Methods of Data Analytics for Social Media by Considering Social Interaction Patterns and Post Evolving Trends

Abstract

Social media provides a platform for people to share their life experiences for grouping friends together for advertising business products and for governments to announce information With the various services provided by social media the number of users continues to gradually increase However determining how best to understand users’ demands and perspectives from their information sharing in social media is an important challenge Therefore this dissertation focuses on the development of data analytic technologies based on the implicit information of post relative content and user behaviours hiding in social media Firstly the sentiment analysis method related to social media users based on integrating their posts of textual opinion and social interactions is proposed herein With this method a social opinion graph which indicates users’ social actions and relationships is constructed; the sentiment guiding matrix denotes the influential strength between users’ sentiments the textual sentiment classifier is built for classifying textual opinion and social enthusiasm is considered as the care degree between two users The method is applied to the real cases of Taiwan’s presidential election and hot social issues The experimental results of the integrated sentiment classification achieve better accuracy compared to previous research for both cases Secondly the early forecast method of post-based evolution trajectory (TEF) on social media is presented herein In model generation phase the historical data of post forwarding and responding activities on social media are collected in consecutive time for forming the post-based evolution trajectory Then the classification function of each trend type is defined for classifying post-based evolution trajectory which is then labeled as one of the trend types the L-type B-type D-type and G-type The post-based evolution trajectory model is then generated by random forests for further forecasting In trajectory forecast phase the real-time data of post forwarding and responding activities are collected in consecutive time for forming the target trajectory which is forecasted according to the stages of the classification estimation correlation evaluation and distance calculation between the target trajectory and post-based evolution trajectory model TEF is applied on the social media posts achieving excellent performance in trend type forecasting and forecasting the trend type into four classifications while also reducing the time consumed in data processing Thirdly the integrated social data analytics hub is built with an integrated content viewer and social data analytic services In this hub user’s files which are stored in different social media platforms are backed up in the local network attached storage The files such as video pictures or document stored in user’s multiple social media spaces are managed only through the viewer of unified file management Moreover the viewer of private social network is constructed for sharing files with close friends and family instead of in public social media spaces Furthermore the microblog information in different social media platforms displays at a time through the viewer of personal social media services In this viewer three social data analytic services are provided including hotly discussed post analysis sentiment analysis and resonance user mining The experimental results show that the hub builds an environment for managing cross-platform information through the integrated content viewer easily and provides various social data analytic services

Date of Award	2016 Feb 16
Original language	English
Supervisor	Yau-Hwang Kuo (Supervisor)