CRA-W CREU Posters
It’s a big challenge for mobile users to pick high quality mobile apps from a large pool of candidates in app markets. In this work, two sentiment analysis techniques, SentiWordNet and a support vector machine (SVM), were used to evaluate app quality from reviews. By aggregating sentiment scores of apps from a developer, we generate a reputation score that can provide users with a more precise measurement of developer reliability.
Our project is interdisciplinary in nature and utilizes computational tools to assess biological problems, specifically tools that can be used for transcriptomic analyses of non-model systems. The class Aves, a non-model system, has no complete reference genome, and is comprised of a long evolutionary history. Using next generation sequencing (NGS) and cloud computing technologies we are working to determine the genetic cause for species divergence in Aves.
Although the words in verbal human communication can be transcribed and parsed for meaning, there is an entire dimension of verbal communicative speech that has remained essentially unexplored with respect to computational “understanding”: prosody. This research project investigates how specific pitch contours are both produced by human speakers and also how each pitch contour is perceived in verbal communication.
Selection in increasingly complex 3D graphics scenes is addressed by picking algorithms which must be fast to minimize delays. The most common picking algorithms are ray casting and color picking. We computationally analyze and compare these two algorithms based on accuracy and speed. Our results indicate that the color picking algorithm consistently outperforms the ray picking algorithm in terms of selection time, and it also appears to be more accurate.
We compare the empirical performance of approximation algorithms for the k-center and online bottleneck matching problems to their theoretical guarantees. In online bottleneck matching, n server-vertices lie in a metric space and n request-vertices arrive over time and must be matched upon arrival, minimizing the maximum distance of any assignment. We show that the performance of approximation algorithms in practical settings can vary greatly from the theoretical guarantees.