Vector Similarity

• Vector similarity measures similarities between

  • 2 docs

  • A question and a doc

  • Doc within a genre

  • Docs with the same sentiment

Ad hoc Information Retrieval (IR)

• Find the doc that best matches the query

• Docs are unordered set

• Each term in the document represents a score

TFIDF = Common weight for vector

• Term frequency: number of times term t occurs in the document

• Inverse document frequency:

$$
log(\frac{# docs}{# of docs containing t})
$$

• TFIDF(t) = TF(t) X IDF(t)

  • Score terms highly that occur frequently in the doc or query

  • Vice versa

Evaluation Metrics: Precision, recall, F-measure

• System Output = answers from a system

• Answer Key = correct answers from humans

• Correct = length (System Output ∩ Answer Key)

• Precision = Correct ÷ length(System Output)

• Recall = Correct ÷ length(Answer Key)

• F = 2/1/precision + 1/recall