Using Similarity Search in Homeopathy Repertorization Software – Part 2

In my earlier article, I had explained how natural language text describing a symptom can be converted to a formal “rubric” by using a vector database. In today’s article, I will show how this can be extended to support multiple symptoms in the input text with automatic repertorisation to identify potential remedies.

Let us consider this text:

A female patient 30 years old visited my clinic. An irritable person. She gets  bursting headache before menses. There is also drawing pain in sacral region. 

It is clear that this text has 4 sentences and 3 of them are symptom descriptions. How can we repertorise this? See this figure:

Example Case – 1

I modified my earlier Python program to support multiple sentences and also to identify appropriate remedies based on the encountered rubrics (I am using Kent Repertory). The program’s output, as it appears in the “Remedies” area, clearly lists the identified rubrics and possible remedies.  

Chamomilla and Natrum Muriaticum are present in all three rubrics. Kali Carb and Nux Vomica are suggested based on their total weight. Of course, as is normally done, one has to go through the Materia Medica and select the best remedy after further study.

Here is another example:

This is a 60 year old male patient. He complains of a distressing cough for the past two days. He additionally describes it as a whooping cough. Cough triggered when lying down. 

Here is the program’s output:

Example Case – 2

Here also, the program has correctly identified the rubrics present in the natural language text and has given its repertorial analysis.

This is indeed promising. 

The two examples discussed here are deliberately simple and are meant to drive home the point that “semantic similarity” can be exploited when building such systems. Of course, no actual case (acute or chronic) is so trivial. The actual case text may run to multiple pages and additionally the homeopath might want to make use of multiple repertories (a rubric might not be present in a particular repertory). An intelligent software must also allow the homeopath to submit photos (for example, tongue picture) and include those in its analysis.

I hope my dream becomes reality soon, thus empowering homeopaths to cure complex cases without much trial-and-error.

Have a nice week!

Tags: homeopathy, Repertorization, Semantic Similarity, vector database