Suggestions

This page describes the Suggestions capability.

Suggestions

The "Suggestions" Section is located on the Right Sidebar.

Every time a Topic appears in the Advisor Workspace, Crystal proposes a set of dynamic suggestions, which include suggestions of Topics related to the one you just asked. These suggestions change based on your questions.

Suggestions serve as a guide for you to continue exploring your data, indicating which questions to ask next to uncover fresh insights.

If you are new on Crystal, Suggestions can also inspire you to formulate more effective questions for the Advisor, since the Topics inside the cards are expressed in the form of questions.

Explore Suggestions

Exploring Suggestions is easy: simply click on the suggestion of your interest and you will immediately find it open and ready to explore in the Advisor.

Each time you ask a question, the suggestions will be refreshed based on the last Topic you asked about.

Please Note

Whenever possible, the algorithm will suggest up to six related Topics. The first four will generally be the more similar Topics, while the last two might be the least similar.

By showing a mix of similar and different Topics, Crystal offers a balance between informed guidance and out-of-the-box tips that may spark curiosity for alternative data exploration paths.

How Does It Work?

The Suggestions capability uses an algorithm called a Recommender System, which is powered by the Business Knowledge Graph, a data model that provides a personalized representation of the data behind your Crystal Project.

Deep Dive on the Business Knowledge Graph

The Business Knowledge Graph (BKG) serves as the foundational knowledge base that Crystal leverages to answer your questions.

It is similar to a network of interconnected information and metadata, built upon the private data linked to Crystal.

For each Crystal Project, the Business Knowledge Graph is unique and isolated, tasked with a very clear objective: to extract all the information AI models need from the metadata, to transform information into conversation.

The construction of the BKG begins with Crystal’s configuration, where data sources are selected and Business Entities are defined, along with their semantic properties.

The BKG interprets both the content of the data connected to Crystal and the decisions made during configuration to establish semantic links between elements of the corporate knowledge base.

Through this process, the BKG learns the corporate lexicon, its taxonomy, and specific expressions, achieving a deep understanding of the business.

Additionally, the BKG adapts and evolves based on the questions asked to Crystal, identifying new correlations among data and learning from users, ensuring that the results produced are always accurate, certified, and contextual.

This is one of the core technologies underlying Crystal.

Based on your customized Business Knowledge Graph, the Recommender System analyzes the dataset and selects the Topics that are more relevant to the Topic asked based on the similarity of their content.

In particular, the algorithm compares Topics based on their Entities, Objectives, or Data Sources, like this:

  1. Entities Match: first, it searches for Topics containing the same Entities as those in the Advisor’s Topic.

  2. Objective Alignment: in absence of similar Entities, it looks for Topics with matching Objectives.

  3. Source Correlation: if there aren't any matching Objectives, it extends the search to Topics from the same Table or Data Source.

Whenever possible, the algorithm will suggest up to six related Topics. The first four will be the most similar ones, while thee last two Topics will be the least similar.

Rules & Limitations

The rules and limitations to consider when using the Suggestion capability are as follows:

  • Suggestions Based on Configured Topics: Suggestions only consider Configured Topics, i.e. the ones that have been manually created by the Admin, whereas auto-generated Topics are not included.

  • Dependency on Focused Topic: The activation of the algorithm requires a “Topic in focus”, i.e. a Topic open in Advisor.

  • A Minimum Number of Topics: There must be at least seven Topics configured and published in Console for the algorithm to suggest six correlated Topics.

  • Potential for Random Suggestions: When the configuration isn't optimal, the algorithm defaults to random suggestions.

Keep exploring your data!


Last updated

Was this helpful?