We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
A MULTICLASS SENTIMENT CLASSIFICATION USING SKIP-GRAM EMBEDDING WITH SUPPORT VECTOR MACHINE-STOCHASTIC GRADIENT DESCENT (SVM-SGD).
- Authors
K.-K. A, Abdullah; S. M, Sodimu; T. J., Odule; O. O, Solanke
- Abstract
N-gram feature is used to represent documents in natural language processing but leads to curse of dimensionality. Sentiment classification based on word embedding to represent document with an incremental method reduce dimensionality with dense vector representation of words. This works focus on the skip-gram model with negative sampling for vector representation. The text class is predicted by maximising the probabilities of embedding vectors of words under the class. However, a Support Vector Machines with an extension of Stochastic Gradient Descent (SVM-SGD) is employed for effective classification of datasets into multiclass. This is achieved by maximising the margin between hyperplane of every two class pair using online learning as well as controlling the constraints and minimise the regularisation error. This reduces the effect of imbalanced classes in training the classifier parameters. Hence, solve a quadratic programming problem while running SGD for chosen iterations and returning the average point in the number of classes in terms of accuracy and computational cost.
- Subjects
NATURAL language processing; SUPPORT vector machines; QUADRATIC programming; EMBEDDINGS (Mathematics); HYPERPLANES
- Publication
Annals. Computer Science Series, 2019, Vol 17, Issue 2, p234
- ISSN
1583-7165
- Publication type
Article