96 Context beyond simple rules: enhancing the natural language processing pipeline for sentiment and theme prediction in family and friends test feedback

Le résumé fourni par la source

Background In this work, we present our outcomes on operationalising a Natural Language Processing (NLP) pipeline which helps the Patient Experience team by reducing their efforts towards a time-consuming manual approach. Our pipeline automates and provides an accurate summary of NHS Friends and Family Test (FFT) feedback closer to a human prediction by using the contextual information within the feedback. Our approach also differs from the current pipeline, which uses a traditional AI technique known as Bag-of-Words (BoW) approach and fails to capture the nuances of patient sentiment and the specific themes outlined in the NHS Patient Experience Framework. Addressing these issues, our implemented NLP pipeline is used, and we have trained GOSH-specific Small Language Models (SLM) using our in-house data validated by the Patient Experience team.Methods Our implemented pipeline is used for training AI models, specifically SLMs known for its ability to understand complex and context-rich language for capturing the contextual meaning of the feedback. Further, these models are designed to categorise patient feedback according to two target variables: Sentiment (Positive, Neutral, Negative) and Theme (10 categories based on the NHS Patient Experience Framework, including Respect for Patient-Centred values, Physical Comfort, amongst others). To build these models, a validated dataset of 11,360 patient feedback was collected and anonymised.Results Our GOSH-specific AI model performs 92% accurately in comparison with the currently used AI model that performs only 40% accurately using a traditional BoW approach, and thus can effectively process the nuanced language found in patient feedback.Conclusion This study underscores the advantages of advanced NLP techniques for automating tasks in healthcare settings, particularly for analysing complex patient feedback. Our results also suggests that this approach can lead to more actionable insights, supporting better patient care and more reactive service improvements across the NHS.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.