43 An Expert System Technique for Sentiment Analysis of Opinions 1 Sabreena Nawaz Anee Fatima Dept of Computer Science UET, Lahore 2 Rabia Aslam Khan Dept of IT Lahore Garrison University 3 Kashaf Adoja Dept of Software Engineering Lahore Garrison University Abstract: To help the users and the product owners it is quite necessary to extract aspects from the online reviews, their sentiment polarities, and associations between them. There is a great deal of work done in the field of sentiment analysis. Lexical and learning-based systems can be combined to separate the assessments from online opinions and reviews. In learning-based techniques, the Gaussian mixture model can be used for getting probabilistic results for polarities against aspects and naïve baize classifiers for the problem of spam comments which produced better and competitive results against previous techniques. Keywords – Opinions, mining, reviews, sentiments, web, social data Sabreena et al LGURJCSIT 2019 LGU Research Jounral for Computer Sciences & IT ISSN: 2521-0122 (Online) ISSN: 2519-7991 (Print) 1. INTRODUCTION Nowadays, an extensive number of sentiment related reports are placed on the different forums on the internet. Mostly individuals post item surveys, state their political perspectives, and share their emotions. The capacity to concentrate assumptions from such sources can give precious data about individuals' perspectives on different points. The feeling found inside the remarks, input or investigates give valuable pointers to a wide range of purpose. Sentiment is basically an opinion, thought, belief or attitude regarding any situation or occasion. To judge these ideas, scrutiny is required which is mainly describes as opinion mining. Ideas scrutiny is done to harvest and classify instinctive information in provided material by using analysis techniques on text, processing of natural language and rules of the language. Additionally, the objective of sentiment analysis is to induce the thinking of orator or writer towards specific subjects and to check the positivity and negativity of the reviewer related to that article. Generally, the attitude of the reviewer is based on assessment or judgment and state of mind in which the reviewer may be at the time of writing or reading. The main problem that is going to be considered is to distinguish between negative and positive feedbacks, filtering spam comments and avoiding the problem of ellipsis. The main approaches used earlier were the method-based on the lexicon approach and machine learning approach that is also called a statistical approach. The orientation of the file or article can be calculated by the lexicon-based approach by measuring the semantic orientation of terminologies and phrases in the file. To proceed further in this method manually or automatically dictionaries of words are generated. Mostly adjectives are used as pointers for the orientation of content that is usually text and this orientation is done semantically. The statistical approach that classifies text majorly comprise on building machine classifiers that are trained on distinct data set of text or phrases that may be speech labeled or no. Unigrams and bigrams are Vol. 3 Issue 3, July - September 2019 LGU Research Jounral for Computer Sciences & IT 3(3) LGURJCSIT