Free-form, unstructured and semi-structured textual data has become increasingly more prevalent in the telecommunications industry, with service and equipment providers alike. Some typical examples include textual data from customer care tickets, machine logs, alarm and alerting systems, and diagnostics. There is a growing business need to rapidly and automatically understand the underlying key topics and categories of this bulk collection of text. With the present mode of operation of relying on domain experts to analyze textual data, there is a clear need to apply text analytics to automate the process. Difficulties arise due to the jargon-filled and fragmented, incomplete nature of textual data in this field. In this paper, we propose a domain-agnostic, unsupervised approach that deploys a multi-stage text processing pipeline for automatically discovering the key topics and categories from free-form text documents. Using anonymized datasets retrieved from actual customer care tickets and system logs, we show that our approach outperforms traditional text mining approaches, and performs comparably to manual categorization tasks that were undertaken by domain experts with full system knowledge.

In this paper, we address the detection of face-to-face social interactions between people using unobtrusive, wearable iBeacon devices, as well as the determination of groups and communities that these individuals form. This is a challenging problem since this type of sensor data is very noisy, often incomplete with many missing values, and easily perturbed by motion and nearby obstacles, especially in very dynamic indoor environments. The key idea of our approach is to transform the original noisy sensor data into specific feature vectors by a series of statistical methods designed to overcome the challenges in the data. These features vectors, reflecting the interaction between people, are then clustered to reveal communities without the need to rely on localization information or pre-defined interaction models and parameters. Our data-driven approach provides robust community detection that is not sensitive to noise and missing signals, and automatically captures the dynamic interaction between people. Our approach is scalable for application in real-time.

This paper presents a collection of approximation formulas that allow a network planner to quickly estimate the size of a mesh optical network with limited inputs.In particular, it provides a set of equations that relate number of sites, average fiber connectivity, demand load and capacity for various mesh protection architectures.These results can be used to easily and quickly estimate the amount of traffic that can be carried over a given network, or, conversely, given the traffic to be supported, to assess the characteristics of the topology required (in terms of number of nodes, connectivity).Finally, this analysis can be used to estimate the restoration performance that can be expected without requiring any extensive simulation studies.

The complexity of modern networks and the stringent quality of service requirements that result in very small important event probabilities can render standard Monte Carlo (MC) simulation intractable. Accelerated simulation using splitting can potentially overcome these difficulties, but conventional methods cannot be successfully applied to systems for which the occurrence of the important event does not temporally coincide with the conditions that lead to it. Cell delay experienced through single and tandem switches exhibits this behavior. We develop two enhanced splitting methods based on a splitting technique which was previously used to estimate rate delay probabilities through a single switch. The enhanced methods accurately capture the queueing behavior that leads to excessive cell delay through tandem switches. We use the enhanced methods to efficiently estimate rate delay probabilities for tagged traffic traversing tandem switches in the presence of background traffic. The enhanced methods penetrate into a significantly higher delay (lower probability) region compared to the previous method. Speedup over standard MC simulation is observed to be inversely proportional to the probability being estimated.

Importance sampling has been used as an efficient method for estimating rare probabilities (on the order of 10/sup -6/ or less) relating to the performance of communication systems and networks. Parametric importance sampling methods are not very effective in cases where the input processes are characterized by uniform probability distributions (e.g., random delays), which arise frequently. We present a conditional importance sampling scheme for systems with input processes that can be characterized by uniform distributions. The scheme adaptively modifies an initial biasing strategy as samples are taken. Changing a problem specific component enables the algorithm to be applied to a diverse set of systems. The overall approach is more effective than parametrically biasing the uniform input distributions. We use the conditional biasing algorithm to estimate rare jitter probabilities in ATM switches for CBR sources multiplexed with heterogeneous background traffic. For the experimental systems considered, we observe that the improvement in simulation efficiency is inversely proportional to the probability being estimated.

For ATM networks with heterogeneous input traffic, namely constant bit rate (CBR) and variable bit rate (VBR) traffic, we consider the aggregate probability that the cell transfer delay exceeds a given threshold (delay threshold probability). We use the connection traffic descriptors standardized by the ATM Forum to characterize the input traffic. In this paper, we extend our importance sampling simulation model, which was previously used for homogeneous traffic, to heterogeneous traffic. Again, we observe that the improvement in simulation efficiency is inversely proportional to the probability being estimated.