(improvement)(Chat) Simplify processor in Headless and Chat (#822)

Co-authored-by: jolunoluo
This commit is contained in:
LXW
2024-03-15 12:39:14 +08:00
committed by GitHub
parent aad8bb4367
commit 988a025cdf
19 changed files with 85 additions and 186 deletions

View File

@@ -70,6 +70,11 @@ public class Agent extends RecordInfo {
return !CollectionUtils.isEmpty(getParserTools(AgentToolType.NL2SQL_LLM));
}
public boolean containsNL2SQLTool() {
return !CollectionUtils.isEmpty(getParserTools(AgentToolType.NL2SQL_LLM))
|| !CollectionUtils.isEmpty(getParserTools(AgentToolType.NL2SQL_RULE));
}
public Set<Long> getDataSetIds() {
Set<Long> dataSetIds = getDataSetIds(null);
if (containsAllModel(dataSetIds)) {

View File

@@ -15,6 +15,7 @@ public class PluginExecutor implements ChatExecutor {
return null;
}
PluginSemanticQuery query = PluginQueryManager.getPluginQuery(parseInfo.getQueryMode());
query.setParseInfo(parseInfo);
return query.build();
}

View File

@@ -6,7 +6,7 @@ import com.tencent.supersonic.chat.server.util.ComponentFactory;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import java.util.List;
public class Text2PluginParser implements ChatParser {
public class NL2PluginParser implements ChatParser {
private final List<PluginRecognizer> pluginRecognizers = ComponentFactory.getPluginRecognizers();

View File

@@ -7,10 +7,13 @@ import com.tencent.supersonic.headless.api.pojo.request.QueryReq;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import com.tencent.supersonic.headless.server.service.ChatQueryService;
public class Text2SqlParser implements ChatParser {
public class NL2SQLParser implements ChatParser {
@Override
public void parse(ChatParseContext chatParseContext, ParseResp parseResp) {
if (!chatParseContext.enableNL2SQL()) {
return;
}
QueryReq queryReq = QueryReqConverter.buildText2SqlQueryReq(chatParseContext);
ChatQueryService chatQueryService = ContextUtils.getBean(ChatQueryService.class);
ParseResp text2SqlParseResp = chatQueryService.performParsing(queryReq);

View File

@@ -12,7 +12,6 @@ import com.tencent.supersonic.chat.server.plugin.event.PluginAddEvent;
import com.tencent.supersonic.chat.server.plugin.event.PluginDelEvent;
import com.tencent.supersonic.chat.server.plugin.event.PluginUpdateEvent;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.chat.server.service.AgentService;
import com.tencent.supersonic.chat.server.service.PluginService;
import com.tencent.supersonic.common.config.EmbeddingConfig;
import com.tencent.supersonic.common.util.ComponentFactory;
@@ -54,9 +53,7 @@ public class PluginManager {
public static List<Plugin> getPluginAgentCanSupport(ChatParseContext chatParseContext) {
PluginService pluginService = ContextUtils.getBean(PluginService.class);
AgentService agentService = ContextUtils.getBean(AgentService.class);
Agent agent = agentService.getAgent(chatParseContext.getAgentId());
Agent agent = chatParseContext.getAgent();
List<Plugin> plugins = pluginService.getPluginList();
if (Objects.isNull(agent)) {
return plugins;

View File

@@ -1,6 +1,7 @@
package com.tencent.supersonic.chat.server.pojo;
import com.tencent.supersonic.auth.api.authentication.pojo.User;
import com.tencent.supersonic.chat.server.agent.Agent;
import com.tencent.supersonic.headless.api.pojo.SchemaMapInfo;
import com.tencent.supersonic.headless.api.pojo.request.QueryFilters;
import lombok.Data;
@@ -9,9 +10,16 @@ import lombok.Data;
public class ChatParseContext {
private String queryText;
private Integer chatId;
private Integer agentId;
private Agent agent;
private User user;
private QueryFilters queryFilters;
private boolean saveAnswer = true;
private SchemaMapInfo mapInfo = new SchemaMapInfo();
public boolean enableNL2SQL() {
if (agent == null) {
return true;
}
return agent.containsNL2SQLTool();
}
}

View File

@@ -1,25 +1,25 @@
package com.tencent.supersonic.headless.server.processor;
package com.tencent.supersonic.chat.server.processor.parse;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.common.util.ContextUtils;
import com.tencent.supersonic.headless.api.pojo.DataSetSchema;
import com.tencent.supersonic.headless.api.pojo.EntityInfo;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import com.tencent.supersonic.headless.core.chat.query.QueryManager;
import com.tencent.supersonic.headless.core.pojo.ChatContext;
import com.tencent.supersonic.headless.core.pojo.QueryContext;
import com.tencent.supersonic.headless.server.service.impl.SemanticService;
import org.springframework.util.CollectionUtils;
import java.util.List;
/**
* EntityInfoProcessor fills core attributes of an entity so that
* users get to know which entity is parsed out.
*/
public class EntityInfoProcessor implements ResultProcessor {
public class EntityInfoProcessor implements ParseResultProcessor {
@Override
public void process(ParseResp parseResp, QueryContext queryContext, ChatContext chatContext) {
public void process(ChatParseContext chatParseContext, ParseResp parseResp) {
List<SemanticParseInfo> selectedParses = parseResp.getSelectedParses();
if (CollectionUtils.isEmpty(selectedParses)) {
return;
@@ -30,10 +30,9 @@ public class EntityInfoProcessor implements ResultProcessor {
return;
}
//1. set entity info
DataSetSchema dataSetSchema =
queryContext.getSemanticSchema().getDataSetSchemaMap().get(parseInfo.getDataSetId());
SemanticService semanticService = ContextUtils.getBean(SemanticService.class);
EntityInfo entityInfo = semanticService.getEntityInfo(parseInfo, dataSetSchema, queryContext.getUser());
DataSetSchema dataSetSchema = semanticService.getDataSetSchema(parseInfo.getDataSetId());
EntityInfo entityInfo = semanticService.getEntityInfo(parseInfo, dataSetSchema, chatParseContext.getUser());
if (QueryManager.isTagQuery(queryMode)
|| QueryManager.isMetricQuery(queryMode)) {
parseInfo.setEntityInfo(entityInfo);

View File

@@ -1,28 +0,0 @@
package com.tencent.supersonic.chat.server.processor.parse;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import lombok.extern.slf4j.Slf4j;
import org.apache.commons.collections4.CollectionUtils;
import java.util.List;
/**
* RespBuildProcessor fill response object with parsing results.
**/
@Slf4j
public class RespBuildProcessor implements ParseResultProcessor {
@Override
public void process(ChatParseContext chatParseContext, ParseResp parseResp) {
parseResp.setChatId(chatParseContext.getChatId());
parseResp.setQueryText(chatParseContext.getQueryText());
List<SemanticParseInfo> parseInfos = parseResp.getSelectedParses();
if (CollectionUtils.isNotEmpty(parseInfos)) {
parseResp.setState(ParseResp.ParseState.COMPLETED);
} else {
parseResp.setState(ParseResp.ParseState.FAILED);
}
}
}

View File

@@ -1,19 +1,18 @@
package com.tencent.supersonic.headless.server.processor;
package com.tencent.supersonic.chat.server.processor.parse;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import com.tencent.supersonic.headless.core.pojo.ChatContext;
import com.tencent.supersonic.headless.core.pojo.QueryContext;
import lombok.extern.slf4j.Slf4j;
/**
* TimeCostProcessor adds time cost of parsing.
**/
@Slf4j
public class TimeCostProcessor implements ResultProcessor {
public class TimeCostProcessor implements ParseResultProcessor {
@Override
public void process(ParseResp parseResp, QueryContext queryContext, ChatContext chatContext) {
public void process(ChatParseContext chatParseContext, ParseResp parseResp) {
long parseStartTime = parseResp.getParseTimeCost().getParseStartTime();
parseResp.getParseTimeCost().setParseTime(
System.currentTimeMillis() - parseStartTime - parseResp.getParseTimeCost().getSqlTime());

View File

@@ -7,6 +7,7 @@ import com.tencent.supersonic.chat.api.pojo.request.ChatExecuteReq;
import com.tencent.supersonic.chat.api.pojo.request.ChatParseReq;
import com.tencent.supersonic.chat.api.pojo.request.PageQueryInfoReq;
import com.tencent.supersonic.chat.api.pojo.response.ShowCaseResp;
import com.tencent.supersonic.chat.server.agent.Agent;
import com.tencent.supersonic.chat.server.executor.ChatExecutor;
import com.tencent.supersonic.chat.server.parser.ChatParser;
import com.tencent.supersonic.chat.server.persistence.dataobject.ChatDO;
@@ -18,10 +19,12 @@ import com.tencent.supersonic.chat.server.persistence.repository.ChatRepository;
import com.tencent.supersonic.chat.server.pojo.ChatExecuteContext;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.chat.server.processor.parse.ParseResultProcessor;
import com.tencent.supersonic.chat.server.service.AgentService;
import com.tencent.supersonic.chat.server.service.ChatService;
import com.tencent.supersonic.chat.server.util.ComponentFactory;
import com.tencent.supersonic.chat.server.util.QueryReqConverter;
import com.tencent.supersonic.common.util.BeanMapper;
import com.tencent.supersonic.common.util.ContextUtils;
import com.tencent.supersonic.common.util.JsonUtil;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import com.tencent.supersonic.headless.api.pojo.request.DimensionValueReq;
@@ -73,7 +76,7 @@ public class ChatServiceImpl implements ChatService {
@Override
public ParseResp performParsing(ChatParseReq chatParseReq) {
ParseResp parseResp = new ParseResp();
ParseResp parseResp = new ParseResp(chatParseReq.getChatId(), chatParseReq.getQueryText());
ChatParseContext chatParseContext = buildParseContext(chatParseReq);
for (ChatParser chatParser : chatParsers) {
chatParser.parse(chatParseContext, parseResp);
@@ -102,6 +105,9 @@ public class ChatServiceImpl implements ChatService {
private ChatParseContext buildParseContext(ChatParseReq chatParseReq) {
ChatParseContext chatParseContext = new ChatParseContext();
BeanMapper.mapper(chatParseReq, chatParseContext);
AgentService agentService = ContextUtils.getBean(AgentService.class);
Agent agent = agentService.getAgent(chatParseReq.getAgentId());
chatParseContext.setAgent(agent);
QueryReq queryReq = QueryReqConverter.buildText2SqlQueryReq(chatParseContext);
MapResp mapResp = chatQueryService.performMapping(queryReq);
chatParseContext.setMapInfo(mapResp.getMapInfo());

View File

@@ -2,9 +2,7 @@ package com.tencent.supersonic.chat.server.util;
import com.tencent.supersonic.chat.server.agent.Agent;
import com.tencent.supersonic.chat.server.pojo.ChatParseContext;
import com.tencent.supersonic.chat.server.service.AgentService;
import com.tencent.supersonic.common.util.BeanMapper;
import com.tencent.supersonic.common.util.ContextUtils;
import com.tencent.supersonic.headless.api.pojo.request.QueryReq;
public class QueryReqConverter {
@@ -12,11 +10,7 @@ public class QueryReqConverter {
public static QueryReq buildText2SqlQueryReq(ChatParseContext chatParseContext) {
QueryReq queryReq = new QueryReq();
BeanMapper.mapper(chatParseContext, queryReq);
if (chatParseContext.getAgentId() == null) {
return queryReq;
}
AgentService agentService = ContextUtils.getBean(AgentService.class);
Agent agent = agentService.getAgent(chatParseContext.getAgentId());
Agent agent = chatParseContext.getAgent();
if (agent == null) {
return queryReq;
}

View File

@@ -3,7 +3,11 @@ package com.tencent.supersonic.headless.api.pojo.response;
import com.google.common.collect.Lists;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import lombok.Data;
import org.apache.commons.collections4.CollectionUtils;
import java.util.Comparator;
import java.util.List;
import java.util.stream.Collectors;
@Data
public class ParseResp {
@@ -20,4 +24,33 @@ public class ParseResp {
FAILED
}
public List<SemanticParseInfo> getSelectedParses() {
selectedParses = selectedParses.stream()
.sorted(Comparator.comparingDouble(SemanticParseInfo::getScore))
.collect(Collectors.toList());
generateParseInfoId(selectedParses);
return selectedParses;
}
public ParseState getState() {
if (CollectionUtils.isNotEmpty(selectedParses)) {
this.state = ParseResp.ParseState.COMPLETED;
} else {
this.state = ParseState.FAILED;
}
return this.state;
}
private void generateParseInfoId(List<SemanticParseInfo> selectedParses) {
for (int i = 0; i < selectedParses.size(); i++) {
SemanticParseInfo parseInfo = selectedParses.get(i);
parseInfo.setId(i + 1);
}
}
public ParseResp(Integer chatId, String queryText) {
this.chatId = chatId;
this.queryText = queryText;
parseTimeCost.setParseStartTime(System.currentTimeMillis());
}
}

View File

@@ -1,84 +0,0 @@
package com.tencent.supersonic.headless.server.processor;
import com.tencent.supersonic.headless.api.pojo.SchemaElementMatch;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import com.tencent.supersonic.headless.core.pojo.ChatContext;
import com.tencent.supersonic.headless.core.pojo.QueryContext;
import com.tencent.supersonic.headless.core.chat.query.SemanticQuery;
import com.tencent.supersonic.headless.core.chat.query.rule.RuleSemanticQuery;
import lombok.extern.slf4j.Slf4j;
import org.apache.commons.collections.CollectionUtils;
import java.util.ArrayList;
import java.util.List;
import java.util.stream.Collectors;
/**
* QueryRankProcessor ranks candidate parsing results based on
* a heuristic scoring algorithm and then takes topN.
**/
@Slf4j
public class QueryRankProcessor implements ResultProcessor {
private static final int candidateTopSize = 5;
@Override
public void process(ParseResp parseResp, QueryContext queryContext, ChatContext chatContext) {
List<SemanticQuery> candidateQueries = queryContext.getCandidateQueries();
candidateQueries = rank(candidateQueries);
queryContext.setCandidateQueries(candidateQueries);
}
public List<SemanticQuery> rank(List<SemanticQuery> candidateQueries) {
log.debug("pick before [{}]", candidateQueries);
if (CollectionUtils.isEmpty(candidateQueries)) {
return candidateQueries;
}
List<SemanticQuery> selectedQueries = new ArrayList<>();
if (candidateQueries.size() == 1) {
selectedQueries.addAll(candidateQueries);
} else {
selectedQueries = getTopCandidateQuery(candidateQueries);
}
generateParseInfoId(selectedQueries);
log.debug("pick after [{}]", selectedQueries);
return selectedQueries;
}
public List<SemanticQuery> getTopCandidateQuery(List<SemanticQuery> semanticQueries) {
return semanticQueries.stream()
.filter(query -> !checkFullyInherited(query))
.sorted((o1, o2) -> {
if (o1.getParseInfo().getScore() < o2.getParseInfo().getScore()) {
return 1;
} else if (o1.getParseInfo().getScore() > o2.getParseInfo().getScore()) {
return -1;
}
return 0;
}).limit(candidateTopSize)
.collect(Collectors.toList());
}
private void generateParseInfoId(List<SemanticQuery> semanticQueries) {
for (int i = 0; i < semanticQueries.size(); i++) {
SemanticQuery query = semanticQueries.get(i);
query.getParseInfo().setId(i + 1);
}
}
private boolean checkFullyInherited(SemanticQuery query) {
SemanticParseInfo parseInfo = query.getParseInfo();
if (!(query instanceof RuleSemanticQuery)) {
return false;
}
for (SchemaElementMatch match : parseInfo.getElementMatches()) {
if (!match.isInherited()) {
return false;
}
}
return parseInfo.getDateInfo() == null || parseInfo.getDateInfo().isInherited();
}
}

View File

@@ -1,33 +0,0 @@
package com.tencent.supersonic.headless.server.processor;
import com.tencent.supersonic.headless.api.pojo.SemanticParseInfo;
import com.tencent.supersonic.headless.api.pojo.response.ParseResp;
import com.tencent.supersonic.headless.core.chat.query.SemanticQuery;
import com.tencent.supersonic.headless.core.pojo.ChatContext;
import com.tencent.supersonic.headless.core.pojo.QueryContext;
import lombok.extern.slf4j.Slf4j;
import java.util.List;
import java.util.stream.Collectors;
/**
* RespBuildProcessor fill response object with parsing results.
**/
@Slf4j
public class RespBuildProcessor implements ResultProcessor {
@Override
public void process(ParseResp parseResp, QueryContext queryContext, ChatContext chatContext) {
parseResp.setChatId(queryContext.getChatId());
parseResp.setQueryText(queryContext.getQueryText());
List<SemanticQuery> candidateQueries = queryContext.getCandidateQueries();
if (candidateQueries.size() > 0) {
List<SemanticParseInfo> candidateParses = candidateQueries.stream()
.map(SemanticQuery::getParseInfo).collect(Collectors.toList());
parseResp.setSelectedParses(candidateParses);
parseResp.setState(ParseResp.ParseState.COMPLETED);
} else {
parseResp.setState(ParseResp.ParseState.FAILED);
}
}
}

View File

@@ -40,9 +40,7 @@ public class SqlInfoProcessor implements ResultProcessor {
}
List<SemanticParseInfo> selectedParses = semanticQueries.stream().map(SemanticQuery::getParseInfo)
.collect(Collectors.toList());
long startTime = System.currentTimeMillis();
addSqlInfo(queryContext, selectedParses);
parseResp.getParseTimeCost().setSqlTime(System.currentTimeMillis() - startTime);
}
private void addSqlInfo(QueryContext queryContext, List<SemanticParseInfo> semanticParseInfos) {

View File

@@ -121,7 +121,7 @@ public class ChatQueryServiceImpl implements ChatQueryService {
@Override
public ParseResp performParsing(QueryReq queryReq) {
ParseResp parseResult = new ParseResp();
ParseResp parseResult = new ParseResp(queryReq.getChatId(), queryReq.getQueryText());
// build queryContext and chatContext
QueryContext queryCtx = buildQueryContext(queryReq);
@@ -157,6 +157,9 @@ public class ChatQueryServiceImpl implements ChatQueryService {
resultProcessors.forEach(processor -> {
processor.process(parseResult, queryCtx, chatCtx);
});
List<SemanticParseInfo> parseInfos = queryCtx.getCandidateQueries().stream()
.map(SemanticQuery::getParseInfo).collect(Collectors.toList());
parseResult.setSelectedParses(parseInfos);
return parseResult;
}

View File

@@ -55,7 +55,7 @@ public class SemanticService {
}
public EntityInfo getEntityInfo(SemanticParseInfo parseInfo, DataSetSchema dataSetSchema, User user) {
if (parseInfo != null && parseInfo.getDataSetId() > 0) {
if (parseInfo != null && parseInfo.getDataSetId() != null && parseInfo.getDataSetId() > 0) {
EntityInfo entityInfo = getEntityBasicInfo(dataSetSchema);
if (parseInfo.getDimensionFilters().size() <= 0 || entityInfo.getDataSetInfo() == null) {
entityInfo.setMetrics(null);

View File

@@ -10,8 +10,8 @@ com.tencent.supersonic.headless.core.chat.parser.SemanticParser=\
com.tencent.supersonic.headless.core.chat.parser.QueryTypeParser
com.tencent.supersonic.chat.server.parser.ChatParser=\
com.tencent.supersonic.chat.server.parser.Text2PluginParser, \
com.tencent.supersonic.chat.server.parser.Text2SqlParser
com.tencent.supersonic.chat.server.parser.NL2PluginParser, \
com.tencent.supersonic.chat.server.parser.NL2SQLParser
com.tencent.supersonic.chat.server.executor.ChatExecutor=\
com.tencent.supersonic.chat.server.executor.PluginExecutor, \
@@ -42,10 +42,7 @@ com.tencent.supersonic.headless.core.cache.QueryCache=\
com.tencent.supersonic.headless.server.processor.ResultProcessor=\
com.tencent.supersonic.headless.server.processor.ParseInfoProcessor, \
com.tencent.supersonic.headless.server.processor.QueryRankProcessor, \
com.tencent.supersonic.headless.server.processor.SqlInfoProcessor, \
com.tencent.supersonic.headless.server.processor.TimeCostProcessor, \
com.tencent.supersonic.headless.server.processor.RespBuildProcessor
com.tencent.supersonic.headless.server.processor.SqlInfoProcessor
com.tencent.supersonic.headless.core.chat.parser.llm.DataSetResolver=\
com.tencent.supersonic.headless.core.chat.parser.llm.HeuristicDataSetResolver
@@ -60,8 +57,9 @@ com.tencent.supersonic.chat.server.plugin.recognize.PluginRecognizer=\
com.tencent.supersonic.chat.server.plugin.recognize.embedding.EmbeddingRecallRecognizer
com.tencent.supersonic.chat.server.processor.parse.ParseResultProcessor=\
com.tencent.supersonic.chat.server.processor.parse.RespBuildProcessor,\
com.tencent.supersonic.chat.server.processor.parse.QueryRecommendProcessor
com.tencent.supersonic.chat.server.processor.parse.QueryRecommendProcessor,\
com.tencent.supersonic.chat.server.processor.parse.EntityInfoProcessor,\
com.tencent.supersonic.chat.server.processor.parse.TimeCostProcessor
com.tencent.supersonic.chat.server.processor.execute.ExecuteResultProcessor=\
com.tencent.supersonic.chat.server.processor.execute.MetricRecommendProcessor,\

View File

@@ -18,7 +18,7 @@ com.tencent.supersonic.headless.server.processor.ResultProcessor=\
com.tencent.supersonic.headless.server.processor.ParseInfoProcessor, \
com.tencent.supersonic.headless.server.processor.QueryRankProcessor, \
com.tencent.supersonic.headless.server.processor.SqlInfoProcessor, \
com.tencent.supersonic.headless.server.processor.TimeCostProcessor, \
com.tencent.supersonic.chat.server.processor.parse.TimeCostProcessor, \
com.tencent.supersonic.headless.server.processor.RespBuildProcessor
com.tencent.supersonic.headless.core.chat.parser.llm.DataSetResolver=\