Mining product opinions with most frequent clusters of aspect terms

This paper addresses the problem of more accurately mining product aspect opinions from Twitter posts, in the presence of spam and noisy posts, by proposing an algorithm called Microblog Aspect Miner (MAM). MAM takes a three step approach of classifying the microblog posts into subjective and objective posts using opinion scores of words from SentiWordNet. MAM then represents frequent nouns of subjective posts as vectors in such a way that nouns semantically similar to the products have a similar vector value using the WordVec algorithm. K-Means clustering algorithm is used to obtain the cluster of aspects relevant to the product to separate the noisy aspects so that the most relevant aspects are ranked using proposed Aspect-Product Similarity Threshold based on cosine similarities. Experiments show that this improves accuracy of obtaining relevant aspects of products from microblog posts in comparison to such existing aspect based opinion mining (ABOM) systems as Twitter Aspect Classifier (TAC)