Together with United Partners, one of the leading international corporate and marketing communications agencies, Identrics created a weekly overview of relevant stories from the PR and marketing landscape across multiple countries.

We coupled topic modeling technologies with social media engagement measurement to give domain experts a quick and rich breakdown of the hot topics of the industry.

The result is an automated process, which flags the most engaging stories from last week, and systematises and presents them in a clear and appealing newsletter.

Initial Dataset

We started off with our data lake where we index over 1.7 million documents per day. From here, United Partners pinpointed industry relevant sources per their expertise. 

Next we described broad industry topics revolving around PR and marketing agencies, marketing awards, industry trends and opinions, and translated them into language specific search strings.

The aim here was to limit the initial dataset while casting a wide net in order to be sure that no relevant information was left out. However, this also allowed for irrelevant content to pass through, which we address in the next steps.

Article Translation

Since we served an international audience and wanted experts from different countries to understand the situations in all countries within the scope of the newsletter, we translated all articles to English.

We did this with last generation multilingual AI translators trained in more than 50 languages. 

Topic Classification

We structured the newsletter in three main categories:

  • Campaigns, branding and reputation; 
  • Companies, projects and activities; 
  • Trends, opinion and awards.

To keep the format more efficient, we limited each category to its three most engaging articles.

The process:
  1. We used a custom pipeline starting with a binary classification distinguishing between marketing and non-marketing articles.
  2. We combined the binary results with an unsupervised state-of-the-art topic modeling algorithm which forms clusters of articles exhibiting specific similarities. Each cluster is described by auto generated keywords which are then used to select the articles relating to the industry. 

These two steps assured that only PR and Marketing articles made it to the newsletter.

Then, we finished with a fine-grained classifier to distribute the articles into the three predetermined categories.

Article Engagement

For this case Facebook interactions were prevalent, so we focused on those. For each article we measured the amount of engagement it had (shares, comments, likes and reactions) via the Facebook plugin. This resulted in a quantifiable representation of which articles were most popular in the past week.

We used this information to select the most engaging articles, which were to be included in the newsletter. 

To learn more, explore our Topic engagement services.

Article Lead

In order to give as much information as possible in a single screen, we automatically included each article lead in the newsletter.

The Newsletter

The structured results are exported in a machine readable format which is easily imported into any CRM and marketing platform.

Result

With the help of this technological solution, we help our clients to communicate important information from their industry to their audience in a quick, easy and efficient way. 

Notes

This stack of technologies and general approach can be adopted for any topic and delivery formats and mixed with other Identrics’ solutions