As the size of the sentiment data increases, it is difficult to predict the word polarity of an aspect term in a sentence in real-time applications. Most of the conventional models have a dependency of one or more aspect terms for classification problem. However, these models require high computational time for word embedding process and aspect level sentiment classification. Also, these models have high true negative rate and misclassification rate on large aspect databases. In this work, a hybrid multiple word embedding methods and classification approach are implemented in the CNN framework on the large databases. Experimental results show that the proposed aspect level word embedding based classification approach has better efficiency in terms of true positive rate , runtime and accuracy than the conventional aspect level classification approaches.

Aspect sentiment, filtering, classification , polarity.