Our methods enable us to quickly and reliably identify trends from constantly growing volumes of unstructured data. To do this, we use an in-house procedure that continuously processes various data sources in a range of languages – such as RSS feeds, homepages, newsletters, posts on social media platforms, scientific publications, and patent applications.

Research field I: Natural Language Processing

Natural Language Processing is the framework within which unstructured speech data is transformed into a structured data format. This enables machines to identify and understand language and text in order to subsequently generate relevant answers. The sub-section, Natural Language Understanding (NLU), focuses on the pure understanding of natural language. NLU is primarily focused on machine reading comprehension. To this end, mainly grammar and context is analysed to identify the meaning and significance of a sentence. To generate natural language, the second sub-section, Natural Language Generation (NLG) comes into play. NLG primarily addresses text construction. A machine can construct texts in different languages based on an established data set.

The following areas are related to FE research:

Source: Eggers, W., Mali, N., Gracie, M. (2019): Using AI to unleash the power of unstructured government data. Deloitte. Online: https://www2.deloitte.com/us/en/insights/focus/cognitive-technologies/natural-language-processing-examples-in-government-data.html

Research field II: Semantic Web

The entities and relations determined in previous steps in the NLP process are saved in a knowledge graph. The basic technologies used in constructing knowledge graphs have been established for a long time. Graph-based data structures permit the linking of differently formatted data sets and dynamic expansion across additional classes. To be able to semantically link extracted information in a knowledge graph, a domain-specific system of rules is needed (ontology). This will describe the relational classes (e.g. procurement, research, use), entity types (product, technology, organisation, etc.), and meta information (timestamp, information source, etc.) that are permittable for a market. It is also possible to integrate additional knowledge from external knowledge graphs (Diffbot, Wikidata, Dimensions.ai, etc.).

The following areas are related to FE research:

Illustration based on Eggers, W., Mali, N., Gracie, M. (2019): Using AI to unleash the power of unstructured government data. Deloitte. Online: https://www2.deloitte.com/us/en/insights/focus/cognitive-technologies/natural-language-processing-examples-in-government-data.html