| Title: | Session Reconstruction and Analysis |
|---|---|
| Description: | Functions to reconstruct sessions from web log or other user trace data and calculate various metrics around them, producing tabular, output that is compatible with 'dplyr' or 'data.table' centered processes. |
| Authors: | Os Keyes |
| Maintainer: | Os Keyes <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 2.0.3 |
| Built: | 2025-02-28 03:01:06 UTC |
| Source: | https://github.com/ironholds/reconstructr |
Calculates the "bounce rate" within a set of sessions - the proportion of sessions consisting only of a single event.
bounce_rate(sessions, user_id = NULL, precision = 2)bounce_rate(sessions, user_id = NULL, precision = 2)
sessions |
a sessions dataset, presumably generated with
|
user_id |
a column that contains unique user IDs. NULL by default; if set, the assumption will be that you want per-user bounce rates. |
precision |
the number of decimal places to round the output to - set to 2 by default. |
either a single numeric value, representing the percentage of sessions
overall that are bounces, or a data.frame of user IDs and bounce rates if
user_id is set to a column rather than NULL.
sessionise for session reconstruction, and
session_length, session_count and
time_on_page for other session-related metrics.
#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall bounce rate rate <- bounce_rate(sessions) # Calculate bounce rate on a per-user basis per_user <- bounce_rate(sessions, user_id = uuid)#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall bounce rate rate <- bounce_rate(sessions) # Calculate bounce rate on a per-user basis per_user <- bounce_rate(sessions, user_id = uuid)
sessionreconstruct provides functions to aid in reconstructing and analysing user sessions. Although primarily designed for web sessions (see the introductory vignette), its session approach is plausibly applicable to other domains.
Oliver Keyes <[email protected]>
link{session_count} counts the number of sessions in a sessionised
dataset, producing either a count for the overall dataset or on a per-user
basis (see below).
session_count(sessions, user_id = NULL)session_count(sessions, user_id = NULL)
sessions |
a dataset of sessions, presumably generated by
|
user_id |
the column of |
either a single integer value or a data.frame (see above).
#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall bounce rate count <- session_count(sessions) # Calculate session count on a per-user basis per_user <- session_count(sessions, user_id = uuid)#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall bounce rate count <- session_count(sessions) # Calculate session count on a per-user basis per_user <- session_count(sessions, user_id = uuid)
an example dataset of events, for experimenting with session reconstruction and analysis
session_datasetsession_dataset
a data.frame of 63,524 rows consisting of:
Hashed and salted unique identifiers representing 10,000 unique clients.
timestamps, as POSIXct objects
URLs, to demonstrate the carrying-along of metadata through the sessionisation process
The uuid and timestamp columns come from an anonymised dataset of
Wikipedia readers; the URLs are from NASA's internal web server, because space is awesome.
Calculate the overall length of each session.
session_length(sessions)session_length(sessions)
sessions |
a dataset of sessions, presumably generated with
|
a data.frame of two columns - session_id, containing unique
session IDs, and session_length, containing the length (in seconds)
of that particular session.
Please note that these lengths should be considered a minimum; because of how sessions behave, calculating the time-on-page of the last event in a session is impossible.
sessionise for session reconstruction, and
time_on_page, session_count and
bounce_rate for other session-related metrics.
#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate session length len <- session_length(sessions)#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate session length len <- session_length(sessions)
sessionise takes a data.frame of events
(including timestamps and user IDs) and sessionises them,
returning the same data.frame but with two additional columns -
one containing a unique session ID, and one the time difference between
successive events in the same session.
sessionise(x, timestamp, user_id, threshold = 3600)sessionise(x, timestamp, user_id, threshold = 3600)
x |
a data.frame of events. |
timestamp |
the name of the column of |
user_id |
the name of the column of |
threshold |
the number of seconds to use as the intertime threshold - the time that can elapse between two events before the second is considered part of a new session. Set to 3600 (one hour) by default. |
x, ordered by userID and timestamp, with two new columns -
session_id (containing a unique ID for the session a row is in)
and delta (containing the time elapsed between that row's event,
and the previous event, if they were both in the same session).
bounce_rate, time_on_page,
session_length and session_count - common metrics
that can be calculated with a sessionised dataset.
# Take a dataset with URLs and similar metadata and sessionise it - # retaining that metadata data("session_dataset") sessionised_data <- sessionise(x = session_dataset, timestamp = timestamp, user_id = uuid, threshold = 1800)# Take a dataset with URLs and similar metadata and sessionise it - # retaining that metadata data("session_dataset") sessionised_data <- sessionise(x = session_dataset, timestamp = timestamp, user_id = uuid, threshold = 1800)
time_on_page generates metrics around the mean (or median)
time-on-page - on an overall, per-user, or per-session basis.
time_on_page(sessions, by_session = FALSE, median = FALSE, precision = 2)time_on_page(sessions, by_session = FALSE, median = FALSE, precision = 2)
sessions |
a sessions dataset, presumably generated with
|
by_session |
Whether to generate time-on-page for the dataset overall (FALSE), or on a per-session basis (TRUE). FALSE by default. |
median |
whether to generate the median (TRUE) or mean (FALSE) time-on-page. FALSE by default. |
precision |
the number of decimal places to round the output to - set to 2 by default. |
either a single numeric value, representing the mean/median time on page
for the overall dataset, or a data.frame of session IDs and numeric values if
by_session is TRUE.
sessionise for session reconstruction, and
session_length, session_count and
bounce_rate for other session-related metrics.
#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall time on page top <- time_on_page(sessions) # Calculate time-on-page on a per_session basis per_session <- time_on_page(sessions, by_session = TRUE) # Use median instead of mean top_med <- time_on_page(sessions, median = TRUE)#Load and sessionise the dataset data("session_dataset") sessions <- sessionise(session_dataset, timestamp, uuid) # Calculate overall time on page top <- time_on_page(sessions) # Calculate time-on-page on a per_session basis per_session <- time_on_page(sessions, by_session = TRUE) # Use median instead of mean top_med <- time_on_page(sessions, median = TRUE)