modularize abstract video list header and implement video hotness recommendation variant

1 files changed, 38 insertions, 4 deletions

diff --git a/server/models/video/video-query-builder.ts b/server/models/video/video-query-builder.ts
index 9e5b6febb..65b72fe1c 100644
--- a/server/models/video/video-query-builder.ts
+++ b/server/models/video/video-query-builder.ts
@@ -32,6 +32,8 @@ export type BuildVideosQueryOptions = {
   videoPlaylistId?: number
 
   trendingDays?: number
+  hot?: boolean
+
   user?: MUserAccountId
   historyOfUser?: MUserId
 
@@ -239,14 +241,46 @@ function buildListQuery (model: typeof Model, options: BuildVideosQueryOptions)
     }
   }
 
-  // We don't exclude results in this if so if we do a count we don't need to add this complex clauses
+  // We don't exclude results in this so if we do a count we don't need to add this complex clause
   if (options.trendingDays && options.isCount !== true) {
     const viewsGteDate = new Date(new Date().getTime() - (24 * 3600 * 1000) * options.trendingDays)
 
     joins.push('LEFT JOIN "videoView" ON "video"."id" = "videoView"."videoId" AND "videoView"."startDate" >= :viewsGteDate')
     replacements.viewsGteDate = viewsGteDate
 
-    attributes.push('COALESCE(SUM("videoView"."views"), 0) AS "videoViewsSum"')
+    attributes.push('COALESCE(SUM("videoView"."views"), 0) AS "score"')
+
+    group = 'GROUP BY "video"."id"'
+  } else if (options.hot && options.isCount !== true) {
+    /**
+     * "Hotness" is a measure based on absolute view/comment/like/dislike numbers,
+     * with fixed weights only applied to their log values.
+     *
+     * This algorithm gives little chance for an old video to have a good score,
+     * for which recent spikes in interactions could be a sign of "hotness" and
+     * justify a better score. However there are multiple ways to achieve that
+     * goal, which is left for later. Yes, this is a TODO :)
+     *
+     * note: weights and base score are in number of half-days.
+     * see https://github.com/reddit-archive/reddit/blob/master/r2/r2/lib/db/_sorts.pyx#L47-L58
+     */
+    const weights = {
+      like: 3,
+      dislike: 3,
+      view: 1 / 12,
+      comment: 6
+    }
+
+    joins.push('LEFT JOIN "videoComment" ON "video"."id" = "videoComment"."videoId"')
+
+    attributes.push(
+      `LOG(GREATEST(1, "video"."likes" - 1)) * ${weights.like} ` + // likes (+)
+      `- LOG(GREATEST(1, "video"."dislikes" - 1)) * ${weights.dislike} ` + // dislikes (-)
+      `+ LOG("video"."views" + 1) * ${weights.view} ` + // views (+)
+      `+ LOG(GREATEST(1, COUNT(DISTINCT "videoComment"."id") - 1)) * ${weights.comment} ` + // comments (+)
+      '+ (SELECT EXTRACT(epoch FROM "video"."publishedAt") / 47000) ' + // base score (in number of half-days)
+      'AS "score"'
+    )
 
     group = 'GROUP BY "video"."id"'
   }
@@ -372,8 +406,8 @@ function buildOrder (value: string) {
 
   if (field.toLowerCase() === 'random') return 'ORDER BY RANDOM()'
 
-  if (field.toLowerCase() === 'trending') { // Sort by aggregation
-    return `ORDER BY "videoViewsSum" ${direction}, "video"."views" ${direction}`
+  if ([ 'trending', 'hot' ].includes(field.toLowerCase())) { // Sort by aggregation
+    return `ORDER BY "score" ${direction}, "video"."views" ${direction}`
   }
 
   let firstSort: string