PHP
suaranetijen
SuaraNetijen mengindeks opini publik di internet terhadap suatu entitas, mengubah opini relevan menjadi sentimen positif/netral/negatif, lalu mengagregasikannya menjadi Sentimen Netijen dan ranking.
Detail proyek
suaranetijen
SuaraNetijen mengindeks opini publik di internet terhadap suatu entitas, mengubah opini relevan menjadi sentimen positif/netral/negatif, lalu mengagregasikannya menjadi Sentimen Netijen dan ranking.
README
SuaraNetijen
Sebelum pilih, cek kata netizen.
Search-first public sentiment index for Indonesia. Search a brand, product, or service and see whether public conversation about it leans positive or negative.
Situs langsung: suaranetijen.id · Pasar: Indonesia · Lisensi: MIT
Apa yang bisa dilakukan, dengan data nyata:
- Cari merek, produk, atau layanan dan lihat Sentimen Netijen (0-100) dari opini publik yang dikumpulkan dari forum, YouTube, dan portal ulasan, beserta jumlah opini dan sebaran positif/netral/negatif. Contoh: Samsung, ranking smartphone.
- Lihat tema yang paling sering dipuji dan dikeluhkan per entitas (frekuensi, tanpa skor per tema).
- Bandingkan dua merek berdampingan, mis. Oppo vs Vivo; dua skor ditampilkan terpisah dan tidak digabung.
- Rating Netijen (bintang dari pengguna) tetap metrik terpisah. Cara hitung dan batasnya ada di metodologi dan sumber data.
Stack: Laravel 13, Inertia v3 + Vue (SSR), PostgreSQL, Redis + Horizon, tersebar di satu host web dan tiga worker.
Backlog dan keputusan arsitektur ada di docs/; rencana SEO terkini di
docs/31-seo-plan.md.
Open source & lisensi
SuaraNetijen adalah proyek open source untuk publik. Kode sumber dan dokumentasi tersedia di GitHub dan kode proyek ini dirilis di bawah MIT License. Lisensi ini berlaku untuk kode dan dokumentasi repository; dependensi serta data/konten pihak ketiga tetap mengikuti lisensi dan ketentuan sumber masing-masing.
Product definition
SuaraNetijen indexes public opinion about an entity, turns each relevant opinion into a positive / neutral / negative classification, then aggregates that into Sentimen Netijen and category rankings.
Third-party content is processing input, not republished product. The durable assets are the entity graph, sentiment observations, historical aggregates, rankings, and first-party ratings.
Public metrics
| Metric | Range | Source |
|---|---|---|
| Sentimen Netijen | 0-100 | aggregated crawler sentiment |
| Top Suara Netijen | ranked themes | most-discussed themes, by frequency, from the same opinion data (docs/25) |
| Rating Netijen | 1-5 | first-party user star ratings |
| Jumlah Opini | count | relevant observations in the period |
| Distribusi Sentimen | % | positive / neutral / negative split |
| Trend Sentimen | time series | schema supports it; not required for first launch |
Sentimen Netijen, Top Suara Netijen, and Rating Netijen are never merged — provenance has to stay obvious for each. Top Suara Netijen is theme frequency, never a numeric per-theme score — it is not aspect scoring.
Score formula v1
score = 100 * (positive_count + 0.5 * neutral_count) / opinion_count
Equivalent to positive=100, neutral=50, negative=0. A score goes public at >= 30 opinions and
becomes ranking-eligible at >= 100 (examples/score-config.yaml).
MVP frozen scope
- ~200 seed entities, types
brand/product/service. - Search-first, mobile-first, SEO-first.
- Laravel modular monolith, Inertia + Vue + Tailwind.
- PostgreSQL as source of truth and as the MVP search engine (
pg_trgm+ FTS). - Redis from day one for queue, cache, locks, and rate limits; Horizon for queue observability.
- Source adapters: YouTube, KASKUS, DiskusiWebHosting, SerayaMotor, IndoForum (selective), Bluesky, LowEndTalk (targeted), MediaKonsumen, Mojok.co (esai), plus first-party ratings.
- Top Suara Netijen: theme extraction/normalization and Top 5 frequency ranking per entity
(
docs/25) — no numeric per-theme score, no per-category aspect taxonomy. - Out: specification database, price comparison, aspect scores, expert scores, benchmarks, verified purchase, opinion fact checking, and political entities.
Stack as installed
| Layer | Installed |
|---|---|
| PHP | 8.4 (composer requires ^8.3) |
| Framework | Laravel 13 |
| Frontend | Inertia v3 + Vue 3.5 + Tailwind 4, Vite 8 |
| Auth | Laravel Fortify (incl. two-factor) |
| Routing types | Laravel Wayfinder |
| Database | PostgreSQL |
| Queue / cache | Redis + PhpRedis |
| Queue operations | Laravel Horizon 5.x |
| Tests | Pest |
| Lint / static analysis | Pint, PHPStan, vue-tsc |
Horizon runs the documented supervisor-critical, supervisor-crawl, supervisor-analysis, and
supervisor-maintenance groups. The test suite keeps SQLite/array/sync defaults unless an explicit
integration run supplies PostgreSQL and Redis settings.
Getting started
composer setup # install deps, .env, key, migrate, npm install, build
composer dev # serve app, queue worker, and Vite together
Other useful scripts:
composer test # config clear + pint check + phpstan + artisan test
composer lint # pint --parallel
composer types:check # phpstan
composer ci:check # npm check + vue-tsc + test
Current implementation status
Phase 0 (foundation), Phase 1 (search), Phase 2 (sentiment substrate), Phase 3 (first
observations), Phase 4 (public score), Phase 5 (coverage expansion), Phase 6 (first-party
rating), and Phase 7 (public launch readiness) are implemented and verified against real
PostgreSQL and Redis, not just SQLite tests.
Admin access uses an authenticated access-admin Gate,
non-admin users receive 403, the ~200-entity seed CSV imports cleanly (209 entities), and local
lint, static analysis, and tests pass.
Search (/, /search, GET /api/search) implements PRD acceptance criteria 1 and 2 — typo and
multi-word matching — verified against live seed data. One tracked gap: full-text search on
name/category/description (docs/13, ADR-004) is not implemented, only pg_trgm similarity,
exact/prefix matching, and token-based matching.
Epic 3 (sentiment data model) and Epic 4 (adapter framework) implement docs/17's Definition of
Done exactly: sentiment_observations rejects a duplicate (entity_id, source_item_id) at the
database level (verified live on Postgres); the aggregate/ranking services reproduce the
60/20/20 -> 70.0 worked example from docs/11; FakeSourceAdapter runs
discover -> fetch -> extract -> temp storage -> expiry end-to-end via queued jobs, and one
adapter's simulated failure doesn't affect another running in parallel; SourceRateLimiter goes
through real Redis at runtime. A gap was found and fixed during this verification pass: public
score / ranking thresholds and the formula version were hard-coded PHP constants instead of
reading examples/score-config.yaml — now sourced from config/scoring.php
(SCORING_PUBLIC_MIN_OPINIONS, SCORING_RANKING_MIN_OPINIONS, SCORING_FORMULA_VERSION env
overrides).
Epic 5 (source adapters wave 1) and the Phase-3 slice of Epic 7 (entity matching, relevance,
sentiment) implement docs/17's Definition of Done: DiskusiWebHostingAdapter,
SerayaMotorAdapter, IndoForumAdapter, and BlueskyAdapter all reach healthy preflight and
produce at least one CandidateOpinion against fixtures (never live network, per docs/22);
EntityMatcher never lets a brand match cascade to a service observation; unresolved or
non-evaluative mentions go to unmatched_mentions, never a best-guess entity;
UpsertSentimentObservationJob is idempotent on (entity_id, source_item_id), confirmed live on
Postgres. Two gaps were found and fixed during this verification pass: IndoForumAdapter didn't
actually enforce its forum-ID allowlist (any numeric ID would still crawl); and the sentiment
classifier had never been evaluated against docs/22's Indonesian test set as the Epic 7 DoD
requires — now measured at 16/16 on a curated formal/slang/typo/mixed-English/emoji/negation/
sarcasm set, which also surfaced and fixed an emoji-stripping bug that made emoji-only sentiment
always return null.
Epic 8 (public score) and Epic 12 (Top Suara Netijen) implement docs/17's Definition of Done:
PRD acceptance criteria 3-4, 8, 11, and 12 pass — score only shown at >= 30 opinions, deterministic
recomputation from raw counts (60/20/20 -> 70.0), /top/{slug} reuses the Epic 3 ranking query
unchanged (score desc, opinion_count desc, name asc, no popularity bonus), no AggregateRating
schema.org markup, and Top Suara Netijen shows theme frequency with the correct empty-state copy
below threshold, no numeric per-theme score. A gap was found and fixed during this pass:
ThemeExtractor derived a theme's sentiment purely from its canonical-key suffix with no negation
handling, mis-stamping negated mentions (e.g. "servernya gak cepat sama sekali") as positive —
fixed with an Indonesian negation-marker check. The noindex gap noted here for below-threshold
entities was closed in Phase 7 (Epic 10).
Epic 6 (source adapters wave 2) implements docs/17's Definition of Done: KaskusAdapter,
YouTubeAdapter, and LowEndTalkAdapter all reach healthy preflight and produce at least one
CandidateOpinion against fixtures, never live network; KaskusAdapter and YouTubeAdapter each
report an explicit policy_disabled state (robots disallow-all for KASKUS, missing
YOUTUBE_API_KEY for YouTube) instead of guessing; all three sources are seeded enabled: false
so wave 2 stays off until an operator records a live preflight result. A gap was found and fixed
during this pass: KaskusAdapter and LowEndTalkAdapter both accept an operator-supplied
thread_urls override for targeted backfill, and resolved it with no host check — unlike the
neighboring forum_ids/category_urls allowlists, an absolute URL on any host would have been
crawled as-is. Fixed with a shared isSameHost() guard on AbstractHttpSourceAdapter.
Epic 9 (Rating Netijen) implements docs/17's Definition of Done: PUT/DELETE /api/entities/{id}/rating upsert on (user_id, entity_id), never insert-only, with
RatingAggregator recomputing rating_snapshots synchronously on every write and a regression
test confirming rating writes never alter sentiment_snapshots (ADR-007/011). docs/12's minimum
anti-abuse list (auth, rate limit, CSRF, unique constraint, burst logging, account ban) is
implemented; a gap was found and fixed during this pass — "account ban/admin disable" was the one
item from that list with no implementation at all, so users.is_banned was added and enforced in
the rating request's authorize(). No admin UI toggles the flag yet (Epic 11's admin views cover
crawl/source operations, not users). The unique constraint was confirmed to reject a duplicate
insert with UniqueConstraintViolationException live on Postgres.
Epic 10 (UX/SEO) and Epic 11 (Operations) implement docs/17's Definition of Done: noindex
meta on below-threshold entity pages plus a /sitemap.xml that excludes the same entities, the
category page and static/trust pages, admin diagnostics (crawl_states, ingestion_failures,
unmatched_mentions) with failure replay, a source kill switch that stops discovery/fetch with no
deploy, and encrypted daily Postgres backups plus a scheduled monitor:metrics alert command.
Three gaps were found and fixed during this pass: the sitemap's eligibility check wasn't scoped to
the same 365d/all periods used for lastmod, so a 30d/90d-only eligible entity could slip
through inconsistently; a discovery-stage ingestion failure had no retry path in the admin panel
(neither a document nor an item exists yet at that stage) and silently marked itself resolved
without re-queuing anything; and docs/16's monthly restore-test requirement had no scheduled
--verify run. Known gaps, not fixed here: backups are local-disk only, and PRD acceptance
criterion 9 (360px usability) was checked by pattern consistency with already-verified pages, not
a live Lighthouse/browser run.
The Admin, Entities, Search, Sources, Ingestion, Sentiment, Themes, and Ratings
domain modules are present. Implementation order lives in docs/17-implementation-backlog.md.
Phase status
Phases and epic mapping match docs/18-roadmap.md.
| Phase | Epics | Status |
|---|---|---|
| 0. Foundation | 0, 1 | Done |
| 1. Findability | 2 | Done |
| 2. Sentiment substrate | 3, 4 | Done |
| 3. First observations | 5, 7 (partial) | Done |
| 4. Public score | 8, 12 | Done |
| 5. Coverage expansion | 6 | Done |
| 6. First-party rating | 9 | Done |
| 7. Public launch readiness | 10, 11 | Done |
Documentation
docs/ holds 25 numbered documents and is the source of truth for product, schema, and
architecture decisions. CLAUDE.md / AGENTS.md carry the condensed version for coding agents.
| File | Purpose |
|---|---|
docs/01-product-vision.md |
Vision, positioning, product principles |
docs/02-prd-mvp.md |
MVP PRD, user stories, acceptance criteria |
docs/03-scope-entity-taxonomy.md |
Entity types, hierarchy, categories |
docs/04-ux-information-architecture.md |
Search-first IA, mobile UX rules |
docs/05-technical-architecture.md |
Stack and modular monolith layout |
docs/06-domain-data-model.md |
Tables and raw-content policy |
docs/07-source-strategy-registry.md |
Source selection, roles, governance |
docs/08-source-adapter-specs.md |
Per-adapter contract and health states |
docs/09-crawler-ingestion-indexer.md |
Pipeline, dedupe, incremental crawl |
docs/10-entity-matching-opinion-filter.md |
Entity resolution and relevance filter |
docs/11-sentiment-scoring-ranking.md |
Score formula, thresholds, ranking |
docs/12-first-party-rating-moderation.md |
Star rating and anti-abuse minimum |
docs/13-search-seo.md |
Search ranking, SEO page model, sitemap |
docs/14-api-jobs-scheduler.md |
Endpoints, queues, jobs, schedule |
docs/15-security-privacy-governance.md |
Security, privacy, source guardrails |
docs/16-observability-deployment.md |
Deployment, metrics, logs, backup |
docs/17-implementation-backlog.md |
Epics, order, launch gate |
docs/18-roadmap.md |
Product, source, and scale roadmap |
docs/19-success-metrics.md |
Product, index, and source KPIs |
docs/20-risk-register.md |
Risks and mitigations |
docs/21-architecture-decisions.md |
ADR-001..011, frozen decisions |
docs/22-testing-strategy.md |
Unit, adapter fixture, NLP, E2E strategy |
docs/23-seed-entity-strategy.md |
~200 seed entity plan |
docs/24-current-reference-baseline.md |
Externally validated facts (2 Sep 2026) |
docs/25-top-suara-netijen.md |
Theme Index / Top Suara Netijen: pipeline, data model, ranking, scope |
Config examples live in examples/: score-config.yaml, source-registry.yaml,
queue-topology.yaml, .env.example.
Implementation rule
Build a modular monolith. Separate a component only after a measured bottleneck or ecosystem requirement makes separation clearly beneficial.
No Kafka, Kubernetes, dedicated search cluster, separate Python service, or microservices in the MVP without evidence of need.