Semantic relatedness measures how strongly terms are associated by comparing their distributions across document sets. A term is related to a foreground query when it appears there more often than in an unrelated background collection. Common terms such as “the” are discounted despite frequent co-occurrence. section 5.4.4
A relatedness calculation typically uses:
- A foreground set containing documents matching the primary query. section 5.4.4
- A background set representing the whole collection or a random sample. section 5.4.4
- A normalized score, commonly mapped from −1.0 to 1.0, where 0.0 indicates no relation. section 5.4.4
Pain and ibuprofen
If the foreground set contains documents matching “pain,” “ibuprofen” receives a high score when it occurs there more often than in random documents. The same comparison can identify terms related to “advil,” including “motrin,” “aleve,” and “naproxen.” section 5.4.4
