基於實時通話行為建模之詐騙電話偵測研究

Abstract

本篇論文主要目的在提出一個即時辨識電話號碼是否為詐騙電話的模型。在辨識電話號碼是否為詐騙電話時，有兩個問題需要處理，一個是訓練好的模型無法適用於新的資料，而另一方面，可以對新出現的電話號碼進行辨識的模型又準確率不高。 我們提出一個模組化的通話表徵與辨識模型，藉由兩階段的訓練，學習產生通話表徵以及用通話表徵進行辨識。第一階段的通話行為預測訓練讓模型學會產生含有豐富資訊的通話表徵，有了通話表徵，就可以訓練一個簡單的分類器進行辨識是否為詐騙集團。模型在實驗中表現遠高於隨機分類並擊敗對通話行為沒有建模的基準模型。 在未來工作方面，可以考慮同時對多個電話號碼建模，因為有些詐騙行為是同時運用多個電話號碼協作完成。

The main purpose of this thesis is to propose a model that can detect whether a phone number is a fraud in real-time. There are two problems in detecting fraud. Some methods can only apply at the same time interval as training data. On the other hand, a model that can apply to a new phone number have low precision. We propose a modularized call representation and detection model. By two-phases training, our model can generate call representations and uses the call representations to detect fraud. In the first phase, call behavior prediction training allows model generating call representation containing rich information. We then train a simple classifier to detect fraud based on the call representation. Our model outperforms the random baseline and beats baseline model which lacking the call behavior module. As for future work, multi phone number modeling can be used to detect complex fraud because Some fraud is cooperating between several phone numbers.