Font Size: a A A

Research On Analysis And Prediction Of User’s Behavior In Social Networks

Posted on:2017-03-29Degree:DoctorType:Dissertation
Country:ChinaCandidate:M M WangFull Text:PDF
GTID:1108330482994777Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
With the rapid development of Internet, users’ participation in network activities has been unprecedently increased in both depth and breadth. Social network is not only a tool for communication and self-expression, but also a means of information releasing and public relations marketing for governments, enterprises and organizations. It is becoming a main carrier of information sharing, experience exchange and public opinion. The emergence of social networks subverts the original form of information dissemination, and has brought fundamental changes on people’s previous mode of production, life style, mode of communication and thinking style, such as information acquisition, interpersonal communication, entertainment activities, and so on. Social network is composed of nodes and edges that represent different objects and various types of relationships respectively. As a platform, which shares, disseminates, and accesses information, social network is originality, timeliness, grassroots, randomness, debris, and so forth. Information flow is the driving forth of formation and development of social network, its propagation process is mainly affected by network structure. Since users are main actors in social networks, therefore, information flow which is triggered and propagated via users, freighted with user’s behavior and affects other users through users’ interactive behavior. Consequently, effective analysis and mining behavior rules and interest changes behind user’s behavior in social networks can not only help to deeply understand formation and evolution mechanism of social networks, but also help to improve user experience, customer relationship management, product marketing and precision marketing. It follows that research on analyzing and predicting user’s behavior in greater depth is of great theoretical and practical significance.Base on the characteristics of social networks, the studies in this thesis focus on a number of related problems of analyzing and predicting user’s behavior in social networks, the main contributions and innovations are summarized as follows:(1) Research on User’s Retweeting Behavior Based on Weighted Nonnegative Matrix Factorization AlgorithmRetweeting function in microblogging provides a convenient way to air user’s opinion, as well as a way for users to communicate with each other. Thanks to its great flexibility, this way of information transmission is favored by communicators; meanwhile, it brings a viral spread of a microblog through retweeting behavior between different users for its being nonmandatory, targeted and personalized, so a profound study on user’s retweeting behavior can contribute to understanding propagation process of microblog in the network. Hence a weighted nonnegative matrix factorization model for retweeting behavior prediction is proposed in this thesis. Different from previous methods which predicted retweeting behavior without taking user’s emotions into consideration, this thesis put forward a new concept, emotion difference, which represents difference between the emotion reflected in user’s recent contents and a certain microblog’s sentiment. And along with URL and hashtag, emotion difference is regarded as a content-based factor in the problem of retweeting behavior prediction so as to gain performance. In order to be applied to dynamic networks better, on the basis of time series of user’s contents and user’s network topological information, this thesis considers network as a dynamic flow of time slices and distributes different weights to different time slices to blend temporal information in retweeting behavior prediction algorithm, so as to capture dynamic evolution process of information and network structure. Traditional Salton metrics which calculated adjacent degree between nodes based on network topological information was appropriate for undirected networks, according to directivity of link, this thesis improves traditional Salton metrics and combines it with frequency of interactions to measure relationship-based factor in a social network. Employing social relationship factor to constrain objective function, this thesis casts the predicting problem into solutions of nonnegative matrix factorization from user-based dimension and nonnegative matrix factorization from content-based dimension, respectively, so as to reduce complexity effectively. Experiments on real-world Sina microblog dataset demonstrate the effectiveness of the proposed framework, in addition, further experiments are conducted to understand the importance of features’ weights and temporal information in user’s retweeting behavior prediction.(2) Research on User’s Customizing Social Circles Behavior Based on Improved Fast Clustering AlgorithmCommunity structure, which is a significant property of social networks, often represents specific organized groups of users with similar attributes, hobbies or closer relationships. And community structure is not only very important for understanding the characteristics of complex network, discovering latent topology, predicting network evolution and so on, besides, identifying community structure can also facilitate many tasks such as following/follower recommendation, task allocation, proximity alignment, maximizing influence, retweeting behavior prediction, mining cyber criminal networks and so forth. In recent years, a novel function has been provided in some major social networks: users can categorize their friends into social circles which can be used to filter status updates posted by distant acquaintances, hide personal information from coworkers. As a kind of community structure defined by users themselves, social circle discovery has gradually attracted many researchers’ attention, therefore, this thesis proposes an improved fast clustering method for social circle discovery. Put forward in-link Salton metric and out-link Salton metric according to directionality of linkages to achieve a better representation on adjacent degree between users in directed social networks. Improve a fast clustering method with novel density estimator and extra social circle integration step in order to better adapt to large statistical errors, followed by employing it to detect overlapping social circles in social networks. Evaluate the proposed framework on real-world Facebook, Google+, Twitter dataset and elaborate the importance of different parameters and different features on social circle discovery.(3) Research on Spammer’s Behavior Based on Improved Artificial Immune AlgorithmSince social network which rides with tremendous opportunities, so it has been the main target of spammers, online social network has become susceptible to the actions of malicious users. In social networks, spammers often pry privacy information, promote business, gain high popularity and propagate mendacious opinions, which can result in bad user experiences, as well as misleading or even trapping users. To a certain extent, spammers hamper the healthy development of business marketing activities in online social network platforms. Consequently, an improved artificial immune algorithm for spammer identification is proposed in this thesis. Take user’s individual emotion as a metric to recognize spammer for the first time. Blend temporal information in user’s behavior pattern features on the basis of time series of user’s contents and network topological information so as to capture dynamic evolution process of information and network structure. Improve artificial immune algorithm with novel affinity measurement, multifarious affinity thresholds and normally distributed mutation operator, followed by employing it which is able to better adapt to individual diversity to detect spammers in dynamic social networks. Evaluate the proposed framework on real-world Twitter dataset and elaborate the importance of different behavior pattern features and temporal information on spammer identification.(4) Research on User’s Behavior of Big Five Personality Traits Based on Improved ML-KNN AlgorithmThe explosion of users’ generated data provides a potentially very rich source of information, which plays an important role in helping online researchers understand user’s behaviors deeply. User’s personality traits are the driving force of their behaviors, and individual differences in personality traits may have an impact on user’s online activities, research on user’s personality traits in depth can help to better understand user’s behavior in social networks. Additionally, user’s personality traits can be used to predict early adoption about Facebook, optimize search results, manifest social influence, distinguish individuals who have some common properties in the crowd, predict customer satisfaction and loyalty, therefore, this thesis proposes an improved ML-KNN algorithm for multidimensional personality traits recognition. Group users according to different degrees of user’s personality traits and adopt two correlation metrics to demonstrate the existence of weak correlations between different degrees of user’s personality traits for the first time. Exploit correlations between features and user’s personality traits via correlation analysis, so as to quantify importance of each feature and construct weighted feature set. Update threshold of each personality trait dynamically rather than adopt a unified threshold; On the basis of weighted feature set, dynamic personality traits’ thresholds and prior knowledge of correlations between user’s personality traits, this thesis deploys an improved ML-KNN algorithm to recognize user’s Big Five personality traits. Experiments on real-world Facebook dataset demonstrate the effectiveness of the proposed framework, besides, further experiments are conducted to validate the existence of correlations between user’s personality traits and understand the importance of features’ weights, dynamic thresholds of personality traits and correlations between user’s personality traits in user’s personality traits recognition.To sum up, the studies in this thesis provide a new train of thought for solving related problems in social networks through utilizing machine learning algorithm, artificial immune algorithm and non negative matrix factorization algorithm to analyze and predict user’s behavior. Although the studies in this thesis have a certain theoretical and practical significance, some key technologies still need to be further improved and perfected.
Keywords/Search Tags:Retweeting behavior prediction, Overlapping social circle discovery, Spammer identification, Personality traits recognition, Machine learning algorithm, Artificial immune algorithm, Nonnegative matrix factorization algorithm, Social networks
PDF Full Text Request
Related items