Abstract
Millions of people use Internet for developing new skills, booking online tickets, socialising, etc. Out of the sundry activities, giving online reviews by customers has become very customary these days and the fastest medium to make one’s voice heard. With the advent of analytics, more specifically, text mining, the online reviews of the customers have made a huge difference in shaping the future strategies of the companies and have also helped them to study the customer responses of their rivals. In an effort to help hospitals analyse the patient’s reviews present online on various social media platforms, this paper analyses the 659 reviews of people across the nation, on one of the best medical college and hospital of India, All India Institute of Medical Sciences, New Delhi. An attempt is made in this article to develop fuzzy sentiment analysis model with integration of naïve base classifier, which helps to analyse reviews of different hospitals and can come up with their own social media competitive analysis strategy. The results reveal the value text mining can bring to the table for any hospital and the immense business value that it holds.
Introduction
The Internet, nowadays, finds its place in almost every urban household and has already started penetrating in rural households too. With easily accessible Internet, people are now, more active than before, on almost all social media platforms (Constantinides & Fountain, 2008). People socialise, entertain themselves, receive information and, hence, keep themselves engaged. Almost all companies have their social presence, and there are multiple platforms where the customers can give their reviews on the products/services that they use in their day-to-day lives. These reviews and all other data present online turn out to be the gold mine of textual information (Dai et al., 2011).
Organisations are trying to change their present by being more data-driven and the future by predicting what could happen in the years to come by using all the data available. By using social media, the companies cannot just see their customer’s reactions, but they also get the opportunity to gain business values by increasing customer loyalty and by customer retention. Companies can build their brand using social media platforms and increase awareness (Godes & Mayzlin, 2004). Also, they get a chance to reach out to their customers by responding to their issues online, and this way, they can create a community for their products/services.
Many hotel and pizza chains are already leveraging the social media platforms by connecting with their customers, responding to their feedbacks, addressing the complaints and issues of the customers (He et al., 2013). By analysing the responses of the customers on social media platforms, the companies can come up with different social media strategies, and if need be, they can build new products/services, keeping in mind the sentiments and needs of the customers. Companies can, thus, enhance their efficacy and build on customer satisfaction (Culnan et al., 2010).
Though many industries are now digging deep into understanding the usage of data analytics, still the healthcare sector, especially the hospitals, are far from realising the importance of utilising social media responses of patients. Hospitals are broadly not familiar with the competitive intelligence of social media and the procedures involved in mining the data from various social media platforms.
Healthcare had always been the talk of the town for every nation, may it be the private healthcare system or the government healthcare sector. The healthcare system of any nation is one of those major players, which decides upon the development of the nation. Many types of research are carried out pertaining to various developments in the field of medicine/healthcare, but a limited number of researchers have used data analytics to gauge the sentiment of patients and, hence, use it to comprehend and come up with a competitive strategy over the other healthcare players in the same domain.
Health is considered to be an important contributor to the economic development of our country and very essential for the internal stability of the nation too. A major part of the hospital industry depends upon medical tourism, and hence, having an online presence, a website, is very common these days when it comes to hospitals. People, nowadays, come along, make a community online and share their reviews, feelings and experiences with the people of the community. The hospital industry has also seen a drastic change, wherein, the people/patients share their experiences on these social media sites and even try to cater to the queries of different people. Hence, these data, available online, have been a very rich resource for competitive analysis by the hospitals; they can do predictive analysis and come up with competitive strategies for their organisations. Thus, this study is conducted for mining the textual information present on various social media sites for hospitals and analysing them using sentiment analysis. A huge amount of data is available online because of blogs, e-mails, online reviews, etc., that contain a lot of textual information. To extract valuable patterns and information from the huge amount of data is surely a challenging task. Thus, text mining came into the picture—a machine learning tool, which can be applied to any amount of data, and hence, the trends and patterns from the data can be retrieved. The data that are pulled are either semi-structured or unstructured, and text mining is applied on this set of data, and hence, further analysis is carried out and patterns are found.
Many researchers have successfully used the text-mining tool to analyse data in various domains. Tane et al. (2004) used text mining, and they grouped the e-learning resources and documents as per the similitude among different topics. Literature is also reported (Abdous & He, 2011), which uses the text mining technique to analyse the various online questions that were posted by many video streaming students, and hence, they found out patterns, trends and various technological issues from the data. Text mining has mighty capabilities, and hence, it is believed that fascinating insights can be brought into light with respect to human behaviour (Godble & Roy, 2008; Lucini et al., 2017).
In this particular study, one of the best government institutions—All India Institute of Medical Sciences (AIIMS), New Delhi—is studied, and fuzzy sentiment analysis is applied on 659 reviews of different patients all across India on various social media sites. Thus, fuzzy sentiment analysis is carried out on the unstructured text content, and this study tries to answer the sentiment behind the reviews of patients. Also, an attempt has been made to generalise this model so that it can be applied to reviews of any other hospital.
Methodology
The reviews are collected from different online platforms and filtered. The filtered data are stored in the corpus, and the term frequency matrix is developed. The study comprises two tasks—the first part involves finding the association of frequent words with all other words present in the reviews. This association between words is found with the help of the R language. The second part is to generate sentiment behind the reviews, using fuzzy sentiment analysis. The preliminaries used to generate the sentiment in reviews are as follows:
The traditional approach of sentiment analysis uses Equation (1) to generate sentiment in reviews, whereas fuzzy sentiment analysis determines membership of emotions in reviews using Equation (2). After calculating individual membership of emotions in each review, the model uses Equation (3) to generate sentiments hidden in the reviews.
Results and Discussion
Reviews Collected from Various Sources.
Details of Frequent Words.
The association of the word ‘Doctor’ with other frequent words was found out, wherein, the correlation cut-off chosen for association was 0.22. Thus, all the words associated with the word ‘Doctor’ with a correlation value are presented in Table 3.
Association of Words with Other Words.
The sentiments behind the reviews are generated using the traditional approach by R and fuzzy sentiment analysis. The sentiment is classified in three categories—Positive, Negative and Neutral. A comparison of sentiment for some reviews generated by traditional approach and fuzzy approach is presented in Table 4. In the third column of Table 4, the sentiments generated by the traditional approach are presented, and in columns 4 and 5, the percentage of the positive and negative sentiment behind the reviews are presented. Column 6 of Table 4 shows the final sentiment behind the review. If the percentage of positive sentiment is greater than 0.5, then it is classified as positive; if the percentage of negative sentiment is greater than 0.5, it is classified as negative; or if both are equal to 0.5, then it is classified as neutral. As per the observation, there is a huge difference in the sentiments of the same review, using the two methods.
Comparison of Sentiment by R and Fuzzy Sentiments.
Using the traditional sentiment analysis approach, in total, around 80 reviews come out to be of negative emotion, whereas, by using fuzzy sentiment analysis approach, the number of reviews that give negative emotion, comes out to be around 280.
Similarly, the number of reviews with positive emotion comes out to be around 415, using the traditional sentiment analysis approach, whereas, using the fuzzy approach, the positive emotions come out to be 220 in number. The number of reviews with neutral emotion does not make a major difference when analysed using both the methods.
In total, 10 emotions are taken for fuzzy sentiment analysis, and the membership values of each emotion is calculated and presented in Table 5. As per Table 5, after analysing the entire set of reviews, we come to know that the ‘Trust’ emotion has the highest membership value of 0.23, which means that most people have trust on the hospital, and this is the emotion that comes out very prominently after analysing the reviews. Similarly, the trust emotion is followed by positive emotion, with a value of 0.21, which means that largely, people think/talk about the hospital positively. Positive emotion is followed by fear with a value of 0.144, which again means that people do have fear in them when it comes to treatment in the hospital, but less as compared to trust and positive emotion, still, more than the rest of the emotions.
Emotions and Their Membership Values.
The emotions of all the 564 (did not include the reviews with no emotion) reviews are plotted in the graph and presented in Figure 1. The green colour signifies percentage of positive emotion in the reviews, whereas the blue colour signifies the percentage of negative emotion in the reviews. The x-axis reflects the review number, whereas the y-axis shows the percentage of emotion (negative and positive) of each review.
Performance Matrix of Naïve Bayes Classifier.
Out of 564 reviews (564 reviews remain after removing the reviews that have no emotion), 400 reviews are taken as the training data set, whereas 164 reviews are taken as the testing data set, by dividing the entire data set into 70:30. The confusion matrix is then developed, and it is found that the actual negative sentiments, which were previously 60, come out to be 54 by our model. The positive sentiments that previously were 104 in number come out to be 110 by our model.
The actual negative sentiments were 60 in number, whereas the model has correctly predicted 48 reviews as negative and hence the true negative value is 48. A total of 12 reviews, which were negative previously, are classified as positive by our model; thus, we can say that the false-positive value is 12.
Similarly, the actual positive sentiments were 104 in number, whereas the model predicted 98 reviews as positive; hence, true positive is 98, and 6 reviews, which were previously positive, are classified as negative by the model; thus, the false-negative value reported by the model is 6.
Hence, the accuracy of this model is 89.02%. The sensitivity and the specificity of the model is 94.23% and 80%, respectively. With this accuracy, the model can be used to predict the sentiments of any other hospital.
Conclusion
Sentiment analysis has proved itself to be one of the best methods to get an overview of the public opinion on different products or services. Many industries have been using the traditional approach of sentiment analysis and at the same time have been facing the loopholes, which the traditional method carries with itself. Hence, fuzzy sentiment analysis, as used in this research for the hospital industry, can be replicated by other industries, and we can easily do away with the gaps of the traditional method, and alongside that, we can get more accurate results too. The model proposed in this article performs with an accuracy of 89.02% and is capable to handle all the forms of uncertainty.
Thus, this model can be used across the hospital industry to get the sentiments of the patients, and as a result, the hospital can work to improve their services and work upon the pain points of the patients.
Footnotes
Acknowledgements
The authors are highly grateful to the Goa Institute of Management, Goa, for carrying out this work.
Authors’ Contributions
The corresponding author devised the project, the main conceptual ideas and proof outline. The co-author gathered the data from various sources. Both the authors have designed and performed the experiments, derived the models, analysed the data and interpreted the results.
Availability of Data and Material
Data will be provided on request.
Declaration of Conflicting Interests
Funding
The authors received no financial support for the research, authorship and/or publication of this article.