Unified framework for spam detection and risk assessment in short message communication media / Adewole Kayode Sakariyah

Short message communication media (SMCM), such as mobile and microblogging social networks, have become essential part of many people daily routine. Despite the benefits offered by these communication media, they have become the popular platforms for distributing spam contents. Research in spam message and spam account detection in SMCM has received growing interests in the recent years, mainly focusing on introducing separate frameworks that can identify spam message or spam account. There are hundreds of published works related to spam message and spam account detection that aim to identify effective detection methods. While spam message and spam account studies have recently advanced, there are still areas available to explore, mostly with respect to introduction of unified method that can detect spam message and spam account within a single framework as well as identifying risk levels of spam accounts. Existing content-based methods for spam detection degraded in performance due to many factors. For instance, unlike contents posted on social networks like Facebook and Renren, SMS and microblogging messages have limited size composed using many domain-specific words such as idioms and abbreviations. In addition, microblogging messages are unstructured and noisy. These distinguished characteristics posed challenges to existing approaches for spam message detection. The state-of-the-art solutions for spam accounts detection have faced different evasion tactics in the hands of intelligent spammers. Thus, the need to investigate features, which can be used to identify spam message and spam account in SMCM. This study is concerned with introduction of a unified framework that can detect spam message and spam account as well as assessing account risk level. To achieve this aim, this study proposed a novel framework, which combines three models: Spam Account Detection Model (SADM), Spam Message Detection Model (SMDM), and Spam Risk Assessment Model (SRAM). Sixty-nine (69) set of features were identified from five main categories to develop the SADM. Additionally, eighteen (18) features were introduced to build the SMDM. The performance of ten (10) machine learning algorithms were evaluated to select the best classifier for both SADM and SMDM. Bio-inspired evolutionary search method was studied to identify the discriminating features for spam account detection. A model to estimate the levels of risk of spam accounts is established using Fuzzy Analytic Hierarchy Process. Four levels of risk were employed with their corresponding response strategies used to map risk levels into different types of response. To assess the performance of the proposed framework, an evaluation study with four stages was undertaken. With promising results being gathered, a proof-of-concept study was conducted using an online assessment mode to demonstrate the applicability of the proposed framework. Based on the results gathered, this study has demonstrated that the proposed framework can be used to detect spam message and spam account as well as assess the risk level of spam accounts in SMCM