Strategies Overview

Owned by Former user (Deleted)
Last updated: May 08, 2023 by Joyce Chu
8 min read

Recommendation Strategies Overview

The Hawksearch service enables online retailers and publishers the ability to drive a rich and compelling user experience. This experience drives visitors to the products and information that they're seeking dynamically. This document describes Strategies (Algorithms) used by the Hawksearch Recommendation Module. For more information on how to configure strategies within a widget refer to Widgets- Adding a Strategy

 

Recommendation Context

Recommendation context is defined for each recommendation strategy to decide the context used to generate recommendation. Available options include: 

  • Page context:

Recommendations will be based on information from the page that a particular user is viewing. 

  • Last viewed item:

Recommendations will be based on the last item that was viewed by a particular user. 

  • Last added to cart:

Recommendations will be based on the last item that was added to cart by a particular user. 

  • Last purchased:

Recommendations will be based on the last item that was purchased by a particular user. 

Recommendation Strategies

Featured Items

The "Featured Items" strategy generates queries on any fields that are available in the data feed. A query is built using the standard Hawksearch query builder control that is also available for search and recommendations. An example of the featured item query below will recommend items that belong to the same brand as another item that the user is looking at (when page context is selected) that are flagged as a sale item and sorted by price in descending order.  

 

More like This

The "More Like This" strategy shows similar items based on their text description. This strategy allows a user to define fields used to determine similarity. An example below is the configuration showing all items that contain the highest number of common words describing an item in 2 fields: Title/Name and Content. Items are then filtered to leave only those belong to the same brand. 

 

Viewed Then Bought

The "Viewed then Bought" strategy recommends products that are similar to the product that users viewed then bought. The machine learning model measures similarity with users' previous actions. Based on the View and Buy event from different users, product A and product B are considered similar if most users who viewed then bought product A also viewed then bought product B. The "Viewed Then Bought" strategy provides Filter Settings that can further narrow recommended items. A good example would be to show recommended items with similar prices. Additional filters to narrow recommended items are available for this strategy.

 

 

 

Personalized

The "Personalized" strategy provides a 100% one-on-one personalization experience based on the user’s previous View and Buy event. This can build a stronger connection with the audience and help lead to better customer retention. While other strategies require one or more items to be in the view context, the "Personalized" strategy only requires a unique Visitor Identifier. This strategy finds user-to-item relationships in real-time and doesn’t require training the model. It is a great strategy to engage new users of the website. Additional filters to narrow recommended items are available for this strategy. 

 

 

Also Added To Cart

The "Also Added to Cart" strategy recommends products that are similar to the product that the user added to the cart. The machine learning model measures similarity with users' previous actions. Based on the Add to Cart event from different users, product A and product B are considered similar if most users who add product A to the cart also add product B to the cart. Additional filters to narrow recommended items are available for this strategy. A good example would be to show recommended items with similar prices. See the example below. 

 

Add To Cart For Me

The “Add To Cart For Me” strategy is a personalized strategy that recommends a user other items to add to the cart. This machine learning model uses the user’s data and returns items that the user should also add to the cart. This strategy finds user-to-item relationships in real-time and doesn’t require training the model. Additional filters to narrow recommended items are available for this strategy.

 

 

Also Bought

The "Also Bought" strategy recommends products that are similar to the product that the user bought. The machine learning model measures similarity with users' previous actions. Based on the Buy event from different users, product A and product B are considered similar if most users who bought product A also bought product B. Additional filters to narrow recommended items are available for this strategy. 

 

 

 

Buy For Me

The “Buy For Me” strategy is a personalized strategy that recommends a user other items to buy. This machine learning model uses the user’s data and returns items that the user should also buy. This strategy is a user-to-item relationship that works in real-time and doesn’t require training the model. Additional filters to narrow recommended items are available for this strategy.

 

 

 

Also Viewed

The "Also Viewed" strategy recommends products that are similar to the product that the user viewed. The machine learning model measures similarity with users' previous actions. Based on the View event from different users, product A and product B are considered similar if most users who viewed product A also viewed product B. Additional filters to narrow recommended items are available for this strategy. 

 

 

 

View For Me

The “View For Me” strategy is a personalized strategy that recommends a user other items to view. This machine learning model uses the user’s data and returns items that the user should also view. This strategy is a user-to-item relationship that works in real-time and doesn’t require training the model. Additional filters to narrow recommended items are available for this strategy.

 

 

 

Frequently Bought Together

The “Frequently Bought Together” strategy looks at the attributes of multiple items and suggests other items that users might like based on their current choices. This can help guide the user to discover recommended products that complement the one they chose or might have missed. For example, if a customer has 10 items in their cart, this strategy can recommend the 11th item that the customer might want to buy together. Additional filters to narrow recommended items are available for this strategy.

 

 

 

Best Sellers

The "Best Sellers" strategy recommends the top-selling product based on real-time sales data. This strategy can help open up users to a whole new range of up-sell and cross-sell opportunity. While defining the “Most Popular” strategy properties, Hawksearch allows the user to define a Time Period. The popularity of items takes the Buy event into consideration. To avoid a cold start, Hawksearch also provides an option to import historical sales data. Additional filters to narrow recommended items are available for this strategy. 

 

 

Hot Now

The "Hot Now" strategy recommends popular products within a short time period based on real-time sales data. While defining the Hot Now strategy properties, Hawksearch shows 2

settings:  

1. Primary Time Period  

2. Secondary Time Period  

Time periods are used to configure the "now" aspect of the sales data – for lower traffic sites, the time period can be set to a higher value, and for higher traffic sites, the time period can be set to a lower value. A user can configure the Time Period to update in short minute intervals. Hot now calculates the rank based on the following events: Buy, Add to Cart, and View. For each event, the product ranking is calculated by selecting products from both time periods (Primary and Secondary) and multiplying their quantities with defined multipliers. In the default multiplier, events in the Primary Time Period contribute the most, and events in the Secondary Time Period have respectively lower impacts on the ranking. Products with the highest rankings will then become the final recommendations. Additional filters to narrow recommended items are available for this strategy. 

 

Most Popular

The "Most Popular" strategy recommends the most popular products based on real-time sales data. This strategy adds a social proof element to the recommendation and helps introduce categories that are new to users. While defining the “Most Popular” strategy properties, Hawksearch allows the user to define a Time Period. The popularity of items is decided based on the following events: Buy, Add to Cart, and View – each event is ranked differently. Buy events contribute the most, and Add to Cart and View events have respectively lower impact on popularity. Additional filters to narrow recommended items are available for this strategy. 

 

 

Trending Items

The "Trending Items" strategy utilizes combined strategies to recommend trending items. While defining the "Trending Items" Strategy properties, Hawksearch allows the user to define a Time Period. A "Trending" factor is calculated using data from other existing strategies:   

  • Most Popular   

  • Hot Now   

  • Best Sellers   

Additional filters to narrow recommended items are available for this strategy. 

 

Recently Searched

The "Recently Searched" strategy shows a list of items that were recently displayed in the search result for a particular user– items are displayed in a random order and contain items from various searches (not only the last one)
Configuration allow to define, view context, number of recent searches to take into consideration and number of top items from these search results that should be used. Additional filters to narrow recommended items are available for this strategy.

 

Recently Viewed

The "Recently Viewed" strategy shows a list of recently viewed items – items are displayed in descending order (last viewed item will be displayed first). Additional filters to narrow recommended items are available for this strategy.