Description
Encodes each user session as a fixed-length numeric vector of counts, durations, sequences or learned embeddings, so sessions can be compared, clustered and modelled.
Sessions are variable-length sequences of events, but most analytical and generative models require fixed-length input. Representations range from hand-crafted aggregates (event counts, durations, depth, error counts), through n-gram and transition-matrix features capturing order, to learned embeddings from sequence models such as word2vec-style event embeddings, recurrent networks or transformers. The choice determines what similarity means: two sessions with the same events in different order are identical under counts but different under sequence representations. Rigour requires defining session boundaries consistently (e.g. 30 minutes inactivity), choosing representations that retain the behaviours of interest, checking that similar vectors correspond to qualitatively similar sessions, and documenting the encoding.
Attributes
- Discipline
- Computational & Machine Learning
- Loss type
- Latent
- Stage
- Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Define session boundaries consistently, commonly a 30-minute inactivity timeout or app foreground/background events.
- Select a representation: aggregates, bag of events, n-grams or transition probabilities, or learned embeddings.
- Compute vectors and normalise (e.g. TF-IDF weighting for event counts, standardisation for numeric features).
- Sanity-check by retrieving nearest-neighbour sessions for samples and reviewing them as replays.
- Evaluate representation quality on a downstream task (e.g. predicting task success) and compare alternatives.
- Document the encoding and version it alongside event schema changes.
Sources
- Deep Generative Modeling (Jakub Tomczak, Springer)
- Deep Learning (Ian Goodfellow, Yoshua Bengio & Aaron Courville)
- Methods Atlas (Vivek Gopinath)
- Grbovic et al. (2015) E-commerce in your inbox: product recommendations at scale
- Deep Generative Modelling (Springer)
- Deep Learning (Goodfellow)
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 53 — Generative Models & Synthetic User Testing
Also known as
- Session embedding
- Session vectorisation
Categories
- Analytics & Behaviour
- Computational & Machine Learning