Multimodal Sentiment Analysis using Kernel Based Support Vector Machine

There has been a steady expansion in the number of social websites, electronic media outlets, and blogging services, all of which contribute a large number of user messages such as comments, reviews, and opinions from customers. The quantity of these signals, which are used to evaluate the quality of a wide variety of goods and services, is rather substantial. As a result, it is challenging to monitor and identify the feelings of customers. As a result, a method known as Sentiment Analysis (SA), also known as Opinion Mining (OM), was developed. This method gathers data from the internet by using Natural Language Processing (NLP) strategies and then analyses people's thoughts. It does this by extracting subjective information from the words, assessing opinions, and quantifying positive and negative evaluations. The end result is a conclusion about the specific entity, which may be an organisation, the public, or a product. The NLP first gathers the views, and then those opinions are sorted into positive, neutral, and negative categories. Both the marketing promotion and the customer's ability to choose the most appropriate product for their needs may be improved by doing an analysis. The purpose of this study is to evaluate the use of different machine learning and swarm intelligence optimization methods to perform sentiment analysis on data derived from movie reviews. This study helps many businesses increase their profits while simultaneously lowering the number of problems caused by a certain product. Within the scope of this investigation, three distinct models of sentiment analysis were investigated. An effective method for doing sentiment analysis known as a Hybrid Kernel based Support Vector Machine (HKSVM) has been presented. The preprocessing that is described here is meant to cut down on the data inconsistencies and get rid of the data that isn't needed. After that, the Expectation Maximization (EM) technique is used in order to identify an outlier. After that, the Multivariate Autoregressive Model (MAR) scheme is used for the purpose of extracting potentially useful characteristics, and the Modified Cat Swarm Optimization (MCSO) strategy is offered for the purpose of locating the greatest opinion score. In conclusion, the high score accumulated during the opinion review step is carried over to the HKELM classification phase. Within this KELM function, the Particle Swarm Optimization (PSO) algorithm is used for the purpose of improving the kernel parameters and classifying the opinions in an efficient manner. It was found that obtaining an accuracy rate of 96.82 percent, a sensitivity rate of 97.14 percent, a specificity rate of 91.24 percent, a precision rate of 96.23 percent, a recall rate of 90.24 percent, and an F-Measure rate of 89.78 percent yielded a high level of performance.