If someone likes a local Italian restaurant, predicting that they might like another, similar Italian restaurant isn’t exactly rocket science (or, should we say, data science).
What’s a bit more impressive is figuring out that the same person might like a certain fashion brand and have a proclivity for jazz music because they’re looking at a discussion thread about upscale pizza restaurants. Those are the sort of findings that Reddit is hoping to uncover with its new partnership with predictive inference company Qloo (pronounced “clue”).
The partnership, announced on Tuesday, will provide Reddit with access to over 10 years of Qloo’s proprietary data, helping it serve more relevant ads to users.
Qlooed in
In 2019, Qloo acquired TasteDive, a multi-industry review and recommendation platform. The data it collects through TasteDive is a key part of its catalogue, Qloo CEO and Founder Alex Elias told AdExchanger. The catalogue tracks correlations between different interests (or “entities”), like restaurants and musical artists, based on shared attributes.
Qloo determines similarities by categorizing entities based on a range of data points from sources, including TasteDive reviews, metadata from the open-source knowledge base Wikidata and data licensed from third parties.
For instance, the famous Italian restaurant in New York I Sodi isn’t just categorized as “Italian food,” Elias said. Rather, Qloo’s knowledge base also pays attention to signals that describe its architecture and ambiance (which Elias described as “Scandinavian minimalism”) so it can be linked to, say, fashion brands with a similar aesthetic – not just comparable Italian restaurants.
One jump ahead
Understanding the shared attributes that wouldn’t be obvious at first glance is one of the main reasons Reddit has elected to partner with Qloo, said Virgilio Pigliucci, Reddit’s senior director of engineering.
Users can opt into personalized ads on Reddit, which are particularly helpful when a user is scrolling through their personal feed and engaging with a variety of content, said Pigliucci. But when someone is on a particular subreddit, Reddit generally aims to serve them ads that are “contextually similar” to the posts they’re looking at, he added.
Qloo’s data helps Reddit find connections between products that might not seem to have much in common.
The signals that Reddit gets from its own out-of-the-box data usually suggest advertising a product that is directly related to the post a user is reading, said Pigliucci, like an espresso machine ad on a post about the best coffeemakers.
Using Qloo’s data, Reddit saw that people who are in the market for an espresso machine are also often looking to buy a table that the machine can live on. Now, Reddit might serve an ad for a tall desk on that same thread, rather than echoing recommendations that are already organically cropping up in the conversation.
Seizing the moment
Historically, it could take days or weeks for Reddit to optimize a new advertiser’s campaign because it needed to track the way users engaged with the ads, Pigliucci said. Now, Reddit has a better understanding of connections between different categories and products and can “optimize against the signals” provided by Qloo by finding a close proxy to the advertiser or ingesting its prior data – no user engagement data necessary.
Plus, Qloo’s data allows for more resonant advertising without requiring any personal data, Pigliucci said. The model draws correlations between entities without having to know the user’s search query, browsing history or even what subreddit they’re in.
It’s a win-win for advertisers and clients. Client privacy is protected, and advertisers see results “right away,” said Pigliucci, “instead of having to wait a little bit,” as was the standard with its previous machine learning capabilities.
