Micro-blog And Online Shopping BasedUser Psychological Stress Detection

In the modern highly-competitive society,growing people experience severe stress from different aspects from school to work,from family to inter-personal relationship,which negatively influences their mental and physical health.It is of great significance being timely and accurately aware of users' psychological stress and providing effective measures to help them cope with stress.Fortunately,in the big data era,users' rich online behavior leaves powerful cues for sensing their psychological status.State-of-theart work demonstrates the feasibility to detect users' stress from their micro-blog tweets.However,social network based psychological stress detection is still in its infancy and many issues are waiting to be addressed.Firstly,feature space of the existing work upon tweet-wise stress detection still need further completion,and the detection result is rather rough.Secondly,high level stress patterns such as periodicity have not been investigated yet,while they play a much more significant role in stress analysis and future stress prediction than tweet-wise detection.Thirdly,data sparsity dramatically hurts microblog based stress detection,while so far no work makes any investigation upon this.In this dissertation,we propose a series of effective methods to address these problems.Specifically,the contributions of this dissertation are summarized as follows.1.A systematic exploration for teens' stress detection involving social interactions: Since a tweet(140 characters limited)is too short to express stressful status in a comprehensive way,this dissertation first time combines the user's social interactions(i.e.,comments and responses,etc.)under the tweet to supplement the tweet-wise adolescent stress detection.A systematically exploration of micro-blog feature space is constructed,comprised of four kinds of features,tweeting content features,posting features,interaction behavior features and comment-response content features,for teenagers' stress category and stress level detection.Furthermore,leveraging linguistic rules,we map the features under different stress categories within the tweet expression and realize the multi-category and multi-level tweet-wise stress detection.2.Detecting adolescent periodic stress via micro-blog: Based on the tweet-wise stress detection,we further investigate a high level stress pattern,periodic stress.Investigating fine-grained stressors,we first decompose the teenager's stress series into stressorspecific series.Considering the continuity of stress emotion,a trend-based method is leveraged to model the discrete stress series into the sequence of alternative stress/nonstress intervals.Calculating the similarity between such intervals,the stress periodicity is finally identified by extending the symbol based DTW distance with the WARP algorithm.This is the first work defining and detecting stress periodicity,and also the first work exploring user's high level stress patterns.3.Integrating online shopping with micro-blog for stress detection: To settle the data sparsity issue in micro-blog based stress detection,this dissertation novelly exploits user's online shopping behavior as a new convenient channel to discover user's psychological status.Inspired by psychological studies,by analyzing the user's searching,browsing and purchasing behaviors in e-commerce sites,based on user's shopping interests and purchase indulgence,we originally define and extract a series of stressrelated outlier features(e.g.,the indulgently expensive cost,impulsive order frequency,etc.)from user's shopping indulgences to detect whether the user suffers stress.Considering the fact that shopping data lacks of semantic rules,we further design fusion strategies to integrate the detection result from both online shopping and micro-blog to stress category and stress level detection.