Author Email: email@example.com
Social network usage has been increasing among common people day by day, which has lead to an outburst of social data. This social data is a reflection of the user’s behaviour and can be used to predict the patterns with which the user will interact with the social community at large. In this work we propose a novel technique for prediction of user behaviour on the twitter platform which combines natural languages processing (NLP) ,session based analysis, sentiment analysis and behaviour pattern analysis. In this work first we perform NLP based pre-processing which clean all the tweets, for this we used POS tagging which remove all stop words, whitespaces etc. and extract variations of nouns , adjectives and adverbs, these module compares the tweets, and converts them into positive and negative polarity via fuzzy comparison ,and finally a session based engine is used to divide the tweet polarities into sessions and analyse the session-by-session polarity pattern to predict the next tweet behaviour of the user .Our work shows more than 80% accuracy when tested on real time data sets, and is significantly lightweight in terms of processing speed than standard systems.
Polarity, Sentiment Analysis, Social Networks, Sessions Based Analysis, User Behaviour Prediction.
Social pattern analysis has come up recently as an excellent field of study for observing user behaviour both online and in real life. Experts suggest that social network websites like Facebook and Twitter are one of the best tools to observe and in some cases predict the user behaviour by using pattern analysis and data mining. Now-a-days twitter has become most popular social networking site where users send and read posts of up to 140 characters and the posts are either image based or text based. On social networking sites, users share and post lots of knowledge and useful information which is most beneficial for another user in many ways.
User’s action on social media generates behavioural data which show user tastes, involvement, opinions and relationships. Social behaviour of user means the user’s opinion, tweeting or commenting positively or negatively and what kind of language (or word) they used on social media.
On social networking site there are different types of people where they exchange their opinions, some user has positive opinion and some have negative opinion .
Muhammad Al-Quraishi et al. proposed a work to analyse and anomalous behaviours and proposed an integrated social media content analysis that leverages three levels of features, i.e. user generated content, social graph connections and user profile activities. For this also collected a large number of user profile files from twitter and YouTube and this system is based on multiple layers(four layers) which are mutually related and every module having direct communication with every other module. For this experiment, trained five supervised algorithms and used one computer for the experiment and assessed several classifiers i.e. decision tree J48,classification by regression (ClsReg),a support vector machine(SVM) ,random forest (RF)and OIR and trained the classifier model using both normal user and malicious user and the classifier perform well for the tested dataset, the F-measure for the RF and OIR algorithm were the same i.e. 95%,94% for J48,95% for ClsReg and the SVM attained a rate of only 84%.
In our work, we consider only twitter data or twitter user profile and we will be analysing the social behaviour of the users from their tweets and evaluate either the user is malicious or not and predict whether the user will be tweeting positively or negatively based on his/her history. For this work we used corpus based (or rule based model) and also implement by using support vector machine (SVM) algorithm.
Our proposed session based approach outperforms all other classification based approaches, and thus can be used in real time with large datasets. Our session based approach demonstrates a mean performance of more than 80% accuracy, and by using SVM algorithm we achieve accuracy more than 90%, which is a big jump from the conventional classification approaches, and thus is suitable to be integrated into real life social networks like twitter, Facebook or LinkedIn.
67 total views, 1 views today