MindNet

What it is MindNet is a Microsoft Research knowledge-representation project that automatically builds semantic networks from dictionaries, encyclopedias, and free text. Its pipeline sentence-splits and parses text into semantic dependency graphs, aggregates them, and weights subgraphs probabilistically by corpus frequency.

Key concepts Broad-coverage parsing; Logical Form semantic dependency graphs; graph aggregation; corpus-frequency weights; path search, sorting, and similarity.

How you’d use it Build or inspect a large semantic network derived from text, then search, rank, or compare paths for work aimed at world knowledge and common-sense reasoning; a small number of sample MindNets are available for online browsing.

LLM angle none stated

Pitfalls & lessons Only a small number of sample MindNets are described as available for online browsing; the page does not state a general download or current production interface.

Verdict A clear example of automatic, corpus-scale semantic-network acquisition, with the fetched page emphasizing the construction pipeline more than present-day access.

Sources consulted

• https://www.microsoft.com/en-us/research/project/mindnet/