Voice Spam Detecting Technique Based on User Behavior Pattern Model

Abstract

Voice spam is a major serious issue that would lead the people's lives a few inconveniences in recent years. In this paper, a model based on user behavior pattern is proposed to design an anti spam calls technique. The basic idea for the technique is that spammers with revenue motivation behave significantly distinct to legitimate callers. The characteristic parameters representing user behavior pattern can be joined to help filter spam calls among phone system. The proposed scheme is applicable for detecting spam calls under different backgrounds and it is more simple and effective compared to existing spam calls defending techniques.

Unsupervised Clustering-based SPITters Detection Scheme

“…The works are mainly divided into three research categories: (1) features-based SPITters detection scheme e.g., Refs. [1], [3], [5], [6], [7], (2) SPITters detection based on social network trustworthiness e.g., Refs. [8], [9], [10], [11] and (3) content-based SPIT detection e.g., Refs.…”

“…Wang et al proposed call/receive ratio and normalized call frequency based features CI and F CD which are input into the kmeans clustering algorithm [6]. The scheme finds the center mass of a legitimate callers and classifies each caller by comparing the distance between the caller and a common reference model with the trained threshold.…”

“…Hence, we can leverage the ratio between incoming calls and outgoing calls as a feature for discrimination. Although many features characterize this fact, e.g., IR, HR, SR [5], CI [6], the ratio between outgoing and incoming calls [16], BDR (Bi-Directional Ratio) and IOR [19], we use IOR in this paper. Both BDR and IOR focus on "the number of callees" instead of "the number of outgoing calls" used in IR, HR, SR, and CI.…”

“…Although there exist many other feature-based SPIT detection methods e.g., Refs. [3], [5], [6], and [16], they give ambiguous setting for the threshold, reference model, or training method: thus, we do not compare them in order to avoid inaccurate comparisons. Both LTD and PMG need threshold tuning.…”