Disambiguation Methods

Overview

To incorporate a disambiguation method there are two core steps:

1. Leverage an LLM to identify all the "topics" within a particular data corpus and its relevant entities.
2. Build watsonx assistant actions for each identified "topic" in the data corpus and create a step to provide all the relevant entities as options to the user when a topic is mentioned by the end user.
3. Combine the user's selected option with their document search request and present the clarified answer to them.

How to Implement

1. Leverage a python script and the watsonx sdk to determine all the topics within a data corpus and the relevant entities.

   Example Topics/Entities:

   Topic	Entities
   Browser	Chrome, Safari, Firefox
   Devices	iPhone, Android
   Printer	model A, model B

2. Build watsonx Assistant actions for each topic and build a step to surface the entities as options to the user

   

3. Build another step within the watsonx Assistant action to store the selected option as a variable, for example as "selected_option"

   

4. Combine the user's "selected option" with the document search request

   For example, if the variable "user_query" holds the user's original question, such as "How do I clear my browser cache?", you'll need to append the selected option to the "user_query" variable before sending the clarified query to the "Document Search" method.

   Example)

   user_query.append(" on").append(selected_option)

   This will result in the the final query to the "Document Search" method to look like: "How do I clear my browser cache on Safari" which will then return the most appropriate answer.