The Complete Guide to Recommendations and Library Discovery
A practical discovery system for moving among related suggestions, direct search, and category browsing while preserving version, availability, and viewer intent.
In short: Begin with a question, then choose the right route. Use search for a known title or attribute, category browsing for a defined range, and recommendations for a related but open-ended path. Validate why a suggestion fits, confirm its exact version and availability, and stop after a small number of meaningful hops with a focused shortlist.
Library discovery should end in a confident viewing decision or a useful shortlist. It should not become an endless sequence of attractive cards whose relationships remain unexplained.
Understand the three discovery routes
| Route | Best starting condition | Useful outcome |
|---|---|---|
| Search | You know a title, person, term, or attribute | Direct retrieval |
| Browse | You know the range, such as a category or year | Structured comparison |
| Recommendations | You have a relevant seed title but want alternatives | Related exploration |
Use the route-selection comparison when the starting condition is unclear. A viewer can switch routes, but each switch should solve a problem: search identifies the seed, browsing broadens the field, and recommendations reveal relationships.
Know what a recommendation can mean
Norva publicly describes recommendations derived from the compatible source connected by the user. The product does not provide the media library itself. Available suggestions therefore depend on the source, its media, and its metadata.
A suggestion may appear related by genre, creator, series relationship, release context, language, or another available field. Do not claim a particular internal weighting or a global recommendation system without verified product evidence. Source-based recommendations explains the safe scope in detail.
Start with a discovery question
Write a one-sentence brief before opening the first related row:
- “Find a film from the same period with a different genre.”
- “Find another series with the same creator, if metadata supports it.”
- “Find a version with the required language and subtitles.”
- “Find an older title related to this recent seed.”
The brief gives every candidate a pass or fail condition. “Show me something good” cannot be audited because “good” has no shared criteria.
Inspect the relationship evidence
For each candidate, record the seed, candidate, visible shared fields, meaningful differences, and confidence. DCMI metadata terms distinguish title, creator, subject, date, type, format, language, and relation. EIDR’s public work describes hierarchies among works and versions. These frameworks help readers separate a work-level relationship from a version-level match.
Do not infer a relationship from artwork proximity alone. Use the related-title evaluation method to state what is visible and what remains unknown.
Verify the candidate before shortlisting
Open the detail page without starting playback where possible. Confirm exact title, media type, release context, series or episode relationship, source availability, version, language, subtitles, and intended viewer. A recommendation can be conceptually relevant while the available version does not meet the session’s needs.
Languages, subtitles, offline access, and media availability depend on the source, media, supported device, and associated rights. Discovery does not override those conditions.
Use a three-hop stop rule
A hop moves from one seed to a related candidate. Allow up to three meaningful hops in one session:
- choose a verified seed;
- inspect one related set;
- select the most promising candidate as the next seed only when the relationship adds a new direction;
- stop after the third hop and compare the candidates already collected.
This is a decision aid, not a universal usability limit. It protects against loops where each card merely repeats the same genre or franchise. A later guide explains the complete three-hop discovery session.
Turn discovery into a shortlist
Keep only candidates that answer the brief and are sufficiently verified. For each one, write a short reason, a version or access constraint, and the intended viewer or occasion. Separate “interesting to investigate” from “ready to watch.”
Do not add every suggestion to favorites. A focused shortlist is temporary; a favorite represents a durable retrieval or preference purpose. Review the shortlist after the session and remove candidates whose question has been answered.
Audit relevance without pretending to know the algorithm
Sample a small set of suggestions and classify each relationship as visible, plausible but incomplete, or unsupported by available metadata. Record which fields were present and which would have helped. Use the recommendation relevance audit for a reproducible method.
An irrelevant suggestion is still evidence: it may reveal an overly broad seed, sparse metadata, a version mismatch, or a discovery question that needs refinement. It does not prove a specific internal fault.
Original evidence: route-and-hop card
Create a card with six fields: Question, Route, Seed, Candidate, Relationship Evidence, Decision. Run one session with no more than three hops and ask another reviewer to explain why each candidate survived.
Record how many candidates were ready, needed investigation, or failed the brief. These counts describe one library session; they are not product performance statistics. The card’s value is a transparent path from intent to shortlist.
Common mistakes and limitations
- Starting without a discovery question.
- Treating card order as evidence of relationship strength.
- Claiming an internal recommendation method without proof.
- Ignoring version, language, or subtitle differences.
- Saving every candidate as a durable favorite.
- Following related rows until the original brief is forgotten.
- Assuming incomplete source metadata can be repaired by the discovery interface.
Frequently asked questions
Are recommendations better than search?
Neither is universally better. Search retrieves known targets; recommendations explore relationships; browsing compares a defined range.
How many related titles should I inspect?
Use the smallest set that answers the brief. The three-hop rule limits drift, not the exact number of visible cards.
Why can an apparently relevant title be unusable?
The available version, language, subtitles, device context, or source rights may not meet the viewer’s requirements. Verify readiness separately from relevance.
Your next step
Explore Norva's Product Experience