@seojin08/skill-finder 0.0.0-stage → 0.3.2

package/server/src/search/koreanTerms.js

@@ -0,0 +1,284 @@
+import { GENERIC_WORDS } from "./rank.js";
+// Korean words → the English words skill descriptions actually use. Rule-based only (no AI, user decision):
+// the search suggests the English query as a button, and the README recommendation maps Korean READMEs with it.
+export const KO_EN = {
+    // xlsx and excel first: the words skill names and descriptions use most (spreadsheet narrowed the search)
+    엑셀: ['xlsx', 'excel', 'spreadsheet'],
+    스프레드시트: ['spreadsheet'],
+    표: ['table'],
+    추출: ['extract'],
+    변환: ['convert'],
+    요약: ['summarize'],
+    번역: ['translate'],
+    리액트: ['react'],
+    워드: ['docx', 'word'],
+    파워포인트: ['pptx', 'powerpoint'],
+    피피티: ['pptx', 'slides'],
+    슬라이드: ['slides'],
+    발표: ['presentation'],
+    테스트: ['testing'],
+    단위: ['unit'],
+    브라우저: ['browser'],
+    데이터베이스: ['database', 'sql'],
+    배포: ['deploy', 'deployment'],
+    보안: ['security'],
+    취약점: ['vulnerability'],
+    디자인: ['design'],
+    문서: ['document', 'documentation'],
+    이미지: ['image'],
+    서버: ['server'],
+    깃허브: ['github'],
+    깃: ['git'],
+    커밋: ['commit'],
+    리뷰: ['review'],
+    검토: ['review'],
+    풀리퀘스트: ['pull', 'request'],
+    이슈: ['issue'],
+    자동화: ['automation'],
+    크롤링: ['scraping'],
+    스크래핑: ['scraping'],
+    웹: ['web'],
+    앱: ['app'],
+    모바일: ['mobile'],
+    프론트엔드: ['frontend'],
+    백엔드: ['backend'],
+    성능: ['performance'],
+    최적화: ['optimization'],
+    디버깅: ['debugging'],
+    디버그: ['debug'],
+    로그: ['logging'],
+    리팩토링: ['refactoring'],
+    리팩터링: ['refactoring'],
+    마이그레이션: ['migration'],
+    차트: ['chart'],
+    그래프: ['chart'],
+    시각화: ['visualization'],
+    분석: ['analysis'],
+    데이터: ['data'],
+    이메일: ['email'],
+    메일: ['email'],
+    캘린더: ['calendar'],
+    일정: ['calendar'],
+    영상: ['video'],
+    동영상: ['video'],
+    음성: ['audio'],
+    회의록: ['meeting', 'notes'],
+    릴리스: ['release'],
+    릴리즈: ['release'],
+    노트: ['notes'],
+    마크다운: ['markdown'],
+    파이썬: ['python'],
+    자바스크립트: ['javascript'],
+    타입스크립트: ['typescript'],
+    도커: ['docker'],
+    쿠버네티스: ['kubernetes'],
+    인프라: ['infrastructure'],
+    클라우드: ['cloud'],
+    결제: ['payment'],
+    인증: ['authentication'],
+    로그인: ['login'],
+    검색: ['search'],
+    기획: ['planning'],
+    글쓰기: ['writing'],
+    블로그: ['blog'],
+    피그마: ['figma'],
+    노션: ['notion'],
+    슬랙: ['slack'],
+    // second batch (search debate, panel 7): words real Korean queries use that the first list lacked
+    코드: ['code'],
+    메시지: ['message'],
+    점검: ['audit'],
+    감사: ['audit'],
+    웹사이트: ['website'],
+    사이트: ['website'],
+    크롤러: ['scraper'],
+    프롬프트: ['prompt'],
+    에이전트: ['agent'],
+    스킬: ['skill'],
+    버그: ['bug'],
+    수정: ['fix'],
+    오류: ['error'],
+    에러: ['error'],
+    타입: ['type'],
+    로고: ['logo'],
+    알림: ['notification'],
+    회의: ['meeting'],
+    문서화: ['documentation'],
+    생성: ['generate'],
+    한글: ['hwp'],
+    파일: ['file'],
+    플러그인: ['plugin'],
+    터미널: ['terminal'],
+    스크립트: ['script'],
+    템플릿: ['template'],
+    대시보드: ['dashboard'],
+    컴포넌트: ['component'],
+    함수: ['function'],
+    린트: ['lint'],
+    포맷: ['format'],
+    정규식: ['regex'],
+    브랜치: ['branch'],
+    머지: ['merge'],
+    충돌: ['conflict'],
+    변경로그: ['changelog'],
+    보고서: ['report'],
+    리포트: ['report'],
+    이력서: ['resume'],
+    계약서: ['contract'],
+    자막: ['subtitle'],
+    썸네일: ['thumbnail'],
+    아이콘: ['icon'],
+    폰트: ['font'],
+    접근성: ['accessibility'],
+    마케팅: ['marketing'],
+    광고: ['ads'],
+    게임: ['game'],
+    유니티: ['unity'],
+    안드로이드: ['android'],
+    스위프트: ['swift'],
+    코틀린: ['kotlin'],
+    자바: ['java'],
+    러스트: ['rust'],
+    스프링: ['spring'],
+    장고: ['django'],
+    넥스트: ['nextjs'],
+    테일윈드: ['tailwind'],
+    수파베이스: ['supabase'],
+    파이어베이스: ['firebase'],
+    모니터링: ['monitoring'],
+    백업: ['backup'],
+    머신러닝: ['machine', 'learning'],
+    챗봇: ['chatbot'],
+    웹훅: ['webhook'],
+    큐: ['queue'],
+    캐시: ['cache'],
+    쿼리: ['query'],
+    스키마: ['schema'],
+    랜딩페이지: ['landing', 'page'],
+    크롬: ['chrome'],
+    확장: ['extension'],
+    유튜브: ['youtube'],
+    트위터: ['twitter'],
+    인스타그램: ['instagram'],
+    주식: ['stock'],
+    회계: ['accounting'],
+    세금: ['tax'],
+    논문: ['paper'],
+    연구: ['research'],
+    그림: ['image'],
+    사진: ['photo'],
+    음악: ['music'],
+};
+/** Endings a Korean noun takes in a query ("엑셀로", "표를", "정리해줘" → the noun before them). */
+const PARTICLES = ['', '을', '를', '이', '가', '은', '는', '의', '에', '에서', '로', '으로', '와', '과', '도', '만', '까지', '부터', '하기', '하는', '해', '해줘', '하고', '용'];
+/** More endings, for splitting a query word into dictionary words (koreanSuggestion). */
+// (no "화": "한글화" is localization, not 한글 + an ending; "문서화" has its own entry)
+const ENDINGS = [...PARTICLES.slice(1), '에게', '한테', '랑', '이랑', '처럼', '들', '들을', '할', '한', '하자', '해서', '하려면', '시키기'];
+/**
+ * Words that say what to do rather than what about ("정리", "만들기"): dropped from the English query rather
+ * than blocking it. Those in REQUEST_WORDS are not even worth showing as dropped.
+ */
+const EMPTY_WORDS = ['정리', '작성', '만들기', '만들어', '만들어줘', '만드는', '찾기', '찾아줘', '해줘', '해', '하기', '그리기', '좀', '관련', '방법', '내용', '도구', '자동', '자료'];
+const REQUEST_WORDS = new Set(['해줘', '해', '하기', '좀', '만들어', '만들어줘', '찾아줘']);
+/** At most this many dropped words of at most this many characters are listed (a pasted sentence stays short). */
+const DROPPED_MAX = 5;
+const DROPPED_CHARS = 12;
+/**
+ * The words the README recommendation maps (the specific ones: "데이터", "웹" or "검색" in a README say little
+ * about which skills fit). It also takes any ending ("문서화", "테스트합니다"), as it always has.
+ */
+export const README_TERMS = new Set(['엑셀', '스프레드시트', '리액트', '워드', '파워포인트', '슬라이드', '테스트', '브라우저', '데이터베이스', '배포', '보안', '디자인', '문서', '이미지', '서버', '깃허브']);
+/**
+ * Longest dictionary word the token starts with, when what follows it is only an ending ("엑셀로" → 엑셀,
+ * "표준" is not 표); else null. `readme`: only README_TERMS, and anything may follow.
+ */
+export function koreanTerm(token, readme = false) {
+    let best = null;
+    for (const k of Object.keys(KO_EN)) {
+        if (readme ? !README_TERMS.has(k) : !PARTICLES.includes(token.slice(k.length)))
+            continue;
+        if (token.startsWith(k) && (!best || k.length > best.length))
+            best = k;
+    }
+    return best ? KO_EN[best] : null;
+}
+/** Words longer than this are not split (a pasted sentence without spaces would only waste time). */
+const SPLIT_MAX_CHARS = 40;
+/**
+ * Splits a Korean word into dictionary words and empty words, each optionally followed by one ending
+ * ("코드리뷰" → 코드 + 리뷰, "엑셀표추출해줘" → 엑셀 + 표 + 추출 + 해줘), in the fewest pieces. Null unless
+ * the whole word is covered: "표준" is not 표 + something, "서버리스" is not 서버.
+ */
+function splitKorean(word) {
+    if (word.length > SPLIT_MAX_CHARS)
+        return null;
+    // longest keys first: with as many pieces either way, the first found stays, so "문서화" is one word
+    // (documentation) rather than 문서 + an ending, whatever order the dictionary lists them in
+    const keys = [...Object.keys(KO_EN), ...EMPTY_WORDS].sort((a, b) => b.length - a.length);
+    const best = Array(word.length + 1).fill(null);
+    best[0] = [];
+    for (let i = 0; i < word.length; i++) {
+        const here = best[i];
+        if (!here)
+            continue;
+        for (const k of keys) {
+            if (!word.startsWith(k, i))
+                continue;
+            for (const e of ['', ...ENDINGS]) {
+                if (e && !word.startsWith(e, i + k.length))
+                    continue;
+                const j = i + k.length + e.length;
+                const piece = { key: k, en: EMPTY_WORDS.includes(k) ? null : KO_EN[k] };
+                if (!best[j] || here.length + 1 < best[j].length)
+                    best[j] = [...here, piece];
+            }
+        }
+    }
+    return best[word.length];
+}
+/**
+ * An English query for a Korean one, word by word: Latin words stay ("pdf에서" → "pdf"), dictionary words are
+ * replaced (two synonyms at most, and not "deploy deployment"), other Korean words are dropped and listed.
+ * Null unless at least one Korean word was translated. Rule-based only (no AI, user decision).
+ */
+export function koreanSuggestion(query) {
+    if (!/\p{Script=Hangul}/u.test(query))
+        return null;
+    const out = [];
+    const dropped = [];
+    let translated = false;
+    for (const token of query.toLowerCase().match(/[\p{L}\p{N}]+/gu) ?? []) {
+        out.push(...(token.match(/[a-z0-9]+/g) ?? []).filter((w) => w.length >= 2));
+        const korean = token.replace(/[a-z0-9]+/g, ''); // "pdf에서" → "에서", an ending: nothing to translate
+        if (!korean || ENDINGS.includes(korean))
+            continue;
+        const pieces = splitKorean(korean);
+        if (!pieces) {
+            dropped.push(korean);
+            continue;
+        }
+        for (const p of pieces) {
+            if (!p.en) {
+                if (!REQUEST_WORDS.has(p.key))
+                    dropped.push(p.key);
+                continue;
+            }
+            // two words at most, the second only if it is not a form of the first: more would lower each result's
+            // share of matching words and narrow the all-words search
+            const [a, b] = p.en;
+            out.push(a);
+            if (b && !b.startsWith(a) && !a.startsWith(b))
+                out.push(b);
+            translated = true;
+        }
+    }
+    const words = [...new Set(out)];
+    // "스킬 만들기" → "skill" alone would only lead to the "too broad" notice
+    if (!translated || words.length === 0 || words.every((w) => GENERIC_WORDS.has(w)))
+        return null;
+    return { query: words.join(' '), dropped: [...new Set(dropped)].slice(0, DROPPED_MAX).map((w) => (w.length > DROPPED_CHARS ? `${w.slice(0, DROPPED_CHARS)}…` : w)) };
+}
+/** koreanSuggestion's query alone. */
+export function englishSuggestion(query) {
+    return koreanSuggestion(query)?.query ?? null;
+}