While the wealth of readership data that the Memo platform provides is certainly exciting, it is hard to extract those insights without having a way to “slice and dice” the numbers.
This is why our app assigns each article a “topic,” a label that provides an instant understanding of that article’s content.
Say you want to hone in on coverage about hiring and workplace benefits – perhaps to see which outlets get high readership on the subject, or how your share of readership compares to competing employers. You can filter by topic in the app, in this case Business News – Hiring, Wages, Benefits, and view readership stats specific to this theme.
Or conversely, if a topic is creating a lot of noise in your media coverage, you can filter it out to focus on the readership data most relevant to you. Pretty handy, right?
This month, we’re excited to release a new machine learning-powered classification system that improves both the categorization process and the topic taxonomy.
We’re calling this launch “universal topics.” And while a faster, more consistent topic schema might not seem flashy, I promise that it’s an actual game changer for surfacing deeper readership insights at scale. Here are the details:
Memo’s article topics are built for how PR & Comms teams view media coverage
Unlike tools that generalize the subject of a piece of content, Memo tags articles with topics that are relevant to the coverage categories PR & Comms teams think about every day. An article about the new iPhone wouldn’t just be “consumer electronics” or “iPhone.” Rather, Memo might label it Product – Launch – iPhone 14 so that the user knows this was product-related press about the launch of the new iPhone.
Universal topics follow a pattern of Primary Topic (all) – Secondary Topic (most) – Tertiary Topic (some). For many longtime Memo users, these labels will look familiar. Here’s how the taxonomy works:
First, every article is assigned one of 13 possible Primary Topics:
- Advice: How-to, instructional, and advice-driven articles
- Business News: Corporate news, earnings coverage, executive news, etc.
- Celebrity: Celebrity interviews, paparazzi coverage, gossip, and partnerships
- Content Availability: Coverage of how, when, and where to access content
- Corporate Initiatives: ESG-related initiatives by companies
- Deals & Promos: Sales announcements and promotional offers
- Event: Summits, conferences, sporting events, awards ceremonies, etc.
- Human Interest: Feature stories that portray people in an emotional way
- Incident: Coverage of crimes, deaths, cyberattacks, etc.
- Industry Trends: General industry news and commentary
- Issue: Coverage of large societal issues
- Product: News about a company’s products, including launches and reviews
- Merchandising: Articles promoting the availability of retail goods
Content and industry nuances are captured with subtopics
Once the Primary Topic has been identified, most articles will receive a Secondary Topic, and some will receive a Tertiary Topic.
For certain Primary Topics, there are finite lists of corresponding Secondary Topics. All of the possible Secondary Topics for Business News, for example, are the following:
- Business News – Earnings
- Business News – Expansion
- Business News – Hiring, Wages, Benefits
- Business News – Leadership
- Business News – M&A & Partnerships
- Business News – Stocks & Markets
- Business News – Thought Leadership
For other Primary Topics, there are infinite Secondary Topics. For instance, the primary topic Event has endless possibilities. In these cases, we use machine learning to pull out named entities that identify the specific event. An article that discusses the Emmy awards will be labeled Event – Emmy Awards. An article about the Academy Awards will be labeled Event – Academy Awards.
The primary topic Product is a combination of finite and infinite: the Secondary Topics are always either News, Review, Launch, or Roundup. Then we use entity extraction to pull a custom Tertiary Topic, typically the name of the product. So an article titled “Apple Launches iPhone 14 and 14 Plus” would be categorized as Product – Launch – iPhone 14.
The new topic schema enables easier readership analysis
A major benefit to the newly launched topic conventions is that the consistent structure makes it easier for brands to hone in on the coverage they care about for a given analysis.
In the iPhone example above, a brand like Apple could zoom in on iPhone 14 launch coverage to measure the readership of that campaign; they could look at the entire product news cycle for the iPhone 14 (the launch, product reviews, product promos, etc) to track readership over time; or they could compare all product launches across different devices to find the best outlets for the next device launch.
Universal topics balance article fidelity with industry flexibility
Previously, topics were assigned at the brand level, meaning an article’s topic might differ across accounts.
By acknowledging that the topic of the article is a singular entity, universal topics create more value by prescribing an identity to that article, rather than assigning a topic in the context of the account’s dashboard.
The launch of Brand tags in our platform in 2021 (labels that indicate which brands are included in an article) removed the need to clarify the brand context in articles with topics. In the past, for instance, when an article only mentioned an account’s keywords in passing, Memo assigned it the topic Mention. While that is the context of the article in relation to the brand, it doesn’t actually tell us anything about the contents of the article!
With universal topics, each article’s topic is a direct representation of what that article is about. As described above, Secondary and Tertiary Topics provide more detailed entity information to make the topic relevant to the industry and brand. Altogether, this structure ensures consistency in reporting and allows Memo to provide deeper insights at scale.
Universal topics are also processed faster with new AI models
An added benefit to this launch is a standardized delivery time of the topic assignments. The granular customization at the brand level was possible with our legacy keyword-based topic assignment system, but it posed a hurdle as we migrated to more advanced, scalable machine learning models. And frankly, it took a long time to process.
Now, when you log into the dashboard first thing in the morning, all of the articles for the previous day will have a topic in addition to readership, every time! No more waiting for updates at 12pm EST or later.
We have been working hard over the past several months to revamp our topic system, and we are so excited to share these improvements with our clients. The scalability of this system paves the way for deeper and faster insights, reports, and product features – so this is just the beginning.