会话划分与事件排序

A clickstream dataset records every user action: page views, button clicks, and purchases — each as a row with a user_id, session_id, event_type, and timestamp. Before computing any metrics, you must understand the schema. Run df.dtypes and df.sample(5) to see representative rows and confirm the timestamp column parsed as datetime rather than string.

Event analysis is meaningless if rows are not in time order. Sort by user_id and then timestamp so all events for the same user appear consecutively and in the correct sequence. Use sort_values(['user_id', 'timestamp']) and reset the index to get a clean integer sequence reflecting the sorted order.

If sessions are not pre-labelled, you can define them by a 30-minute inactivity gap: a new session begins whenever the time since the previous event for the same user exceeds 30 minutes. Use groupby('user_id')['timestamp'].diff() to compute the gap between consecutive events, then flag rows where the gap exceeds the threshold as session starts.

Session duration is the time between the first and last event in a session. Group by user_id and session_id and compute max - min on the timestamp column. Convert the resulting Timedelta to minutes with .dt.total_seconds() / 60 for a human-readable metric. Sessions with zero duration are single-event visits.

The number of events in a session is a proxy for user engagement. A session with 20 events likely indicates deeper exploration than one with 2. Use groupby(['user_id', 'session_id']).size() to count rows per group and name the result event_count. Merge this summary back to the sessions table for a complete session-level dataset.

The first event in a session shows the entry point (e.g. homepage vs. search results page) and the last event shows exit behaviour (purchase vs. bounce). Use groupby().first() and groupby().last() on the sorted DataFrame. These entry and exit columns are essential inputs for funnel and retention analysis.

Assigning a sequential event number within each session enables analysis of common event sequences. Use groupby(['user_id', 'session_id']).cumcount() + 1 to create an event_rank column. Filtering on event_rank == 1 gives the landing event; filtering on the maximum rank gives the exit event for each session.

Inter-event time — the gap between consecutive events within a session — reveals how quickly users move through the product. Compute it with groupby(['user_id', 'session_id'])['timestamp'].diff(). Large gaps between specific event types suggest hesitation or confusion at that step, which is a signal for UX improvement.

The number of sessions per user measures how often each user returns. Power users with many sessions drive retention KPIs. Compute sessions per user with df.groupby('user_id')['session_id'].nunique(). Segment users into one-time visitors, occasional users, and frequent users using pd.qcut() on the session count.

Understanding which days of the week generate the most sessions helps schedule product releases, email campaigns, and server scaling. Extract day_of_week from the session start timestamp and group by it. Mapping integer values (0=Monday, 6=Sunday) to day names makes the output more readable in reports and charts.

Combine all computed metrics — duration, event count, entry event, exit event, and day of week — into a single session-level summary DataFrame. This table is the foundation for funnel analysis, retention modelling, and machine learning feature generation. Each row represents one session rather than one raw event.

Test your understanding of Data Analysis concepts from this lesson.

In this lesson you learned: sorting events chronologically and inferring session boundaries from time gaps, computing session duration and event counts per session, and building a session-level summary table for downstream analysis. Next up we explore funnel analysis to measure conversion between product steps.