Limits...
A systematic exploration of the micro-blog feature space for teens stress detection.

Zhao L, Li Q, Xue Y, Jia J, Feng L - Health Inf Sci Syst (2016)

Bottom Line: Different classifiers are employed to detect potential stress categories and corresponding stress levels.Experimental results show that all the features in the feature space positively affect stress detection, and linguistic negative emotion, proportion of negative sentences, friends' caring comments and teen's reply rate play more significant roles than the rest features.Involving comments and responses under the tweet supplement the detection and improves the detection accuracy of 16.8 %.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science and Technology, Institute of Data Science, Centre for Computational Mental Healthcare Research, Tsinghua University, 100084 Beijing, China.

ABSTRACT

Background: In the modern stressful society, growing teenagers experience severe stress from different aspects from school to friends, from self-cognition to inter-personal relationship, which negatively influences their smooth and healthy development. Being timely and accurately aware of teenagers psychological stress and providing effective measures to help immature teenagers to cope with stress are highly valuable to both teenagers and human society. Previous work demonstrates the feasibility to sense teenagers' stress from their tweeting contents and context on the open social media platform-micro-blog. However, a tweet is still too short for teens to express their stressful status in a comprehensive way.

Methods: Considering the topic continuity from the tweeting content to the follow-up comments and responses between the teenager and his/her friends, we combine the content of comments and responses under the tweet to supplement the tweet content. Also, such friends' caring comments like "what happened?", "Don't worry!", "Cheer up!", etc. provide hints to teenager's stressful status. Hence, in this paper, we propose to systematically explore the micro-blog feature space, comprised of four kinds of features [tweeting content features (FW), posting features (FP), interaction features (FI), and comment-response features (FC) between teenagers and friends] for teenager' stress category and stress level detection. We extract and analyze these feature values and their impacts on teens stress detection.

Results: We evaluate the framework through a real user study of 36 high school students aged 17. Different classifiers are employed to detect potential stress categories and corresponding stress levels. Experimental results show that all the features in the feature space positively affect stress detection, and linguistic negative emotion, proportion of negative sentences, friends' caring comments and teen's reply rate play more significant roles than the rest features.

Conclusions: Micro-blog platform provides easy and effective channel to detect teenagers' psychological stress. Involving comments and responses under the tweet supplement the detection and improves the detection accuracy of 16.8 %.

No MeSH data available.


Related in: MedlinePlus

Detection performance of different feature combinations on the data subset containing tweets with comments
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4837598&req=5

Fig9: Detection performance of different feature combinations on the data subset containing tweets with comments

Mentions: To eliminate data imbalance, we also examine the performance of different feature combinations on the data subset, containing tweets with comments. It is interesting to see that this time, features of social interaction and comment-response play positive roles in stress detection, as illustrated in Fig. 9. Relying only on tweeting contents FW cannot provide enough information. Together with the posting behavior features (FW + FP), the performance improves around 5.63 % than merely FW. After adding social interaction and comment-response features (FW + FP + FI + FC), the average F-measure reaches 75 %, outperforming FW by 23.3 %, and (FW + FP) by 16.8 %. This verifies the significance of social interactions in the micro-blog feature space construction.Fig. 9


A systematic exploration of the micro-blog feature space for teens stress detection.

Zhao L, Li Q, Xue Y, Jia J, Feng L - Health Inf Sci Syst (2016)

Detection performance of different feature combinations on the data subset containing tweets with comments
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4837598&req=5

Fig9: Detection performance of different feature combinations on the data subset containing tweets with comments
Mentions: To eliminate data imbalance, we also examine the performance of different feature combinations on the data subset, containing tweets with comments. It is interesting to see that this time, features of social interaction and comment-response play positive roles in stress detection, as illustrated in Fig. 9. Relying only on tweeting contents FW cannot provide enough information. Together with the posting behavior features (FW + FP), the performance improves around 5.63 % than merely FW. After adding social interaction and comment-response features (FW + FP + FI + FC), the average F-measure reaches 75 %, outperforming FW by 23.3 %, and (FW + FP) by 16.8 %. This verifies the significance of social interactions in the micro-blog feature space construction.Fig. 9

Bottom Line: Different classifiers are employed to detect potential stress categories and corresponding stress levels.Experimental results show that all the features in the feature space positively affect stress detection, and linguistic negative emotion, proportion of negative sentences, friends' caring comments and teen's reply rate play more significant roles than the rest features.Involving comments and responses under the tweet supplement the detection and improves the detection accuracy of 16.8 %.

View Article: PubMed Central - PubMed

Affiliation: Department of Computer Science and Technology, Institute of Data Science, Centre for Computational Mental Healthcare Research, Tsinghua University, 100084 Beijing, China.

ABSTRACT

Background: In the modern stressful society, growing teenagers experience severe stress from different aspects from school to friends, from self-cognition to inter-personal relationship, which negatively influences their smooth and healthy development. Being timely and accurately aware of teenagers psychological stress and providing effective measures to help immature teenagers to cope with stress are highly valuable to both teenagers and human society. Previous work demonstrates the feasibility to sense teenagers' stress from their tweeting contents and context on the open social media platform-micro-blog. However, a tweet is still too short for teens to express their stressful status in a comprehensive way.

Methods: Considering the topic continuity from the tweeting content to the follow-up comments and responses between the teenager and his/her friends, we combine the content of comments and responses under the tweet to supplement the tweet content. Also, such friends' caring comments like "what happened?", "Don't worry!", "Cheer up!", etc. provide hints to teenager's stressful status. Hence, in this paper, we propose to systematically explore the micro-blog feature space, comprised of four kinds of features [tweeting content features (FW), posting features (FP), interaction features (FI), and comment-response features (FC) between teenagers and friends] for teenager' stress category and stress level detection. We extract and analyze these feature values and their impacts on teens stress detection.

Results: We evaluate the framework through a real user study of 36 high school students aged 17. Different classifiers are employed to detect potential stress categories and corresponding stress levels. Experimental results show that all the features in the feature space positively affect stress detection, and linguistic negative emotion, proportion of negative sentences, friends' caring comments and teen's reply rate play more significant roles than the rest features.

Conclusions: Micro-blog platform provides easy and effective channel to detect teenagers' psychological stress. Involving comments and responses under the tweet supplement the detection and improves the detection accuracy of 16.8 %.

No MeSH data available.


Related in: MedlinePlus