AMRITA_CEN-NLP@SAIL2015: Sentiment Analysis in Indian Language Using Regularized Least Square Approach with Randomized Feature Learning

Aspect Sentiment Identification using Random Fourier Features

“…The resultant unique vocabulary list was written as a row vector where each word in a text is mapped to the word in the row vector, if present it is assigned a 1 to its corresponding position else 0. It denotes a one-hot vector representation [24]. For a review text which has multiple occurrences of similar words, its term frequency (number of times a particular word has occurred) was calculated.…”

“…Since the matrix that we prepared was sparse in nature, a random Fourier mapping was applied to find a lowdimensional dense representation [23][24]. Therefore, a concise feature list was created wherein a linear classification could be applied [23][24].…”

“…Since the matrix that we prepared was sparse in nature, a random Fourier mapping was applied to find a lowdimensional dense representation [23][24]. Therefore, a concise feature list was created wherein a linear classification could be applied [23][24]. The main aim was to outline the input which contains the feature set into a concise feature space using a randomized function map as shown in Equation 1, where z is as an explicit mapping function, which maps the data to a finite dimensional space…”

“…When the data dimension is large, the implicit mapping (via random mapping technique) has a problem in kernel computation as the computational cost will be heavy. However, the kernel computation will not be heavy when the data is mapped using explicit mapping method [24]. A nonlinear function map z (.)…”

“…In [23], the investigators explain the RKS algorithm using radial basis function (RBF) as shown in Equation 4 The Fourier transform, if scaled correctly, , will become a probability distribution as shown in Equation (6), where Ω is described as a random variable which follows a Gaussian distribution [23][24].…”