Keyword Mask Feature

The Keyword Mask function detects sentences of keyword stuffing in your uploaded documents and subsequently masks them. The keyword stuffing is a listed keyword that is deliberately designed to gain an unfair ranking advantage in document searches, making responses of LLMs less trustworthy. This function detects keyword stuffing injected by adversaries. Besides, it detects unintentional keyword lists, such as the pages of "Table of Contents" and "Indexes". These result in making RAG-integrated LLM applications more secured and accurate.

How to use

Select "Keyword Mask" when uploading documents

You can choose applying this function to documents when registering retrievers with the document management function of this application.

Specifically, in the document upload phase, select "Create retriever" and "Create new retriever", and then check the "Keyword Mask" button. Afterward, by clicking on the "Upload File" button, a retriever composed of the masked documents will be registered.

（You cannot select "Keyword Mask" when "Create retriever" and "Create new retriever" are not clicked on.)

To confirm which retrievers Keyword Mask is applied to, see the registered retriever list. The mask icon indicates that Keyword Mask is applied to the corresponding retriever.

Select the registered retriever when creating a chat

When you create a chat, you must choose the registered retriever where the Keyword Mask function is activated. Then, you can have a chat based on the retriever.

Let's try comparing LLM responses using the retrievers with or without Keyword Mask.