Vector Space Models in Information Retrieval and for Word Meaning

Question 1

Vector Space Models

Answer 1

Assume that meaning can be modeled by context

Represent the topic by counting how often each word occurs

Word counts represented as vectors

Similarity in meaning as proximity in space

Question 2

Applications of vector space models

Answer 2

• Web search

- Automatic thesaurus creation

Question 3

Euclidean distance

Answer 3

distance (P, Q)
= sqrt (sum from 1 to n (Pi-Qi)^2)

Does not normalize the number of counts

Question 4

Cosine similarity

Answer 4

Look at the angle between the vectors

Built-in normalization