Overview

Knowledge Search enables the Moveworks AI Assistant to answer a user’s questions directly in chat, by understanding the questions and responding instantly with a summary from retrieved snippets, which can be viewed in the Citations page with links to articles in the knowledge base, or links to external articles taken from trusted knowledge sources. The content of these answers come from the knowledge bases that your team already manages in repositories like ServiceNow Knowledge Management, Microsoft SharePoint, Confluence and more.

How does Knowledge search work in Moveworks?

The most important thing to note is that the Moveworks AI Assistant leverages the same semantic and keyword search that is used in Classic, with a query rephrasing step at the start to expand and anchor the query in the current context and make it more complete and specified.

Search for knowledge (KBAs, files) or forms works in a three-step process:

1. Plugin selection: The relevant search plugin, i.e. knowledge search or forms search plugin must first be selected for serving the query by the reasoning engine. This is a prerequisite for any results to be returned.
2. Retrieval and ranking: Once the search plugin is called, the user’s query is passed to the plugin and it evaluates the query against the available resources using a combination of semantic and keyword search. The most relevant pieces of content are then ranked into a list based on their utility for the query and the top 8 resources are returned to the reasoning engine. Note that any content must have a relevance score higher than an predefined threshold, and If no resources cross that threshold, then no results are returned.
3. Summarization: Based on the resources provided by the search plugin, the reasoning engine creates a summarized response for the user taking into account their question, the available content and any clarifications or follow up questions. All the resources returned by the search plugin are not necessarily used for response generation; the Moveworks AI Assistant chooses only the ones that were required to produce the response, and provides citations that link to those reference resources.

If no results were found by calling the selected plugin, then the reasoning engine may try a different plugin, or may decide to offer the user a handoff option so that they can file a ticket or get help through other configured options for their request.

FAQ

1. Do I still need to format my articles for snippetization with Moveworks AI Assistant?

   Snippetization is still a part of how we ingest and break down articles in Moveworks AI Assistant. And while the Moveworks AI Assistant is able do summarization of articles, summarization is dependent on the number and quality of snippets. Additionally, unsnippetized articles may not fit in the maximum possible context length. So articles that are optimized for snippetization will function better than articles that are not.

2. How do ServiceNow Knowledge Blocks work within Moveworks AI Assistant?

   For the same reason as above, Knowledge Blocks help create properly formatted snippets which in turn create good summarizations within the Moveworks AI Assistant. Also, snippets are still viewed within the Citations page and need to be optimally rendered.

3. Does Moveworks support meta tags?

   Moveworks incorporate meta tags as one of the signals to our search algorithm, and use them as part of the relevance calculation. However, note that relevance is mainly influenced by the article title and body, as well as the user's query.

   The meta tags on a document are used to help increase the documents relevance based on matches with the keywords of the query. Tags that are about specific related entities are given more weight while generic / broad terms are not as useful. Meta tags that are also highly similar to a specific query are given a boost as well. The meta tags do not have to appear in the title or body to be considered important. For example, if

   your query is forgot my Outlook account passwordand the relevant document is How to reset your email password

   then

   1. meta tags such as Outlook or forgot account password would be helpful ✅
   2. meta tags such as email or reset would have less impact

In summary, the two scenarios in which tags have a high impact on relevance are when: (1) tags closely overlap with the query and (2) closely related tags to the entities being referenced