The main steps in creating a recommender using the Xineoh Consumer Behaviour API are:
Choose an application type
When creating an application, you decide the application type for your use case from one of the following two types.
Upload and Import your data to Xineoh.
You have two options when uploading and importing your item, user and interaction records. You can either use bulk S3 upload or HTTPS upload. For large quantities of historical data, bulk upload is ordinarily the better choice. HTTPS is usually the preferred option for real-time updates unless the amount of data is substantial.
The data you import will be dependent on your application type.
When you do a bulk upload, it will follow our base schema and S3 file structure. Once the data is uploaded to your bucket, you will provide import and process instructions to Xineoh via an API query. The data posted through HTTPS will match the applicable JSON schema.
Create Your Recommender
Once the Xineoh API has processed your data, you will be able to start customising and implementing your own recommendation engine.
You then build, train, test and optimise your models with only a single API query for each task. You can also set your preferences for how cold starts should be handled and filter out unwanted results.
Xineoh offers you the ability to perform item and user segmentation on your data through clustering. We provide you with the total MSE (Mean Square Error) for each number of clusters, allowing you to choose the optimal quantity of clusters.
The model can then be deployed, and you have built your own recommendation engine.
Now you can get real-time recommendations via HTTPS queries. You can integrate the queries into different applications, including: