Automated selection and news aggregation for business reporting
Project Profile
The client recognized our potential in handling large volumes of text and engaged us to achieve savings in their customer-facing aggregation of media articles. The savings of the previously manually performed processing steps shows the potential for our technology to perform natural language processing (NLP) with an unsupervised approach.
At a glance - essential project data
| Duration | From 11/1/2020 to 4/30/2021 with about 6 months of full engagement | |
|---|---|---|
| Data and Tools | Market - News/Media Sources • National media databases 10 million articles • 800 media • 2 years article scope | |
| Integration | • Web API - for metadata enrichment with model generated data • Quality - Reporting: Power BI • ML-Ops: Coupling with customer systems for continuous transfer from and to our models | |
| AI Methods | • NLP • DeepLearning |
Engagement Use-Case
Extending the full text query system with contextual results to the "perfect" result of an editor (automated press kit).
Client motivation / Solution aims
Saving of manual summarization/selection activities
Improve quality of results and verifiability of previous results of involved resources
Monitor quality, improve prediction accuracy
Automate the ML-Ops process to a fully automated report
AI Approach
| AI key technology | for contexts and content unsupervised learning without ontology Mixed Model Approach result optimization | |
|---|---|---|
| Solution Approach | • Individualized Word Model (German) • SmartSearch - Licencing • Lemmatizing • Entity Recognition (Names, Places, Persons, Roles) • Deep-Learning | |
| Project Approach | Simply agile / Demand Driven | |
| Project Type | Operations/ML Ops | |
| ML Integration and ML Operations | • Operations Integration API • VisualizationAPI Power-BI |
Insights and Details
In the approach we look for different models to achieve the project goals.
We compare previous approaches with our models in order to achieve comparable goals in the unsupervised approach as previous workers and processors do. This expert knowledge is then extended and improved by us with trained approaches
Blending of the models with the ACTUAL results.
As ML-Ops support we deliver reports on single events on an ongoing basis.
Here is an example of inline documentation of a METADATA process for the customer systems (item can/must be omitted or included). How to provide sensitivity limits for the final customer decision regarding the result
