Understanding the Power of Words
Illuminate the true potential of words in healthcare with Talking Medicines, the global leader in patient intelligence. Through our advanced models and data science methodologies, we shed light on the vital importance of patient and healthcare professional voices. Experience the gold standard in understanding patient insights and empowering healthcare decision-making. Through analyzing social media, we capture the voice of the patient, painting a comprehensive picture of patients and their ecosystems. Join us in improving health outcomes by harnessing the true power of patient and healthcare professional voices.
Background
Patient-based analysis of social media is a growing research field with the aim of delivering precision medicine, but it requires accurate classification of posts relating to patients’ experiences. We motivate the need for this type of classification as a pre-processing step for further analysis of social media data in the context of related work in this area. In this paper we present experiments for a three-way document classification by patient voice, professional voice or other. We present results for our proprietary model classifiers on English data from two difference data sources and two domains (cardiovascular and skin diseases).
Conclusion
The main conclusion resulting from this work is that using more data for training a classifier does not necessarily result in best possible performance. In the context of classifying social media posts by patient and professional voice, we showed that it is best to train separate models per data source instead of a model using the combined training data from both sources. We also found that it is preferable to train separate models per domain (cardiovascular and skin) while showing that the difference to the combined model is only minor (0.01 accuracy). Our highest overall F1-score (0.95) obtained for classifying posts as patient voice is a very good starting point for further analysis of social media data reflecting the experience of patients.
Acknowledgements
We would like to thank the Talking Medicines Limited annotators who worked extremely hard to create the data needed for model training and validation.
To read the full article follow the link below:
BMC Medical Informatics and Decision Making
Authors: Beatrice Alex, Donald Whyte, Daniel Duma, Roma English Owen, and Elizabeth A.L. Fairley.
