Cosine Similarity Metrics for Estimating Symptom Similarity in Homeopathy

Homeopathic prescribing is fundamentally based on the principle of similarity, where the physician compares a patient’s symptoms with remedy symptoms described in the materia medica. Traditionally, repertories are used for this comparison, which organize symptoms into rubrics and generate a ranked list of possible remedies. While repertories are indispensable, they are prone to error 1 their use depends heavily on the physician’s understanding of rubric structure and interpretation, and they are prone to subjectivity and structural limitations 2. These factors can affect consistency and accuracy in remedy selection. This paper introduces a novel method to quantify symptom similarity using cosine similarity 3, combined with the modern natural language processing model Sentence-BERT (SBERT) 4. In simple terms, this approach converts symptom descriptions into numerical representations – vectors - that capture their meaning rather than just individual words. Cosine similarity is then used to mathematically estimate how close a patient’s symptom description is to a remedy’s symptom description5. The resulting similarity values range from 0 (no similarity) to 1 (perfect similarity) and can be easily interpreted as percentages for clinical use. A higher cosine similarity score indicates greater resemblance. We developed a Python-based software tool that automates this process and allows direct comparison between a patient’s structured symptoms and remedy symptom profiles from the materia medica. The software outputs a ranked list of the most similar remedies without relying on manual repertorization or rubric selection. This approach offers a reproducible and objective supplement to classical homeopathic practice. It does not replace clinical judgment but supports individualized prescribing by reducing subjectivity, improving scalability, and enabling rapid comparison across large remedy databases.