Conversational search

Before you begin

Conversational search supports English, Spanish, Brazilian Portuguese, or French. To do so, you need to set up Elasticsearch or other search provider you use through a custom service to work for the language that your content is on. You can set up Elasticsearch to use dense vector retrieval, which can provide highly effective semantic search for languages that includes Spanish, Brazilian Portuguese, and French. For more information, see enablement material.

The prebuild PDF upload feature with Elasticsearch integration is only supported in English.

You must configure the search integration to enable the conversational search feature. For more information about configuring IBM Watson™ Discovery search integration, see Discovery search integration setup. For more information about configuring Elasticsearch integration, see Elasticsearch search integration setup.

Enable conversational search

• Go to the Search Integration window.
• Set the Conversational search toggle to On.
• In the Define the text for the citation title, type How do we know?. The Define the text for the citation title is enabled only when Conversational search toggle is switched to On.

The Citation title feature is not supported in web chat integration.

The citations are quotations from the documents which relate to the generated responses. The citation title in the **Conversational search** shows how the responses are generated from various sources and displays the sources of the searched responses. The default text in the citation title appears in between the conversational answer and the citations. For example, if you type `How do you know?` in the **Define the text for the citation title**, then `How do you know?` is displayed between the conversational answer and the citations. 

• In the Tendency to say I don’t know, select the tendency to use. Less often is selected as the default option.

• Click Save to apply your configuration.

  

Behavioral tuning of Conversational search

You can tune your conversational search behavior with the Tendency to say I don’t know option. By tuning the assistant's tendency to fall back to the “I don’t know” answer, you can help your assistant to reduce Large language model (LLM) hallucinations and provide higher fidelity answers for conversational search. The higher you set the tendency, the more cautious the assistant is about answering. When the assistant admits that it does not know something, the assistant uses the Fallback action.

In the Tendency to say I don’t know, you can select the tendency to use: - Rarely: The assistant never admits that it does not know something. - Less often The assistant sometimes admits that it does not know something. - More often The assistant usually admits that it does not know something. - Most often The assistant admits all the time that it does not know something.

Route your assistant to use conversational search

After you enable Conversational search on Search integration, you can enable Search routing to use conversational Search. For more information about search routing, see Configuring the search routing when no action matches.

Test conversational search

You can test the conversational search in actions preview, the preview page, or by using the preview link.

In this example, the user asks, Tell me about a custom extension. Search results are pulled from your knowledge base when the conversational search is Off. In this case, the answer is returned as a list of cards that are relevant to custom extensions.

When the conversational search is On, the same search results are pulled from your knowledge base. The results are passed to an IBM watsonx generative AI model. This model produces a conversational reply to the user's question, in the form of a text response about custom extensions.

The citation title feature is not supported in the web chat integration. So, the default text in the citation title is not displayed in the preceding image.