PSYCH OpenIR  > 社会与工程心理学研究室
基于社交媒体数据的心理特征自动识别新方法研究
Alternative TitleA Methodology study of Automatic Identification of Psychological Traits Utilizing Social Media Data
刘明明
Subtype硕士
Thesis Advisor朱廷劭
2019-06
Degree Grantor中国科学院大学
Place of Conferral中国科学院心理研究所
Degree Name理学硕士
Degree Discipline应用心理学
Keyword生态化识别 生活事件 家庭暴力 失独 方法论
Abstract

自我报告法自提出以来一直是心理研究中应用最为广泛的测量方法,具有操作简便、信效度高等优点,但也存在难以进行回溯追踪研究以及需要测量对象积极配合等不足。基于此,本研究提出一种基于人工智能的心理特征识别新方法一生态化识别。生态化识别指采用生态化方式采集的数据,使用机器学习计算模型,对用户的心理特征进行自动识别,从而实现无侵扰测量,提高生态效度。利用生态化识别方法能够非接触地自动识别测量对象的心理特征,可以基于历史数据进行回溯追踪研究。

针对自我报告法难以进行回溯研究、追踪流失率较高以及需要被试配合等问题,本研究通过家庭暴力、失去独生子女及小说主人公的升学、结婚事件三个研究,分别检验生态化识别方法在回溯事件即时影响、追踪事件长期影响轨迹以及识别无法进行主观报告样本的心理特征等三方面的可行性。研究结果显示:

(1>通过生态化识别获取用户回溯事件前的心理状态,实现对于事件影响的准确刻画。家庭暴力受害者的抑郁程度显著高于对照组(t =4.08, p<0.01 >,自杀可能性显著高于对照组(t=2.138 p<0.01>,生活满意度显著低于对照组用户(t =3.087, p<0.01,同时家庭暴力受害者的宜人性(t(1282)=7.071, p<0.01 >,尽责性较低(t(1282)=5.351, p<0.01>,神经质较高(t(1282)=-7.268, p<0.01>。该结果与以往研究结论相符。同时,由于生态化识别可以确认用户在经历事件前的心理状态,因此可以得出,受害者在家庭暴力后出现心理健康的消极变化,以及尽责性、神经质的变化;而受害者在受到家庭暴力前就己经展现出较低的宜人性及开放性。

(2)通过生态化识别方法测量失独用户的语言使用变化趋势,实现对失独用户的心理变化的纵向追踪。失独用户的心理健康状态略差于对照组用户,具体表现为使用更少的积极情绪词(t(452)=3.643, p<0.001,更多的消极情绪词尤其是悲伤词与死亡词(t(452)=3.643, p<0.001,该发现与以往研究结论相符。进一步,研究发现失独用户使用更多的家庭词(t(452)=3.643, p<0.001,且家庭词在失独后逐渐减少;失独用户的消极情绪词逐渐减少,而悲伤词维持不变(p>0.05 > ;失独用户在失独后短期内高频使用宗教词(t(166.053)=4.287, p<0.001,并在1年后恢复与对照组相同频率。

(3>通过生态化识别,实现对小说虚拟人物的心理发展分析。由于无法通过自我报告获取小说人物的心理特征,利用生态化识别,通过对《平凡的世界》中主人公孙少安和孙少平的性格特征的自动识别,发现孙少安具有极强的外向性,较强的尽责性和开放性;孙少平具有很强的开放性,较强的尽责性和外向性。进一步检验生活事件对小说主人公性格发展的影响,孙少安结婚后宜人性下降、神经质增加;孙少平升学后开放性增加,尽责性降低,与国外相同生活事件影响的研究结论一致。

以上研究表明生态化识别在回溯研究、追踪研究以及难以接触被试的情景下对心理特征自动识别的可行性。与传统心理测量方式相比,生态化识别方法具有非接触性,可回溯性等特点,适用于传统研究方法难以开展的样本较特殊分散、研究对象不配合、生活事件前的状态难以回溯等研究场景。将生态化识别方法与传统测量方式相结合,采用社交网络数据对于用户的心理特征进行自动识别,能够有效提升心理测量的应用范围与测量效率。

Other Abstract

The self-reporting method is the most widely used and most controversial measurement in psychological research. Although the self-reporting method is easy to operate and with high reliability and validity, it couldn't avoid the influence of the participants' motivation, the degree of cooperation, and the low reliability on retrospective research. This study proposes a new method of automatic recognizing psychological traits utilizing machine learning algorithm一Ecological Recognition (ER). ER refers to use ecological data and the machine learning algorithm to predict the psychological traits without contacting the participants. The new method can overcome the shortcomings of the traditional measurement methods, especially the studies of major life events. Through three horizontal researches, the current study verified the validity of ER on examining short-term effects, tracking the trajectory of the long-term impact, and identifying of psychological traits when self-report was not doable. Major life events like domestic violence, loss of the only child, and graduation and marriage of the novel protagonists were studied.

The results of the three studies are shown as follows:

(1)Study 1 examined the validity of ER by identifying the short-term effect of domestic violence. By measuring the psychological traits exactly before the event, we found a consistent conclusion with previous studies that the DV victims have higher level of depression and suicide probability, and a lower level of life satisfaction. The DV victims were lower scored in Agreeableness and Conscientiousness, while higher in Neuroticism. Further, as we measured the exact mental state before the event, it can be inferred that the victim had a poorer mental health after the domestic violence, as well as a decrease in Conscientiousness and an increase in Neuroticism afterwards.

(2) Study 2 examined the validity of ER by tracking the long-term trajectory of mental status after the loss of the only child. The mental status of the lost-only-child users were slightly worse than that of the control group, which is characterized by the use of fewer positive emotional words, and more negative emotional words, especially sad words and death words. Further, the study found that the lost-only-child users used more family words, and it gradually decrease after the loss; the use of negative emotional words gradually decrease, while the use of sad words remain unchanged; the lost-only-child users temporarily increased the frequency of using religious words and returned to normal after 1 year.

(3) Study 3 intended to test the validity of ER by identifying the character development of the protagonists in the "Ordinary World", Sun Shao'an and Sun Shaoping, whom cannot self-report for sure. Consistent with the novel scene, Shaoan has supreme extroversion, high conscientiousness and openness; Shaoping has high openness, conscientiousness and extroversion. Further, we examined the impact of graduation and marriage on personality development. Shaoan decreased in Agreeableness and increased in Neuroticism after got married; Shaoping got higher openness and lower Neuroticism after graduation, which is consistent with the conclusions of the similar life events in foreign studies.

The above three studies utilized several life events to prove the validity of the ER method. Compared with the traditional psychological measurement methods, the ER method can measure the psychological traits retrospectively and without contact, which is suitable for studies where the samples are hard to gain, the participants are not cooperated, and the mental status before the life events are difficult to backtrack. Combining with the traditional method, the ER method could effectively enlarge the application range and improve efficiency of the psychometry.

Pages72
Language中文
Document Type学位论文
Identifierhttp://ir.psych.ac.cn/handle/311026/29296
Collection社会与工程心理学研究室
Recommended Citation
GB/T 7714
刘明明. 基于社交媒体数据的心理特征自动识别新方法研究[D]. 中国科学院心理研究所. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
刘明明-硕士学位论文.pdf(3178KB)学位论文 限制开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[刘明明]'s Articles
Baidu academic
Similar articles in Baidu academic
[刘明明]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[刘明明]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.