Stars-894

+----------------+          +-------------------+          +-------------------+
|   Content      |  POST    |   Tag Suggestion |  GET     |   Taxonomy Service|
|   Editor UI    |--------->|   Service (Node) |--------->|   (REST API)      |
+----------------+          +-------------------+          +-------------------+
        ^                         ^   ^   ^
        |                         |   |   |
        |  GET (suggestions)      |   |   |
        +-------------------------+   |   |
                                      |   |
                               +------+---+------+
                               | NLP Model (Python)|
                               +-------------------+
  • NLP Layer – Lightweight transformer (e.g., distilbert-base-uncased) fine‑tuned for keyword extraction. Hosted in a separate Python container, exposing a gRPC endpoint ExtractKeywords(text) → [term, score].
  • Analytics – Simple event collector (/api/analytics/tag-suggestion) writes to the existing Snowflake event table.

    • | ID | As a… | I want to… | So that… | |----|-------|------------|----------| | STARS‑894‑US1 | Content author | See a list of suggested tags while I type | I can quickly add the most relevant tags without searching | | STARS‑894‑US2 | Content author | Accept or reject each suggestion with a single click or keyboard shortcut | I retain full control over the final tag set | | STARS‑894‑US3 | Content author | View why a tag was suggested (highlighted snippet) | I can trust the recommendation and understand its relevance | | STARS‑894‑US4 | Product analyst | Export acceptance/rejection data to the analytics dashboard | I can measure the impact of the feature and spot gaps in the taxonomy |

    | Risk | Impact | Mitigation | |------|--------|------------| | NLP model returns irrelevant terms (low precision) | Poor author trust, extra manual work | Start with a conservative confidence threshold (e.g., 0.75); adjust based on analytics. | | Increased latency on slow connections | Frustrating editing experience | Debounce, cache recent suggestions, fallback to “unavailable” banner after 1 s timeout. | | Taxonomy drift (new terms not in taxonomy) | Suggestions cannot be mapped → no UI display | Show raw term in tooltip with “Add to taxonomy” link (future ticket). | | Over‑reliance on suggestions leading to tag homogenization | Loss of niche tags | Encourage authors to add custom tags; keep the UI for free‑form tag entry. | | Service outage of NLP micro‑service | Feature unavailable | Graceful fallback; continue publishing without suggestions. | STARS-894

    If you want, I can:

    | Item | Details | |------|----------| | Report Title | STARS‑894: Mission Overview, Technical Assessment, and Preliminary Findings | | Prepared for | [Funding Agency / Stakeholder] | | Prepared by | [Project Team / Department] | | Date | 10 April 2026 | | Confidentiality Level | [Unclassified / Restricted / Confidential] | NLP Layer – Lightweight transformer (e

    | Goal | Benefit | |------|---------| | Reduce manual tagging effort | Faster authoring workflow; less repetitive work | | Increase tag accuracy & consistency | Better content organization, easier navigation, improved search relevance | | Boost SEO & content discoverability | More precise metadata leads to higher organic traffic | | Capture emerging topics automatically | Dynamically surface new terms that may need taxonomy expansion | | ID | As a… | I want

  • Keyboard – Arrow‑up/down to navigate, Enter to accept, Esc to dismiss.
  • Persisted State – Accepted tags are added to the article’s tag list instantly; dismissed suggestions disappear only for that session (they may re‑appear if the article is edited later).
    • меню