Sentiment Mining from Online Patient Experience using Latent Dirichlet Allocation Method

Background/Objectives: The paper is focusing on the problem of mining sentiments in the health care text. The attempt here is to apply sentiment analysis technique to extract the feelings of patients with various emotion labels like happiness, sadness and surprise about the healthcare. Methods/Statistical Analysis: In this paper the connectivity between social emotions and affective terms are predicted from the patient experience automatically using a joint emotion-topic model by augmenting Latent Dirichlet Allocation (LDA) along with a layer for emotion modeling. The following six modules like Preprocessing, Topic Generation, Polarity Classification, Sentiment Classification, Sentiment Analysis and Aspect Ranking are identified in our system. The set of latent topics is generated from emotions initially. From each of the latent topic affective terms are generated. Finally K-means clustering is applied to detect the emotion. Using aspect ranking technique the weightage of the document is calculated. Findings: An intricate description about sentiments reflected in the reviews of patient experience is not provided by many of the sentiment prediction approaches. Experimental results proved that the meaningful latent topics for each emotion are successfully identified by the proposed model. The identified emotions are useful to categorize the document and assist the online users to select required healthcare based on their emotional preferences. Application/Improvements: The machine learning process is able to make a careful determination of patient opinion about the various administration aspects of a hospital based on the prediction accuracy that have been achieved. Various machine learning predictions are correlated with results of more conventional surveys. It will be interesting to generate more efficient algorithms based on topic models in several other opinion mining systems and for large-scale data sets.