Anca A. Chandra

Automatic Discovery of Service Name Replacements Using Ledger Data

Recent studies have illustrated historical financial data could be used to predict future revenues and profits. Prediction models would be accurate when long-run data that traces back for multiple years is available. However, changes in service structures often result in alteration of the nomenclatures of the services, making the streams of financial transactions associated with affected services discontinue. Manually inquiring the history of changes can be tedious and unsuccessful especially in large companies. In this paper, we propose a machine learning based algorithm for automatically discovering service name replacements. The proposed methodology draws heterogeneous features from financial data available in most ledger databases, and hence is generalizable. Our proposed methodology is shown to be effective on ground-truth synthesized data generated from real-world IBM service delivery ledger database.

Forecasting Workloads in Multi-step, Multi-route Business Processes

This paper presents a technique developed to forecast workloads in a business process. Business processes such as the process of engaging on a service contract consist of multiple steps that are not necessarily sequential. There can also be multiple routes that work can take in transition. In order to forecast workloads at different steps of such business processes, one needs to predict dynamic movements of process instances within the system as well as the arrival of new instances from outside. By analyzing transition log data, we construct a Markov chain, which models the movement of process instances across different steps of the business process. Our approach takes into account the fact that an instances prior trajectory may affect its future transitions. Via numerical studies, we demonstrate the overall performance of the proposed forecasting method. We also investigate how the performance of the forecasting method changes as various characteristics of the business process change. The proposed technique is general, and can be applied to a large class of business processes.

Forecasting Service Profitability

This paper describes ways to connect ledger cost behavior of a service delivery project with cost estimations derived at the time of contractual agreement. The purpose of this connection is to improve the management of the service life cycle, providing long range forecasting of the profitability of various service offerings. We emphasize cost, but our methods apply also to revenue, and consequently to profit. In a perfect financial world the connection between the cost estimate and the actual costs would be maintained in a tight feedback loop. The complex real world of multi-year and multi-country service delivery projects requires accommodation of (1) historical changes in accounting practices and (2) missing data. We describe the content of a cost case, the content of the ledger, and methods for forecasting profitability of parts of delivery projects using both the cost case and historical ledger experience of other similar projects. This paper reports a work in progress. We limit discussion of accuracy measurements to one benchmark, and we discuss potential improvements we have not yet implemented.

An Empirical Study on Hybrid Recommender System with Implicit Feedback

The amount of data generated by systems is growing quickly because of the appearance of mobile devices, wearable devices, and The Internet of Things (IoT), to name a few. Because of that, the importance of personalized recommendations by recommender systems becomes more important for consumers inundated with vast amount of choices. Many different types of data are generated implicitly (for example, purchase history, browsing activity, and booking history), and less intrusive recommendation systems can be built upon implicit feedback. There are previous efforts to build a recommender system with implicit feedback by estimating the latent factors or learning the personalized ranking but these approaches do not fully take advantage of various types of information that can be created from implicit feedback such as implicit profiles or a popularity of items. In this paper, we propose a hybrid recommender system which exploits implicit feedback and demonstrate better performance of the proposed recommender system based on the expected percentile ranking and a precision-recall curve against two state-of-the-art recommender systems, Bayesian Personalized Ranking (BPR) and Implicit Matrix Factorization methods, using hotel reservation data.

Discovering Discontinuity in Big Financial Transaction Data

Business transactions are typically recorded in the company ledger. The primary purpose of such financial information is to accompany a monthly or quarterly report for executives to make sound business decisions and strategies for the next business period. These business strategies often result in transitions that cause underlying infrastructures and components to change, including alteration in the nomenclature system of the business components. As a result, a transaction stream of an affected component would be replaced by another stream with a different component name, resulting in discontinuity of a financial stream of the same component. Recently, advancement in large-scale data mining technologies has enabled a set of critical applications to utilize knowledge extracted from a vast amount of existing data that would otherwise have been unused or underutilized. In financial and services computing domains, recent studies have illustrated that historical financial data could be used to predict future revenues and profits, optimizing costs, among other potential applications. These prediction models rely on long-term availability of the historical data that traces back for multiple years. However, the discontinuity of the financial transaction stream associated with a business component has limited the learning capability of the prediction models. In this article, we propose a set of machine learning–based algorithms to automatically discover component name replacements, using information available in general ledger databases. The algorithms are designed to be scalable for handling massive data points, especially in large companies. Furthermore, the proposed algorithms are generalizable to other domains whose data is time series and shares the same nature as the financial data available in business ledgers. A case study of real-world IBM service delivery retrieved from four different geographical regions is used to validate the efficacy of the proposed methodology.

A Policy-Driven Framework for Document Classification and Enterprise Security

In an enterprise, accessing data on the go using several ubiquitous devices is fast becoming the norm. This brings its own set of challenges when dealing with access to sensitive resources. The pervasive nature of these devices makes them ideal candidates for cyber-attacks. The challenge arises in ensuring that devices conform to proper access control requirements. Due to the nature of the cyber-attacks on ubiquitous devices, there is a need to provide dynamic access control to enterprise resources. A system that automatically classifies documents by sensitivity level would be essential to every enterprise. This helps reduce manual document classification that might be prone to errors. To this effect, we propose a framework that automates the security profile detection, access pattern of enterprise resources using text mining, user context, enterprise policies. We also develop a prototype system to elevate the effectiveness of our framework.