Follow

Data Feed Weights Explained: Unweighted, Popularity, Reduced Heat, and Heat

To clarify first, there are two types of Data feeds -- Fresh and Evergreen -- that determine how content is pulling into the feed. Fresh will pull in content from a certain amount of time back and Evergreen pulls content from the entire content library. How the content is weighted for these two feeds uses the same formula.

Adding a weight to content allows you to give more or less priority for the content. For example, your feed contains only new content, but you may want to prioritize the popular content over content with minimal to no views. You may choose the Popularity Weight to add a higher weight score on the content with more pageviews. This weight score is taken into consideration in addition to a user's interest in the content as well.

Below is a helpful guide to understand how weight score is determined:

  • Unweighted – Each content item is given an equal weighting. Content is not given any additional priority as the weight is equal across all the content, regardless of age and pageviews. When you preview the feed for an unweighted feed, the weight score for all content will be set as . Therefore, content recommendations will be based exclusively on only a user's interest score.

 

  • Popularity – Weighting correlates to the total number of pageviews of the item, regardless of its age.
    • Weight is only based on the total pageviews for the content. For example, if you have a product that has a weight set to "weight": 25241, it is because the content has 25,241 pageviews.

 

  • Reduced Heat – Weighting favors popular content (based on total pageviews) but to a lesser extent also takes into account the recency of published content.
    • (log(pageviews)/time) - since pageviews are held to a lesser extent, the logarithm is taken of the value for pageviews, divided by the time since the content was published.
    • For example: you have a product called Organic Dried Banana. It has 1,026 views and it has been 36 days since the content was published. The log base of 1,026 views is 3.01. Then 3.01 divided by the 36 days is 9.71. The value is then multiple by 100000 to get your the weight value of .0962 that you would see in the feed.
    • On the other hand, you have a product called Chicken Broth. It has 33,149 views and it's been 11 months since the article was published.The log base of 33,149 views is 4.52. Then 4.52 divided by about 11 months is 5.13. Multiplied by 100000 brings the value to the .0513 weight you would see in the feed.
    • If a user was interested in both Organic Dried Banana and Chicken Broth, the user is more likely see Organic Dried Banana first in their recommendations because of the higher weight on the product.

 

  • Heat – favors content that is new and has a high number of pageviews for its age.
    • (pageviews/time) would be the formula to determine the weight, therefore the same formula as Reduced Heat, but without the logarithm. Using the same examples as Reduced Heat:
    • Organic Dried Banana: (1026 / 36 days) * 100000 = 32.797
    • Chicken Broth: (33149 / 11 months) * 100000 = 376.620
    • If a user was interested in both Organic Dried Banana and Chicken Broth, the user is more likely see Chicken Broth first in their recommendations because of the higher weight on the product. 

HOW DOES WEIGHT IMPACT USER INTEREST SCORES

The weight score added to each content is then multiplied with the interest score a user has for each tag on the content. The content is then sorted by horizon_select by content with the highest score for a given tag (the content just needs a high interest score for a single tag on the content, not all the tags). 

For the sake of example, a data feed has the following two items:

Item 1:

  • Title: Organic Dried Banana
  • Tags: food, organic, vegetarian
  • Your interest score for "food" is 0, "organic" is 10, and "vegetarian" is 25
  • Content was published 36 days ago.
  • Content has 1,026 page views.

Item 2: 

  • Title: Chicken Broth
  • Tags: food, chicken, soup
  • Your interest score for "food" is 0, "chicken" is 0, and "soup" is 5
  • Content was published 11 days ago.
  • Content has 33,149 page views.

Here is what the final interest score for each content would look like after the weight is added: 

Unweighted:

Item 1: 

  • food: 0 * 1 = 0
  • organic: 10 * 1 = 10
  • vegetarian: 25 * 1 = 25

Item 2: 

  • food: 0 * 1 = 0
  • chicken: 0 * 1 = 0
  • soup: 5 * 1 = 5

RESULT: The interest score is multiplied by the weight, which is "1" on all content, effectively ranking interest in content by just the interest score. Item 1 would be recommended over Item 2 because the tag "vegetarian" has the highest score of all the tags.

Popularity:

Item 1: 

  • food: 0 * 1,026 = 0 
  • organic: 10 * 1,026 = 10,260
  • vegetarian: 25 * 1,025 = 25,625

Item 2: 

  • food: 0 * 33,149 = 0
  • chicken: 0 * 33,149 = 0
  • soup: 5 * 33,149 = 165,745

RESULT: The interest score is multiplied by the weight, which is total views on the content. Item 2 would be recommended over Item 1 because the tag "soup" has the highest score of all the tags.

 

Reduced Heat:

Using the Reduced Heat score calculated above.

Item 1: 

  • food: 0 * .0962 = 0
  • organic: 10 * .0962 = .962
  • vegetarian: 25 * .0962 = 2.405

Item 2: 

  • food: 0 * .0513 = 0
  • chicken: 0 * .0513 = 0
  • soup: 5 * .0513 = .2565

RESULT: The interest score is multiplied by the result from the formula. Item 1 would be recommended over Item 2 because the tag "vegetarian" has the highest score of all the tags. 

 

Heat:

Using the Heat score calculated above.

Item 1: 

  • food: 0 * 32.797 = 0
  • organic: 10 * 32.797 = 327.97
  • vegetarian: 25 * 32.797 = 819.925

Item 2: 

  • food: 0 * 376.620 = 0
  • chicken: 0 * 376.620 = 0
  • soup: 5 * 376.620 = 1883.1

RESULT: The interest score is multiplied by the result from the formula. Item 2 would be recommended over Item 1 because the tag "soup" has the highest score of all the tags. 

 

 With that said, make sure to test the different Weights to find the right fit for your business / campaign!

Top