Introduction

One AI Analytics allow transformation of a large pool of language inputs into meaningful hierarchical groups. You can then use the clustered data structure for analytics, or you can automate actions based on querying new inputs against the clustered data.

What is it good for

  • Cluster customer support tickets, analyze them, and respond automatically.
  • Cluster subject lines of articles, analyze them, and post to different twitter tags.
  • Cluster sales conversation highlights, analyze them, rank employees accordingly.

Analytics can be helpful in any case where you have large amounts of language data which need to be sorted, or grouped, so you can execute different actions on each group, without a prior knowledge of the exact groups you want to split the language data into.

Basic Example

As an example, let’s say you have plenty of support tickets, and you want to extract the most significant cases that repeat themselves. This can easily be done using Analytics; You first create a collection named “customer feedback”, and add all the tickets to the collection as items. Our clustering engine will cluster the collection for you, and you will be able to identify different groups of items.

Let’s say we have the following 5 sentences (=items):

  • Your service was great
  • You helped me a lot
  • Can I have a discount?
  • Your service is too expensive
  • Why can’t your service be cheaper?

We create a collection named “customer feedback” and add those 5 items to it. The clustering engine will understand that there are two clusters here: one for praising the service, and one for complaining about the service price.

Your service was greatYour service is too expensive
You helped me a lot.Can I have a discount?
Why can’t your service be cheaper?

The engine is smart enough to pick the clusters itself. It can pick as little as two clusters or as many as n clusters, where n is the number of items, which is limitless. As the “title” of each cluster one of the items in the cluster is randomly selected and written in bold.

Phrases

There are two layers of groupings while clustering; The more granular of them is grouping items into phrases. For example, “Your service is very expensive” and “Your service is so expensive”, are very strongly related and can be treated the same, so they are grouped together under the phrase “Your service is very expensive”. These groupings of phrases are then clustered.

Let’s enhance our example to demonstrate phrases. We’ll add two sentences to the previous list of items (new items underlined):

  • Your service was great
  • Your service was awesome
  • You helped me a lot
  • Can I have a discount?
  • Your service is too expensive
  • Your service is very expensive
  • Why can’t your service be cheaper?

Now when we cluster, the engine will first find phrases:

PhraseItem
Your service was greatYour service was great
Your service was awesome
You helped me a lotYou helped me a lot
Can I have a discount?Can I have a discount?
Your service is too expensiveYour service is too expensive
Your service is very expensive
Why can’t your service be cheaper?Why can’t your service be cheaper?

The original list of 7 items resulted in 5 phrases. Now on that list of phrases, clustering will be done.

ClusterPhraseItem
Your service was greatYour service was greatYour service was great
Your service was awesome
You helped me a lotYou helped me a lot
Your service is too expensiveCan I have a discount?Can I have a discount?
Your service is too expensiveYour service is too expensive
Your service is very expensive
Why can’t your service be cheaper?Why can’t your service be cheaper?

In this way, from an original list of 7 items, we ended up with 5 phrases and 2 clusters


Did this page help you?