Siemens – Natural Language Query

Saving $10m per year


Siemens increase sales and reduce service costs through Natural Language Queries of complex, heterogeneous enterprise data


Their Goal:

Provide a system to allow untrained users who have no knowledge of the underlying information sources to get answers to complex domain-specific questions using natural language.

Their Challenges:

  • Poor meta-data
  • Data sources: Dozens of internal and external sources are needed
  • Natural language query processing with an understanding of the domain-specific dictionaries, synonyms, line-of-business and region of business specific abbreviations, and shorthand notation.
  • Structured and un-structured data
  • Data growth – 800% annually growth in information
  • Real-time response to queries – 0.8 seconds

Their Solution:

The queries include semantic technology for contextual inferred understanding – for example “show me all active units in the a/p region” requires a contextual understanding of what the ‘units’ are, in this case generators and also to know the business users abbreviate the Asia Pacific region as ‘a/p’. A significant design goal to meet user expectations of responsiveness was to provide answers in less than 1 second regardless of complexity of the query or span of database sizes and locations

  • Allegrograph Semantic platform:
    • Energy domain specific concepts and terminology
      • 12,000+ regular expressions used to identify organization names, serial numbers embedded in text
      • Linked Open Data
        • To provide look-up sources of common information that resolve  contextual information in documents
        • Connections to public databases in semantic –
          • DBpedia
          • FreeBase
          • GeoNames

 The Benefits:

  • Reduce the time to find client information by 90%
  • Savings of $10mill in per year in direct and indirect savings
  • Better business insights into customer needs and potential sales

AllegroGraph turns complex data into actionable business insights