Sequential Pattern Mining

Description

Finds frequent ordered subsequences of actions across many sessions, optionally with gaps, to reveal common behavioural paths and those that discriminate outcomes.

Sequential pattern mining finds subsequences (not necessarily contiguous) that occur in at least a minimum proportion of sequences. Algorithms such as GSP, SPADE and PrefixSpan scale to millions of sessions. Constraints on gaps, window length and pattern length keep results meaningful. Contrast or discriminative pattern mining compares support between groups (e.g. converters vs abandoners) to find sequences characteristic of each. Compared with path diagrams, which show only contiguous next steps, this finds patterns like 'viewed pricing ... opened docs ... invited teammate' wherever they occur, giving hypotheses for qualitative follow-up.

Attributes

Loss type
Unmigrated
Stage
Post-launch
Effort
Days

Steps

  1. Encode sessions or user histories as sequences of abstracted events.
  2. Set minimum support (e.g. 2-5%) and constraints on maximum gap and pattern length.
  3. Run PrefixSpan or SPADE to extract frequent patterns.
  4. For outcome comparison, compute support in each group and rank patterns by lift or statistical tests, correcting for multiple testing.
  5. Filter redundant patterns (closed or maximal patterns) and review top patterns with session replays.
  6. Turn discriminative patterns into hypotheses and test them experimentally.

Sources

  • Srikant & Agrawal, Mining sequential patterns: Generalizations and performance improvements (1996), EDBT
  • Pei et al., PrefixSpan (2001), ICDE
  • Fournier-Viger et al., A survey of sequential pattern mining (2017), Data Science and Pattern Recognition

Also known as

  • PrefixSpan
  • GSP
  • SPADE
  • Frequent sequence mining

Categories

  • Analytics & Behaviour
  • Computational & Machine Learning