The consulting boutique commsLAB supports organizations to position themselves in a profile-compliant way. The reputation of an organization and its dynamics over time are analyzed using reputation-relevant media articles so that trends can be identified in good time and sensible measures can be taken. These measures support the organizations in achieving or maintaining a credible and attractive public profile in the long term.
To reach this goal, commsLAB classifies hundreds of thousands of media articles into several dimensions, including a dozen business-relevant fields of action such as “Strategy” or “Products and services”.
These extensive classifications, made by media and communication experts, form the raw material of commsLAB's business. Trained AI models now support this “human coding” to further raise the qualitative standard of this classification work and to maintain it across the impressive mass of text data.
The automated recognition of topics in a text is known as “topic modeling” and is part of “Natural Language Processing” (“NLP”). As is typical for the machine learning process, careful data management (DM), feature engineering (FE) and thorough exploratory data analysis (EDA) carefully prepared the creation of several models in this project. Among other things, DistilBERT, one of the classic NLP transformers available since 2017, was trained to accurately classify text according to fields of action in fractions of a second.