A session is a bounded sequence of interactions associated with a user or client context during a period of activity. In search, it groups requests closely enough to interpret their shared context and behavior. section 4.1.4
Session identity sits between longer-lived identifiers and individual requests. A user ID may persist across devices; device and browser IDs persist across broader scopes; a session ID persists only through one session; a request ID lasts for one request. Cookies can link requests to sessions and sessions to browser identities. section 4.1.4
Search-session records commonly retain:
- A session identifier, query, ranked results, and a click indicator for each result. section 11.1.3
- The presented ranking, which may differ between sessions because the index or relevance algorithm changed. section 11.1.3
- Enough repeated session data to calculate result-level click-through rates and derive judgments for learning to rank. section 11.1.3
A search session as a data row group
For the query "transformers dark of the moon," session 3 records results at ranks 0 through 14 and whether each result was clicked. Grouping many such sessions by document produces click counts, session counts, and click-through rates. section 11.1.3
Listing 11.3 Examining a search session
query = "transformers dark of the moon" sessions = get_sessions(query) #1 print(sessions.loc[3]) #2
Inspect a session by querying the data for a search term and selecting one session identifier. section 11.1.3
