Using the Recommender System on ReFocus
With the release of v0.4.0, the ReFocus application has the ability to train a Recommendation System model on your data.
Applications that you use every day (Netflix, Youtube, Facebook, Instagram, etc.) all use Recommender Systems to power their content delivery. And now you can use this powerful tool on your data too!
Applications that you use every day (Netflix, Youtube, Facebook, Instagram, etc.) all use Recommender Systems to power their content delivery. And now you can use this powerful tool on your data too!
Behind the scenes, Recommender Systems are a grouping of algorithms and statistical methods that, when combined, offer relevant results and recommendations of items or products to users. Towards Data Science has an excellent introductory article about the technical aspects of Recommender Systems if you would like to learn more.
Training a Recommender System on Your Data in ReFocus
Training the Recommender System within ReFocus is extremely easy, and will take about five minutes to train.
First, if you have not already done so, upload a dataset that you want to train the Recommender System on. ReFocus encourages you to have at least 6,000 data points and 15 attributes for each of those data points. It is also required that you have at least one yes/no, 1/0, or other binary valued column in your dataset.
Next, navigate to the Models tab and click on the Create New Model tab in the bottom right corner of the page. This will take you through the new model training wizard introduced in the v0.4.0 release.
Click Get Started.
This screen displays the two different model training options that you have at your disposal. The second option, What is my best choice? is the Recommender System train. Select this option to move forward.
Now you are able to select the data that you wish to train the model on. Select your recently uploaded data, or another dataset that was already uploaded into the ReFocus system.
Once selected, you will be shown a preview of your data, along with the new ReFocus Data Score in the top right corner that was introduced in v0.4.0. Ensure that your data looks correct before moving on. It is also encouraged that you have a ReFocus Data Score of at least 80/100 to guarantee as accurate of a model as possible.
The next screen is for data categorization. Our backend system makes the best effort to categorize your columns for you without the need for your input, but sometimes we do not categorize every column perfectly. Use this step to make any changes that are necessary.
Once categorized, there is one more data bucketing step necessary. Each column in your dataset needs to be set as either a column pertaining to a customer, a product, or the outcome that you are trying to get recommendations for. There are a few rules to this bucketing process:
- Two of your columns must belong to a customer
- Two of your columns must belong to a product
- Only one of your columns can be an output
Once bucketed, click on the orange button in the bottom right corner to move to the next step.
Here you will add identifiers for your customer data and product data. These identifiers will be used for future reference when ReFocus shows your recommendation predictions.
Once identified, you will move on to the last step - naming your model, categorizing your model, and giving your model a description.
After filling this information in, give the large orange button a click and wait about 1 minute for your data to be preprocessed and your model to start training!
Once your Recommender System has started training, please wait 5-10 minutes for it to complete.
Viewing Recommendations From Your Model
After a successful training of a Recommendation System has been completed in ReFocus you will be able to add the model as a card to your dashboard.
To add the model as a card to your dashboard, navigate to the Dashboard page via the top navigation links.
If you have never added a model to your dashboard before, you will see the screen below. Click Add to Dashboard on the Recommender System that you just trained and then click Generate Dashboard.
If you have added models to your dashboard already, click on the Add More Models button in the bottom right, select your Recommender System model, and click Update Dashboard.
Note: It may take some time for the ReFocus system to generate your model's recommendations. If there are no recommendations immediately shown, check back in a couple of hours while we generate your insights.
What's Next?
The release of the Recommender System model training is just in its initial stages in the v0.4.0 release. In the coming releases, your Recommender System models will be able to:
- Continue to learn with incremental training when you add newer data to your model as time goes on
- Become more accurate for your business with reinforcement training
- Give confidence scores on how likely a recommendation is to become true
The development team is extremely excited to deliver these features in the coming months and look forward to hear your input on the current form of our Recommendation System models!