Ediseases Dataset

The eDiseases dataset contains patient data from the MedHelp health site (http://www.medhelp.org/), where different communities share information and opinions about diseases. Each community consists of a number of conversations; a conversation being a sequence of comments posted by patients. To build the dataset, we automatically extracted 10 conversations from each of the following three communities: allergies, crohn and breast cancer. We selected a set of diseases that, according to medical expert, show high heterogeneity concerning both the degree of medical understanding of the diseases and the profile of the users. The conversations were selected randomly, but we automatically filtered out conversations with less than 10 posts. In total, we extracted 146 posts for allergies, 191 posts for crohn, and 142 posts for breast cancer; which include 983 sentences for allergies, 1780 sentences for crohn, and 1029 sentences for breast cancer, covering a 6 years time interval. Three frequent users of health forums annotated each sentence in the dataset as: Factuality: OPINION, FACT, EXPERIENCE.<br> Polarity: POSITIVE, NEUTRAL, NEGATIVE. In case of doubt, the annotators labeled the sentence as NOT_LABELED. As a result, we collected 967 labeled sentences for allergies, 1,709 labeled sentences 294 for crohn, and 959 labeled sentences for breast cancer.eDiseases数据集收录了来自MedHelp健康平台（http://www.medhelp.org/）的患者数据，该平台下设多个疾病主题社区，供用户分享与各类疾病相关的信息及观点。每个社区包含若干对话，此处的对话特指由患者发布的多条评论所构成的序列。 为构建该数据集，我们从过敏症、克罗恩病以及乳腺癌三个社区中各随机提取10段对话。本次选取的疾病类别经医学专家认定，在疾病的医学认知程度与用户画像两方面均呈现高度异质性。对话选取采用随机抽样方式，但我们自动过滤了评论总数不足10条的对话。最终，过敏症社区共提取146条评论、克罗恩病社区191条、乳腺癌社区142条；对应句子数量分别为983句、1780句与1029句，数据覆盖时长达6年。 三名健康论坛的高频用户对数据集中的每句文本进行标注，标注维度分为两类： 1. 事实属性（Factuality）：可标注为观点（OPINION）、事实（FACT）、经历（EXPERIENCE）； 2. 情感极性（Polarity）：可标注为积极（POSITIVE）、中立（NEUTRAL）、消极（NEGATIVE）。 若存在标注分歧，则将该句子标记为未标注（NOT_LABELED）。最终，我们共收集到过敏症社区967句标注文本、克罗恩病社区1709句标注文本以及乳腺癌社区959句标注文本。