AI-Powered Related Posts in Init Live Search: From Bigram Title Search to Multi-Signal Scoring

Before 1.8.1: Bigram Title Search

• Title parsing into bi-grams, cleaned and normalized by locale.
• search-core ran weighted queries (title, title_tag, title_excerpt…), merging results by score.
• Deterministic output, consistent and strong for semantically rich titles.

Limitations: heavily title-dependent; lacked time and engagement signals; low diversity when many posts were near-duplicates.

1.8.1: Introducing [init_live_search_related_ai] (multi-signal)

• Signals: tag, series, title bigrams (cosine), same_keyword (ACF), category, views, comments, freshness.
• Candidate pool: expanded from recent posts, same series, and same_keyword metadata; extendable via filters.
• Templates: reuses existing related-posts-*.php templates, no new CSS required.
• Filter API: ai_candidates, ai_signals, ai_weights, ai_score.

1.8.2: AI Related Posts Engine v2

• Algorithm v2 with explicit cache versioning for safe invalidation.
• Freshness replaced by two distinct signals: recency (vs. now) and time_gap (vs. source post).
• MMR diversification: balances relevance and diversity, avoiding near-duplicate results.
• Performance: auto-filter publish status, deduplicate, pre-cache posts/meta/terms, optimized scoring loop with safe weighted RNG.
• New filters: ai_half_life_recency, ai_half_life_gap, ai_mmr_lambda, ai_selected.

Quick comparison

• Legacy Related (related_posts): bigram-based, precise when titles are strong, deterministic and fast.
• AI Related (related_ai): multi-signal, context-aware (taxonomy, engagement, time), results both relevant and diverse.

Usage

Bigram (legacy, still supported):

[init_live_search_related_posts count="6" template="default"]

AI (recommended):

[init_live_search_related_ai count="6" post_type="post" template="default" css="1" schema="1"]

• count: number of posts
• post_type: post types
• template: reuses existing related-posts-*.php
• css/schema: toggle CSS and schema output

Developer filters (core)

• init_plugin_suite_live_search_ai_candidates: add/replace candidate pool (ACF, taxonomy…)
• init_plugin_suite_live_search_ai_signals: extend/override signals (e.g., same author boost)
• init_plugin_suite_live_search_ai_weights: adjust weights (core normalizes)
• init_plugin_suite_live_search_ai_score: tweak final candidate score
• init_plugin_suite_live_search_ai_half_life_recency, init_plugin_suite_live_search_ai_half_life_gap: control decay of time-based signals
• init_plugin_suite_live_search_ai_mmr_lambda: balance relevance vs. diversity
• init_plugin_suite_live_search_ai_selected: override or reorder final selection after MMR

Migration notes (1.8.1 → 1.8.2)

• If you previously weighted freshness, switch to recency and time_gap in ai_weights.
• Templates and shortcode builder remain unchanged; no UI updates required.
• Tune ai_mmr_lambda to increase or reduce output diversity.

Conclusion

AI-Powered Related Posts takes related content beyond bigram matching: multi-signal scoring, time-aware decay, and MMR diversification. With [init_live_search_related_ai], you keep your templates, gain more accurate and diverse results, and have complete extensibility through filters—all with minimal overhead.