Co-occurrence measures how often two terms appear in the same document, contextual unit, or user activity. In search, unusually frequent shared contexts can indicate semantic relatedness: “pain” and “ibuprofen” co-occur more meaningfully than either term and common words such as “the” or “and.” section 5.4.4 section 6.3.1
A practical co-occurrence pipeline turns search signals into related-term scores:
- Compare raw co-occurrence with individual term counts using pointwise mutual information (PMI), which reduces the influence of generally popular terms. section 6.3.1
Related product and topic queries
A query-log aggregation finds 244,876 keyword pairs searched by at least one common user. Examples include “iphone 4s” with “iphone,” “laptop” with “hp laptop,” and “bose” with “beats.” Raw counts identify popular pairs, while PMI distinguishes association from simple popularity; sparse pairs still require caution because low shared-user counts can introduce noise. section 6.3.1
