Research On Human Dynamics Based On Social Relationships

The heterogeneous nature of human behaviors contributes to various complex phenomena in the human social system.The understanding and exploration of human behaviors have long been the direction of scholars in various fields.In recent years,with the development of information technology and the advent of the era of big data,more and more digital footprints of human behaviors have left on the Internet,which provides an opportunity for quantitative analysis of human behaviors.The study on the spatialtemporal statistical laws of human behaviors and the dynamic mechanisms has attracted extensive attention from scholars in many fields and achieved fruitful results.However,when human behaviors expand from offline real physical space to online virtual cyber space,the current researches have not given a clear description of the coupling law,strategy selection,time evolution and other characteristics between different behavior spaces.Therefore,by mining social media data recording real human behaviors which includes offline check-in data and online social network data,this thesis will carry out systematic research on the coupling association and behavioral strategy of human behaviors based on social relations from the aspects of similarity,diversity,spatial mobility,and online friend-making.Firstly,the correlation between the similarity of human online and offline behaviors is analyzed by mining the dataset of QQ users which containing the social relationship and check-in records.A set of metrics is defined to measure the similarity of social network structure and mobile check-in trajectory between individuals.The social similarity can be described by whether the two individuals are friends,whether they have common friends,the number of common friends and the diversity of common friends,while the mobility similarity can be obtained by calculating cosine similarity between trajectory vectors.The statistical results show a positive correlation between social similarity and mobility similarity of pairs of individuals.Specifically,the mobile similarity between friends and friends with common friends was higher than that between non-friends and friends without common friends.What's more,the diversity of common friends,rather than the number of them,shows a highly positive correlation with the pair's mobility similarity.It also found that individuals with the same behavioral preference,such as preferring to visit new places that they have never been to,also have higher mobility similarity.Finally,the control experiments constructed based on the network null model,the co-location rate index which measures the mobility similarity,and the edge connectivity index which measures the social similarity all show that our statistical results are valid and robust.Secondly,the correlation between the diversity of human online and offline behaviors was analyzed through the dataset of Gowalla check-in data.The online social diversity was measured by the number of users' friends in the network space and Shannon entropy based on the network structure,and the offline spatial diversity was measured by the users' visited locations in the physical space and Shannon entropy based on the visiting frequency.Then,the correlation between the diversity of human behaviors is analyzed from the perspectives of self-correlation,social correlation and topological correlation in these two behavior spaces.The statistical results show that the diversity of human behavior shows a positive correlation which is independent of the diversity metrics.This positive correlation not only exists in the same behavior space of one individual,but also exists between different behavior spaces,and even extends to his/her direct and indirect social relations.This correlation at the social level decays with the increase of the topological distance,and almost disappear after the topological distance exceeds 3 hops.In addition,the analysis of the evolutionary pattern of diversity finds that spatial diversity would increase over time.Certain differences are observed between such growth pattern of active and inactive users,suggesting that the use of Internet products by social network users shows an accelerating trend of differentiation.Then,using the location category(POI)information in the QQ data,the human spatial movement behavior is analyzed from two perspectives,i.e.,pattern and strategy.In the research on behavior pattern,strong heterogeneity feature is found in both visiting pattern of single POI and transition pattern between two consecutive POI.Most of the visiting and transition are within the scope of a few of the most frequently visited POI.The visiting and growth pattern of POI follow Zipf's law and Heaps' law respectively,and a universal behavioral pattern is observed in groups of users with different amounts of visiting POI.In the research on behavior strategy,by comparing the individuals' visiting pattern of locations and POI,it is found that the visiting pattern of POI is more heterogeneous and bounded,with a higher social repetition rate than that of locations.Inspired by the above empirical analysis,we believe that the visiting pattern of POI can be explained by three strategies,including exploring a new POI,returning to a visited POI,and staying at the current POI.The simulation results show that the dynamic model constructed according to the three strategies can well reproduce the phenomena observed in the empirical analysis.Finally,by analyzing the behavior of the users of a large online social network,Renren,we investigate the strategies and their evolution of online friend-making behavior.Empirical research shows that homophily,triadic closure,and preferential attachment are three popular strategies for online social network users to make friends,which can explain more than 90% of the total relations.Among them,the homophily and triadic closure strategies are verified by the fact that the probability that individuals make friends with who are from the same province or school as them,or with friends' friends is significantly higher than that of the random case.Preferential attachment strategy is proved by the fact that both users who already exist in network and users who just enter network prefer to connect to the popular users with lots of friends.In most cases,social network users adopt multiple strategies to choose new friends,rather than a single strategy.In addition,such strategies change over time.Users who just enter the network tend to use homophily strategy,while over time,they will use the other two strategies more frequently.Homophily is the most frequently used single strategy,and the combination with triadic closures is the most frequently used combined strategy.