Concepts
RSSMonster combines a conventional feed reader with optional analysis and recommendation features. Understanding the following distinctions helps explain why an article appears, how it is grouped, and what changes when you read it.
Articles and revisions
An article is an entry collected from a syndication feed or an HTML + XPath source. Stable publisher identity is checked before duplicate-content matching. When a publisher updates an existing entry, RSSMonster can revise the stored article while preserving read state, bookmarks, clicks, and manual tags. A metadata correction or extraction repair is not necessarily a new revision.
RSSMonster stores source and presentation data separately. Raw source content (contentOriginal), sanitized presentation HTML (contentHtml), canonical plain text (contentText), and description fields have distinct purposes. Historical contentStripped data is not interchangeable with all of these representations. API JSON omits raw contentOriginal; use the returned normalized content fields.
Duplicates, events, and interests
| Concept | What it represents | Example |
|---|---|---|
| Duplicate | Another record representing the same content, based on deterministic identity/content evidence | A syndicated copy |
| Event | Coverage of a particular occurrence | Several reports about one product announcement |
| A broader semantic theme connecting events or behavior | Developments in battery technology | |
| Interest Island | A recurring personal interest inferred from engagement | The user’s sustained interest in electric transport |
Articles about the same event can provide different reporting and remain distinct. Semantic similarity alone is not duplicate evidence. Likewise, an event and a personal interest are not interchangeable “clusters.” Event grouping can show a representative article and continuing developments while keeping coverage available through story-source controls.
Read-state behavior depends on the selected view and grouping. See Marking Articles Read and grouped-event reading; reading one article does not universally mark every related article read.
Quality, originality, and trust
Article quality combines stored quality, sentiment, and advertising scores into a normalized 0–1 value. Higher advertising scores mean less promotional content. AI analysis supplies estimates when enabled; otherwise default values can be used. These are ranking signals, not verification of facts.
Originality evidence in FeedTrust comes from actual duplicate links. It does not measure how much novel reporting every article adds to its semantic event. FeedTrust combines recent article quality, exposure-backed engagement, deterministic originality, and explicit negative feedback. Sparse evidence is pulled toward a neutral value of 0.75.
FeedTrust is user-specific source history. It is separate from crawl health and personal affinity. A reliable HTTP endpoint is not necessarily a valuable source, and a low personal interest match is not necessarily a low-quality article.
Ranking and filtering
Ranking orders an eligible collection. Newest and Oldest order by date; Quality combines article quality with FeedTrust; Recommended incorporates personal relevance; Top Stories emphasizes recent event coverage.
Filtering decides which articles qualify. Score thresholds, the selected status, feed/category scope, and search expressions can exclude articles before you see the ranking. Grouping can further reduce visible rows without deleting coverage. Chronological ordering alone therefore does not remove every active filter.
Reading feedback
Read state, exposure, clicks, bookmarks, and explicit More like this or Not interested feedback provide different signals. Interest Islands use behavioral evidence for personal relevance; FeedTrust uses its own bounded engagement model. Fresh installations may have little useful personal evidence, and zero related recommendations is a valid outcome.
Organization and automation
Categories organize subscriptions. Tags label articles. Smart Folders store queries, while generated feeds expose query results as RSS. These features complement one another.
Actions match article fields with regular expressions during crawling. They can tag, bookmark, change state, override scores, or store filtered articles. Discarding through an Action is not deletion. Feed item filters control acceptance of future entries from a particular feed; search filters operate on stored articles.
Processing and availability
Crawling extracts and normalizes entries, resolves identity and revisions, applies duplicate/filter rules, and persists results. Optional processing adds analysis, vectors, and semantic organization. Work is split across web, crawl-worker, and AI-worker processes; see How RSSMonster Works.
The default SQLite deployment is a lightweight reader. The MySQL deployment supports the background AI workflow. Per-feed switches cannot override disabled server capabilities. Model choice and provider availability affect semantic results, so inspect the configuration before interpreting missing summaries or recommendations as missing articles.