Trend and market analysis

Applying our methods, we are able to quickly and reliably identify trends from steadily growing amounts of unstructured data. For this purpose, we use an in-house developed procedure that processes various data sources such as RSS feeds, homepages, newsletters, contributions on social media platforms, scientific publications or patent applications on an ongoing basis and across multiple languages. Using Semantic Web tools, relationships, events and facts are automatically recognised and annotated, linked by means of Linked Open Data principles and mapped in a dynamic knowledge graph. Building on this, we identify market- and technology-specific trends, derive industry-specific scenarios and analyse relevant indicators over time. The following diagram shows the key steps of our analysis process:

  • Data collection: During the current project stage, we focus on text data which are continuously being collected from more than 1,400 different data sources. Sources include news articles via Google and RSS feeds, specific company websites, social media platforms, regional and subject-specific information platforms, and databases with scientific publications or patent applications.
  • Text analysis: Original text data from any language area are automatically translated into English for further evaluation in order to obtain standardised information. Relevant mentions in the texts, such as companies, technologies, or countries, are then identified through Natural Language Processing (NLP), represented as unique entities, and linked to other knowledge graphs. This makes it possible to identify mentions across different sources and to automatically link them with implicit information.
  • Knowledge graph development: In order to structure and analyse the knowledge gathered from the texts, we develop topic-specific knowledge graphs. These reflect the connections and contexts of specific mentions and contents. The structure of the knowledge graph enables incremental, flexible growth of the knowledge base as well as traceability of the origin of processed information.
  • Trend exploration: At this stage, it is possible to direct specific questions to the knowledge graph (e.g. "Which companies offer hydrogen powered commercial vehicles?”). Furthermore, it is also possible to search more broadly for previously unknown developments, which can be identified using methods such as topic modelling.
  • Visualisation: Not only is the knowledge base continually expanding, but also market and technology environments are dynamic and constantly changing. Therefore, the results are visualised using modern business intelligence tools, which ensure an up-to-date, interactive presentation of results and facilitate further analysis.

 

We conduct concrete research on various methods, including the following:

  • Text mining for the automated development of large, diversified amounts of data
  • Natural Language Processing (NLP) and Disambiguation through Semantic Web / Linked Open Data in order to prepare and process data as well as to develop dynamic, semantic data structures
  • Application of Named Entity Recognition (NER), Named Entity Linking (NEL) and Entity Matching, also regarding previously unknown entities such as start-ups or new technologies
  • Sentiment analyses to capture customer attitudes
  • Topic modeling to identify weak trend signals
  • Scenario techniques for embedding identified trends in scenarios in a specific corporate context