Text mining helps to convert big text data into a small amount of most relevant data for a particular problem, and also helps providing knowledge provenance and interpreting patterns. Text mining has to do with further analysing the relevant data to discover the actionable knowledge that can be directly useful for decision making or many other tasks. In the previous work, authors have used conditional entropy with maxima/minima boundaries in identifying aspects from given text in identifying aspects. In the present work, we concentrated on mining word associations using mutual information in measuring co-occurrence of words. The objective of our research is to propose a novel probabilistic latent aspect rating regression (LARR) model. In which, natural language processing is used to represent the text, discovering and analysing topics using expectation maximisation (EM) algorithm. In addition to that, we compared the performance of the proposed model with multiple ordinal logistic regression classifier (MLRC) and fuzzy the rating obtained from the model. The finding in our research is the use of mutual information to mine synatagmatic relations helps in identifying the aspect at document level at greater extend, intern helps to multi level sentiment polarity prediction. Copyright © 2019 Inderscience Enterprises Ltd.