【作者】 何孟欣； 刘红云；

【Author】 Mengxin He;Hongyun Liu;Faculty of Psychology,Beijing Normal University;

【机构】 北京师范大学心理学部；

【摘要】 随着我国大众对于心理健康话题的兴趣日益增加与新媒体的兴起,以社交网络平台为基础的心理科普得到了充足发展。然而,目前对于网络心理科普的传播内容与传播效果的研究相对较少,且停留在定性、主观分析的层面,不利于进一步促进和指导心理科普的发展。本文通过网络爬虫技术收集"知乎"网站上6个知名心理科普账号的1160篇心理科普文章,使用K-means算法和潜在狄利克雷分布(Latent Dirichlet Allocation, LDA)主题模型对文章主题行分类与挖掘,从而了解当前心理科普的内容分布与特点,为心理科普发展提供参考。K-means算法是一种常用的聚类分析算法,可将对象根据其属性间的距离自动地分为K类。通过比较不同K值时点到其所属类中心的距离,可以确定最佳分类数。LDA主题模型是一种文档主题生成模型,用于挖掘文本中潜在的主题信息。使用LDA主题模型对语料库进行训练,可以得到指定数量的主题,每个主题可用一系列关键词的分布进行描述。本研究基于Python语言,首先使用jieba模块对收集到的心理科普文章进行分词,并根据《哈工大停用词表》去除文本中的停用词。对预处理后的文本,使用scikit-learn模块实现文本向量化与K-means聚类算法,比较K值不同时文章的聚类效果,确定最佳聚类数为18。随后使用gensim模块训练指定主题数为18的LDA模型。训练结果显示18个主题的第一第二主题词分别为1)孩子、妈妈;2)抑郁、自恋;3)父母、失恋;4)努力、伴侣;5)女性、内向;6)对方、边界;7)女性、自恋;8)自恋、对方;9)出轨、朋友圈;10)女性、抑郁症;11)孩子、父母;12)爱、孩子;13)道歉、父母;14)产后、孩子;15)父亲、孩子;16)野生、心理咨询;17)父母、孩子;18)对方、分手。结果说明:目前网络平台中的心理学科普主要关注家庭、亲密关系、性别、心理健康等话题,体现出应用为导向、贴近生活的特点,为读者的现实生活提供帮助;不同主题的文章中,关键词的重合程度较高,体现了心理科普媒体对热点话题的重视;主题覆盖范围较为狭窄,缺乏对认知心理学、人事与组织心理学等心理学子领域的介绍,不利于大众对心理学科产生全面的了解与印象。更多还原

【Abstract】 With the increasing interest for mental health and the development of new media in China, psychological science communication based on social media internet platform has developed continuously. However, there are few studies about the context and the effect of online psychological science communication. And the existed studies were limited to qualitative methods, which were limited to guide the further development of psychological science communication. In the present studies, we gathered 1160 articles on Zhihu website from 6 the most famous psychological science communication profiles through a web crawler program. Then we used K-means algorithm and Latent Dirichlet Allocation(LDA) topic model to classify and extract topics of the articles, to learn about the context and the characteristics of psychological science communication and offer advices to its further development. K-means algorithm is a type of common cluster analysis algorithm, which can divide the objects into K types by their own attributes. Through comparing the different sum of squares for error with different K, a best K can be confirmed to do the cluster. LDA topic model is a type of document topic generation model, which can be used to extract the latent topics of text. By using LDA model to training the corpus, we can get the specified number of topics of articles. Each topic will be described with a distribution of a series of keywords. The present study was conducted by Python language. After text segmentation by jieba module and removing all the stopwords based on HIT Stopwords List, we compared the effect of cluster analysis with different K by scikit-learn module. The results showed the cluster effect was the best when K was equal to 18. Then, a LDA topic model was trained based on genism module, whose number of topics was set to 18. The results of LDA model showed that the top two keywords of 18 topics were: 1) children, mother; 2) depression, narcissism; 3) parents, break-up; 4) effort, mate; 5) women, introversion; 6) each other, edge; 7) women, narcissism; 8) narcissism, each other; 9) infidelity, Circle of Friends; 10) women, depression; 11) children, parents; 12) love, children; 13) apology, parents; 14) postpartum, children; 15) father, children; 16) wild, consulting; 17) parents, children; 18) each other, break-up. The results showed that present online psychological science communication mostly focused on the topics about family, intimate relationship, gender, mental health, etc. Their articles has showed the features of application-oriented, closed to daily life and helpful to audience’s life. There was a large amount of overlaps between top 20 keywords of different topics, which showed that psychological science communication medias did valued those popular issues. However, their articles overly focused on above topics, causing a lack of introduction of different fields of psychology. It was not helpful to popularize the overall psychology to the public.更多还原