Merge branch 'feature_lqc_20240926_rank_vov' of algorithm/recommend-server into master

recommend-server-service/src/main/java/com/tzld/piaoquan/recommend/server/service/rank/strategy/RankStrategy4RegionMergeModelV562.java

@@ -32,6 +32,10 @@ public class RankStrategy4RegionMergeModelV562 extends RankStrategy4RegionMergeM
     @ApolloJsonValue("${rank.score.merge.weightv562:}")
     private Map<String, Double> mergeWeight;
 
+
+
+
+
     @Autowired
     private FeatureService featureService;
 
@@ -332,32 +336,102 @@ public class RankStrategy4RegionMergeModelV562 extends RankStrategy4RegionMergeM
 
         List<RankItem> items = ScorerUtils.getScorerPipeline("feeds_score_config_20240807.conf")
                 .scoring(sceneFeatureMap, userFeatureMap, rankItems);
+
+
+        // 获取VoV预测模型参数
+        // 融合权重
+        double alpha_vov = mergeWeight.getOrDefault("alpha_vov", 1.0);
+
+        double vov_thresh = mergeWeight.getOrDefault("vov_thresh", 0.1);
+
+        double view_thresh = mergeWeight.getOrDefault("view_thresh", 1535.0);
+
+        double level50_vov = mergeWeight.getOrDefault("level50_vov", 0.123);
+
+        double level_95_vov = mergeWeight.getOrDefault("level_95_vov", 0.178);
+
+        double beta_vov = mergeWeight.getOrDefault("beta_vov", 100.0);
+
+        List<Double> weightList = new ArrayList<>(7);
+        weightList.add(mergeWeight.getOrDefault("d2_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("d1_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("h48_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("h24_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("h3_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("h2_ago_vov_w", 0.0));
+        weightList.add(mergeWeight.getOrDefault("h1_ago_vov_w", 0.0));
+
+
+
         Map<String, Map<String, String>> vid2MapFeature = this.getVideoRedisFeature(vids, "redis:vid_hasreturn_rov:");
-        Map<String, Map<String, String>> vid2VovFeatureMap = this.getVideoRedisFeature(vids, "redis:vid_vov_1d3d:");
-        double alpha_vov = mergeWeight.getOrDefault("alpha_vov", 2.0);
+        Map<String, Map<String, String>> vid2VovFeatureMap = this.getVideoRedisFeature(vids, "redis:vid_vovhour4rank:");
         List<Video> result = new ArrayList<>();
 //        String hasReturnRovKey = mergeWeight.getOrDefault("hasReturnRovKey", 1.0) < 0.5 ? "rate_1" : "rate_n";
 //        Double chooseFunction = mergeWeight.getOrDefault("chooseFunction", 0.0);
 
         for (RankItem item : items) {
             double score = 0.0;
-            double recommend_rate_1d = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
-                    .getOrDefault("recommend_rate_1d", "0"));
-            double recommend_exp_per_1d = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
-                    .getOrDefault("recommend_exp_per_1d", "0"));
-            double vorScore =  recommend_rate_1d * recommend_exp_per_1d;
-            item.getScoresMap().put("recommend_rate_1d", recommend_rate_1d);
-            item.getScoresMap().put("recommend_exp_per_1d", recommend_exp_per_1d);
-            item.getScoresMap().put("vorScore", vorScore);
-            item.getScoresMap().put("alpha_vov", alpha_vov);
-            double hasReturnRovScore = Double.parseDouble(vid2MapFeature.getOrDefault(item.getVideoId() + "", new HashMap<>())
-                    .getOrDefault("rate_n", "0"));
-            item.getScoresMap().put("hasReturnRovScore", hasReturnRovScore);
+            // 获取其他模型输出score
             double fmRovOrigin = item.getScoreRov();
             item.getScoresMap().put("fmRovOrigin", fmRovOrigin);
             double fmRov = restoreScore(fmRovOrigin);
             item.getScoresMap().put("fmRov", fmRov);
-            score = fmRov * (1 + hasReturnRovScore) * (1.0 + alpha_vov * recommend_rate_1d);
+
+
+            // 获取VoV输入特征
+            double h1_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h1_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double h2_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h2_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double h3_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h3_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double h24_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h24_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double h48_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h48_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double d1_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("d1_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+            double d2_ago_vov = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("d2_ago_vov", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+
+            double h1_ago_view = Double.parseDouble(vid2VovFeatureMap.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("h1_ago_view", "-2")); // 如果没有时，默认为多少？？ 需要考虑
+
+            item.getScoresMap().put("h1_ago_vov", h1_ago_vov);
+            item.getScoresMap().put("h2_ago_vov", h2_ago_vov);
+            item.getScoresMap().put("h3_ago_vov", h3_ago_vov);
+            item.getScoresMap().put("h24_ago_vov", h24_ago_vov);
+            item.getScoresMap().put("h48_ago_vov", h48_ago_vov);
+            item.getScoresMap().put("d1_ago_vov", d1_ago_vov);
+            item.getScoresMap().put("h1_ago_view", h1_ago_view);
+            item.getScoresMap().put("alpha_vov", alpha_vov);
+            item.getScoresMap().put("view_thresh", view_thresh);
+            item.getScoresMap().put("vov_thresh", vov_thresh);
+
+
+
+            List<Double> featureList = new ArrayList<>(7);
+            featureList.add(d2_ago_vov);
+            featureList.add(d1_ago_vov);
+            featureList.add(h48_ago_vov);
+            featureList.add(h24_ago_vov);
+            featureList.add(h3_ago_vov);
+            featureList.add(h2_ago_vov);
+            featureList.add(h1_ago_vov);
+
+            // todo 线性加权 预测VoV
+
+
+            double vov_p = calculateScore(featureList, weightList,item,  vov_thresh, view_thresh, h1_ago_view,level50_vov,level_95_vov,beta_vov);
+
+
+            double hasReturnRovScore = Double.parseDouble(vid2MapFeature.getOrDefault(item.getVideoId() + "", new HashMap<>())
+                    .getOrDefault("rate_n", "0"));
+            item.getScoresMap().put("hasReturnRovScore", hasReturnRovScore);
+            score = fmRov * (1 + hasReturnRovScore)  * (1.0 + alpha_vov * vov_p);
+
+
+            item.getScoresMap().put("vov_p", vov_p);
 
             Video video = item.getVideo();
             video.setScore(score);
@@ -380,6 +454,55 @@ public class RankStrategy4RegionMergeModelV562 extends RankStrategy4RegionMergeM
         return result;
     }
 
+
+    private  double calculateScore(List<Double> featureList, List<Double> weightList,RankItem rankItem,
+                                        double vov_thresh, double view_thresh, double h1_ago_view,double level50_vov,double level_95_vov,double beta_vov) {
+        // 检查 h1_ago_view 条件
+        if (h1_ago_view == -2 || h1_ago_view == -1 || h1_ago_view < view_thresh) {
+            rankItem.getScoresMap().put("origin_vov_p", 0d);
+            return 0;
+        }
+
+        // // 检查 featureList 是否全为 -1
+        // if (featureList.stream().allMatch(f -> f == -1)) {
+        //     rankItem.getScoresMap().put("origin_vov_p", 0d);
+        //     return 0;
+        // }
+
+        // 计算有效特征的总权重和得分
+        double score = 0;
+        List<Integer> validIndices = new ArrayList<>();
+
+        for (int i = 0; i < featureList.size(); i++) {
+            if (featureList.get(i) != -1) {
+                validIndices.add(i);
+            }
+        }
+
+        // 如果没有有效特征，返回 0
+        if (validIndices.isEmpty()) {
+            rankItem.getScoresMap().put("origin_vov_p", 0d);
+            return 0;
+        }
+
+        // 计算得分，动态调整权重
+        for (int index : validIndices) {
+            double weight = weightList.get(index);
+            score += featureList.get(index) * weight;
+        }
+        rankItem.getScoresMap().put("origin_vov_p", score);
+        // 调整vov
+        if (score < vov_thresh) {
+            score = 0;
+        } else {
+            double term1 = 1 / (1 + Math.exp(-1*beta_vov * (score - level50_vov)));
+            double term2 = 1 + Math.exp(-1*beta_vov * (level_95_vov - level50_vov));
+            score = term1 * term2;
+        }
+        return score;
+    }
+
+
     private void readBucketFile() {
         InputStream resourceStream = RankStrategy4RegionMergeModelV562.class.getClassLoader().getResourceAsStream("20240609_bucket_274.txt");
         if (resourceStream != null) {