Speaker: Gülşen Demiröz (Sabanci University)
Title: Automatic Cost Model Discovery for Combinatorial Interaction Testing of Software Systems
Speaker: İnanç Arın (Sabanci University)
Title: Contribution to Churn Prediction Process by Analyzing the Text Data
Speaker: Ezgi Demirel (Sabanci University)
Title: Hierarchical Knowledge-Rich Semantic Maps for Path Finding
Time: 13:40 -- 14:30
Place: FENS L027
Gülşen Demiröz (Sabanci University)
Automatic Cost Model Discovery for Combinatorial Interaction Testing of Software Systems
Abstract: Testing of highly configurable systems almost always involves sampling enormous configuration spaces and testing only the selected configurations.Combinatorial interaction testing approaches, such as covering arrays, systematically sample the configuration space and test only the selected configurations. My thesis motivation is making covering arrays aware of the different testing costs for each configuration. Therefore, first part of my thesis aims to estimate the cost of carrying out a quality assurance (QA) task, such as building a configuration or running a test case on a configuration, across a whole configuration space, as the estimates can be used for planning the QA process after the samples have been taken as well as for taking cost-aware samples. However, manually creating cost models for this purpose is generally cumbersome and error-prone, thus impractical.
In this talk I will present our automated approaches for cost model discovery in configuration spaces. Given a configuration space, a QA task of interest, and a means of measuring the cost of carrying out the QA task on a configuration, one proposed approach first identifies important effects, i.e., combinations of option settings that affect the cost most, by using a class of highly economical experimental designs, called screening designs, and then uses the important effects to fit a cost model to the observations. The second approach we propose for automatically discovering cost models is by creating generalized linear regression models from the data set with measured times for each test case of a 4-way covering array. The model contains an intercept, 1-factor options, and all products of pairs of distinct options (no squared terms). The resulting cost models, takes as input a possibly previously unseen configuration and estimates the cost of carrying out the QA task on the configuration. We empirically evaluated the proposed approach using two different types of formal screening designs, namely fractional factorial designs and D-optimal designs, with three different QA tasks on two large highly configurable software systems. Then we evaluated the model’s effectiveness by measuring the coefficient of determination (R2) of several tests on exhaustive data sets for screening designs and also on ten different 2,3-way covering arrays. The results of our experiments strongly suggest that the proposed approaches can efficiently and effectively discover cost models for configuration spaces.
Bio: Gülşen Demiröz has received her B.S. and M.S. degrees in Computer Science from Bilkent University in 1995 and 1997. Then she has worked as a Software Design Engineer, Test Lead and Development Lead in various teams at Microsoft Corporation headquarters in Redmond, WA, USA between 1997-2008. She has been an Instructor at Sabancı University since Fall 2008 teaching programming courses with C++ for undergraduates and with C# for IT Master’s program. She has started her PhD studies in Fall 2011 also at Sabancı University with her thesis advisor Cemal Yılmaz. Her research area is software testing of highly configurable systems and combinatorial interaction testing.
İnanç Arın (Sabanci University)
Contribution to Churn Prediction Process by Analyzing the Text Data
Abstract: Churn prediction is one of the major issues of telecommunication companies for their marketing strategies. These companies always try to attract their competitors’ customers; however they give more importance to prevent their own customers to change their operators, because studies across a number of industries have revealed that the cost of keeping an existing customer is around 10% of the cost of acquiring a new one. There are lots of works, both in literature and industry, for churn analysis but they are mostly based on structured Customer Relationship Management (CRM) data (like customer demographic information, purchase history, products, service usage, billing data etc). We contribute to this process with analyzing the text data (in Turkish) obtained from emails, surveys or scripts of phone calls. The text data has been analyzed by using Natural Language Processing (NLP) and Data Mining techniques; and their results have been combined to create a new model for generating churn scores of the customers.
Bio: İnanç Arın is a PhD candidate at Sabanci University Computer Science and Engineering Department. He received his BSc and MSc degree from Sabanci University in 2010 and 2012, respectively.
Ezgi Demirel (Sabanci University)
Hierarchical Knowledge-Rich Semantic Maps for Path Finding
Abstract: Motivated by multiple robots helping/guiding humans to their requested locations in a dynamic environment (e.g., shopping mall), we study path finding and unfolding problems. To solve these problems more efficiently and more intuitively, we introduce a novel mathematical model, called Hierarchical Knowledge-Rich Semantic Map (HSM), that hierarchically represents physical accessibility and qualitative spatial relations between entities of an environment as semantic maps at different levels of abstraction. At each level of abstraction, these semantic are associated with a knowledge base about the entities and the relations at that level. The idea is to combine HSMs with other sorts of knowledge (e.g., commonsense taxonomic knowledge about concepts related to entities, commonsense default knowledge about general assumptions, temporary knowledge extracted from observations and interactions with humans or other robots), to help/guide humans to their goal locations. For that reason, we introduce generic solutions to path finding problems in HSMs using knowledge representation and reasoning methods from Artificial Intelligence, and we introduce an algorithm that recursively solves unfolding problems, taking into account the users’s constraints, preferences, optimization requests, along with other sorts of relevant knowledge. We empirically show the usefulness of computing a path hierarchically with HSMs, compared to computing a path over a flat representation.
Bio: Ezgi Demirel received her B.Sc degree in Computer Science and Engineering from Sabanci University, Turkey in 2014. Currently, she is studying for her master's degree in Sabanci University. Her research interests are in the areas of Artificial Intelligence, Knowledge Representation and Reasoning.