Investigations of deep neural network based learning models for opinion mining

Abstract

In general, opinion mining referred to as sentiment mining which is a potential domain of research that tries to determine the opinion lying underneath a text represented in natural language. This opinion mining is an integrated discipline of computational linguistics and information retrieval. It aids in estimating and extracting subjective information from the related source materials. It concentrates on the process of mining opinions from the collection of source document that portrays significant information about an object. The subjective opinion mining is achieved by extracting attributes of the objects from the social network user comments in order to identify whether the extracted opinions are positive, neutral and negative. Moreover, the degree of interest exhibited by the individual users towards online opinions related to a social event, political happening, products and services plays a vital tool for perceiving the mentality of the people towards the respective issue. These opinions are determined to introduce comprehensive impact for the purpose of opinion sharping, such that the government, political parties, financial institutions, public and private companies and organization can monitor them from effectively perceiving the mind of the common people. However, the process of web monitoring is a complex activity as there exists a diversified number of sources that contain voluminous data. At this juncture, Opinion mining tools are considered as a boon for potentially processing a huge amount of online reviews for the objective of determining the inherent opinions. Furthermore, the machine learning approaches that extracts and explores the opinions on the domain feature level for reliable classification of overall opinion on a multi-scale is essential. The deep neural network architectures of machine learning approaches are identified to be potential in extracting and investigating huge amount of opinions on the level of the domain attributes with the precise classification ofsentiments into differe