|
@@ -11,6 +11,7 @@ import com.tzld.piaoquan.recommend.server.service.recall.strategy.*;
|
|
|
import com.tzld.piaoquan.recommend.server.service.score.ScorerUtils;
|
|
|
import com.tzld.piaoquan.recommend.server.util.CommonCollectionUtils;
|
|
|
import com.tzld.piaoquan.recommend.server.util.FeatureBucketUtils;
|
|
|
+import com.tzld.piaoquan.recommend.server.util.FeatureUtils;
|
|
|
import com.tzld.piaoquan.recommend.server.util.SimilarityUtils;
|
|
|
import lombok.extern.slf4j.Slf4j;
|
|
|
import org.apache.commons.collections4.MapUtils;
|
|
@@ -39,6 +40,8 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
private static final List<String> cfRovList = Collections.singletonList("rovn");
|
|
|
private static final List<String> videoSimAttrs = Arrays.asList("cate1_list", "cate2", "cate2_list",
|
|
|
"keywords", "style", "theme", "title", "topic", "user_value");
|
|
|
+ private static final List<String> videoCateAttr = Arrays.asList(FeatureUtils.cate1Attr, FeatureUtils.cate2Attr, FeatureUtils.festive1Attr);
|
|
|
+ private static final double smoothPlus = 5.0;
|
|
|
|
|
|
@Override
|
|
|
public List<Video> mergeAndRankRovRecall(RankParam param) {
|
|
@@ -47,7 +50,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
//-------------------合-------------------
|
|
|
//-------------------逻-------------------
|
|
|
//-------------------辑-------------------
|
|
|
-
|
|
|
+ long currentMs = System.currentTimeMillis();
|
|
|
List<Video> oldRovs = new ArrayList<>();
|
|
|
oldRovs.addAll(extractAndSort(param, RegionHRecallStrategy.PUSH_FORM));
|
|
|
oldRovs.addAll(extractAndSort(param, RegionHDupRecallStrategy.PUSH_FORM));
|
|
@@ -72,7 +75,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
//-------------------新地域召回------------------
|
|
|
List<Video> v1 = extractAndSort(param, RegionRealtimeRecallStrategyV1.PUSH_FORM);
|
|
|
v1 = v1.stream().filter(r -> !setVideo.contains(r.getVideoId())).collect(Collectors.toList());
|
|
|
- v1 = v1.subList(0, Math.min(mergeWeight.getOrDefault("v1", 5.0).intValue(), v1.size()));
|
|
|
+ v1 = v1.subList(0, Math.min(mergeWeight.getOrDefault("v1", 10.0).intValue(), v1.size()));
|
|
|
rovRecallRank.addAll(v1);
|
|
|
setVideo.addAll(v1.stream().map(Video::getVideoId).collect(Collectors.toSet()));
|
|
|
//-------------------scene cf rovn------------------
|
|
@@ -119,6 +122,9 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
Map<String, String> c8 = featureOriginUser.getOrDefault("alg_mid_feature_sharecf", new HashMap<>());
|
|
|
Map<String, String> c9 = featureOriginUser.getOrDefault("alg_mid_feature_returncf", new HashMap<>());
|
|
|
|
|
|
+ // time feature
|
|
|
+ addTimeFeature(currentMs, userFeatureMapDouble);
|
|
|
+
|
|
|
if (!c1.isEmpty()) {
|
|
|
userFeatureMapDouble.put("playcnt_6h", Double.parseDouble(c1.getOrDefault("playcnt_6h", "0")));
|
|
|
userFeatureMapDouble.put("playcnt_1d", Double.parseDouble(c1.getOrDefault("playcnt_1d", "0")));
|
|
@@ -215,6 +221,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
double share = tuple4.second.isEmpty() ? 0 : Double.parseDouble(tuple4.second.getOrDefault("share_pv_" + prefix2, "0.0"));
|
|
|
double returns = tuple4.third.isEmpty() ? 0 : Double.parseDouble(tuple4.third.getOrDefault("return_uv_" + prefix2, "0.0"));
|
|
|
|
|
|
+ double f0 = ExtractorUtils.calLog(exp);
|
|
|
double f1 = ExtractorUtils.calDiv(share, exp);
|
|
|
double f2 = ExtractorUtils.calLog(share);
|
|
|
double f3 = ExtractorUtils.calDiv(returns, exp);
|
|
@@ -222,6 +229,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
double f5 = f3 * f4;
|
|
|
double f6 = ExtractorUtils.calDiv(returns, share);
|
|
|
|
|
|
+ String key0 = tuple4.name + "_" + prefix2 + "_" + "EXP";
|
|
|
String key1 = tuple4.name + "_" + prefix2 + "_" + "STR";
|
|
|
String key2 = tuple4.name + "_" + prefix2 + "_" + "log(share)";
|
|
|
String key3 = tuple4.name + "_" + prefix2 + "_" + "ROV";
|
|
@@ -229,6 +237,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
String key5 = tuple4.name + "_" + prefix2 + "_" + "ROV*log(return)";
|
|
|
String key6 = tuple4.name + "_" + prefix2 + "_" + "ROS";
|
|
|
|
|
|
+ featureMap.put(key0, f0);
|
|
|
featureMap.put(key1, f1);
|
|
|
featureMap.put(key2, f2);
|
|
|
featureMap.put(key3, f3);
|
|
@@ -296,6 +305,8 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
addVideoStatFeature(vid, featureOriginVideo, featureMap);
|
|
|
//addVideoCFFeature(vid, featureOriginVideo, featureMap);
|
|
|
addVideoSimFeature(headVideoInfo, videoInfo, featureMap);
|
|
|
+ addVideoAttrFeature(videoInfo, featureMap);
|
|
|
+ addVideoTimeFeature(currentMs, videoInfo, featureMap);
|
|
|
|
|
|
item.featureMapDouble = featureMap;
|
|
|
}
|
|
@@ -303,7 +314,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
// 3 连续值特征分桶
|
|
|
readBucketFile();
|
|
|
Map<String, String> userFeatureMap = new HashMap<>(userFeatureMapDouble.size());
|
|
|
- Map<String, String> norUserFeatureMap = FeatureBucketUtils.bucketFeature("20241209_nor_bucket.txt", userFeatureMapDouble);
|
|
|
+ Map<String, String> norUserFeatureMap = FeatureBucketUtils.noBucketFeature(userFeatureMapDouble);
|
|
|
for (Map.Entry<String, Double> entry : userFeatureMapDouble.entrySet()) {
|
|
|
String name = entry.getKey();
|
|
|
Double score = entry.getValue();
|
|
@@ -331,10 +342,10 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
}
|
|
|
}
|
|
|
item.featureMap = featureMap;
|
|
|
- item.norFeatureMap = FeatureBucketUtils.bucketFeature("20241209_nor_bucket.txt", featureMapDouble);
|
|
|
+ item.norFeatureMap = FeatureBucketUtils.noBucketFeature(featureMapDouble);
|
|
|
}
|
|
|
// 4 排序模型计算
|
|
|
- double fmRovLogBase = mergeWeight.getOrDefault("fmRovLogBase", 3.5);
|
|
|
+ double fmRovLogBase = mergeWeight.getOrDefault("fmRovLogBase", 2.8);
|
|
|
double xgbNorScaleType = mergeWeight.getOrDefault("xgbNorScaleType", 0.0);
|
|
|
double xgbNorBias = mergeWeight.getOrDefault("xgbNorBias", -1.6945);
|
|
|
double xgbNorWeight = mergeWeight.getOrDefault("xgbNorWeight", 1.8968);
|
|
@@ -343,7 +354,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
double vorLogWeight = mergeWeight.getOrDefault("vorLogWeight", 10.0);
|
|
|
double vorLogBase = mergeWeight.getOrDefault("vorLogBase", 2.6);
|
|
|
Map<String, String> sceneFeatureMap = new HashMap<>(0);
|
|
|
- List<RankItem> items = ScorerUtils.getScorerPipeline("feeds_score_config_fm_xgb_20241209.conf").scoring(sceneFeatureMap, userFeatureMap, norUserFeatureMap, rankItems);
|
|
|
+ List<RankItem> items = ScorerUtils.getScorerPipeline("feeds_score_config_fm_xgb_20250218.conf").scoring(sceneFeatureMap, userFeatureMap, norUserFeatureMap, rankItems);
|
|
|
// 5 排序公式特征
|
|
|
Map<String, Map<String, String>> vid2MapFeature = this.getVideoRedisFeature(vids, "redis:vid_hasreturn_vor:");
|
|
|
List<Video> result = new ArrayList<>();
|
|
@@ -419,17 +430,20 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
List<String> infoPeriod = tuple3.second;
|
|
|
Map<String, String> infoMap = tuple3.third;
|
|
|
for (String period : infoPeriod) {
|
|
|
+ double view = getVideoOneInfo("view_" + period, infoMap);
|
|
|
double share = getVideoOneInfo("share_" + period, infoMap);
|
|
|
double return_ = getVideoOneInfo("return_" + period, infoMap);
|
|
|
double view_hasreturn = getVideoOneInfo("view_hasreturn_" + period, infoMap);
|
|
|
double share_hasreturn = getVideoOneInfo("share_hasreturn_" + period, infoMap);
|
|
|
- double ros = getVideoOneInfo("ros_" + period, infoMap);
|
|
|
- double rov = getVideoOneInfo("rov_" + period, infoMap);
|
|
|
- double r_cnt = getVideoOneInfo("r_cnt_" + period, infoMap);
|
|
|
- double r_rate = getVideoOneInfo("r_rate_" + period, infoMap);
|
|
|
- double r_cnt4s = getVideoOneInfo("r_cnt4s_" + period, infoMap);
|
|
|
- double str = getVideoOneInfo("str_" + period, infoMap);
|
|
|
-
|
|
|
+ double ros = ExtractorUtils.smoothDiv(return_, share, smoothPlus);
|
|
|
+ double rov = ExtractorUtils.smoothDiv(return_, view, smoothPlus);
|
|
|
+ double r_cnt = ExtractorUtils.smoothDiv(return_, view_hasreturn, smoothPlus);
|
|
|
+ double r_rate = ExtractorUtils.smoothDiv(view_hasreturn, view, smoothPlus);
|
|
|
+ double r_cnt4s = ExtractorUtils.smoothDiv(return_, share_hasreturn, smoothPlus);
|
|
|
+ double str = ExtractorUtils.smoothDiv(share, view, smoothPlus);
|
|
|
+ double s2r_rate = ExtractorUtils.smoothDiv(share_hasreturn, share, smoothPlus);
|
|
|
+
|
|
|
+ featureMap.put(infoType + "_" + period + "_" + "view", ExtractorUtils.calLog(view));
|
|
|
featureMap.put(infoType + "_" + period + "_" + "share", ExtractorUtils.calLog(share));
|
|
|
featureMap.put(infoType + "_" + period + "_" + "return", ExtractorUtils.calLog(return_));
|
|
|
featureMap.put(infoType + "_" + period + "_" + "view_hasreturn", ExtractorUtils.calLog(view_hasreturn));
|
|
@@ -440,6 +454,7 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
featureMap.put(infoType + "_" + period + "_" + "r_rate", r_rate);
|
|
|
featureMap.put(infoType + "_" + period + "_" + "r_cnt4s", r_cnt4s);
|
|
|
featureMap.put(infoType + "_" + period + "_" + "str", str);
|
|
|
+ featureMap.put(infoType + "_" + period + "_" + "s2r_rate", s2r_rate);
|
|
|
}
|
|
|
}
|
|
|
}
|
|
@@ -496,6 +511,43 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
}
|
|
|
}
|
|
|
|
|
|
+ private void addVideoAttrFeature(Map<String, String> rankInfo, Map<String, Double> featureMap) {
|
|
|
+ if (!rankInfo.isEmpty()) {
|
|
|
+ for (String attr : videoCateAttr) {
|
|
|
+ String attrVal = rankInfo.getOrDefault(attr, "");
|
|
|
+ int attrId = FeatureUtils.getAttrId(attr, attrVal);
|
|
|
+ if (attrId > 0) {
|
|
|
+ String key = String.format("%s@%s@%d", "r", attr, attrId);
|
|
|
+ featureMap.put(key, 1.0);
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+ }
|
|
|
+
|
|
|
+ private void addVideoTimeFeature(long currentMs, Map<String, String> rankInfo, Map<String, Double> featureMap) {
|
|
|
+ double createTime = 1D;
|
|
|
+ try {
|
|
|
+ if (!rankInfo.isEmpty() && rankInfo.containsKey("gmt_create_timestamp")) {
|
|
|
+ String createMsStr = rankInfo.get("gmt_create_timestamp");
|
|
|
+ long createMs = Long.parseLong(createMsStr);
|
|
|
+ createTime = FeatureUtils.getCreateTime(currentMs, createMs);
|
|
|
+ }
|
|
|
+ } catch (Exception e) {
|
|
|
+ log.error("get video crate time error", e);
|
|
|
+ }
|
|
|
+ featureMap.put("createTime", createTime);
|
|
|
+ }
|
|
|
+
|
|
|
+ private void addTimeFeature(long currentMs, Map<String, Double> featureMap) {
|
|
|
+ Calendar calendar = Calendar.getInstance();
|
|
|
+ calendar.setTimeInMillis(currentMs);
|
|
|
+
|
|
|
+ int week = calendar.get(Calendar.DAY_OF_WEEK);
|
|
|
+ int hour = calendar.get(Calendar.HOUR_OF_DAY) + 1;
|
|
|
+ featureMap.put("week", week * 1.0);
|
|
|
+ featureMap.put("hour", hour * 1.0);
|
|
|
+ }
|
|
|
+
|
|
|
private double norCalibration(double scaleType, double polyBias, double polyWeight, double powerWeight, double powerExp, double score) {
|
|
|
if (scaleType < 1) {
|
|
|
return norPolyCalibration(polyBias, polyWeight, score);
|
|
@@ -506,8 +558,8 @@ public class RankStrategy4RegionMergeModelV564 extends RankStrategy4RegionMergeM
|
|
|
|
|
|
private double norPolyCalibration(double bias, double weight, double score) {
|
|
|
double newScore = bias + weight * score;
|
|
|
- if (newScore < 1E-8) {
|
|
|
- newScore = 0;
|
|
|
+ if (newScore < score) {
|
|
|
+ newScore = score;
|
|
|
}
|
|
|
return newScore;
|
|
|
}
|