Prof. K. Venkataraghavan, Ph.D, Dean-Administration, IIM Kashipur emphasises that the mechanism of searching for doctors, like any other sector/field, relies upon public-generated content in online websites, search engines, and apps in current times. However, understanding and predicting these ratings is complex due to the subjective nature of patient opinions and various factors that influence their assessments
What do you do when you have health complications that a regular family doctor cannot fully understand or remedy? The answer is to search for a “good” doctor who can be a specialist or a general physician who can help diagnose and cure the complication. Similarly, if a physician proposes some surgery, we may want to look for a surgeon and maybe a doctor near our residence for post-recovery precautions. The mechanism of searching for doctors, like any other sector/field, relies upon public-generated content in online websites, search engines, and apps in current times.
Improvements in communication technologies have increased the accessibility of the internet and online platforms. Browsing, seeking, and sharing information online have become almost second nature. Online platforms are popular for people seeking travel, education, entertainment, or health information. People search for information about travel, lifestyle, education, and careers on websites or social media. In recent times, online platforms have become popular among patients. Patients share their experiences with doctors and hospitals on such platforms. Scientific papers report that 7 per cent of all searches on the internet are related to health, and 3/4th of people consider online reviews in their quest to find a good doctor.
The experiences are shared as reviews and are usually accompanied by a rating. These ratings can be helpful for other patients seeking medical assistance. However, understanding and predicting these ratings is complex due to the subjective nature of patient opinions and various factors that influence their assessments. For example, a patient may have rated a hospital highly because it was affordable. Likewise, another patient may have rated the same hospital highly because it had less waiting time. Both these ratings are useful to patients who prioritise affordability and waiting time. However, such ratings and accompanying reviews may not be useful to patients who accord high priority to the experience of the doctors or the treatment quality. In short, it is necessary to understand the factors that drive the ratings.
Several analytical methods under the umbrella and natural language processing techniques can uncover hidden insights from textual reviews. However, these methods vary widely in their ability to discover hidden insights. One of the easiest ways, the “Bag-of-words” technique, generates insights from the most often repeating words in the sentences. While this technique is simple, it reduces a complex phenomenon to a rather simple one. Communication is a complex task in which people express their thoughts cogently over several words and sentences. Continuity of thought manifests itself as hierarchically related words that are in turn, conveyed across multiple sentences. But the disadvantage of the “Bag-of-words” technique is that it treats each word independently, disregarding the relationships and nuances conveyed by the entire text. The implication is that the recommendations made by such techniques would not be relevant to a patient, and consequently, the patient may lose faith in such recommendations.
Deep learning, a branch of artificial intelligence, offers superior techniques and gaining popularity over conventional methods for predictive tasks. Deep-learning techniques have found acceptance in various contexts, such as agriculture, tourism, e-commerce, manufacturing, finance, insurance, banking, stock markets, driverless cars, and robotics. The deep-learning technique was employed in the research.
The results of their experiments demonstrate the effectiveness of the proposed deep learning model in predicting online doctor ratings. The model achieves a high accuracy in its predictions, which suggests that it can be a valuable tool for patients, healthcare providers, and other stakeholders to gain insights into doctor performance and patient satisfaction. Using a class of deep learning networks called Recurrent Neural Network coupled with an attention mechanism for this purpose. A specific variant of recurrent neural networks has shown a remarkable ability to make predictions from sequential data such as textual reviews. The attention mechanism improves upon the recurrent neural network by “paying more attention” to certain sections of the texts, which it believes to impact the prediction significantly. In the research, this approach was tested with more than 35000 reviews and ratings gathered from a popular rating website and found it predicted the rating better compared to traditional approaches.
The research findings have several implications. From the perspective of a rating website, it can move from merely offering reviews and ratings to providing recommendations based on a more accurate deep-learning model. From a patient’s point of view, such recommendations would be deemed more relevant.
From a larger perspective, the Digital India vision is “to transform India into a digitally empowered society and knowledge economy is also in line with increasing behavior of the public in general”. The future of the government and private health sector presents tremendous opportunities to digitise the processes and focus on e-health records. The Ministry of Health and Family Welfare (MoHFW) defines and promotes these e-health records. A truly inclusive society is when health services, including doctor consultations, are rated and reviewed by the large public in general. This behavior of reviewing and rating doctors exists worldwide and is enabled via websites, apps, and online forums. Just as e-health records are necessary for any doctor to know their patient comprehensively with history and reference, in the same way, reviews and ratings by a patient for a doctor visited in past and ongoing treatments may be used to recommend personalised choices of doctors which can be consulted in case of any incumbent requirements. This is particularly helpful when multiple options result in a dilemma. Deep learning and artificial intelligence make these choices personalised and rank them based on experiences shared by several people through reviews and ratings.