Abstract:
Every online product selling applications are having review systems for their customers to review the products that they have purchased. Customers' reviews about the product will be either negative or positive and some reviews will give the meaning explicitly and some reviews will have implicit meaning. Nowadays most of the people do purchasing through online as a result there are thousands of reviews for a single product. On the other hand, these reviews will be useful for other customers to decide whether to purchase the product or not by going through the reviews. Mining implicit features from the customer reviews is a fundamental requirement for extracting customers' opinions and summarizing. This research focuses on extracting implicit features from reviews for opinion mining using a word embedding model. It removes noisy words and learn the model parameters automatically and extract the implicit features from customer reviews. Most of the existing researches have focused on implicit feature extraction from Chinese web reviews and only few attempts are made to extract implicit features from English web reviews. Implicit feature extraction was done through supervised, semi-supervised and unsupervised learning approaches. This research focuses on supervised aspect extraction using deep learning. This research proposes a novel and yet simple CNN model employing two types of pre-trained embeddings for aspect extraction: general-purpose embeddings and domain-specific embeddings associated with a Word Embedding based Correlation (WEC) model by integrating advantages of both the translation model and word embedding to extract implicit features. WEC model can score their correlation score for each word in review and feature. Then the CNN is used to identify the feature where the input for CNN is similarity matrix generated using the correlation scores. CNN gives the matching score of the review feature pair as the output and the review’s corresponding feature will be identified from the feature set.
