Search engines not only assist normal users, but also provide information that hackers and other malicious entities can exploit in their nefarious activities. With carefully crafted search queries, attackers can gather information such as email addresses and misconfigured or even vulnerable servers. We present SearchAudit, a framework that identifies malicious queries from massive search engine logs in order to uncover their relationship with potential attacks. SearchAudit takes in a small set of malicious queries as seed, expands the set using search logs, and generates regular expressions for detecting new malicious queries. For instance, we show that, relying on just 500 malicious queries as seed, SearchAudit discovers an additional 4 million distinct malicious queries and thousands of vulnerable Web sites. In addition, SearchAudit reveals a series of phishing attacks from more than 400 phishing domains that compromised a large number of Windows Live Messenger user credentials. Thus, we believe that SearchAudit can serve as a useful tool for identifying and preventing a wide class of attacks in their early phases. 1

Abstract not found

Predicting potential advertisement clicks of users are important for advertisement recommendation, advertisement placement, presentation pricing, and so on. In this paper, several machine learning algorithms including conditional random fields (CRF), support vector machines (SVM), decision tree (DT) and backpropagation neural networks (BPN) are developed to learn user’s click behaviors from advertisement search and click logs. In addition, four levels of features are extracted to represent user search and click intents. Given a user’s search session and a query, machine learning algorithms along with different features are proposed to predict if the user will click advertisements displayed for the query. We further study the impact of feature selection algorithms on the prediction models. Random subspace (RS), F-score (FS) and information gain (IG) are employed to search for a predictive subset of features. The experiments show that CRF model with the random subspace feature selection algorithm achieves the best performance.