Saad Saleh

Post Summarization of Microblogs of Sporting Events

Every day 645 million Twitter users generate approximately 58 million tweets. This motivates the question if it is possible to generate a summary of events from this rich set of tweets only. Key challenges in post summarization from microblog posts include circumnavigating spam and conversational posts. In this study, we present a novel technique called lexi-temporal clustering (LTC), which identifies key events. LTC uses k-means clustering and we explore the use of various distance measures for clustering using Euclidean, cosine similarity and Manhattan distance. We collected three original data sets consisting of Twitter microblog posts covering sporting events, consisting of a cricket and two football matches. The match summaries generated by LTC were compared against standard summaries taken from sports sections of various news outlets, which yielded up to 81% precision, 58% recall and 62% F-measure on different data sets. In addition, we also report results of all three variants of the recall-oriented understudy for gisting evaluation (ROUGE) software, a tool which compares and scores automatically generated summaries against standard summaries.

Analytical Modeling of End-to-End Delay in OpenFlow Based Networks

OpenFlow enabled networks split and separate the data and control planes of traditional networks. This design commodifies network switches and enables centralized control of the network. Control decisions are made by an OpenFlow controller, and locally cached by switches, as directed by controllers. This can significantly impact the forwarding delay incurred by packets in switches, because controllers are not necessarily co-located with switches. Only very few studies have been conducted to evaluate the performance of OpenFlow in terms of end-to-end delay. In this paper, we develop a stochastic model for the end to end delay in OpenFlow switches based on measurements made in Internet-scale experiments performed on three different platforms, i.e., Mininet, the GENI testbed, and the OF@TEIN testbed.

A stochastic model for transit latency in OpenFlow SDNs

Software defined networks (SDNs) introduced the concept of decoupling control and data planes which is a paradigm shift. The OpenFlow protocol is one of a number of technologies that enables this decoupling and, in effect, commodifies network equipment. As of now, there is still limited work that has been done towards modeling the transit delay across OpenFlow switches experienced by network traffic. In this work we develop a stochastic model for the path latency in Open vSwitch (used together with a POX controller) based on measurements made in experiments performed on three different platforms which include 1) Mininet, 2) MikroTik RouterBoard 750GL and 3) GENI testbed softswitch. We propose a log-normal mix model (LNMM) and show that it offers a R2 value of greater than 0.90 for most of our experiments. We also demonstrate how the M/M/1 models proposed in earlier studies is a poor fit.

Improving QoS of IPTV and VoIP over IEEE 802.11n

Tremendous growth rates of Internet Protocol Television (IPTV) and Voice over Internet Protocol (VoIP) have demanded the shift of paradigm from wired to wireless applications. Increased packet loss with continuously varying wireless conditions make the transmission a challenging task in wireless environment. Our study investigates and proposes improvement in the transmission of combined IPTV and VoIP over the IEEE 802.11n WLAN. Our major contributions include the analytical and experimental investigations of (1) transport layer protocol UDP/TFRC for IPTV and VoIP, (2) optimal physical layer parameters for IPTV and VoIP, (3) proposition of wireless enhancement of TFMCC (W-TFMCC) to enhance the capacity and Quality of Service (QoS) of wireless IPTV and VoIP. Analytical and experimental evaluations show a 25% increase in capacity using TFRC with 167% more bandwidth share to TCP. Our study shows that use of W-TFMCC with optimal parameters can enhance IPTV and VoIP capacity by 44%.

IM session identification by outlier detection in cross-correlation functions

The identification of encrypted Instant Messaging (IM) channels between users is made difficult by the presence of variable and high levels of uncorrelated background traffic. In this paper, we propose a novel Cross-correlation Outlier Detector (CCOD) to identify communicating end-users in a large group of users. Our technique uses traffic flow traces between individual users and IM service providers data center. We evaluate the CCOD on a data set of Yahoo! IM traffic traces with an average SNR of -6.11dB (data set includes ground truth). Results show that our technique provides 88% true positives (TP) rate, 3% false positives (FP) rate and 96% ROC area. Performance of the previous correlation-based schemes on the same data set was limited to 63% TP rate, 4% FP rate and 85% ROC area.

Breaching IM session privacy using causality

The breach of privacy in encrypted instant messenger (IM) service is a serious threat to user anonymity. Performance of previous de-anonymization strategies was limited to 65%. We perform network de-anonymization by taking advantage of the cause-effect relationship between sent and received packet streams and demonstrate this approach on a data set of Yahoo! IM service traffic traces. An investigation of various measures of causality shows that IM networks can be breached with a hit rate of 99%. A KCI Causality based approach alone can provide a true positive rate of about 97%. Individual performances of Granger, Zhang and IGCI causality are limited owing to the very low SNR of packet traces and variable network delays.

Sentiment classification of tweets using hierarchical classification

This paper addresses the problem of sentiment classification of short messages on microblogging platforms. We apply machine learning and pattern recognition techniques to design and implement a classification system for microblog messages assigning them into one of three classes: positive, negative or neutral. As part of this work, we contributed a dataset consisting of approximately 10, 000 tweets, each labeled on a five point sentiment scale by three different people. Experiments demonstrate a detection rate between approximately 70% and an average false alarm rate of approximately 18% across all three classes. The developed classifier has been made available for online use.