符。 conversation_id 代表连续交互会话的聚类ID。 num_characters 分箱后的消息长度（如"1-10""11-50""500+"）。 relative_time_seconds 自观测周期开始起流逝的秒数。 message_type 条目分类（如文本、音频、图片、贴纸、系统消息）。 responds_to_id 回复线程中的父消息ID（若适用）。 array_emojis 原始消息中包含的唯一表情列表。 interrogative 布尔值，指示是否存在问号。 exclamatory 布尔值，指示是否存在感叹号。 sender_hash 发件人的加盐SHA256哈希。 mentioned_phones_hash 文本中通过"@"提及的电话号码的哈希列表。 4. 潜在研究应用 交互动力学：分析数字通信中的响应时长与轮次模式。 非语言沟通：研究不同聊天类型中表情的使用情况与使用频率。 网络拓扑：通过"responds_to"线程（threads）与提及关系映射信息流。 行为建模：检测基于会话的模式与通信爆发特征。