Extend AI Related Posts via Filters — Init Live Search

Architecture Overview

• Candidate Pool: default set includes latest posts and random posts from the same category.
• Signals: features used for scoring such as category, tag, comment, views, title bigrams, recency, time gap.
• Weights: normalized weights (sum = 1) applied to each signal.
• MMR Diversification: reduces redundancy with Max Marginal Relevance.
• Cache: results cached with an algo version key for smooth invalidation.
• Filters: extension points, all prefix-consistent and non-breaking.

Supported Filters

• init_plugin_suite_live_search_ai_candidates: add or remove candidates.
• init_plugin_suite_live_search_ai_signals: add or override signals.
• init_plugin_suite_live_search_ai_weights: adjust weights.
• init_plugin_suite_live_search_ai_score: tweak the final score per candidate.
• init_plugin_suite_live_search_ai_selected: post-process the selected list.
• init_plugin_suite_live_search_ai_half_life_recency: half-life for recency signal.
• init_plugin_suite_live_search_ai_half_life_gap: half-life for time-gap signal.
• init_plugin_suite_live_search_ai_mmr_lambda: lambda factor for MMR relevance/diversity.

Example Extension: Candidates, Signals, and Weights

The snippet below demonstrates adding candidates from same_keyword (meta) and series (taxonomy), enriching the scoring signals, and adjusting weights accordingly.

// + Add candidates: same series (random) + same_keyword (meta_query)
// + Merge with existing candidates
add_filter('init_plugin_suite_live_search_ai_candidates', function($candidates, $post_id, $post_type){
    $candidates = is_array($candidates) ? $candidates : [];

    // ----- SAME_KEYWORD -----
    $same_kw_raw = function_exists('get_field')
        ? (string) get_field('same_keyword', $post_id)
        : (string) get_post_meta($post_id, 'same_keyword', true);

    if ($same_kw_raw !== '') {
        $tokens = array_filter(array_map('trim', preg_split('/[,;|]+/u', $same_kw_raw)));
        if (!empty($tokens)) {
            $meta_query = ['relation' => 'OR'];
            foreach ($tokens as $t) {
                $meta_query[] = [
                    'key'     => 'same_keyword',
                    'value'   => $t,
                    'compare' => 'LIKE',
                ];
            }
            $kw_pool = get_posts([
                'post_type'      => $post_type,
                'post_status'    => 'publish',
                'posts_per_page' => 100,
                'post__not_in'   => [$post_id],
                'fields'         => 'ids',
                'meta_query'     => $meta_query,
                'no_found_rows'  => true,
            ]);
            $candidates = array_merge($candidates, (array)$kw_pool);
        }
    }

    // ----- SERIES (light) -----
    $series_terms = wp_get_post_terms($post_id, 'series', ['fields' => 'ids']);
    if (!empty($series_terms) && !is_wp_error($series_terms)) {
        $series_pool = get_posts([
            'post_type'      => $post_type,
            'post_status'    => 'publish',
            'posts_per_page' => 30,
            'post__not_in'   => [$post_id],
            'fields'         => 'ids',
            'orderby'        => 'rand',
            'tax_query'      => [[
                'taxonomy' => 'series',
                'field'    => 'term_id',
                'terms'    => $series_terms,
            ]],
            'no_found_rows'  => true,
        ]);
        $candidates = array_merge($candidates, (array)$series_pool);
    }

    return array_values(array_unique(array_map('intval', $candidates)));
}, 10, 3);

add_filter('init_plugin_suite_live_search_ai_signals', function($signals, $post_id, $candidate_id){
    // SAME_KEYWORD
    $get_kw = function($pid){
        $raw = function_exists('get_field') ? (string) get_field('same_keyword', $pid)
                                            : (string) get_post_meta($pid, 'same_keyword', true);
        return $raw !== '' ? array_filter(array_map('trim', preg_split('/[,;|]+/u', $raw))) : [];
    };
    $src = $get_kw($post_id);
    $dst = $get_kw($candidate_id);

    $kw_score = 0.0;
    if ($src && $dst) {
        $inter = array_intersect($src, $dst);
        $kw_score = count($inter) / max(count($src), 1);
    }
    $signals['same_keyword'] = $kw_score;

    // SERIES (binary, light)
    $src_series = wp_get_post_terms($post_id, 'series', ['fields' => 'ids']);
    $dst_series = wp_get_post_terms($candidate_id, 'series', ['fields' => 'ids']);
    $signals['series'] = (!empty(array_intersect($src_series, $dst_series))) ? 1.0 : 0.0;

    return $signals;
}, 10, 3);

add_filter('init_plugin_suite_live_search_ai_weights', function($weights){
    $weights['tag']            = 0.25;
    $weights['series']         = 0.20;
    $weights['title_bigrams']  = 0.15;
    $weights['same_keyword']   = 0.15;
    $weights['category']       = 0.08;
    $weights['views']          = 0.07;
    $weights['comment']        = 0.05;
    $weights['freshness']      = 0.05;
    return $weights;
}, 10, 1);

Step-by-Step Guide

1. Open the theme/plugin file where you want to add the filters.
2. Paste the example code into a suitable PHP file (e.g. an extension module or functions.php).
3. Ensure the series taxonomy and same_keyword meta field exist and contain data.
4. Adjust posts_per_page limits to balance quality and performance.
5. Tune the weights in ..._ai_weights filter to achieve the desired ranking.

Performance and Cache

• Prime caches: core calls _prime_post_caches and update_object_term_cache to avoid N+1 queries.
• Pool limits: keep reasonable candidate limits for meta/tax queries to avoid DB load.
• TTL: results are cached via transient with TTL configurable using ..._ai_cache_ttl filter.
• Algo version: bump $algo_ver when making significant logic changes for smooth cache invalidation.

Advanced Ranking Tweaks

• Use init_plugin_suite_live_search_ai_score to adjust final scores, e.g., boosting posts with high conversions.
• Use init_plugin_suite_live_search_ai_selected to exclude or reorder specific posts after ranking.
• Adjust init_plugin_suite_live_search_ai_mmr_lambda to balance relevance vs. diversity.

Testing and Measurement

• Create test sets with posts sharing series, keywords, and categories to observe ranking changes.
• Compare CTR/Time on Page before and after enabling custom filters.
• Monitor logs and performance metrics for candidate pool size and processing time.

Implementation Notes

• Always check the existence of taxonomy/meta before querying.
• Return only IDs and set no_found_rows to reduce query overhead.
• Exclude $post_id from candidates and keep only published posts.
• Normalize and deduplicate with array_unique and intval.

Conclusion

With its open filter system and cache-friendly design, AI Related Posts in Init Live Search enables flexible extension while maintaining high performance. Start with the example in this guide, then fine-tune signals and weights to match your content strategy.