Using AI Nodes

Graze has a whole bunch of Artificial Intelligence features built in so you can create amazing feeds with less work.

All of the following nodes will run an AI against the text of a post:

• Content Moderation - Determine if the content of a post relates to 5 different categories (ex: sexual, violence, etc) or if it's overall "ok"
• Topic Analysis - Determine if a post relates to 1 of 20 different top level post categories
• Language Analysis - Determine the language of a post
• Sentiment Analysis - Determine if a post is positive, negative or neutral
• Toxicity Analysis - Determine if a post is toxic
• Emotion Analysis - Determine which emotion a post conveys
• Financial Sentiment Analysis - Determine if a post has is talking about finance in a positive, negative or neutral tone
• Text Similarity - Determine if the text content of a post is similar to a bit of text you supply (more on this below)

Using Analysis/Moderation Nodes

Most of the nodes all work in the same way, so they're really easy to use!

Each of these nodes will us an "AI model" to determine something about the post. Its output will be a number from 0-1 communicating how likely the AI thinks the text is a match to the model. You can think of that number as "percent that this post is likely a match.

Take the following example, it states "Include the post if it's 30% likely that it relates to science and technology."

Or you can use the "Content moderation" node to not include things that are likely to be sexual. This post reads, "Include the post is if less that 80% likely it's sexual".

Getting the right values

This is somewhat of an art! While AI is advanced, it can still make mistakes and might categorize a post incorrectly. You should use the post debugging tools in our editor to see how these nodes work when tested against real content. That will help you settle on values that work for your feed.

Generally the following values are good starting points:

• 0.3 - Likely the post is a match
• 0.6 - Very likely the post is a match
• 0.9 - Absolutely is a post that matches (this is rare)

Using Text Similarity

The text similarity node is a powerful tool, but it can be a bit unwieldy!

In short it does the following:

Check if the post's text is similar to the provided text

With this there is a lot of room for creativity.

Structure

The first way you can use this node is by comparing "structure" of a post's text.

In the following example we provide "I took my dog to the park". In theory this will match posts that roughly match the structure of the input text.

It could match:

• I took my dog to the mall
• I walked my cat at the park

Since both of those roughly look like the text we provided, they both might get added to your feed

Content

The other main use case for the text similarity node is using it to match specific content. A limitation of nodes like "Word List" is that it is only looking for the exact words you provide. If a post is talking about related words (ex: you want posts about dogs but the post only talks about yorkies), then it wouldn't be added to your feed. With the "Text Similarity" node you can roughly do this!

If we change our above example to just the following we might achieve that! The AI we use probably knows that a yorkie is a dog and will guess that post is more likely to be about dogs, and will include the post in your feed.

This still won't be perfect. For example a post might say, "I got that dog in me". We probably don't want to include that in a a dog feed. You will need to add other logic to exclude things you might not want, but the "Text Similarity" node provides a good general tool to include related posts.