Search

This research attempts to rate document classification algorithms used in two prominent problem domains - Automated Grading of Essays and Sentiment Analysis. These algorithms are variants of support vector machine based algorithms. The motivation was to find an algorithm that could be used by a web-based application. Ideally, this would provide real-time results whilst maximizing accuracy. Additionally, a new technique called Inclusive-Class is proposed to model the problem space for the Automated Grading of Essays.The algorithms that were evaluated included Support Vector Regression (LibSVR), Kernel Ridge Regression (KRR), Generalized Minimum Norm Problem (GMNP), Multi Kernel Learning (MKL) and K-Nearest Neighbors (KNN).For the automated grading of essays Multi-Class experiment, KRR algorithm performed the best with an accuracy of 57.15% and execution time of 0.03s. Using the Inclusive-Class technique, the GMNP algorithm performed the best with an accuracy of 56.59% and execution time of 0.10s.For the sentiment analysis Multi-Class experiment, LIBSVR algorithm performed the best with an accuracy of 78.78% and execution time of 9.9s.

Line width parameters for a number of spectral lines in the pure rotational spectrum of formaldehyde (CH20) and formic acid (HCOOH) have been measured using a sourcemodulated microwave spectrograph. All transitions studied in this investigation were of the type ΔJ=O (i.e. Q-branch transitions), with ΔK-1=0 and ΔK+1 =+l. The center frequencies of the measured lines varied from 8662.0 MHz to 48612.70 MHz. The experimentally determined collision diameters for self broadening interactions involving HCOOH and CH2 Q molecules were found to be 2 - 27 per cent less than those calculated by the Murphy-Boggs theory of collision broadening. Much better agreement between a simplified broadening scheme for symmetric top molecules and the observed foreign-gas collision diameters is obtained by using Birnbaum's theory.