A Comparative Study Of Stochastic Clustering Techniques In Harvesting Emerging Trends From Social Data

Mohd. Safar, Shariff (2015) A Comparative Study Of Stochastic Clustering Techniques In Harvesting Emerging Trends From Social Data. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) A Comparative Study Of Stochastic Clustering Techniques In Harvesting Emerging Trends From Social Data 24 Pages.pdf - Submitted Version Download (107kB)

Abstract

Social media is an emerging area of interest and attracts many researchers as the latter provide tremendous amounts of data readily available that could be exploited for various reasons such as social networking, decision making and marketing. An emerging field in the area of social media data mining is known as topics detection from social data. A topic is harvested from social data by using an unsupervised machine learning task also known as clustering to group similar social data and recognize the importance of the grouped social data to provide a general distinction which will be known as topic. In this work, two clustering techniques known as the hierarchical clustering technique and density based clustering technique which shares stochastic capability is compared using dataset retrieved from Twitter which corresponds to real world events. The classes present in the dataset is restricted to four main topics and the dataset is then used to test the performance of the clustering algorithms. The performance evaluation being used to evaluate the clustering performance of the clustering algorithms are the V-measure which is the harmonic mean of homogeneity and completeness score of the clustering performance of a clustering algorithm. The results shows that the hierarchical clustering technique outperforms the density based clustering technique in determining the correct number of clusters and assigning the data to their respective clusters reliably. Apart from the comparative studies discussed in this project, an analysis tool based on social data is developed to address the problems related to the social data analysis.

Actions (login required)

View Item