kakaotalk-chat-analyzer 0.16.5 → 0.18.3

package/README.md

@@ -186,6 +186,36 @@ kca --help
 
 </details>
 
+<details>
+<summary><strong>분석 preset·기능 (0.18+)</strong></summary>
+
+| preset | 용도 | 90k 메시지 목표 | 시맨틱 | 감정 | LLM |
+|--------|------|-----------------|--------|------|-----|
+| `speed` | RAM·시간 최소 | ~3분 | 끔 | 끔 | 끔 |
+| `balanced` | 기본 권장 | ~5분 | e5-small | 자동 | 끔 |
+| `quality` | 한국어·서사 최대 | ~6분 | ko-v2 | KLUE | 2B/4B |
+| `custom` | 기능 직접 지정 | — | env/플래그 | env | `KCA_LLM=1` |
+
+```bash
+kca capabilities                    # RAM·추천 preset
+kca ./chat.csv --preset balanced
+kca ./chat.csv --preset quality --local
+kca llm pull 2b                     # GGUF (optional node-llama-cpp)
+KCA_LLM_BACKEND=ollama KCA_LLM=1 kca ./chat.csv --preset custom
+```
+
+환경 변수: `KCA_PRESET`, `KCA_SEMANTIC_MODEL`, `KCA_SENTIMENT_MODEL`, `KCA_LLM`, `KCA_LLM_MOCK`, `KCA_ONNX_GPU`, `KCA_EMBED_BATCH`, `KCA_SENTIMENT_BATCH`, `KCA_KIWI_WORKERS`, `KCA_NO_KIWI_WORKERS`, `KCA_PROFILE_PHASES`, `KCA_BENCH_CSV`, `KCA_KEYWORD_SUMMARY_TOP`, `KCA_SHOP_SEARCH_TOP`.
+
+**속도(품질 유지):** 대용량 CSV는 Kiwi worker pool(`KCA_KIWI_WORKERS`, RAM≥8GB 기본 2–4)·시맨틱/감정을 키워드 패스와 겹쳐 실행. `KCA_PROFILE_PHASES=1`로 단계별 ms. quality에서 GPU 가속: `onnxruntime-node` 설치 후 `KCA_ONNX_GPU=metal`(macOS)·`cuda`(Linux)·`dml`(Windows).
+
+**키워드:** 요약은 `KCA_KEYWORD_SUMMARY_TOP`(기본 12)·**빈도 순**; ④ 차트에서 **빈도/특이어** 탭 전환. 전체 ~120개는 집계 상한.
+
+**주제 맵:** graph(공기 군집)·keyword(상위 키워드 시드)·semantic(임베딩 클러스터) 3레인 RRF 병합 — 대용량 방에서 의미 테마 최대 12장. `KCA_TOPIC_MAX`, `KCA_TOPIC_MIN_THEMES`.
+
+**LLM (`quality` / `KCA_LLM=1`):** 주제 제목·서사 + `topicProposals`(키워드 화이트리스트) + 인사이트 bullet·샵검색/상호작용 한 줄(원문 미전송).
+
+</details>
+
 <details>
 <summary><strong>성능·키워드·벤치 (개발·파워유저)</strong></summary>
 
@@ -197,11 +227,13 @@ kca --help
 ```bash
 npx kcachat@latest "./chat.csv" --profile --no-worker
 npm run bench:stream -- 100000   # 저장소 클론 후
+npm run bench:preset             # speed/balanced SLA 스모크
+KCA_BENCH_COMPARE=1 npm run bench:semantic
 ```
 
 </details>
 
-**버전 고정:** `npx kakaotalk-chat-analyzer@0.16.1` · 최신은 `kcachat@latest`가 매번 본체를 받습니다. 리포트 사이드 카드·`#kca-provenance`로 실제 생성 버전을 확인할 수 있습니다.
+**버전 고정:** `npx kakaotalk-chat-analyzer@0.18.2` · 최신은 `kcachat@latest`가 매번 본체를 받습니다. 리포트 사이드 카드·`#kca-provenance`로 실제 생성 버전을 확인할 수 있습니다.
 
 **로컬 개발:**
 
@@ -216,6 +248,11 @@ cd kakaotalk-chat-analyzer && npm install && npm run build && npm test
 
 | 버전 | 요약 |
 |------|------|
+| **0.18.2** | 주제 맵 3레인(graph·키워드·임베딩) 병합·테마 6~12·LLM `topicProposals` |
+| **0.18.1** | 키워드 빈도/특이어 dual-view·샵검색 통계·dyad 셀 숫자·LLM 인사이트 필드 |
+| **0.18.0** | preset(speed/balanced/quality)·5분 예산 skip·LLM 서사·KLUE 감정·dual-lane 툴팁·CI Playwright |
+| **0.17.2** | `kca llm pull`·provenance `llmUsed`·분석 예산 라우터 |
+| **0.16.6** | 글자 수 랭킹·비속어 패턴 통계·transformers 감정 분석(자동/선택) |
 | **0.16.5** | 상호작용 히트맵: 말 많은 사람 축 상단·지연 로드·로딩 스켈레톤 |
 | **0.16.4** | 대용량 방 키워드: minDf 스케일·메시지 수 우선 정렬·시맨틱은 BM25 후보만 보강 |
 | **0.16.3** | 기본 **품질 우선** 프로필(메인 스레드·시맨틱 샘플 확대·RRF 완화·임베딩 주제). 가속은 `--worker` / `--fast` |

package/data/korean-profanity.txt

@@ -0,0 +1,33 @@
+# 한국어 채팅 비속·욕설 패턴(부분 일치, normalize 후 검사). 줄 앞 # 은 주석.
+시발
+씨발
+씨팔
+좆
+지랄
+병신
+븅신
+미친
+미친놈
+개새
+개새끼
+개쉐
+개씨
+개같
+닥쳐
+꺼져
+엿먹
+엿먹어
+죽어
+죽을
+좆같
+좆만
+ㅅㅂ
+ㅆㅂ
+ㅄ
+ㅂㅅ
+ㅈㄹ
+ㅁㅊ
+ㅗㅗ
+fuck
+shit
+bitch

package/dist/src/aggregator.d.ts

@@ -1,4 +1,5 @@
-import type { ChatRecord, EncodingName, PrivacyMode, ReportData } from "./types.js";
+import type { ChatRecord, EncodingName, PrivacyMode, ReportData, SentimentStats } from "./types.js";
+import type { BuildReportOptions } from "./analyze-pool.js";
 export interface FinalizeSourceMeta {
     filePath: string;
     encoding: EncodingName;
@@ -7,6 +8,7 @@ export interface FinalizeSourceMeta {
 }
 export interface FinalizeOptions {
     usedSemanticKeywords?: boolean;
+    usedSentimentAnalysis?: boolean;
     koreanPrimary?: boolean;
     useEmbeddingTopics?: boolean;
     semanticSupplementRrfWeight?: number;
@@ -16,7 +18,9 @@ export declare function semanticSupplementHitCap(corpusMessages: number): number
 export interface AggregatorOptions {
     /** 시맨틱 키워드용 메시지 샘플 수집 */
     semanticSamples?: boolean;
-    /** 시맨틱 리저보어 상한 추정(스트리밍 시 생략 가능) */
+    /** 감정 분석용 메시지 샘플 수집 */
+    sentimentSamples?: boolean;
+    /** 시맨틱·감정 리저보어 상한 추정(스트리밍 시 생략 가능) */
     estimatedMessages?: number;
 }
 export declare class ReportAggregator {
@@ -73,11 +77,18 @@ export declare class ReportAggregator {
     private roomSubManagerMessages;
     private roomManagerMessages;
     private roomShopSearchMessages;
+    private shopSearchUntaggedNotices;
+    private readonly shopSearchMissSamples;
     private roomPhotoBundleMessages;
     private pureLaughMessages;
     private openChatBoilerplateExcluded;
     private semanticThemeCandidates;
     private readonly semanticReservoir;
+    private readonly sentimentReservoir;
+    private readonly profanityCounter;
+    private sentimentStats;
+    /** stats pass에서 리저보어를 채웠으면 keyword pass 중복 push 방지 */
+    private samplesCollectedInStatsPass;
     private prevMs;
     private prevSender;
     private runSender;
@@ -85,9 +96,21 @@ export declare class ReportAggregator {
     private firstDate;
     private lastDate;
     constructor(filePath: string, privacy: PrivacyMode, top: number, options?: AggregatorOptions);
-    drainSemanticSamples(): string[];
+    /** 스트리밍 1패스 후 실제 건수로 리저보어 상한 보정(추정치 과소 시) */
+    ensureSampleCaps(messageCount: number): void;
+    drainSemanticSamples(buildOptions?: BuildReportOptions): string[];
+    drainSentimentSamples(): {
+        text: string;
+        sender: string;
+    }[];
+    applySentimentStats(stats: SentimentStats): void;
+    senderAliasMap(): Map<string, string>;
     messageCount(): number;
     resetKeywordPipeline(): void;
+    markSamplesCollectedInStatsPass(): void;
+    applyKeywordTokens(kwTokens: string[], monthKey: string): void;
+    private pushAnalysisSamples;
+    private pushSemanticSample;
     private consumeKeywords;
     applySemanticKeywordBoost(items: {
         label: string;
@@ -102,6 +125,7 @@ export declare class ReportAggregator {
     consume(record: ChatRecord, opts?: {
         keywordsOnly?: boolean;
         skipKeywords?: boolean;
+        collectSamples?: boolean;
     }): void;
     private bumpSystemNotice;
     finalize(meta: FinalizeSourceMeta, finalizeOpts?: FinalizeOptions): ReportData;

package/dist/src/aggregator.js

@@ -1,6 +1,7 @@
 import { formatDate, formatDateTime, partsToUtcMs, weekdayIndex } from "./date.js";
 import { maskPartialDisplayName, parseChatRoomNameFromExportPath, safeInputName } from "./analysis-labels.js";
 import { GapStreamStats, SessionGapStats } from "./gap-stats.js";
+import { keywordTokensForRecord } from "./keyword-record-tokens.js";
 import { tokenizeForKeywords } from "./keyword-tokenize.js";
 import { adaptiveMinCount, StreamingTfidfKeywords } from "./streaming-tfidf-keywords.js";
 import { TopicMapAccumulator } from "./topic-map.js";
@@ -8,8 +9,13 @@ import { extractHashtagKeywords } from "./korean-hashtags.js";
 import { buildKeywordStopwords } from "./keyword-stopwords.js";
 import { buildTopicStopwords } from "./topic-stopwords.js";
 import { MessageReservoir } from "./message-reservoir.js";
-import { semanticReservoirCap, semanticSampleCap, subsampleSemanticMessages } from "./semantic-policy.js";
-import { mergeKeywordRankings } from "./keyword-merge.js";
+import { SenderMessageReservoir } from "./sender-message-reservoir.js";
+import { ProfanityCounter } from "./profanity.js";
+import { sentimentReservoirCap, sentimentSampleCap, subsampleSentimentRecords, } from "./sentiment-policy.js";
+import { effectiveSemanticSampleCap, semanticReservoirCap, subsampleSemanticMessages, } from "./semantic-policy.js";
+import { getAttachmentMarkers, shouldExtractKeywords } from "./keyword-eligibility.js";
+import { mergeDualLaneKeywords } from "./keyword-rank-dual.js";
+import { shopSearchDisplayTop } from "./report-config.js";
 import { isNoiseKeyword } from "./keyword-quality.js";
 import { formatCompactNumber, formatReplyGapMinutes } from "./report-util.js";
 import { KeywordCounter } from "./keyword-counter.js";
@@ -22,7 +28,9 @@ import { buildEventSpine } from "./event-spine.js";
 import { buildRoomNarrative } from "./room-narrative.js";
 import { buildPeriodCompare } from "./period-compare.js";
 import { buildBenchmarkBandsFromValues } from "./benchmark-bands.js";
-import { mergeEmbeddingThemes } from "./embedding-topics.js";
+import { semanticItemsToTopics } from "./embedding-topics.js";
+import { buildKeywordSeedTopics } from "./keyword-seed-topics.js";
+import { mergeTopicLanes } from "./topic-merge.js";
 const ATTACHMENT_MARKERS = [
     "사진",
     "동영상",
@@ -106,11 +114,18 @@ export class ReportAggregator {
     roomSubManagerMessages = 0;
     roomManagerMessages = 0;
     roomShopSearchMessages = 0;
+    shopSearchUntaggedNotices = 0;
+    shopSearchMissSamples = [];
     roomPhotoBundleMessages = 0;
     pureLaughMessages = 0;
     openChatBoilerplateExcluded = 0;
     semanticThemeCandidates = [];
     semanticReservoir;
+    sentimentReservoir;
+    profanityCounter;
+    sentimentStats = null;
+    /** stats pass에서 리저보어를 채웠으면 keyword pass 중복 push 방지 */
+    samplesCollectedInStatsPass = false;
     prevMs = null;
     prevSender = null;
     runSender = null;
@@ -125,14 +140,44 @@ export class ReportAggregator {
         this.semanticReservoir = options?.semanticSamples
             ? new MessageReservoir(semanticReservoirCap(options?.estimatedMessages))
             : null;
+        this.sentimentReservoir = options?.sentimentSamples
+            ? new SenderMessageReservoir(sentimentReservoirCap(options?.estimatedMessages))
+            : null;
+        this.profanityCounter = ProfanityCounter.create();
+    }
+    /** 스트리밍 1패스 후 실제 건수로 리저보어 상한 보정(추정치 과소 시) */
+    ensureSampleCaps(messageCount) {
+        if (messageCount <= 0)
+            return;
+        const semNeed = semanticReservoirCap(messageCount);
+        const sentNeed = sentimentReservoirCap(messageCount);
+        if (this.semanticReservoir && this.semanticReservoir.capacity() < semNeed) {
+            this.semanticReservoir.growTo(semNeed);
+        }
+        if (this.sentimentReservoir && this.sentimentReservoir.capacity() < sentNeed) {
+            this.sentimentReservoir.growTo(sentNeed);
+        }
     }
-    drainSemanticSamples() {
+    drainSemanticSamples(buildOptions) {
         const raw = this.semanticReservoir?.drain() ?? [];
         if (raw.length === 0)
             return raw;
-        const cap = semanticSampleCap(Math.max(this.total, raw.length));
+        const cap = effectiveSemanticSampleCap(Math.max(this.total, raw.length), buildOptions);
         return subsampleSemanticMessages(raw, cap);
     }
+    drainSentimentSamples() {
+        const raw = this.sentimentReservoir?.drain() ?? [];
+        if (raw.length === 0)
+            return raw;
+        const cap = sentimentSampleCap(Math.max(this.total, raw.length));
+        return subsampleSentimentRecords(raw, cap);
+    }
+    applySentimentStats(stats) {
+        this.sentimentStats = stats;
+    }
+    senderAliasMap() {
+        return buildSenderLabels([...this.senderStats.keys()], this.privacy);
+    }
     messageCount() {
         return this.total;
     }
@@ -140,27 +185,47 @@ export class ReportAggregator {
         this.keywordStream = new StreamingTfidfKeywords();
         this.topicMap = new TopicMapAccumulator();
     }
-    consumeKeywords(record) {
-        const split = splitMessageForAnalysis(record.message);
-        const msg = split.userText.length > 0 ? split.userText : record.message;
-        const messageLength = msg.length;
-        if (split.notices.length > 0 && split.userText.length === 0)
-            return;
-        const foundAttachments = getAttachmentMarkers(msg);
-        if (isOpenChatBoilerplate(msg)) {
-            this.openChatBoilerplateExcluded += 1;
-            return;
+    markSamplesCollectedInStatsPass() {
+        this.samplesCollectedInStatsPass = true;
+    }
+    applyKeywordTokens(kwTokens, monthKey) {
+        this.keywordStream.addDocumentTokens(kwTokens);
+        this.topicMap.addMessage(kwTokens, monthKey);
+        let monthBucket = this.monthlyKeywordBuckets.get(monthKey);
+        if (!monthBucket) {
+            monthBucket = new KeywordCounter();
+            this.monthlyKeywordBuckets.set(monthKey, monthBucket);
         }
-        if (messageLength < 2 || !HAS_TOKEN_CHAR_RE.test(msg) || !shouldExtractKeywords(msg, foundAttachments)) {
+        for (const t of kwTokens)
+            monthBucket.add(t);
+    }
+    pushAnalysisSamples(msg, sender, messageLength, isPureSystem) {
+        if (isPureSystem || isOpenChatBoilerplate(msg))
             return;
+        if (this.sentimentReservoir && messageLength >= 12) {
+            this.sentimentReservoir.push(msg, sender);
         }
-        const kwTokens = tokenizeForKeywords(msg);
-        this.keywordStream.addDocumentTokens(kwTokens);
-        const monthKey = `${record.date.year}-${pad2(record.date.month)}`;
-        this.topicMap.addMessage(kwTokens, monthKey);
+    }
+    pushSemanticSample(msg, messageLength) {
         if (this.semanticReservoir && messageLength >= 12)
             this.semanticReservoir.push(msg);
     }
+    consumeKeywords(record) {
+        const row = keywordTokensForRecord(record);
+        if (!row) {
+            const split = splitMessageForAnalysis(record.message);
+            const msg = split.userText.length > 0 ? split.userText : record.message;
+            if (isOpenChatBoilerplate(msg))
+                this.openChatBoilerplateExcluded += 1;
+            return;
+        }
+        this.applyKeywordTokens(row.tokens, row.monthKey);
+        if (!this.samplesCollectedInStatsPass) {
+            const split = splitMessageForAnalysis(record.message);
+            const msg = split.userText.length > 0 ? split.userText : record.message;
+            this.pushSemanticSample(msg, msg.length);
+        }
+    }
     applySemanticKeywordBoost(items) {
         const valid = items.filter((item) => !isNoiseKeyword(item.label));
         this.semanticThemeCandidates = valid.map((item) => ({
@@ -199,6 +264,14 @@ export class ReportAggregator {
             this.bumpSystemNotice(kind, dayKey);
         for (const tag of split.shopSearchTags)
             increment(this.shopSearchTopics, tag);
+        if (split.notices.includes("shopSearch") && split.shopSearchTags.length === 0) {
+            this.shopSearchUntaggedNotices += 1;
+            if (this.shopSearchMissSamples.length < 8) {
+                const sample = record.message.trim().slice(0, 120).replace(/\s+/g, " ");
+                if (sample)
+                    this.shopSearchMissSamples.push(sample);
+            }
+        }
         const msg = split.userText.length > 0 ? split.userText : record.message;
         const messageLength = msg.length;
         const isPureSystem = split.notices.length > 0 && split.userText.length === 0;
@@ -279,6 +352,15 @@ export class ReportAggregator {
                 for (const domain of foundDomains)
                     increment(this.domains, domain);
             }
+            if (!isPureSystem && !isOpenChatBoilerplate(msg)) {
+                this.profanityCounter.add(msg, record.sender);
+                if (opts?.collectSamples) {
+                    this.pushAnalysisSamples(msg, record.sender, messageLength, isPureSystem);
+                }
+                else if (this.sentimentReservoir && messageLength >= 12) {
+                    this.sentimentReservoir.push(msg, record.sender);
+                }
+            }
             if (isOpenChatBoilerplate(msg)) {
                 this.openChatBoilerplateExcluded += 1;
             }
@@ -287,16 +369,7 @@ export class ReportAggregator {
                 shouldExtractKeywords(msg, foundAttachments)) {
                 if (!opts?.skipKeywords) {
                     const kwTokens = tokenizeForKeywords(msg);
-                    this.keywordStream.addDocumentTokens(kwTokens);
-                    const monthKey = `${record.date.year}-${pad2(record.date.month)}`;
-                    this.topicMap.addMessage(kwTokens, monthKey);
-                    let monthBucket = this.monthlyKeywordBuckets.get(monthKey);
-                    if (!monthBucket) {
-                        monthBucket = new KeywordCounter();
-                        this.monthlyKeywordBuckets.set(monthKey, monthBucket);
-                    }
-                    for (const t of kwTokens)
-                        monthBucket.add(t);
+                    this.applyKeywordTokens(kwTokens, `${record.date.year}-${pad2(record.date.month)}`);
                 }
                 if (!opts?.keywordsOnly) {
                     const kwOpts = {
@@ -308,8 +381,15 @@ export class ReportAggregator {
                     }
                     if (messageLength >= 12)
                         this.repeatPhraseCounter.add(msg, dayKey);
-                    if (this.semanticReservoir && messageLength >= 12)
-                        this.semanticReservoir.push(msg);
+                    if (opts?.collectSamples) {
+                        this.pushSemanticSample(msg, messageLength);
+                    }
+                    else if (!this.samplesCollectedInStatsPass) {
+                        this.pushSemanticSample(msg, messageLength);
+                    }
+                }
+                else if (opts?.collectSamples) {
+                    this.pushSemanticSample(msg, messageLength);
                 }
             }
         }
@@ -375,10 +455,11 @@ export class ReportAggregator {
             prevStat.maxConsecutive = Math.max(prevStat.maxConsecutive, this.runLen);
         }
         const total = this.total;
+        const totalChars = this.totalCharacters;
         const aliases = buildSenderLabels([...this.senderStats.keys()], this.privacy);
-        const participantStats = [...this.senderStats.entries()]
-            .map(([raw, stat]) => {
+        const allParticipants = [...this.senderStats.entries()].map(([raw, stat]) => {
             const sharePercent = total > 0 ? round((stat.messages / total) * 100, 1) : 0;
+            const characterSharePercent = totalChars > 0 ? round((stat.characters / totalChars) * 100, 1) : 0;
             return {
                 alias: aliases.get(raw) ?? "???",
                 messages: stat.messages,
@@ -387,12 +468,17 @@ export class ReportAggregator {
                 attachmentMessages: stat.attachmentMessages,
                 linkMessages: stat.linkMessages,
                 sharePercent,
+                characterSharePercent,
                 nightMessages: stat.nightMessages,
                 maxConsecutive: stat.maxConsecutive,
             };
-        })
+        });
+        const participantStats = [...allParticipants]
             .sort((a, b) => b.messages - a.messages)
             .slice(0, this.top);
+        const participantsByCharacters = [...allParticipants]
+            .sort((a, b) => b.characters - a.characters || b.messages - a.messages)
+            .slice(0, this.top);
         const sortedDays = [...this.daily.keys()].sort();
         const longestStreak = longestDateStreak(sortedDays);
         let peakHour = null;
@@ -453,17 +539,24 @@ export class ReportAggregator {
         }
         const keywordStop = buildKeywordStopwords();
         const keywordLimit = Math.max(120, this.top * 3);
-        const wordRankItems = this.keywordStream.extractKeywordItems({
+        const minDocFreq = adaptiveMinCount(total, finalizeOpts?.koreanPrimary !== false);
+        const keywordCandidates = this.keywordStream.collectKeywordCandidates({
             stopwords: keywordStop,
-            limit: keywordLimit,
-            minDocFreq: adaptiveMinCount(total, finalizeOpts?.koreanPrimary !== false),
+            minDocFreq,
         });
-        this.applySemanticSupplementForRanked(wordRankItems);
-        const keywords = mergeKeywordRankings(wordRankItems, this.keywordSupplement, keywordLimit, finalizeOpts?.semanticSupplementRrfWeight ?? 0.5);
-        let topics = this.topicMap.buildTopics(total, buildTopicStopwords());
-        if (finalizeOpts?.useEmbeddingTopics && this.semanticThemeCandidates.length > 0) {
-            topics = mergeEmbeddingThemes(topics, this.semanticThemeCandidates, total);
-        }
+        const bm25LaneForSemantic = [...keywordCandidates]
+            .sort((a, b) => b.score - a.score || b.messageHits - a.messageHits)
+            .slice(0, Math.min(200, Math.floor(80 + Math.sqrt(Math.max(total, 1)))));
+        this.applySemanticSupplementForRanked(bm25LaneForSemantic);
+        const kwMerged = mergeDualLaneKeywords(keywordCandidates, this.keywordSupplement, total, keywordLimit, finalizeOpts?.semanticSupplementRrfWeight ?? 0.5);
+        const keywords = kwMerged.byFrequency;
+        const keywordsDistinctive = kwMerged.distinctive;
+        const graphTopics = this.topicMap.buildTopics(total, buildTopicStopwords());
+        const keywordTopics = buildKeywordSeedTopics(keywords, keywordsDistinctive, total, this.topicMap);
+        const semanticTopics = finalizeOpts?.useEmbeddingTopics && this.semanticThemeCandidates.length > 0
+            ? semanticItemsToTopics(this.semanticThemeCandidates, total)
+            : [];
+        let topics = mergeTopicLanes({ graph: graphTopics, keyword: keywordTopics, semantic: semanticTopics }, total);
         const burstDetectionMethod = resolveBurstDetectionMethod();
         const keywordTop1SharePercent = top1ShareFromCounts(keywords, total);
         let attachmentMarkerSum = 0;
@@ -661,9 +754,13 @@ export class ReportAggregator {
                 nightSharePercent,
                 emojiMessages: this.emojiMessages,
                 usedSemanticKeywords: finalizeOpts?.usedSemanticKeywords === true,
+                usedSentimentAnalysis: finalizeOpts?.usedSentimentAnalysis === true,
             },
             insights,
             participants: participantStats,
+            participantsByCharacters,
+            profanity: this.profanityCounter.buildProfanityStats(total, aliases),
+            sentiment: this.sentimentStats,
             daily: dailySorted,
             hourly: this.hourly,
             weekdays: this.weekdays.map((count, index) => ({
@@ -674,6 +771,7 @@ export class ReportAggregator {
             attachments: topCounts(this.attachments, this.top),
             domains: topCounts(this.domains, this.top),
             keywords,
+            keywordsDistinctive,
             topics,
             roomEvents: buildRoomEventStats(total, {
                 join: this.roomJoinMessages,
@@ -687,9 +785,14 @@ export class ReportAggregator {
                 manager: this.roomManagerMessages,
                 shopSearch: this.roomShopSearchMessages,
                 photoBundle: this.roomPhotoBundleMessages,
+            }, {
+                tagExtractions: [...this.shopSearchTopics.values()].reduce((a, n) => a + n, 0),
+                uniqueTags: this.shopSearchTopics.size,
+                untaggedNotices: this.shopSearchUntaggedNotices,
             }),
             repeatedPhrases: this.repeatPhraseCounter.top(8, 3),
-            shopSearchTopics: topCounts(this.shopSearchTopics, 10),
+            shopSearchTopics: topCounts(this.shopSearchTopics, shopSearchDisplayTop()),
+            shopSearchMissSamples: process.env.KCA_DEBUG_SHOP === "1" ? [...this.shopSearchMissSamples] : undefined,
             pureLaughMessages: this.pureLaughMessages,
             conversationPace,
             burstDays,
@@ -771,29 +874,7 @@ function getParticipantStat(stats, sender) {
     stats.set(sender, created);
     return created;
 }
-function shouldExtractKeywords(message, attachmentMarkers) {
-    const trimmed = message.trim();
-    if (trimmed.length === 0)
-        return false;
-    if (attachmentMarkers.length === 1 && trimmed === attachmentMarkers[0])
-        return false;
-    if (attachmentMarkers.length > 0 && trimmed.length <= 16) {
-        const onlyMarkers = attachmentMarkers.every((m) => trimmed === m || trimmed.includes(m));
-        if (onlyMarkers && !/[가-힣A-Za-z]{3,}/.test(trimmed.replace(/[^\p{L}\p{N}]/gu, ""))) {
-            return false;
-        }
-    }
-    return true;
-}
-function getAttachmentMarkers(message) {
-    const found = ATTACHMENT_MARKERS.filter((marker) => message.includes(marker));
-    const t = message.trim();
-    if (PHOTO_BUNDLE_RE.test(t) && !found.includes("사진")) {
-        found.push("사진");
-    }
-    return found;
-}
-function buildRoomEventStats(total, c) {
+function buildRoomEventStats(total, c, shopExtra) {
     const sum = c.join +
         c.leave +
         c.deleted +
@@ -817,6 +898,9 @@ function buildRoomEventStats(total, c) {
         subManagerCount: c.subManager,
         managerCount: c.manager,
         shopSearchCount: c.shopSearch,
+        shopSearchTagExtractions: shopExtra?.tagExtractions ?? 0,
+        shopSearchUniqueTags: shopExtra?.uniqueTags ?? 0,
+        shopSearchUntaggedNotices: shopExtra?.untaggedNotices ?? 0,
         photoBundleCount: c.photoBundle,
         total: sum,
         joinSharePercent: pct(c.join),