Bangda Sun

Practice makes perfect

Stanford NLP (Coursera) Notes (19) - Question Answering

Introduction to Question Answering.

1. Paradigms for QA

Information Retrieval based approaches - workflow:

  • Question processing
    • detect question type, answer type (taxonomy), focus, relations, named entites
    • formulate queries to send to search engine
  • Passage retrieval
    • retrieve ranked documents
    • break into suitable passages and rerank
  • Answer processing
    • extract candidate answers
    • rank candidates

There are detailed discussions about answer type detection, passage retrieval, answer extraction (processing, features, machine learning opportunities) in the original slides.

Knowledge based approaches - workflow:

  • Build semantic representation of the queries
    • datetime, location, entities, numerical quantities
  • Map from semantics to query structure
    • geospatial databases
    • ontologies (wikipedia infoboxes, etc)
    • restaurant review sources
    • scientific databases

2. Evaluation Metrics

In addition to commonly used classification metrics in machine learning, mean reciprocal rank also can be used

\[
MRR = \frac{1}{N}\sum^{N}_{i=1}\frac{1}{rank_{i}},
\]

where \(rank_{i}\) is the rank of correct answer for query \(i\).