ASPECT BASED OPINION PREDICTION USING DIVISIVE ANALYSIS FOR THE USER RECOMMENDATION SYSTEM

Social media is a vastly developing technology in the system environment on the internet and this knowledge assistance was useful for many organizations, people or company to make correct decisions about the products, things, and the recently released movie. Opinion mining is used to track the emotions of the public about a specific product and which is one kind of natural language processing or otherwise called as sentiment analysis. But, in case of large reviews about the product, the particular feedback prediction was the major drawbacks. To make a valuable decision about the manufactured product based on the proposed technique of Coherence-based Aspect Opinion Pairs Detection (CAOPD) framework. Initially, preprocessing the input dataset to remove the stop words and extract the relevant keywords (i.e noun, adverb, verb, an adjective based words). By using the Map Reduce (MR) methodology to perform parallel operations with reduce the size of the input data and speed up the system. Then, using the Divisive Analysis (DIANA) method based Nearest-Neighbor Clustering (NNC) algorithm to evaluate the distance and similarity between the extracted keywords and make clusters. This analysis is otherwise called a top-down approach and thus formed a set of active clusters and make the decision between the aspect and opinion of the customer reviews. To split individual reviews (i. e paragraph into the sentence) and applying Part of Speech (PoS) tagging method to extract the aspect and its opinions. Then, finding the coherence range between the aspect opinion pairs based on the Coherence-based Aspect Opinion calculation process. In this work, the fuel and engine recommendation also implemented for suggesting the best fuels and engines used in the machinery. Finally, estimate the relativity of the user review of the similar opinion and aspects word. Therefore, the proposed CAOPD method is compared with the various techniques such as CFACTS-R, FIFS, K-means (TF), K-means (PMI), DF-LDA, L-EM, and the PSM in terms of entropy, purity, precision, recall, accuracy, efficiency. Therefore, the proposed CAOPD system achieves greater performance than the other techniques.