(improvement)(common|headless|chat|auth) 鉴权优化与召回优化

1 修复生成的用户token 一生成就失效的问题
2 如果用户设置的token ,需校验是否数据库存在,因为用户可设置一年的token 有泄露风险
3 结果解析优化, 去除不可以解析的情况,解析问题需要改写后的问,
4 召回样例,用相似度,保住至少有一个样例是高相似度的
5 数据集召回,填加完全匹配格式筛选逻辑
This commit is contained in:
guilinlewis
2025-06-23 09:47:48 +08:00
parent 0721df2e66
commit 7e6639df83
8 changed files with 84 additions and 18 deletions

View File

@@ -18,6 +18,7 @@ import java.util.List;
import java.util.Map;
import java.util.Objects;
import java.util.Set;
import java.util.stream.Collectors;
import static com.tencent.supersonic.common.pojo.Constants.DEFAULT_DETAIL_LIMIT;
import static com.tencent.supersonic.common.pojo.Constants.DEFAULT_METRIC_LIMIT;
@@ -65,12 +66,23 @@ public class SemanticParseInfo implements Serializable {
DataSetMatchResult mr2 = getDataSetMatchResult(o2.getElementMatches());
double difference = mr1.getMaxDatesetSimilarity() - mr2.getMaxDatesetSimilarity();
if (difference == 0) {
if (Math.abs(difference) < 0.0005) { // 看完全匹配的个数,实践证明,可以用户输入规范后,该逻辑具有优势
if (!o1.getDataSetId().equals(o2.getDataSetId())) {
List<SchemaElementMatch> elementMatches1 = o1.getElementMatches().stream()
.filter(e -> e.getSimilarity() == 1).collect(Collectors.toList());
List<SchemaElementMatch> elementMatches2 = o2.getElementMatches().stream()
.filter(e -> e.getSimilarity() == 1).collect(Collectors.toList());
if (elementMatches1.size() > elementMatches2.size()) {
return -1;
} else if (elementMatches1.size() < elementMatches2.size()) {
return 1;
}
}
difference = mr1.getMaxMetricSimilarity() - mr2.getMaxMetricSimilarity();
if (difference == 0) {
if (Math.abs(difference) < 0.0005) {
difference = mr1.getTotalSimilarity() - mr2.getTotalSimilarity();
}
if (difference == 0) {
if (Math.abs(difference) < 0.0005) {
difference = mr1.getMaxMetricUseCnt() - mr2.getMaxMetricUseCnt();
}
}