Help Center

  1. BlueConic Help Center
  2. How Tos
  3. Recommendations
  4. Content recommendations

Content recommendation algorithms

BlueConic offers a variety of content recommendation algorithms to help you deliver personalized, one-to-one content recommendations based on individual customers' or visitors' behaviors, interests, and preferences.

To activate personalization recommendations, you use toolbar plugins for product recommendations or content recommendations.

personalized content recommendations and product recommendations with BlueConic

This article provides notes on how content recommendation algorithms work. To add recommendations to your pages or app, refer to the Deliver 1:1 Content Recommendations Use Case article.

Algorithms overview

Algorithm Description Notes

Breaking news

("RECENT_VIEW")

 Use the breaking news algorithm to boost the articles that are most viewed during a time frame you define.
  • Independent of profile
  • Based on statistics of the last X hours (default is 5 hours)
  • Content items are boosted based on the value of the statistics field “view,” which is incremented whenever the article page is viewed (independent of recommendations).

Viral news

("RECENT_ENTRYPAGE")

 Use the viral news algorithm to boost the articles that are most used as a landing page during the defined time frame.
  • Independent of profile
  • Based on statistics of the last X hours
  • Items are boosted based on the value of the statistics field “entrypage.”
Recent high CTRs ("RECENT_CTR") Use recent high CTR to boost the most clicked-through articles from BlueConic recommendations during the defined time frame.
  • Independent of profile
  • Based on statistics of the last X hours
  • Items are boosted based on the value of the statistics field “click” (someone clicks a content item in a recommendations dialogue) divided by the value of the field “recommendation_view” (an item is recommended).
Same category ("SAME_CATEGORY") Use the same category algorithm to boost articles of the same category as the page the visitor is currently watching.
  • Independent of profile
  • Independent of statistics
  • Items are boosted based on the values of the “categories” metadata field of the “current item” versus the categories of the other items. If items have multiple categories, the score is higher if they have more categories in common.
Look-alike articles ("LOOK_ALIKE") Use the look-alike articles algorithm to boost articles that have similar content to the one someone is currently viewing.
  • Independent of profile
  • Independent of statistics
  • Look-alike algorithms use a TF-IDF weighting factor that measures term frequency and significance to determine which items are most similar to the current item. This determination is based on the words in the content item's “name,” “description,” and “text” metadata fields.

Recency
("RECENCY")

 Use the recency algorithm to boost the most recently added articles.
  • Independent of profile
  • Independent of statistics
  • This personalization algorithm uses the value of your content items' “publicationDate” metadata field to boost items. Newer items score higher than older ones.
Collaborative filtering (“COLLABORATIVE_FILTERING") Use the collaborative filtering algorithms to make personalized content recommendations for individuals based on the browsing behavior of other people who view the same articles.
  • Independent of statistics
  • Dependent on the items the individual has viewed
  • Operates with a ramp-up time, taking time to account for the interests of other viewers
  • This results in an affinity between profiles and items, which is used in a collaborative filtering model.
Same interest
("INTEREST")		 Use the same interest algorithm to boost articles that are similar to those the customer or visitor already viewed, based on the values of the item's “categories” metadata fields.
  • Independent of statistics
  • Dependent on the items the customer or visitor has viewed
  • Operates with a ramp-up time
Seen articles ("RECENTLY_VIEWED") Use the seen articles algorithm to boost articles the visitor recently viewed.
  • Independent of statistics
  • Dependent on the items the visitor has viewed
  • The items the individual most recently viewed are boosted the most. You can use this algorithm to create a “last viewed articles” overview.

Recommendation algorithm considerations for publishers

Algorithm Considerations

Breaking news
  • This algorithm typically has a good CTR rate.
  • Consider using this algorithm for visitors who have a very limited number of page views (i.e., you don’t know what content or specialties this visitor is interested in)
  • This algorithm is good for email recommendations for visitors who have not been to the site.
Recent high CTRs
  • Typically, the most engaged visitors click on the most content recommendations.
  • This algorithm can be useful when targeted to visitors with low engagement (i.e., you don’t know their interests).
Look-alike articles
  • The TF-IDF weighting factor of this algorithm looks for individual terms, not phrases.
  • This algorithm is not recommended for email recommendations because there is no current article to compare against.

Recency
  • This algorithm is not frequently used. Instead, most publishers use the filtering capabilities to only recommend articles that are more recently published.
Collaborative filtering
  • This algorithm is completely agnostic to the content of the article. Think of it like a Venn Diagram. If visitor A has ready article ABC and visitor B has also read article ABC, visitor A would get recommendations for articles that visitor B has read.
  • For email recommendations, visitors without any page views will get the fallback recommendation.
  • This is a useful algorithm for scholarly publishers. Researchers and clinicians often view similar content to their colleagues within a particular specialty.
Same interest
  • This algorithm is good for publishers who have a strong article taxonomy on their site.
  • Readers who view a lot of content within specific category values will be recommended articles with the same category values.
  • For email recommendations, visitors without any page views will get the fallback recommendation.
Seen articles
  • This algorithm is not frequently used by publishers
  • You may want to target highly engaged visitors with “Recently Viewed” articles so that they can refer back to them.
  • For email recommendations, visitors without any page views will get the fallback recommendation.

Related articles

Was this article helpful?
0 out of 0 found this helpful