Feature Request: Personal content filters (mute tags and keywords, with wildcards)

[shaedrich]

TL;DR Grow this from a tag-only deny list into a private, viewer-only content filter: mute a tag or a keyword/phrase (with * wildcards), and hide matching posts from your feed, your recommendations, and (later) search. Silent and one-directional like per-follow mute, but topic-level instead of person-level.

Motivation

Muting a whole tag is blunt. Someone who never wants crypto spam should not have to hunt down every tag it hides behind (#crypto, #bitcoin, #web3), and the stuff they want gone is often in the post text, not a tag at all. So the filter should also match words in the post body, with wildcards, and treat a muted tag as just one kind of entry in the same list.

Where

• Feed: Vutuv.Posts.feed_page/2 (+ feed_post_items/3, feed_repost_items/3). Posts already carry tags via post_tags (Vutuv.Posts.PostTag, has_many :tags, through:), and the feed already drops muted follows (follows.muted) and blocks. This is the same shape one level up.
• Recommendations: "Who to follow" (Vutuv.Social.most_followed_users/1 / Vutuv.Social.PopularUsers) and tag suggestions (Vutuv.Tags).
• Search: VutuvWeb.SearchLive + Vutuv.Search (today has tag: / city: / @handle operators, no negative ones yet).
• Siblings to reuse, not reinvent: per-follow mute (follows.muted, <.mute_button>, the post ⋯-menu "Mute @handle"), block (Vutuv.Social), #hashtag / honor tags (Vutuv.Tags), and the inverse Follow tags (Feature Request: Follow tags #872).

What: one list, two kinds of entry

Each entry is either:

• Tag — an exact tag (crypto hides the tag and #crypto in a body).
• Keyword/phrase — free text matched against the post's rendered plain text, with * wildcards.

Match rules (this is what makes it useful):

• Case-insensitive.
• Default whole-word, so cess does not hide "success" (the classic false-positive trap).
• * = any run of characters, so the user opts into prefix/suffix/substring on purpose: crypto* → cryptocurrency, cryptobro; *coin → bitcoin, altcoin; *maxi* → anywhere.
• A phrase ("machine learning") matches those words adjacent, in order.
• Keyword entries match the post body and its tags/hashtags (muting the word crypto also catches #crypto); a Tag entry matches the tag only.

On a match, hide (do not vanish): collapse the post to a one-line placeholder ("Hidden: matches your filter crypto* · Show anyway") that reveals in place. A silently shorter feed confuses and breaks reply threads; a per-entry "remove completely" option can come later.

Never filter your own posts.

Private and one-directional: the list is yours alone. It never touches the author or anyone else, never notifies, never appears in the agent formats (.md/.txt/.json/.xml). It is sensitive (it reveals what a member dislikes: politics, health, identity), so it is owner-only, belongs in the GDPR export, and is never public.

UI sketch (a /settings page)

Muted words & tags                                   [ + Add ]
─────────────────────────────────────────────────────────────
 #crypto             tag                             Unmute
 crypto*             word · whole-word off           Unmute
 "machine learning"  phrase                          Unmute
 politics            word · until 12 Aug (snoozed)   Unmute
─────────────────────────────────────────────────────────────
 A hidden post shows a "Show anyway" line, never a gap.

Data model (new table)

Do not reuse the existing search_terms table: that is the fuzzy name-search index, unrelated. New table:

content_filters
  id          uuid v7  pk
  user_id     uuid v7  -> users        (owner; private)
  kind        enum     tag | keyword
  pattern     string   tag slug, or keyword/phrase with * wildcards
  whole_word  boolean  keyword only (default true; * overrides locally)
  expires_at  utc_datetime  null = permanent   (optional "snooze")
  timestamps

Entry points

• Settings: a new /settings/filters row ("Muted words & tags"), or folded into Privacy.
• Tag page (/tags/:slug): a "Mute this tag" button (the mirror of Feature Request: Follow tags #872's follow).
• Post ⋯ menu: "Mute #tag" per tag on the post plus a quick "Mute a word…", right beside the existing "Mute @handle".

Engineering notes (the sharp edges)

• Tags filter cheaply in the feed query (exclude posts whose post_tags intersect the muted set), exactly like the existing block/mute subqueries.
• Keywords cannot be a join: compile the viewer's patterns once per request into one regex (glob * → .*, word boundaries unless *), match the post's plain text (strip markdown and mentions so **crypto** still matches), filter in Elixir after hydrate_posts.
• Pagination hazard: filtering after the DB page yields short pages and a wrong "Load more" count. Push tag exclusion into the query and top up keyword-filtered pages inside the cursor loop (or over-fetch), do not just Enum.reject/2 the finished page.
• Cap the list size and pattern length, and validate the glob, so a pathological pattern cannot build a catastrophic regex.

Open questions / extensions

• Search: add negative operators (-word, -tag:x) to Vutuv.Search, and let the filter list pre-apply on /search. (This is the original "tag exclusion on the search page" idea, generalised.)
• Notifications about a muted-topic post: filter them too, or leave? (Lean: leave for v1.)
• A muted tag should also drop out of your own suggested-tags list. (Yes.)

Related

#872 (Follow tags, the inverse), per-follow mute (follows.muted), block (Vutuv.Social). Follow-up to #872.

Written with this GitHub account, but not necessarily by the human behind it — this may just be a note-taking issue that gets deleted later without ever being tackled.