Research On Marketing Strategies Based On Social Media Big Data

With the increasing development of the cloud age, big data has been influencing our daily lives from various perspectives, and the big data trend has been embraced from enterprises to customers, from science to government, etc. As one of the significant data sources and applications of big data, social media has revealed extremely potentials. On one hand, big data is the inevitable consequence of social media. Since users could create, share and exchange large amount of information on various social media platforms, social media data have been growing explosively. On the other hand, big data brings opportunities and challenges for social media as well. Big data enables the probability of data-driven decisions, and therefore how to take advantage of social media big data using data mining techniques and then extract insightful business values has been one of the greatest challenges.In this paper, we take Twitter, a typical social media platform, as an example, to explore the social media marketing applications using social media big data. Generally, we summarize user behaviors in social media as:(1) follow, which leads to connection construction; (2) unfollow, which indicates connection breakup; and (3) mention, which represents social communications including comments, replies, retweets and direct messages through @ symbol. By considering relationship based behaviors (i.e. follow and unfollow) and interaction based behavior (i.e. mention), we can comprehensively understand the behavioral characteristics of social media users, and therefore be prepared for further data analysis.In this paper, considering the user behaviors of social media,we propose to research on several topics from the perspective of social media marketing, and provide the problem formations and solutions for each topic. Particularly, we are investigating three marketing applications, i.e., network public crisis monitoring and analysis, targeted advertising and customer relationship management.Specifically, the research topics and contributions of this paper can be summarized as follows:(1) Category the user behaviors on social media such as Twitter into relationship based behaviors (i.e., follow and unfollow) and interaction based behaviors (i.e., mention), and particularly focus on unfollow and mention. We define some real world marketing application problems from the perspective of social media marketing based on unfollow and mention behaviors, that is, network public crisis monitoring based on unfollow, targeted marketing based on mention and user classification for Twitter audiences.(2) Propose the network public crisis problem based on unfollow. Indeed, unfollow is an interesting behavior because:a) compared to follow which is typically made in a rush, unfollow is a more serious and trustworthy decision; b) unfollow means the breakup of relationships, and thus inherently contains the concept of crisis. To this end, in this paper, we propose the definition of crowd unfollow, referring to the concurrent suddenly group behavior of unfollow, and provide the method of crowd unfollow detection. Moreover, when the crowd unfollow is detected, there might emerge opinion leaders which could be the cause of crowd unfollow crisis.(3) Propose a mention based recommender system, which is built upon the mention mechanism of social media.@ recommendation is defined as the problem of suggesting candidate mention users when a publish user is composing a promotion oriented tweet. We argue that mention and @ recommendation are interesting with two observations. First, publishing tweet by mentioning others could contribute to information sharing and social interactions, and thus benefit customer relationship management over social media. Second, as a pull mechanism, mention could actively trigger and encourage interactions at the time of publishing tweets, and mention users would be explicitly informed about the mention tweet. This is different from simple information posting and the retweeting mechanism, which is indeed a passive push pattern. Specifically, we formulate @ recommendation as a learning-to-rank problem of generating top ranked candidate users given publish user and mention tweet. By comprehensively considering content, social, location and time based features of publishers, tweets and candidate users, we modify the basic Ranking SVM model by introducing two bias with the objective of learning most relevant and responsive candidates.(4) Investigate Twitter audiences classification problem, that is, given specific enterprise or brand user, we try to discover overlapping communities among Twitter audiences, which refers to the whole population of both followers and mention users of target user, by constructing specialized social network based on follow, unfollow and mention behaviors. Besides, we introduce data field theory for user relationship modeling in social networks, including direct and indirect social relationship. Specifically, by calculating the potentials and radiation range of nodes, overlapping communities are partitioned based on user influence. Consequently, by effectively identifying and modeling user characteristics and user relationships, personalized services and solutions could be delivered through Twitter audiences classification.