Mining Users' Leanings In Online Social Networks

User orientation research is one of the most important research directions in the social network field.It is the core technology of analyzing a user's view of a certain thing or time in a social network.The basic data of the user's tendency study is the data that collected from the social media through the API interface.User tendency study is to judge user tendencies by analyzing the text which published by the social media users.The existing methods are usually using sentimental analysis as the foundation,through the finding the specific adjectives using in the text to judge the tendency of users.However,Adjectives with emotional tendencies appear only in a small number of text,using a small part of text to judge users' tendency sonetimes leads to an error result in the final.In view of the user's tendency study,this thesis starts from the words used to break through the limitations of the traditional user's tendency to study.The main work and innovation are as follows:1.Using the specific nous to judge users' tendency.We use the 2016 American election as an example.According to our analysis,there are many un-sentimental words like nouns can also suggest the preferences of online users.For example,when an online user mentions `racism' or `feminlism',this user usually has some negative attitude towards Mr.Donald Trump.We also found these un-sentimental words take a significant share of the total amount of text.Therefore,we are motivated to develop a new analysis based on both un-sentimental words and previous sentimental analysis.The idea is promising to solve the the contribution of our work is proposed a new method to solve the above inaccuracy problem in the polls.2.According to our analysis of users' behaviours,a user tends to retweet or like the text that has similar ideas with them.Meanwhile,text with similar ideas often have high-degree similarity.Therefore,it is a promising work to take the `retweet',`like' or `similarity' as the link relationship to construct the complex network.In the complex network,we willsee every user or tweet as a node,the edges between nodes mean there are relationships between nodes.For example,users who `retweet' or `like' the same text,text which are `retweeted' or `liked' by the same users all have an edge between each other.After constructing the complex network,we can predict election results by dividing the network into several communities through the method of community detection.Compared with traditional community detection problem,we recognize the communities as parts of nodes in the network according to the sentimental analysis.We think that this method could help us in judging users' performance.3.Most of the existing affective analysis using machine learning methods adopt SVM classifier to judge the user's tendency.For supervised learning,a large number of training samples are usually labelling by human beings which is time-consuming and laborious.But there are too many learning tasks in the real world and it is impossible to manually label a large amount of training data for each task.What's worse is that things are always in constant change,so they need to keep marking training samples,which is obviously an impossible task.Long-life machine learning is to create a computer system through continuous data accumulation using model learning and integration.Through the continuously learning and reading the model to identify the users;tendency of text or user texts.This method can effectively solve the above problems.