Social Network Analysis of Calling Data Records for Identifying Influencers and Communities

Nowadays, Telecommunication service providers produce a huge volume of calling data records (CDR) each day. A clear understanding of their customers is a key success of any company. To analyze the behaviors and relationships between customers, social network analysis (SNA) is usually employed to detect influencers and communities along with calling behaviors (profiles). Unfortunately, the graph of CDR is different from that of other social media, e.g., Twitter, Facebook, etc. So, the results should be mistaken and cannot reflect the real customers if SNA is directly applied to CDR, such as, misinterpret "telesales" as "influencer". In this paper, we propose a data cleansing process for CDR in order to filter the anomaly numbers. This can improve the accuracy of the analysis and remove any misinterpreted outcomes. Moreover, a measure is invented to capture influencers based on calling behaviors. The experiment was conducted on 2.5 million calling records of a telecommunication in Thailand. The result showed that using our proposed solution and ranking metrics could detect influencers and communities accurately.