在灾后心理疾病的筛查中，心理问题的早期发现对及时开始治疗至关重要，如果患者能更早被确定在某类症状上存在高风险，有助于及时得到治疗，降低症状恶化发展的可能性。传统的心理测试常用的工具是量表或访谈，在本文中使用了一种新型的问答式测评工具，能够避免传统工具的弊端，施测简便，可以直接收集被试的自述文本。传统的分析方法无法直接对非结构化的文本数据进行分析，本文使用了自然语言处理的文本分类算法，可以将患者的症状特征自动筛查出来，帮助临床医生进行进一步的疾病诊断和治疗。

本文开发的测评工具和自动化筛查技术有助于在创伤事件发生后较快识别创伤后应激障碍患者，尤其是在受灾面积大、受灾者多、施测任务繁重的情况下，能够有效避免传统筛查方式存在的问题。能够区分每位患者的症状特征，有利于医生诊断并根据不同的症状特征设计针对性的治疗方案。

被试选取四川省雅安市芦山县中学七到九年级的学生430名，使用问答式自编问卷收集患者对自身的症状描述；同时使用量表《创伤后应激障碍检查表PCL-5》作为效标问卷，收集有效数据322个，文本共有1484个句子。

研究一在句子基础上进行单标签人工编码，使用K近邻、支持向量机进行建模，识别个体的每句话中描述的创伤后应激障碍症状。研究二在段落层面进行多标签人工编码，使用lasso回归方法，直接从整段个体文本中识别症状特征。研究主要使用准确率（Accuracy）、精确率（Precision）、召回率（Recall）和F1值等评估模型分类效果。结果表明在句子层面建模进行单标签分类时，使用K近邻进行建模的效果相对较好，模型的效果指标均能达到80%以上。在段落层面进行多标签分类时，模型的准确率为28.57%，效果较差。

In the screening of post-disaster mental illness, It is important to detection the mental problems early for timely start of treatment. If patients can identified earlier as certain symptoms at high risk, the earlier they can be diagnosed and treated, This can reduce the likelihood of symptom development. The commonly used tools in traditional psychological tests is a scale or interviews, we using a new kind of question and answer assessment tool in this article, it can avoid the disadvantages of traditional tools, Moreover, it is convenient to directly collect the self-report text of the subjects. It is impossible for traditional methods to analyze unstructured text data directly, we use the text classification algorithm of natural language processing, Text features can be classified to automatically screen out the symptoms of patients, This method makes post-disaster psychological screening more convenient and fast.

The method used in this paper can quickly identify the patients with psychological diseases after the traumatic event, Especially when the large area of disaster leads to more victims, This method can distinguish the symptom characteristics of each patient, which is helpful to arrange targeted treatment according to different symptom characteristics. It is very practical.

In this study, 430 students from grade 7 to grade 9 in Lushan county middle school, YaAn city, SiChuan province, were selected to collect their symptom descriptions by a question and answer assessment tool. At the same time, the scale“Post-Traumatic Stress Disorder Checklist for DSM-5，PCL-5" was used as the standard questionnaire, among which 322 were valid data and 1484 sentences were collected.

In the first study, Performing single-tag manual coding on a sentence basis, Making model use K - nearest neighbor, support vector machine, Distinguish between the symptoms of PTSD that are reflected in each sentence. In the second study, Performing multi-tag manual coding at the paragraph level, Making model use regression methods, The symptom characteristics were analyzed directly from the whole text. In order to evaluate, Accuracy, Precision, Recall and F1 values were used to calculate the classification effect of the model. The results show that when categorizing at the sentence level, The results of modeling with k-nearest neighbor are relatively good, The effect indexes of the model can reach more than 80%. When categorizing at the paragraph level, The model's accuracy was 28.57 %.