Sridhar Varadarajan

An efficient ad recommendation system for TV programs

With broadcast Television (TV) going digital, the number of channels and the programs aired have increased tremendously. Millions of audiences of various categories such as adults, children, youth and families watch these programs. Advertisements (ads) aired during these programs are targeted to reach these varied audiences and are the main revenue earners for TV broadcasters. While TV broadcasters have the task of scheduling hundreds of ads during the various ad breaks of programs, it is important that the ads shown during any ad break have a good impact on the viewers. An intelligent ad recommendation system that takes into account various factors such as ad/program content, viewers’ interests, sponsors’ preferences, program timing, program popularity and the available ad slot that help increasing the ad revenue would be useful for sponsors and broadcasters. We present in this paper a single end-to-end ad recommender system that considers all of these factors and recommends a set of well scheduled and sequenced ads that are the best suited for a given TV ad break. The proposed recommendation system captures the features of the ad video in terms of annotations derived from MPEG-7 descriptions and these annotation keywords are systematically grouped into a number of pre-defined semantic categories by using a categorization technique. A fuzzy categorical data clustering technique is then applied on the categorized data for grouping the best suited ads for a set of pre-defined program classes such as News, Sports, Cartoons etc. The program classes considered are selected to match with the TV program genres proposed in the TV-anytime standard. Since the same ad can be recommended to more than one program depending upon multiple parameters, fuzzy clustering acts as a well suited (and perhaps also the best suited) technique for ad recommendation. The relative fuzzy score called “degree of membership” calculated for each ad is an indicator of the number of program clusters to which the given ad belongs to. The clustered ads are then scheduled using an algorithm that takes into consideration parameters such as program popularity, program timing and available ad slots, to provide the best possible package for sponsors to show their ads. The scheduled set of ads if played randomly during an ad break might make viewers (sponsors) unhappy, for instance, when similar (competing) product ads get played consecutively. Hence, the system employs sequencing algorithm that takes into account the pre- and post-ad sequences in order to better order the scheduled set of ads in any ad break. We show that our proposed recommendation system provides an effective way of recommending the right ads for broadcast TV programs. We also demonstrate that this strategy does indeed help sponsors to attract viewers’ attention while playing their ads during ad breaks of TV programs. The proposed ad recommendation system is compared and evaluated subjectively with the current ad display system, by ten different people, and is rated with a high success score.

Semantic characterization of real world events

Reducing the latency of information delivery in an event driven world has always been a challenge. It is often necessary to completely capture the attributes of events and relationships between them, so that the process of retrieval of event related information is efficient. In this paper, we discuss a formal system for representing and analyzing real world events to address these issues. The event representation discussed in this paper accounts for the important event attributes, namely, time, space, and label. We introduce the notion of sequence templates that not only provides event related semantics but also helps in semantically analyzing user queries. Finally, we discuss the design for our Query-Event Analysis System, which is an integrated system to (a) identify a best sequence template given a user query; (b) select events based on the best sequence template; and (c) determine content related to the selected events for delivering to users.

Hardware-software hybrid packet processing for intrusion detection systems

Security is a major issue in today’s communication networks. Designing Network Intrusion Detection systems (NIDS) calls for high performance circuits in order to keep up with the rising data rates. Offloading software processing to hardware realizations is not an economically viable solution and hence hardware-software based hybrid solutions for the NIDS scenario are discussed in literature. By deploying processing on both hardware and software cores simultaneously by using a novel Intelligent Rule Parsing algorithm, we aim to minimize the number of packets whose waiting time is greater than a predefined threshold. This fairness criterion implicitly ensures in obtaining a higher throughput as depicted by our results.

A Topology Based Localization in Ad Hoc Mobile Sensor Networks

Localization in an ad hoc mobile sensor network is an important requirement as most of the applications that use sensor data require sensor location information to complete the processing. A typical sensor network has over hundred to thousand sensor nodes, and considering the size and cost of a sensor, using only GPS for localization is not very attractive. The mobility of sensor nodes could lead to network topologies wherein accurate computation of absolute position of all the sensor nodes may not be possible. In this paper, we propose a topology based localization approach that suggests a best possible approximate position for sensor nodes for which computation of exact absolute position is not possible. We have identified four basic topological configurations that help compute position with varied degree of accuracy. These atomic configurations have been identified keeping in mind the simplicity of the computational procedures associated with these configurations. In order to put less demand on a computational capability of a sensor node, we suggest that only a pre-defined number of sensor nodes are compute-enabled (c-nodes) in the sense that they have adequate computational power. Similarly, only a pre-defined number of sensor nodes are GPS-enabled. In such a sensor network, the distributed computation of localization is achieved by distributing the computational requirements of individual sensor nodes across the c-nodes. Each sensor node strives to improve its localization by constantly monitoring its neighborhood and requesting an associated c-node to recompute position whenever neighborhood topology changes. We provide some initial results that bring out the merits of the proposed approach.

Priority-based high-speed intelligent rule-checking

Snort rule-checking is one of the most popular forms of network intrusion detection systems (NIDS). Recent work in string matching has focused on offloading string matching to hardware realizations in order to achieve time and space efficiencies. However, any form of implementation would have to maintain a threshold performance in order to keep with the packet data rate. While trying to maintain this threshold, control-unit is forced to either drop some packets or some rules. In this work, we present a packet priority technique that prioritizes the packets such that there is a high probability that the potentially malicious packets are detected within the threshold limit. We also present a novel intelligent string arrangement (ISA) that enables complete packet scanning in a smart and time efficient manner

A shift varying filtering theory for dynamic service guarantees

We extend the recently developed filtering theory under (max,+)-algebra to shift varying setting and develop a calculus for dynamic service guarantees for variable length packet networks. By modifying the Changs traffic characterization to shift varying setting, we introduce two basic network elements for time varying input traffic models: (i) traffic regulators that generate G-regular marked point processes and (ii) dynamic G-server that provide service guarantees for marked point processes. Similar to shift invariant setting under (max,+)-algebra, network elements can be joined by concatenation, “filter bank summation,” and feedback to form a composite network element. We illustrate the use of framework by various examples like time varying G/G/1 queue, segmentation and reassembly, jitter control, damper and window flow control.