Using our recommendation system
Several layers of integration are available when using our RecSys module.
Generally speaking, our RecSys will be used with its front-end integration via a JavaScript recommendation widget.
However, a customer can implement their own widget and use the RecSys API directly.
This way they can customize how recommendations are displayed, i.e. add them to any page (main or article), any number of widgets, with each in a different configuration. E.g. one widget can reflect a visitor's personal recommendations, one focuses on recent articles, and another displays related topics...
Note that some models won't be able to be trained and use until we are able to track those users or get access.
Purple DS HUB
Using any WP instances, our ACM WP Plugin provides a shortcode that can be used directly in the article editor.
Simply add the following segment anywhere:
Note that we don't need to specify the reader_id or item_id since they are derived in the short_code implementation.
See last section for a list of available models and parameters.
Third Party CMS or application using our ACM scripts
To integrate ACM Recommendations into your website you have to complete 3 steps:
- Include the ACM scripts `https://acm.sprylab.com/static/setup/config.js` and `https://acm.sprylab.com/static/acm.js`
- Include a <div id='acm-content-recommendations'/> container at the position where the recommendation should be loaded into
- Call window.acm.recommend(...) anywhere in the code to populate the previously defined div
Here is a possible integration:
See last section for a list of available models and parameters.
Third Party CMS or application using our REST API
If a customer requires full control over the way the content recommendations are displayed, they can use our Recommender API directly and handle all the related HTML/CSS/JS integration by themselves.
Check the Recommender API section of the Rest API Documentation
Note: However, tracking will need to be available. Therefore, we highly recommend the previous method using our ACM scripts since training user-related models won't benefit from tracking data otherwise.
RecSys models and parameters
The current list of available RecSys models is:
- multi_criteria
- content_based (needs review and use assessment)
- entity_based (needs review and use assessment)
- slot
- hybrid
Where the parameters used are:
- reader_id: ID of the reader/visitor (ideally a registered user)
- item_id: ID of the article (which can be taken into account for user-independent recommendation)
- model: name of the recsys model to be used
- limit: maximum number of recommendations returned
- types, sources, categories: filter applied on the corresponding parameter
- w_source_target, w_recency, w_retention_time: weights associated to the corresponding parameter used in the final ranking for some models
- diversify, diversification_criteria: in case the result of our top_k() are returning too similar results, it might be useful to diversify using the parameter described in diversification_criteria (e.g. "model", "type", "source", "category")Note: only used with Hybrid model and needs to be checked properly (source needs update and category needs to be added)