Having a conversation with Crystal means exploring data in depth.

You can break down the information by product, compare them by year, or segment the data by geographical area.

This is where the filters come in!

What's a filter?

Filters help you narrow down the field from more general questions.

For example:

From a question like "What is the sales trend?" to:

  • "What is the sales trend in the UK?" refined by geographical area

  • "What is the sales trend for 2022?” refined by period

  • "What is the sales trend for sneakers?" refined by product segment

How to use a filter

In Crystal, filters can be:

  • Numeric (currencies, item codes, or sizes), either positive or negative;

  • Non-numeric (countries, towns, areas, item names, customer names).

Crystal recognizes words and sentences even if spelled wrong or incomplete and recognizes time filters, too.

For example, no matter if you spell:

  • Month & Day as Numbers (“Total of Sales for 2022 10”, “Total of Sales for 2022 10 7”).

  • Month as a Word ("Total of Sales for 2022 October” or “Total of Sales for October 2022”).

  • Month Abbreviations - (“Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec”).

  • Numbers with comma or dot (1,000 or 1.000)

Crystal will understand.

You can use English translations of time aggregations (day, week, month, quarter, semester, year) and months (January, February, March, etc.) in your sentence even when using the product in another language. Crystal will understand your intent and apply the right filter to the Topics as if they were not in English.

Quick tips

When setting filters, make sure to:

  • Use short and simple values for filter ‘Aliases’;

  • Be specific when adding filters to a question;

  • Remove redundant punctuation and spaces;

  • Do not make values case-sensitive.

As you get used to talking to Crystal, you will be able to specify a single Dynamic Filter and multiple Filter Values at the same time in your conversation in such a way that you will not have to type the Dynamic Filter Name for each Filter Value you wish to apply to the Topic, resulting in a smoother conversation with Crystal!

How to exclude a filter

With Crystal, you can exclude a specific value using filters when asking about a Topic.

For example, you can ask for sales distribution excluding a specific country, like "Show me the distribution of sales excluding Spain."

Another example.

You are monitoring energy production at a plant, and you ask Crystal: "Show me the comparison between the actual budget and budget without energy losses".

To help Crystal understand, use: "other than, without, excluded, except, without considering" before the value you want to exclude.

How to filter in the Topic Card

You can edit Filters directly in the Topic Card as well.

Simply click the ‘Filter button’ on the Topic card and select which filter you want to edit or add (including the Time Filter).

Also, there is a ‘see more' option when at least one filter is applied. Click it to get an overview of all your filters, including the time filter.

This way, you can edit or add any filter in seconds.

Filter Classification

When you request to apply a filter in a conversation, Crystal can:

  • Recognize the filter by exact match, for example, when the question is: "Show me the total sales for France," and the filter is "France."

  • Recognize the filter by partial match, for example, when a user asks: "Show me the total sales for "Italy", but the set filter is "North-Italy."

Let's see how it works.

Exact Match

The text included in the question is the same as in the data source (unless a change is already included in the preprocessing, such as the presence of punctuation marks or special characters).

This is the easiest case for Crystal to recognize.

If the filter is "Italy" and you write "Italy" in the question, the system will recognize it immediately.

Partial Match and Disambiguation

The text in the question is not exactly the same as in the data source, but some parts are missing.

This is considered a ‘partial match’: the system needs help from the machine-learning model NER (Named-entity recognition).

NER is a search engine that can analyze all the words in a sentence and provide a ranking of the most relevant terms, with the highest-scoring words as possible filter values.

Let’s make an example.

By accidentally typing "nrd Italy" instead of "North-Italy", the search engine will find possible filter values and assign a ranking:

  • North-Italy;

  • Italy;

  • South-Italy;

  • Italy.

Since "North Italy" has a higher score, it will be the closest to the value typed and so recognized as the right filter value to select and no need for more questions.

On the other hand, if you are looking for the total sales for the Australian market and type: "Total Sales for Austr" with Australia and Austria in your data source, the NER will find two values with equivalent scores: Australia and Austria.

Since Crystal cannot be sure which of the two you need, the Disambiguate functionality will come into play, asking you to choose which value you want.

Quick tips

The Natural Language Process models consider meaning and sentence structure.

So, it is helpful to introduce the filter value with words such as "for," "in," and so on.

For example, writing "Show me total sales for France" is better than "Total sales France."

Aliases of entities and filter

We can associate Aliases to be used within the Conversation regarding both entities and filters.

Read this article to learn more.

Last updated