Cultivating an interest in the words that matter to you

Devising a method for readers to cultivate their interests

I was able to do this because I was fortunate to work alongside a brilliant group of people to build the Reverb app, a smart reader dedicated to helping people keep up with the topics and news most relevant to them.

A fundamental goal for the app was to engage readers in the experience of continuous reading by understanding an article’s underlying concepts and presenting them as topics to explore for further reading. We achieved this goal by embedding two sliding carousels at the end of every article, powered by our concept extraction service and our recommendation service, respectively.

The first version looked like a stock ticker comprised of ten words: key concepts extracted from the article, ordered by relevance. The second contained a set of links to related articles, represented by metadata similar to Facebook’s Newsfeed: featured image, title, source, author, date, and the story’s first few lines. Tapping on any word in the concept ticker would update the content in the related article carousel to display stories more relevant to that topic. We aspired to create the same effortless experience you get from music recommendation services: read what you like, and find new and interesting content, too.

Initially, we borrowed the same exploration model for our collection feature, but we would learn that collecting reading material is different behavior than binge-reading. Similar to creating a Pandora listening station based on a favorite song, you could create a Reverb reading collection in based on an favorite article. Just save the article as a collection and give it a name. Your collection then appears in your menu as an entry point. Tap it to discover fresh recommendations, or re-read the article at the root of the collection.

Think of the article and its key concepts as the root and branches of the collection. As new articles are published and conceptually evaluated, they can filter through your collection’s concepts and blossom into recommendations. You can return to your collection to discover fresh stories over time. Further, once you’ve created a collection, you can add another article to it to teach the collection more about your interests. When you save a second article to your collection, the key concepts of that collection update to reflect the salient concepts of both articles and consequently send fresh signals to the recommendation service for further reading. We envisioned researchers creating collections to find thesis material and hone their interests.

During development of the collection feature, I cajoled our computational linguists and machine learning experts into a contest to create the most interesting collection. Each day, during scrum, two developers faced off in a bracket to determine the winner.

Collections were measured by three factors: 

1. depth of concepts- how accurate were the keywords extracted?, 
2. breadth of concepts- did the spectrum of keywords surfaced reflect salient interests of the stories rooted in the collection? 
3. relevance of recommended reading

Challenging self-titled "word nerds" to demonstrate their interests to the team in a competition was a bonding experience as well as fuel for many spirited debates on the merits and flaws of our system. 

We quickly realized that our concept extraction service delivered excellent accuracy and relevance for about seven out of ten concepts, but the remaining three tended to be woefully off the mark. Worse, sometimes adding a new article to a collection, ostensibly to give the system more data to parse and better understand the collector’s interest, could result in narrowing the concepts too much (too deep= too specific & therefore not enough breadth to fuel varied reading recommendations). Alternatively, saving multiple articles could result in loosing excellent keywords from the concept ticker. Trust me, you don’t want to hear the cries of a passionate lexicographer when she looses a favorite word from her collection by saving a great read!

In preparation for the contest, I caught one of our developers tweaking the system to remove concepts from an article to improve the recommendations sent to his collection. If the goal of collection recommendations was to provide a feed of reading materiel guided by your interests, why not let our collectors cultivate that feed by pruning their collection’s concepts?

What did collectors need? Since a collection’s concept selection was dictated by the articles saved to that collection, and collectors had no ability to directly manipulate these concepts, the quality of recommended reading for collections had high variance. The only way to manipulate recommendations in a collection was to add or remove articles from that collection, and this proved tedious and unsatisfying. In order to increase the usage, value and meaning of collections, in order to keep collectors coming back for more, we needed to give our collectors control over concepts.

The solution was simple: edit mode. I introduced a collection editing/configuration mode that enabled collectors to add and remove concepts from their collection. Opening edit mode allowed us to look into our pools of interest. I also designed a slightly different user interface where each concept was presented as a word in a bubble. For some of us, it was a joy to behold the concepts that bubbled up from a great reading and collecting session. Tap a concept to pin it to the collection, and it glows. Don’t care for a particular word? Flick its bubble off the screen, and it’s gone from the collection! But why stop there? Missing a wonderful concept? Tap a + button to enter it and pin it to the collection. Save your changes to refresh your collection’s recommended stories. Our talented iOS developers had a ball implementing clustering physics into the bubbles to make them fun to touch and pin. 

After developing my new feature, we ran another collection competition, this time with the entire company. Once the editorial team, biz dev, and our illustrious founder, former Editor of the New Oxford American Dictionary were on board, we knew we were in for a wild ride. We were treated delightful collections from “Urban Farming in the Bay Area,” “Beloved Children’s Books and Authors” and “Smash the Patriarchy” to “Wearable Tech” to “Drone War Strategy” and “March Madness.” Though our company’s favorite was “Absurd Creatures,” Marketing used “Star Wars” as our debut collection.

We learned that collectors are more engaged over time when rewarded with engaging reading recommendations. Our system facilitated a more engaging return to a collection when we enabled collectors to configure their concept preferences. Our initial model didn’t give collectors enough control to cultivate their interests in a way that reliably yielded engaging recommended reading. By devising a way for collectors to fine-tune the interests that drove their reading recommendations, I hoped to help them deepen their interest and develop a conceptual vocabulary around it, supplemented by relevant stories.

While this method truly only benefitted a small subset of our user base, its development and implementation helped our company bond, improved the relevance of our recommendation service through dog-fooding, helped Editorial create interesting collections for Marketing to send as a primary re-engagement tool for the app, and resulted in a service flow change that helped pave the way for a simplified ‘Word Wall’ navigation system that helped Reverb show our readers the smarts behind our system.