家庭暴力施暴者心理特征数据集
收藏资源简介:
首先,对中国的四个主要在线平台(微博、豆瓣、小红书和知乎)进行了初步调查,最终选择知乎作为数据收集的主要来源,因其用户参与度高、回复内容详尽。重点收集了详细描述施暴者特征的回答,涵盖了施暴者的创伤经历、不当行为、人格偏差、情绪失调、认知偏差和暴力倾向等六个维度。数据经过严格筛选和标注,用于建立结构化的数据集,支持家暴风险的多维度分析。采用定制的网络爬虫工具,从知乎平台自动提取相关数据。数据提取后,经过人工筛选和分类,并结合心理学家和专家的意见对数据进行标注,最终形成了包含六个维度的家暴风险评估量表。从知乎平台爬取数据后,对数据进行了分类,分为三个类别:宣传内容/叙述、家暴的基本描述和详细的施暴者特征描述。保留了与施暴者特征相关的591个回答。结合心理学家的临床经验和现有研究,构建并不断迭代家暴风险评估量表。最后,根据量表对数据进行人工标注,形成了一个涵盖六个维度的综合数据集。数据来源可靠,通过知乎平台获取高质量、详细的用户回复。在数据标注阶段,四名具有丰富心理学研究经验的工作人员参与了数据标注,所有数据均经过两位心理学专家的审核。为保证标注质量,每个样本均由所有标注员审核,最终标注一致性Kappa值达到0.708,确保了数据的准确性和可靠性。标注完成后,通过人工检查验证数据的准确性,并利用专家反馈对标注结果进行二次验证。通过随机抽样的方式,检验了数据的分类和标注的一致性,确保所有维度的家暴特征都得到了全面覆盖。
First, a preliminary survey was conducted on four major online platforms in China (Weibo, Douban, Xiaohongshu, and Zhihu), and Zhihu was ultimately selected as the primary source for data collection due to its high user engagement and detailed response content. We focused on collecting responses detailing the characteristics of domestic violence perpetrators, covering six dimensions: the perpetrators' traumatic experiences, misconduct, personality deviations, emotional dysregulation, cognitive biases, and violent tendencies. The data underwent rigorous screening and annotation to establish a structured dataset supporting multi-dimensional analysis of domestic violence risk. A custom web crawler tool was used to automatically extract relevant data from the Zhihu platform. After data extraction, the collected materials underwent manual screening and classification, and were annotated with reference to the opinions of psychologists and experts, ultimately forming a domestic violence risk assessment scale covering the six aforementioned dimensions. After crawling data from Zhihu, the dataset was classified into three categories: propaganda content/narratives, basic descriptions of domestic violence, and detailed descriptions of perpetrator characteristics. A total of 591 responses related to perpetrator characteristics were retained. Combining the clinical experience of psychologists and existing research, the domestic violence risk assessment scale was constructed and continuously iterated. Finally, the data was manually annotated based on the scale, forming a comprehensive dataset covering the six dimensions. The data source is reliable, as high-quality and detailed user responses were obtained via the Zhihu platform. During the data annotation stage, four staff members with rich experience in psychological research participated in the annotation work, and all data were reviewed by two psychological experts. To ensure annotation quality, each sample was reviewed by all annotators, and the final inter-annotator agreement Kappa value reached 0.708, ensuring the accuracy and reliability of the dataset. After annotation was completed, the accuracy of the data was verified through manual inspection, and the annotation results were secondarily validated using expert feedback. Through random sampling, the consistency of data classification and annotation was tested to ensure that all dimensions of domestic violence characteristics were fully covered.




