Knowledge Layer Building

Data sources

• NeedleSeek
• Probase
• Satori
• Freebase
• Raw web pages (from RetroIndex)
• Wikipedia
• ...

Classification of knowledge extraction tasks

• By knowledge type: entities, types, semantic classes, peer similarity, synonymy, antonymy, acronyms, is-a, part-of, attribute names, attribute values, general relations, events...
• By data source: plain sentences, web page contents, page URLs, anchor text, query set, click through, dictionaries, databases...
• By source scope:
  1. for web pages: all pages, pages in a subset of domains (e.g. Wikipedia pages)...
  2. for query set: all queries, queries containing "list of"...
• By techinuqe element: lexical pattern matching, tag pattern matching, co-occurrence calculation, wrapper induction, template generation, context analysis, distributional similarity, graph analysis, bootstrapping, classification, clustering...

Knowledge extraction task items

• IsA extraction from web pages with lexical patterns and produced knowledge iteratively
• Extraction attributes from Wikipedia and other data resources

Target knowledge layers

• Type system
• Attribute system
• Relatedness system

Common Utility Modules

• Common APIs for basic feature requirements
• Query parsing and exploration (QPE)
• Short text conceptualization
• Short text similarity calculation

Applications