feat: ai 임베딩

backend/build.gradle

@@ -40,6 +40,7 @@ dependencies {
     implementation 'org.springframework.boot:spring-boot-starter-actuator'
     implementation 'org.springframework.boot:spring-boot-starter-oauth2-client'
     implementation group: 'org.springframework.boot', name: 'spring-boot-starter-mail', version: '3.0.5'
+    implementation 'org.springframework.boot:spring-boot-starter-batch'
 
     // API Documentation (문서화)
     implementation 'org.apache.commons:commons-lang3:3.18.0'
@@ -78,6 +79,8 @@ dependencies {
     implementation 'org.springframework.ai:spring-ai-starter-model-openai'
     implementation 'org.springframework.ai:spring-ai-advisors-vector-store'
     implementation 'org.springframework.ai:spring-ai-starter-model-chat-memory-repository-jdbc'
+    implementation 'org.springframework.ai:spring-ai-starter-model-ollama'
+    implementation 'org.springframework.ai:spring-ai-starter-model-huggingface'
 
     // Testing (테스트)
     testImplementation 'org.springframework.boot:spring-boot-starter-test'

backend/docker-compose.yml

@@ -58,6 +58,26 @@ services:
       timeout: 5s
       retries: 10
 
+  ollama:
+    image: ollama/ollama:latest
+    container_name: ollama
+    restart: unless-stopped
+    ports:
+      - "11434:11434"
+    volumes:
+      - ollama-data:/root/.ollama
+    entrypoint: [ "/bin/sh", "-c" ]
+    command: >
+      "ollama serve &
+       sleep 5 && 
+       ollama pull daynice/kure-v1:567m && 
+       wait"
+    healthcheck:
+      test: [ "CMD", "curl", "-f", "http://localhost:11434/api/version" ]
+      interval: 10s
+      timeout: 5s
+      retries: 10
+
 volumes:
   mysql-data:
   redis-data:

backend/src/main/java/com/ai/lawyer/domain/chatbot/dto/ExtractionDto.java

@@ -19,7 +19,7 @@ public static class TitleExtractionDto {
     @AllArgsConstructor
     @NoArgsConstructor
     public static class KeywordExtractionDto {
-        private List<String> keyword;
+        private String keyword;
     }
 
 }

backend/src/main/java/com/ai/lawyer/domain/chatbot/service/ChatBotService.java

@@ -57,17 +57,13 @@ public class ChatBotService {
     // 멤버 조회 -> 벡터 검색 (판례, 법령) -> 프롬프트 생성 (시스템, 유저) -> 채팅 클라이언트 호출 (스트림) -> 응답 저장, 제목/키워드 추출
     public Flux<ChatResponse> sendMessage(Long memberId, ChatRequest chatChatRequestDto, Long roomId) {
 
-        if(memberId == null) {
-            log.error("해당 멤버는 존재하지 않거나, accessToken이 만료되거나 잘못되었습니다.");
-        }
-
         Member member = memberRepository.findById(memberId)
                 .orElseThrow(() -> new IllegalArgumentException("존재하지 않는 회원입니다.")
         );
 
         // 벡터 검색 (판례, 법령)
-        List<Document> similarCaseDocuments = qdrantService.searchDocument(chatChatRequestDto.getMessage(), "type", "판례", 3);
-        List<Document> similarLawDocuments = qdrantService.searchDocument(chatChatRequestDto.getMessage(), "type", "법령", 2);
+        List<Document> similarCaseDocuments = qdrantService.searchDocument(chatChatRequestDto.getMessage(), "type", "판례");
+        List<Document> similarLawDocuments = qdrantService.searchDocument(chatChatRequestDto.getMessage(), "type", "법령");
 
         // 판례와 법령 정보를 구분 있게 포맷팅
         String caseContext = formatting(similarCaseDocuments);
@@ -167,18 +163,19 @@ private void handlerTasks(ChatRequest chatDto, History history, String fullRespo
     private void extractAndUpdateKeywordRanks(String message) {
         KeywordExtractionDto keywordResponse = keywordExtract(message, keywordExtraction, KeywordExtractionDto.class);
 
-        for (String keyword : keywordResponse.getKeyword()) {
-            KeywordRank keywordRank = keywordRankRepository.findByKeyword(keyword);
-            if (keywordRank == null) {
-                keywordRank = KeywordRank.builder()
-                        .keyword(keyword)
-                        .score(1L)
-                        .build();
-            } else {
-                keywordRank.setScore(keywordRank.getScore() + 1);
-            }
-            keywordRankRepository.save(keywordRank);
+        KeywordRank keywordRank = keywordRankRepository.findByKeyword(keywordResponse.getKeyword());
+
+        if (keywordRank == null) {
+            keywordRank = KeywordRank.builder()
+                    .keyword(keywordResponse.getKeyword())
+                    .score(1L)
+                    .build();
+        } else {
+            keywordRank.setScore(keywordRank.getScore() + 1);
         }
+
+        keywordRankRepository.save(keywordRank);
+
     }
 
     private void setHistoryTitle(ChatRequest chatDto, History history, String fullResponse) {

backend/src/main/java/com/ai/lawyer/global/batch/BatchScheduler.java

@@ -0,0 +1,25 @@
+package com.ai.lawyer.global.batch;
+
+/*@Slf4j
+@Component
+@EnableScheduling
+@RequiredArgsConstructor
+public class BatchScheduler {
+
+    private final JobLauncher jobLauncher;
+    private final Job dataVectorizationJob;
+
+    @Scheduled(cron = "#{${batch.scheduler.run-every-minute} ? '* * * * * *' : '* * 2 * * *'}")
+    public void runVectorizationJob() {
+        log.info("전체 데이터(판례, 법령) 벡터화 스케줄러 실행...");
+        try {
+            JobParameters jobParameters = new JobParametersBuilder()
+                    .addString("requestDate", LocalDateTime.now().toString())
+                    .toJobParameters();
+
+            jobLauncher.run(dataVectorizationJob, jobParameters); // Job 실행
+        } catch (Exception e) {
+            log.error("전체 데이터 벡터화 배치 작업 실행 중 오류 발생", e);
+        }
+    }
+}*/

backend/src/main/java/com/ai/lawyer/global/batch/DataVectorizationJobConfig.java

@@ -0,0 +1,209 @@
+package com.ai.lawyer.global.batch;
+
+/*@Slf4j
+@Configuration
+@RequiredArgsConstructor
+public class DataVectorizationJobConfig {
+
+    private final JobRepository jobRepository;
+    private final PlatformTransactionManager transactionManager;
+    private final EntityManagerFactory entityManagerFactory;
+    private final VectorStore vectorStore;
+
+    private final JangRepository jangRepository;
+    private final JoRepository joRepository;
+    private final HangRepository hangRepository;
+    private final HoRepository hoRepository;
+
+    private final TokenTextSplitter tokenSplitter = TokenTextSplitter.builder()
+            .withChunkSize(800)
+            .withMinChunkSizeChars(0)
+            .withMinChunkLengthToEmbed(5)
+            .withMaxNumChunks(10000)
+            .withKeepSeparator(true)
+            .build();
+
+    private static final int CHUNK_SIZE = 10; // 배치 처리 시 한 번에 읽어올 데이터 수
+
+    @Value("${batch.page.size.precedent}")
+    private int precedentPageSize; // 하루에 처리할 판례 수
+
+    @Value("${batch.page.size.law}")
+    private int lawPageSize; // 하루에 처리할 법령 수
+
+    @Bean
+    public TaskExecutor taskExecutor() {
+        ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
+        executor.setCorePoolSize(10);
+        executor.setMaxPoolSize(20);
+        executor.setQueueCapacity(100);
+        executor.setThreadNamePrefix("async-thread-");
+        executor.initialize();
+        return executor;
+    }
+
+    // -------------- 전체 데이터 벡터화 정의 --------------
+    @Bean
+    public Job dataVectorizationJob() {
+        return new JobBuilder("dataVectorizationJob", jobRepository)
+                .start(precedentVectorizationStep()) // 판례 벡터화 Step 실행
+                .next(lawVectorizationStep())        // 법령 벡터화 Step 실행
+                .build();
+    }
+
+    // -------------- 판례 벡터화 ---------------
+    @Bean
+    public Step precedentVectorizationStep() {
+        log.info(">>>>>> 판례 벡터화 시작");
+        return new StepBuilder("precedentVectorizationStep", jobRepository)
+                .<Precedent, List<Document>>chunk(CHUNK_SIZE, transactionManager)
+                .reader(precedentItemReader())
+                .processor(precedentItemProcessor())
+                .writer(documentItemWriter())
+                .taskExecutor(taskExecutor())
+                .build();
+    }
+
+    @Bean
+    public JpaPagingItemReader<Precedent> precedentItemReader() {
+        return new JpaPagingItemReaderBuilder<Precedent>()
+                .name("precedentItemReader")
+                .entityManagerFactory(entityManagerFactory)
+                .pageSize(CHUNK_SIZE)
+                .maxItemCount(precedentPageSize)
+                .queryString("SELECT p FROM Precedent p ORDER BY p.id ASC")
+                .build();
+    }
+
+    @Bean
+    public ItemProcessor<Precedent, List<Document>> precedentItemProcessor() {
+
+        return precedent -> {
+            String content = precedent.getPrecedentContent();
+            if (content == null || content.isBlank()) return null;
+
+            Document originalDoc = new Document(content, Map.of(
+                    "type", "판례",
+                    "caseNumber", precedent.getCaseNumber(),
+                    "court", precedent.getCourtName(),
+                    "caseName", precedent.getCaseName()
+            ));
+
+            List<Document> chunkDocs = tokenSplitter.split(originalDoc);
+            List<Document> finalChunks = new ArrayList<>();
+
+            // 청크별로 메타데이터에 인덱스 추가 -> 구분 용도
+            for (int i = 0; i < chunkDocs.size(); i++) {
+                Document chunk = chunkDocs.get(i);
+                Map<String, Object> newMetadata = new HashMap<>(chunk.getMetadata());
+                newMetadata.put("chunkIndex", i);
+                finalChunks.add(new Document(chunk.getText(), newMetadata));
+            }
+            return finalChunks;
+        };
+    }
+
+    // -------------- 법령 백터화 ---------------
+    @Bean
+    public Step lawVectorizationStep() {
+        log.info(">>>>>> 법령 벡터화 시작");
+        return new StepBuilder("lawVectorizationStep", jobRepository)
+                .<Law, List<Document>>chunk(CHUNK_SIZE, transactionManager) // 법령은 한 번에 10개씩 처리
+                .reader(lawItemReader())
+                .processor(lawItemProcessor())
+                .writer(documentItemWriter())
+                .taskExecutor(taskExecutor())
+                .build();
+    }
+
+    @Bean
+    public JpaPagingItemReader<Law> lawItemReader() {
+        return new JpaPagingItemReaderBuilder<Law>()
+                .name("lawItemReader")
+                .entityManagerFactory(entityManagerFactory)
+                .pageSize(CHUNK_SIZE)
+                .maxItemCount(lawPageSize)
+                .queryString("SELECT l FROM Law l ORDER BY l.id ASC")
+                .build();
+    }
+
+    @Bean
+    public ItemProcessor<Law, List<Document>> lawItemProcessor() {
+        return law -> {
+            List<Document> finalChunks = new ArrayList<>();
+
+            List<Jang> jangs = jangRepository.findByLaw(law);
+
+            for (Jang jang : jangs) {
+
+                StringBuilder contentBuilder = new StringBuilder();
+
+                contentBuilder.append(law.getLawName()).append("\n");
+
+                if (jang.getContent() != null && !jang.getContent().isBlank()) {
+                    contentBuilder.append(jang.getContent()).append("\n");
+                }
+
+                List<Jo> jos = joRepository.findByJang(jang);
+                for (Jo jo : jos) {
+
+                    if (jo.getContent() != null && !jo.getContent().isBlank()) {
+                        contentBuilder.append(jo.getContent()).append("\n");
+                    }
+
+                    List<Hang> hangs = hangRepository.findByJo(jo);
+                    for (Hang hang : hangs) {
+                        if (hang.getContent() != null && !hang.getContent().isBlank()) {
+                            contentBuilder.append(hang.getContent()).append("\n");
+                        }
+
+                        List<Ho> hos = hoRepository.findByHang(hang);
+                        for (Ho ho : hos) {
+                            if (ho.getContent() != null && !ho.getContent().isBlank()) {
+                                contentBuilder.append(ho.getContent()).append("\n");
+                            }
+                        }
+                    }
+                }
+
+                // === Jang 단위로 문서화 ===
+                String finalContent = contentBuilder.toString();
+
+                if (!finalContent.isBlank()) {
+                    Map<String, Object> metadata = new HashMap<>();
+                    metadata.put("type", "법령");
+                    metadata.put("lawName", law.getLawName());
+                    metadata.put("jangId", jang.getId());
+
+                    Document originalDoc = new Document(finalContent, metadata);
+
+                    List<Document> chunkDocs = tokenSplitter.split(originalDoc);
+
+                    for (int i = 0; i < chunkDocs.size(); i++) {
+                        Document chunk = chunkDocs.get(i);
+                        Map<String, Object> newMetadata = new HashMap<>(chunk.getMetadata());
+                        newMetadata.put("chunkIndex", i);
+                        finalChunks.add(new Document(chunk.getText(), newMetadata));
+                    }
+                }
+            }
+
+            return finalChunks.isEmpty() ? null : finalChunks;
+        };
+    }
+
+    @Bean
+    public ItemWriter<List<Document>> documentItemWriter() {
+        return chunk -> {
+            List<Document> totalDocuments = chunk.getItems().stream()
+                    .flatMap(List::stream)
+                    .collect(Collectors.toList());
+
+            if (!totalDocuments.isEmpty()) {
+                vectorStore.add(totalDocuments);
+                log.info(">>>>>> {}개의 Document 청크를 벡터 저장소에 저장했습니다.", totalDocuments.size());
+            }
+        };
+    }
+}*/
+

backend/src/main/java/com/ai/lawyer/global/config/AIConfig.java

@@ -3,16 +3,24 @@
 import org.springframework.ai.chat.client.ChatClient;
 import org.springframework.ai.chat.memory.ChatMemoryRepository;
 import org.springframework.ai.chat.memory.repository.jdbc.JdbcChatMemoryRepository;
+import org.springframework.ai.embedding.EmbeddingModel;
+import org.springframework.ai.ollama.OllamaEmbeddingModel;
 import org.springframework.ai.openai.OpenAiChatModel;
-import org.springframework.ai.transformer.splitter.TokenTextSplitter;
 import org.springframework.context.annotation.Bean;
 import org.springframework.context.annotation.Configuration;
+import org.springframework.context.annotation.Primary;
 import org.springframework.jdbc.core.JdbcTemplate;
 import org.springframework.transaction.PlatformTransactionManager;
 
 @Configuration
 public class AIConfig {
 
+    @Bean
+    @Primary
+    public EmbeddingModel primaryOllamaEmbeddingModel(OllamaEmbeddingModel ollamaEmbeddingModel) {
+        return ollamaEmbeddingModel;
+    }
+
     @Bean
     public ChatMemoryRepository chatMemoryRepository(JdbcTemplate jdbcTemplate, PlatformTransactionManager transactionManager) {
         return JdbcChatMemoryRepository.builder()
@@ -26,9 +34,5 @@ public ChatClient openAiChatClient(OpenAiChatModel openAiChatModel) {
         return ChatClient.create(openAiChatModel);
     }
 
-    @Bean
-    public TokenTextSplitter tokenTextSplitter() {
-        return new TokenTextSplitter(500, 150, 5, 10000, true);
-    }
-
 }
+