Skip to content

[DP-463] YouTube 추천 알고리즘 개선 #161

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
nYeonG4001 merged 6 commits into from
May 7, 2026
Merged

[DP-463] YouTube 추천 알고리즘 개선 #161

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
ce3bac2
DP-463: YouTube 추천 알고리즘 개선 — 행동 가중 태그 점수 기반 개인화 + 채널 다양성 페널티 + 탐색 여지
nYeonG4001 May 7, 2026
74b4442
DP-463: 테스트 커버리지 보완 — extractChannel/parseExtra 예외 경로 추가
nYeonG4001 May 7, 2026
c15ae7c
DP-463: buildWeightedTagScores NPE 방지 — Object[] 언박싱 전 null 체크 추가
nYeonG4001 May 7, 2026
d9e5e35
DP-463: applyChannelDiversityPenalty 무한루프 방지 — best null 시 break 추가
nYeonG4001 May 7, 2026
a49a76d
DP-463: NPE 방지 — getTag()/getId() null 체크 및 extractChannel fallback 개선
nYeonG4001 May 7, 2026
99fd08e
DP-463: SonarCloud Reliability 수정 — int→long 캐스트, 테스트 assertion 개선
nYeonG4001 May 7, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 18 additions & 5 deletions src/main/java/com/devpick/domain/content/repository/ContentRepository.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,20 +59,33 @@ List<Content> findLatestExcludingYoutubeAndScrapped(
@Query("SELECT c FROM Content c " +
"WHERE c.isAvailable = true " +
"AND c.source.name = 'YouTube' " +
"AND LOWER(c.title) LIKE LOWER(CONCAT('%', :tagName, '%')) " +
"AND NOT EXISTS (SELECT s FROM Scrap s WHERE s.user.id = :userId AND s.content = c) " +
"ORDER BY c.publishedAt DESC")
List<Content> findYoutubeByTagNameInTitle(
@Param("tagName") String tagName,
List<Content> findLatestYoutubeExcludingScrapped(
@Param("userId") UUID userId,
Pageable pageable);

@Query("SELECT c FROM Content c " +
/** YouTube 추천용: content_tags JOIN, 스크랩 제외 */
@Query("SELECT DISTINCT c FROM Content c JOIN c.contentTags ct " +
"WHERE c.isAvailable = true " +
"AND c.source.name = 'YouTube' " +
"AND ct.tag.id IN :tagIds " +
"AND NOT EXISTS (SELECT s FROM Scrap s WHERE s.user.id = :userId AND s.content = c) " +
"ORDER BY c.publishedAt DESC")
List<Content> findLatestYoutubeExcludingScrapped(
List<Content> findYoutubeByTagIdsExcludingScrapped(
@Param("tagIds") List<UUID> tagIds,
@Param("userId") UUID userId,
Pageable pageable);

/** YouTube 추천용: 탐색 여지 — 관심 태그 외 영역의 YouTube 영상 */
@Query("SELECT DISTINCT c FROM Content c JOIN c.contentTags ct " +
"WHERE c.isAvailable = true " +
"AND c.source.name = 'YouTube' " +
"AND ct.tag.id NOT IN :excludeTagIds " +
"AND NOT EXISTS (SELECT s FROM Scrap s WHERE s.user.id = :userId AND s.content = c) " +
"ORDER BY c.publishedAt DESC")
List<Content> findYoutubeByExcludeTagIdsExcludingScrapped(
@Param("excludeTagIds") List<UUID> excludeTagIds,
@Param("userId") UUID userId,
Pageable pageable);
}
190 changes: 158 additions & 32 deletions src/main/java/com/devpick/domain/content/service/RecommendService.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,8 +25,12 @@
import java.time.LocalDateTime;
import java.time.ZoneId;
import java.time.format.DateTimeFormatter;
import java.time.temporal.ChronoUnit;
import java.util.ArrayList;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
import java.util.HashSet;
import java.util.LinkedHashSet;
import java.util.List;
import java.util.Map;
Expand All @@ -43,11 +47,21 @@ public class RecommendService {

static final int CANDIDATE_LIMIT = 100;
static final int RESULT_SIZE = 8;
static final int TOP_TAGS_LIMIT = 15;
static final int EXPLORE_SIZE = 2;
static final int PERSONALIZED_SIZE = RESULT_SIZE - EXPLORE_SIZE;
static final String REDIS_KEY_PREFIX = "recommend:tags:";
static final long CACHE_TTL_HOURS = 24;
static final String NOT_ENOUGH_MESSAGE = "아직 추천할 글이 부족해요. 더 많은 글을 읽어보세요!";
static final List<String> HISTORY_ACTION_TYPES =
List.of("scrapped", "ai_summary_viewed", "ai_quiz_completed", "content_liked");
static final Map<String, Double> ACTION_WEIGHTS = Map.of(
"ai_quiz_completed", 5.0,
"scrapped", 4.0,
"content_liked", 3.0,
"ai_summary_viewed", 2.0,
"content_opened", 1.0
);
static final ZoneId KST = ZoneId.of("Asia/Seoul");

private final HistoryRepository historyRepository;
Expand Down Expand Up @@ -105,22 +119,6 @@ private List<Content> findByTagNamesInTitle(List<String> tagNames, UUID userId,
return result;
}

private List<Content> findYoutubeByTagNamesInTitle(List<String> tagNames, UUID userId, int limit) {
Set<UUID> seen = new LinkedHashSet<>();
List<Content> result = new ArrayList<>();
for (String tagName : tagNames) {
for (Content c : contentRepository.findYoutubeByTagNameInTitle(
tagName, userId, PageRequest.of(0, limit))) {
UUID id = c.getId();
if (id != null && seen.add(id)) {
result.add(c);
if (result.size() >= limit) return result;
}
}
}
return result;
}

List<UUID> getOrCacheTagIds(UUID userId) {
String today = LocalDate.now(KST).format(DateTimeFormatter.ISO_LOCAL_DATE);
String redisKey = REDIS_KEY_PREFIX + userId + ":" + today;
Expand Down Expand Up @@ -165,36 +163,164 @@ private RecommendContentsResponse buildResponse(
return new RecommendContentsResponse(items, isPersonalized, message);
}

// ─── YouTube 추천 (DP-463) ────────────────────────────────────────────────

@Transactional(readOnly = true)
public YoutubeRecommendResponse getRecommendYoutube(UUID userId) {
List<UUID> tagIds = getOrCacheTagIds(userId);
// 1. 행동 이력 기반 가중 태그 점수맵
Map<UUID, Double> tagScores = buildWeightedTagScores(userId);

if (!tagIds.isEmpty()) {
List<String> tagNames = tagRepository.findAllById(tagIds).stream()
.map(Tag::getName).toList();
if (!tagNames.isEmpty()) {
List<Content> candidates = findYoutubeByTagNamesInTitle(tagNames, userId, CANDIDATE_LIMIT);
if (candidates.size() >= RESULT_SIZE) {
return buildYoutubeResponse(shuffleAndTake(candidates, userId), userId, true, null);
}
}
// 2. 최근 3개월 시청 이력 (재추천 방지)
Set<UUID> viewedIds = new HashSet<>(
historyRepository.findViewedContentIdsSince(userId, LocalDateTime.now(KST).minusMonths(3)));

// 3. 상위 태그 ID (이력 없으면 프로필 태그)
List<UUID> topTagIds = getTopTagIds(tagScores, TOP_TAGS_LIMIT);
boolean isPersonalized = !tagScores.isEmpty();

if (topTagIds.isEmpty()) {
topTagIds = userTagRepository.findByUser_Id(userId).stream()
.filter(ut -> ut.getTag() != null && ut.getTag().getId() != null)
.map(ut -> ut.getTag().getId()).toList();
}

List<String> userTagNames = userTagRepository.findByUser_Id(userId).stream()
.map(ut -> ut.getTag().getName()).toList();
if (!topTagIds.isEmpty()) {
// 4. content_tags JOIN으로 YouTube 후보 조회 (스크랩 제외)
List<Content> candidates = contentRepository.findYoutubeByTagIdsExcludingScrapped(
topTagIds, userId, PageRequest.of(0, CANDIDATE_LIMIT * 2));

if (!userTagNames.isEmpty()) {
List<Content> candidates = findYoutubeByTagNamesInTitle(userTagNames, userId, CANDIDATE_LIMIT);
if (!candidates.isEmpty()) {
return buildYoutubeResponse(shuffleAndTake(candidates, userId), userId, true, null);
// 5. 시청 이력 제외 후 채널 다양성 페널티 적용 랭킹
List<Content> ranked = applyChannelDiversityPenalty(
candidates.stream().filter(c -> !viewedIds.contains(c.getId())).toList(),
tagScores);

if (!ranked.isEmpty()) {
List<Content> result = new ArrayList<>(
ranked.subList(0, Math.min(PERSONALIZED_SIZE, ranked.size())));

// 6. 탐색 여지: 관심 태그 외 영역 영상 2개
Set<UUID> resultIds = new HashSet<>(viewedIds);
result.stream().map(Content::getId).forEach(resultIds::add);
List<Content> explore = findExploreVideos(topTagIds, userId, resultIds, RESULT_SIZE - result.size());
result.addAll(explore);

// 7. 여전히 부족하면 최신 YouTube로 채움
if (result.size() < RESULT_SIZE) {
resultIds = new HashSet<>(viewedIds);
result.stream().map(Content::getId).forEach(resultIds::add);
Set<UUID> finalResultIds = resultIds;
contentRepository.findLatestYoutubeExcludingScrapped(userId, PageRequest.of(0, RESULT_SIZE))
.stream()
.filter(c -> !finalResultIds.contains(c.getId()))
.limit((long) RESULT_SIZE - result.size())
.forEach(result::add);
}

return buildYoutubeResponse(
result.subList(0, Math.min(RESULT_SIZE, result.size())),
userId, isPersonalized, null);
}
}

// 8. Cold start: 최신 YouTube
List<Content> latest = contentRepository.findLatestYoutubeExcludingScrapped(
userId, PageRequest.of(0, CANDIDATE_LIMIT));
return buildYoutubeResponse(shuffleAndTake(latest, userId), userId, false, NOT_ENOUGH_MESSAGE);
}

Map<UUID, Double> buildWeightedTagScores(UUID userId) {
Map<UUID, Double> scores = new HashMap<>();
List<String> actionTypes = new ArrayList<>(ACTION_WEIGHTS.keySet());

List<Object[]> rows = historyRepository.findTagIdActionCountsByUserActionsAfter(
userId, actionTypes, LocalDateTime.now(KST).minusMonths(1));
if (rows.isEmpty()) {
rows = historyRepository.findTagIdActionCountsByUserActionsAfter(
userId, actionTypes, LocalDateTime.now(KST).minusMonths(3));
}

for (Object[] row : rows) {
UUID tagId = (UUID) row[0];
Number rawCount = (Number) row[2];
if (tagId == null || rawCount == null) continue;
String actionType = (String) row[1];
scores.merge(tagId, ACTION_WEIGHTS.getOrDefault(actionType, 1.0) * rawCount.longValue(), Double::sum);
}
return scores;
}

List<UUID> getTopTagIds(Map<UUID, Double> tagScores, int limit) {
return tagScores.entrySet().stream()
.sorted(Map.Entry.<UUID, Double>comparingByValue().reversed())
.limit(limit)
.map(Map.Entry::getKey)
.toList();
}

double computeScore(Content content, Map<UUID, Double> tagScores) {
double tagScore = content.getContentTags().stream()
.filter(ct -> ct.getTag() != null && ct.getTag().getId() != null)
.mapToDouble(ct -> tagScores.getOrDefault(ct.getTag().getId(), 0.0))
.sum();
double recencyBonus = 0;
if (content.getPublishedAt() != null) {
long daysOld = ChronoUnit.DAYS.between(content.getPublishedAt(), LocalDateTime.now(KST));
recencyBonus = Math.max(0, 30 - daysOld) * 0.1;
}
return tagScore + recencyBonus;
}

List<Content> applyChannelDiversityPenalty(List<Content> candidates, Map<UUID, Double> tagScores) {
Map<String, Integer> channelCount = new HashMap<>();
List<Content> pool = new ArrayList<>(candidates);
List<Content> result = new ArrayList<>();

while (!pool.isEmpty() && result.size() < candidates.size()) {
Content best = null;
double bestScore = Double.NEGATIVE_INFINITY;
for (Content c : pool) {
String channel = extractChannel(c);
int count = channelCount.getOrDefault(channel, 0);
double penalty = count >= 1 ? 2.0 * count : 0;
double score = computeScore(c, tagScores) - penalty;
if (score > bestScore) {
bestScore = score;
best = c;
}
}
if (best != null) {
result.add(best);
channelCount.merge(extractChannel(best), 1, Integer::sum);
pool.remove(best);
} else {
break;
}
}
return result;
}

String extractChannel(Content content) {
String fallback = content.getId() != null ? content.getId().toString() : content.getCanonicalUrl();
if (content.getExtra() == null) return fallback;
try {
Map<String, Object> extra = objectMapper.readValue(content.getExtra(), new TypeReference<Map<String, Object>>() {});
Object channelName = extra.get("channelName");
return channelName != null ? channelName.toString() : fallback;
} catch (JsonProcessingException e) {
return fallback;
}
}

private List<Content> findExploreVideos(List<UUID> usedTagIds, UUID userId, Set<UUID> excludeIds, int limit) {
if (limit <= 0 || usedTagIds.isEmpty()) return List.of();
return contentRepository.findYoutubeByExcludeTagIdsExcludingScrapped(
usedTagIds, userId, PageRequest.of(0, limit * 3))
.stream()
.filter(c -> !excludeIds.contains(c.getId()))
.limit(limit)
.toList();
}

private YoutubeRecommendResponse buildYoutubeResponse(
List<Content> contents, UUID userId, boolean isPersonalized, String message) {
List<YoutubeRecommendItem> items = contents.stream()
Expand Down
23 changes: 23 additions & 0 deletions src/main/java/com/devpick/domain/report/repository/HistoryRepository.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,6 +15,29 @@

public interface HistoryRepository extends JpaRepository<History, UUID> {

/** YouTube 추천용: 태그 ID + 액션 타입 + 횟수 (가중 점수 계산용) */
@Query("SELECT ct.tag.id, h.actionType, COUNT(h) FROM History h " +
"JOIN h.content c JOIN c.contentTags ct " +
"WHERE h.user.id = :userId " +
"AND h.actionType IN :actionTypes " +
"AND h.createdAt >= :since " +
"AND h.content IS NOT NULL " +
"GROUP BY ct.tag.id, h.actionType " +
"ORDER BY COUNT(h) DESC")
List<Object[]> findTagIdActionCountsByUserActionsAfter(
@Param("userId") UUID userId,
@Param("actionTypes") List<String> actionTypes,
@Param("since") LocalDateTime since);

/** YouTube 추천용: 사용자가 열람한 콘텐츠 ID (중복 노출 방지) */
@Query("SELECT DISTINCT h.content.id FROM History h " +
"WHERE h.user.id = :userId " +
"AND h.content IS NOT NULL " +
"AND h.createdAt >= :since")
List<UUID> findViewedContentIdsSince(
@Param("userId") UUID userId,
@Param("since") LocalDateTime since);

/** 도서 추천용: 태그명 + 빈도 수 조회 */
@Query("SELECT ct.tag.name, COUNT(ct.tag.name) FROM History h " +
"JOIN h.content c JOIN c.contentTags ct " +
Expand Down
Loading
Loading