Petersen C, Halter R, Kotz D, et al. (2020). Using Natural Language Processing and Sentiment Analysis to Augment Traditional User-Centered Design: Development and Usability Study. JMIR Mhealth Uhealth. 8(8): e16862. doi: 10.2196/16862
Researchers developed a mobile app for a Bluetooth-connected resistance exercise band to help seniors mitigate sarcopenia (age-related loss of muscle mass) and applied natural language processing (NLP) and sentiment analysis to end-user interview data. Sixteen primary care patients (mean age 76) and 6 general internists from a large New Hampshire hospital participated in the user-centered design process. App development progressed in 3 stages: round 1 (baseline), round 2 (5 weeks), round 3 (10 weeks). The research team interviewed patients and clinicians individually about app perceptions at each round, reviewed interview notes and, where feasible, incorporated participant suggestions into app design. Round 1 assessed patient technology use and exercise habits. Round 2 introduced the app prototype. At round 3, the app featured a menu of embedded demo videos of 4 resistance band-based exercises (Bicep Curl, Seated Row, Arm-lifts, Triceps). When users exercise with the Bluetooth-connected resistance band, the app records user movement data, allowing for real-time monitoring and feedback on exercise compliance and treatment progress. Researchers used the Bing sentiment lexicon to analyze sentiment in interview transcripts. The lexicon defines each word as “positive” (+1 value) or “negative” (-1 value). Total average sentiment towards the app improved with each round of development among patients and clinicians, from −0.17 (round 1) to 1.57 (round 2) to 6.00 (round 3). The difference between patient and clinician sentiment was not significant. Researchers also applied NLP-based topic modeling to interview transcripts. NLP revealed 4 major themes in patient transcripts: exercising with technology, difficulty using the app, improving fitness, and help completing a workout. Four different main topics emerged among clinicians: interaction with the app, workouts over time, feature enhancement, and language of instructions. Future research could incorporate insights from NLP and sentiment analysis into app design and test the real-world efficacy of the app and resistance band.