< Back To Portfolio

Client

Scholastic has been delivering literacy resources for kids and outstanding children's books to schools, teachers, and families for more than 90 years.

Goal

Create a data-driven book recommendation engine that harnesses the expertise of tenured Scholastic librarians to improve school library holdings.

Outcomes

  • Adaptive Recommendation Algorithm
  • Narrative Evaluation
  • Visual Benchmark Scorecard
It's astonishing how well it all works.... It is everything we hoped a partnership would be.

Erik Ryle

Director of Paperbacks and Collections

Overview

Our digital lives are rich with recommendation systems. Our phones recommend routes around traffic jams; Facebook and Linkedin recommend friend and colleague connections; and Netflix estimates that 75 percent of viewer activity is driven by movie recommendations.

Behind the scenes, many recommendation systems utilize Collaborative Filtering: by looking at examples of item grouping patterns (say, shopping cart contents) across time, it becomes possible to generate suggestions for what a user might reasonably be interested in. Recommendation systems can do more than say what's popular with other people with similar taste, however. Instead, the technology can actively imbue the ethos and social mission of the recommender - while still leveraging enough popularity analytics to ensure they're appealing.

Scholastic does more than sell books and educational materials to schools, teachers, parents, and children. Since its incorporation in 1920, it has developed a reputation for inspiring kids to read. Scholastic knows, based on their experience and research, what a classroom should contain to provide an ideal range of reading levels and subject diversity for academic principles, as well as spark and maintain a love of reading in students.

Challenge

Scholastic tasked Green River with an interesting challenge. Based on our expertise in building data science software applications for nonprofits and mission driven organizations, they asked us to build a book recommendation engine for teachers that captured the knowledge of the educators at Scholastic. Our colleagues at Scholastic have developed research derived benchmarks for the classroom library - how many read aloud books a kindergarten library should have, what are the reading levels of books for each grade level, how many nonfiction books should the classroom library have, and so on. These benchmarks capture the expertise of scholastic educators, who recommend engaging titles that can both improve academic rigor of the library that teachers will also want to buy and that students will want to read.

Process

In 2015, Green River started working with Scholastic to explore what it would mean to design an algorithm to evaluate classroomlibraries and make compelling recommendations. We felt that with the wisdom of Scholastic’s nearly one hundred years of data collection, combined with our ability designing mission driven business intelligence solutions, that we could develop a software system that could ingest a list of any titles, identify areas of weakness and recommend any additional titles that Scholastic believes would remediate deficiencies. We developed an approach based on a combination of three pillars:

  1. We looked at Scholastic’s benchmarks, finding titles that would best improve certain aspects of a library, like reading level distribution or genre diversity.
  2. We factored in expert recommendations, from "ideal library" sets that Scholastic had curated for a given grade level, based on their philosophical and academic reasons, ensuring the ethos of the Scholastic brand was intact.
  3. And we leveraged the collective wisdom from thousands of teachers using Collaborative Filtering algorithms to suggest books that were popular, or appeared in popular combinations, in other similar libraries and book lists.

The specific parameters of each of the three factors, and their relative weights, were easy to adjust, and the final combination of settings becomes the magic of the system.

Scholastic Recommendations

Solution

The recommendation system continues to learn and adapt automatically as it sees what books teachers are purchasing for their classrooms. This ongoing improvement through machine learning ensures that recommendations always reflect the current needs of students. Built like this, algorithms behind a recommendation engine can be tuned not only to maximize revenue, but to help realize the social missions of our clients.

Scholastic Recommendation Benchmarks

After a classroom library is analyzed, the teacher receives a detailed report that includes a narrative overview, as well as a scorecard demonstrating its relative composition against specific benchmarks Scholastic has defined.