feat: 评论去重 + 热度/新鲜度加权选文
评论去重逻辑: - 查询今日已评论的文章ID,选文时已评论的文章权重降为10% - 若选中已评论文章:改为回复其他用户的评论(虚拟用户互动链) - 若选中未评论文章:正常发新评论,评论成功后随机回复他人评论 热度+新鲜度加权选文规则: - 热度分 = commentNum×3 + praiseNum×2 + readNum×1 - 新鲜度 = 72小时内的新文章获得最高3倍加成,随时间线性衰减 - 综合权重 = (热度分+1) × 新鲜度,确保真实用户互动多的新文章优先被虚拟用户关注
This commit is contained in:
@@ -407,7 +407,54 @@ class NewsPlatformService:
|
||||
if (x.get("recordId") or x.get("id")) not in INVALID_IDS]
|
||||
logger.info(f"[广场新闻] {user.account} 获取到 {len(items)} 条(已过滤本人+无效文章)")
|
||||
import random as _rand
|
||||
return _rand.sample(items, min(count, len(items))) if items else []
|
||||
from datetime import datetime as _dt
|
||||
import math as _math
|
||||
|
||||
# ── 热度 + 新鲜度加权选取 ─────────────────────────────────
|
||||
# 规则:真实用户互动量越大 + 发布时间越新 → 虚拟用户越倾向互动
|
||||
def _hot_weight(a):
|
||||
comment_n = int(a.get("commentNum") or 0)
|
||||
praise_n = int(a.get("praiseNum") or 0)
|
||||
read_n = int(a.get("readNum") or 0)
|
||||
# 热度分:评论权重3倍,点赞2倍,阅读1倍
|
||||
hot_score = comment_n * 3 + praise_n * 2 + read_n
|
||||
|
||||
# 新鲜度衰减:发布时间越近权重越高(72小时内为新鲜文章)
|
||||
freshness = 1.0
|
||||
pub_time_str = a.get("publishTime") or a.get("createTime") or ""
|
||||
if pub_time_str:
|
||||
try:
|
||||
for fmt in ["%Y-%m-%d %H:%M:%S", "%Y-%m-%dT%H:%M:%S"]:
|
||||
try:
|
||||
pub_dt = _dt.strptime(pub_time_str[:19], fmt)
|
||||
hours_old = (_dt.now() - pub_dt).total_seconds() / 3600
|
||||
# 72小时内新鲜文章:新鲜度加成最高3倍
|
||||
freshness = max(1.0, 3.0 - hours_old / 36.0)
|
||||
break
|
||||
except Exception:
|
||||
continue
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# 综合权重:热度 * 新鲜度(基础权重最少为1)
|
||||
return max(1.0, (hot_score + 1) * freshness)
|
||||
|
||||
if len(items) <= count:
|
||||
return items
|
||||
# 加权随机采样
|
||||
weights = [_hot_weight(a) for a in items]
|
||||
selected = []
|
||||
pool = list(range(len(items)))
|
||||
w_pool = list(weights)
|
||||
for _ in range(min(count, len(items))):
|
||||
if not pool:
|
||||
break
|
||||
chosen_idx = _rand.choices(pool, weights=w_pool, k=1)[0]
|
||||
selected.append(items[chosen_idx])
|
||||
i = pool.index(chosen_idx)
|
||||
pool.pop(i)
|
||||
w_pool.pop(i)
|
||||
return selected
|
||||
logger.warning(f"[广场新闻] {user.account} code={d.get('code')} msg={d.get('message')}")
|
||||
except Exception as e:
|
||||
logger.error(f"[广场新闻] {user.account}: {e}")
|
||||
|
||||
@@ -230,10 +230,43 @@ class SchedulerService:
|
||||
)
|
||||
return
|
||||
|
||||
article = random.choice(articles)
|
||||
# ── 文章去重 + 热度加权选取 ─────────────────────────────────
|
||||
# 查询今日已评论过的文章ID(避免重复评论同一篇)
|
||||
from sqlalchemy import func as _func
|
||||
from datetime import date as _date
|
||||
today_str = datetime.now().date()
|
||||
dup_result = await db.execute(
|
||||
select(InteractionRecord.article_id).where(
|
||||
InteractionRecord.user_id == user_id,
|
||||
InteractionRecord.interact_type == "comment",
|
||||
InteractionRecord.status == 1,
|
||||
_func.date(InteractionRecord.executed_at) == today_str,
|
||||
)
|
||||
)
|
||||
already_commented = {r[0] for r in dup_result.all()}
|
||||
|
||||
# 按热度加权:commentNum + praiseNum + readNum 越高权重越大
|
||||
# 同时优先未评论过的文章
|
||||
def _article_weight(a):
|
||||
aid = str(a.get("recordId") or a.get("id", ""))
|
||||
base = (
|
||||
int(a.get("commentNum") or 0) * 3 +
|
||||
int(a.get("praiseNum") or 0) * 2 +
|
||||
int(a.get("readNum") or 0)
|
||||
)
|
||||
# 已评论的文章权重大幅降低(但不为0,还可以点赞/收藏)
|
||||
penalty = 0.1 if aid in already_commented else 1.0
|
||||
return max(1, base) * penalty
|
||||
|
||||
weights = [_article_weight(a) for a in articles]
|
||||
article = random.choices(articles, weights=weights, k=1)[0]
|
||||
|
||||
# 判断是否已评论此文章(用于后续逻辑)
|
||||
news_id = str(article.get("recordId") or article.get("id", ""))
|
||||
already_commented_this = news_id in already_commented
|
||||
|
||||
# 接口返回字段: id/newsTitle/content/digest/createUser
|
||||
# 广场接口字段:recordId=新闻实际ID, id=广场记录ID, title=标题
|
||||
news_id = str(article.get("recordId") or article.get("id", ""))
|
||||
news_title = article.get("title") or article.get("newsTitle") or "未知文章"
|
||||
news_content = article.get("content") or article.get("digest") or news_title
|
||||
news_author = str(article.get("createUser") or "")
|
||||
@@ -278,11 +311,48 @@ class SchedulerService:
|
||||
interactions_done.append("forward")
|
||||
await self._incr_total(db, user_id)
|
||||
|
||||
# ⑤ 评论(AI生成内容,调用 POST /message/comment)
|
||||
if can_comment and random.random() < comment_prob and personality:
|
||||
# ⑤ 评论/回复逻辑(去重:已评论过的文章改为回复他人评论)
|
||||
if can_comment and personality:
|
||||
style_prompt = personality.comment_style_prompt or ""
|
||||
# 字数上限最多80字,避免超出 max_tokens 被截断
|
||||
safe_word_max = min(personality.word_count_max, 80)
|
||||
|
||||
if already_commented_this:
|
||||
# 已评论过此文章 → 改为回复其他用户的评论(虚拟用户互动)
|
||||
if random.random() < reply_prob:
|
||||
existing = await news_service.get_comments(db, user, news_id)
|
||||
if existing:
|
||||
# 优先回复虚拟用户的评论(促进互动),过滤掉自己的评论
|
||||
from app.core.redis_client import get_session as _gs
|
||||
my_sess = await _gs(user.id)
|
||||
my_uid = my_sess.get("platform_uid", "") if my_sess else ""
|
||||
others = [c for c in existing
|
||||
if str(c.get("userId") or c.get("createUser") or "") != my_uid]
|
||||
if others:
|
||||
target = random.choice(others)
|
||||
cid = str(target.get("id") or target.get("commentId") or "")
|
||||
parent_content = target.get("content") or ""
|
||||
if cid:
|
||||
reply_text, r_tokens = await ai_service.generate_reply(
|
||||
db, news_title, parent_content,
|
||||
style_prompt,
|
||||
personality.word_count_min,
|
||||
safe_word_max
|
||||
)
|
||||
if reply_text:
|
||||
r_ok, r_err = await news_service.post_reply(
|
||||
db, user, news_id, cid, reply_text
|
||||
)
|
||||
await self._save_record(
|
||||
db, user, news_id, news_title, "reply",
|
||||
reply_text, r_tokens, r_ok, r_err,
|
||||
parent_comment_id=cid
|
||||
)
|
||||
if r_ok:
|
||||
interactions_done.append("reply")
|
||||
logger.info(f"💬 {user.account} 回复了已评论文章的评论(去重逻辑)")
|
||||
else:
|
||||
# 未评论过此文章 → 正常发评论
|
||||
if random.random() < comment_prob:
|
||||
comment_text, tokens = await ai_service.generate_comment(
|
||||
db, news_title, news_content,
|
||||
style_prompt, personality.word_count_min, safe_word_max
|
||||
@@ -306,30 +376,36 @@ class SchedulerService:
|
||||
)
|
||||
)
|
||||
|
||||
# ⑥ 回复评论(评论成功后,随机回复别人的评论)
|
||||
# ⑥ 评论成功后,随机回复其他用户的评论(互动链)
|
||||
if random.random() < reply_prob:
|
||||
existing = await news_service.get_comments(db, user, news_id)
|
||||
if existing:
|
||||
target = random.choice(existing)
|
||||
cid = str(target.get("id") or target.get("commentId") or "")
|
||||
parent_content = target.get("content") or ""
|
||||
if cid:
|
||||
reply_text, r_tokens = await ai_service.generate_reply(
|
||||
db, news_title, parent_content,
|
||||
from app.core.redis_client import get_session as _gs2
|
||||
my_sess2 = await _gs2(user.id)
|
||||
my_uid2 = my_sess2.get("platform_uid", "") if my_sess2 else ""
|
||||
others2 = [c for c in existing
|
||||
if str(c.get("userId") or c.get("createUser") or "") != my_uid2]
|
||||
if others2:
|
||||
target2 = random.choice(others2)
|
||||
cid2 = str(target2.get("id") or target2.get("commentId") or "")
|
||||
parent_content2 = target2.get("content") or ""
|
||||
if cid2:
|
||||
reply_text2, r_tokens2 = await ai_service.generate_reply(
|
||||
db, news_title, parent_content2,
|
||||
style_prompt,
|
||||
personality.word_count_min,
|
||||
personality.word_count_max
|
||||
safe_word_max
|
||||
)
|
||||
if reply_text:
|
||||
r_ok, r_err = await news_service.post_reply(
|
||||
db, user, news_id, cid, reply_text
|
||||
if reply_text2:
|
||||
r_ok2, r_err2 = await news_service.post_reply(
|
||||
db, user, news_id, cid2, reply_text2
|
||||
)
|
||||
await self._save_record(
|
||||
db, user, news_id, news_title, "reply",
|
||||
reply_text, r_tokens, r_ok, r_err,
|
||||
parent_comment_id=cid
|
||||
reply_text2, r_tokens2, r_ok2, r_err2,
|
||||
parent_comment_id=cid2
|
||||
)
|
||||
if r_ok:
|
||||
if r_ok2:
|
||||
interactions_done.append("reply")
|
||||
|
||||
await db.commit()
|
||||
|
||||
Reference in New Issue
Block a user