0

We build a rails 4 site and use ES for our search travel/accommodation engine. We created a separate ES index for clickstream data, and we store data for non-login(session_id) and login users (user_id). We use the stored data now to show viewed and favorites accommodations on the site.

Now i want based on a click analysis to cluster a visitor (non and login) in a specified cluster. A cluster can be "budget", "Couple", "Family" ect.

These clusters i want to "feed" with the user/session id profiles so i can use it to personalize our channels as site and email.

Can someone guide me. How can we create "rules" so we can assign profiles to a cluster?

Thanks..remco

Remco
  • 681
  • 1
  • 6
  • 20
  • could you share few sample documents? what criteria would you like to use to cluster a user, is there a threshold value to decide this? – ChintanShah25 Jan 22 '16 at 21:06
  • i want to start simpel: Example "Luxury family cluster". A) Viewed + 3000 per week accos (+ 4 guest). B) added + 3000 per week accos to favlist. The if this "pattern" has taken place add this profile in "Luxury family cluster" – Remco Jan 22 '16 at 21:14
  • how does your schema look? do you have click_count, viewed_date etc? – ChintanShah25 Jan 22 '16 at 21:21
  • i let you know tomorrow, because i need to talk to our developer. thanks already! – Remco Jan 22 '16 at 21:28

1 Answers1

0

I suggest you talk to ActionML We have several pieces of software that could apply like our Page Variant Recommender, which will learn from user responses which variant of a page or email works better for certain user profile or segment info. We have the state-of-the-art Universal Recommender which can ingest user clickstream info to make recommendations. All this is build on the PredictionIO Machine Learning Framework, which in turn supports a large number of algorithm plug-ins that can be applied to classification (supervised categorization) or clustering (unsupervized categorization) of users and many other use cases.

ActionML does contract consulting and tries to answer questions about their open source Apache 2 licensed software on the PredictionIO Google Group.

pferrel
  • 5,673
  • 5
  • 30
  • 41