[improvement][Headless] Embedding supports Chinese by default and fixes the issue of abnormal number recognition (#726)

common/src/main/java/com/tencent/supersonic/common/pojo/DataItem.java

@@ -8,7 +8,10 @@ import lombok.Data;
 @Builder
 public class DataItem {
 
-    private Long id;
+    /***
+     * This field uses an underscore (_) at the end.
+     */
+    private String id;
 
     private String bizName;
 
@@ -18,6 +21,9 @@ public class DataItem {
 
     private TypeEnums type;
 
+    /***
+     * This field uses an underscore (_) at the end.
+     */
     private String modelId;
 
     private String defaultAgg;

common/src/main/java/com/tencent/supersonic/common/pojo/SysParameter.java

@@ -75,7 +75,7 @@ public class SysParameter {
         parameters.add(new Parameter("embedding.mapper.number", "5",
                 "批量向量召回文本返回结果个数", "每个文本进行向量语义召回的文本结果个数", "number", "Mapper相关配置"));
         parameters.add(new Parameter("embedding.mapper.distance.threshold",
-                "0.58", "向量召回相似度阈值", "相似度大于该阈值的则舍弃", "number", "Mapper相关配置"));
+                "0.01", "向量召回相似度阈值", "相似度大于该阈值的则舍弃", "number", "Mapper相关配置"));
 
         //parser config
         Parameter s2SQLParameter = new Parameter("s2SQL.generation", "TWO_PASS_AUTO_COT",

common/src/main/java/com/tencent/supersonic/common/util/embedding/InMemoryS2EmbeddingStore.java

@@ -129,7 +129,7 @@ public class InMemoryS2EmbeddingStore implements S2EmbeddingStore {
             List<Retrieval> retrievals = new ArrayList<>();
             for (EmbeddingMatch<EmbeddingQuery> embeddingMatch : relevant) {
                 Retrieval retrieval = new Retrieval();
-                retrieval.setDistance(embeddingMatch.score());
+                retrieval.setDistance(1 - embeddingMatch.score());
                 retrieval.setId(embeddingMatch.embeddingId());
                 retrieval.setQuery(embeddingMatch.embedded().getQuery());
                 Map<String, Object> metadata = embeddingMatch.embedded().getMetadata();