Link Based Overlapping Community Detection and Medical Data Mining Of Social Media for Cancer Prognosis
Ambika P1, Binu Rajan M.R2

1Ambika P, Department of Computer Science & Engineering, SCT College of Engineering, Trivandrum (Kerala), India.
2Binu Rajan M.R, Department of Computer Science & Engineering, SCT College of Engineering, Trivandrum (Kerala), India.

Manuscript received on 10 October 2016 | Revised Manuscript received on 18 October 2016 | Manuscript Published on 30 October 2016 | PP: 70-75 | Volume-6 Issue-1, October 2016 | Retrieval Number: A4749106116/16©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (

Abstract: Social media, ranging from personal messaging to live foras, is providing unlimited opportunities for patients to exchange their views on their experiences with drugs and devices. Here the aim is to understand the correlation between user posts and positive or negative judgment on drugs along with its side effects in cancer patients with particular emphasis on analysing the notion of community detection within this social network by analysing link properties. The proposed system is a two-step analysis framework where positive negative user sentiments are evaluated using data mining tools and techniques followed by identifying overlapping community structures(influential user modules) within the user forum. The two-way process utilizes the comments on internet message boards(cancer research forums) to infer the acceptance and effectiveness of a drug in cancer treatment and maps to the influential user within the network. In the first stage of the current study, opinion labels are developed about each drug based on opinion analysis from user posts and each word is given weightage per node using data mining tools. In the second stage, networks are built from the search results of the forum, a network ranking system reflecting the opinion formation about the drug is developed. Different from traditional algorithms based on node clustering, the proposed method is based on link clustering to discover overlapping communities. Since links usually represent unique relations among nodes, the link clustering will discover groups of links that have the same characteristics. The current approach effectively searches for different levels of organization within the networks and uncovers dense modules using partition density factor. Finally, the accuracy of novel link based overlapping community detection method is compared with the traditional network based community detection model using graph benchmark. Thus the experiment is used to determine opinion from consumer and identify influential users within the retrieved modules using information derived from both term occurrence and word frequency of data and network-based properties in an accurate way.
Keywords: Community detection, Health Informatics, Multi-scale, Markovprocess, Modularity, Overlapping communities, Random walks, Social media, Stability.

Scope of the Article: Data Mining