E-commerce review mining: pulling negative reviews from Taobao, JD and 1688
Three endpoints, paginated by item ID: taobao_comments, jd_comments_v2 and item_review_1688. ¥0.1–0.2 per call, failed calls are not charged. One product = 5 Taobao pages + 5 JD pages + 3 1688 pages = 13 calls ≈ ¥2.6.
Which endpoints to call
| Endpoint | Price | Calls | Subtotal | Purpose |
|---|---|---|---|---|
taobao_comments | ¥0.2 | ×5 | ¥1.0 | Taobao item reviews |
jd_comments_v2 | ¥0.2 | ×5 | ¥1.0 | JD reviews (cheaper than jd_comments at ¥0.3) |
item_review_1688 | ¥0.2 | ×3 | ¥0.6 | 1688 reviews (upstream factory quality feedback) |
| per product (13 calls: 5 Taobao + 5 JD + 3 1688) | ¥2.6 | per full run | ||
| 1,000 products | ¥2,600 | 1000 × ¥2.6 | ||
Working code skeleton
Runs as-is once you set TOKEN. Failed and empty calls are skipped, so they cost nothing.
import time, requests, re
TOKEN, API = "fw_xxx", "https://finewisdom.top/api"
NEG = re.compile(r"broken|fake|refund|waste|cheap|smell|late|damaged|wrong|bad", re.I)
def call(method, **kw):
r = requests.get(API, params={"token": TOKEN, "method": method, **kw}, timeout=60).json()
return r["data"] if r.get("status") == "success" else None
def shop_items(shop_name, max_pages=5):
ids = []
for page in range(1, max_pages + 1):
d = call("taobao_shop_search", shop_name=shop_name, page=page)
if not d:
break
items = d.get("list") or d.get("items") or []
if not items:
break
ids += [i.get("item_id") or i.get("num_iid") for i in items]
time.sleep(1.1)
return [i for i in ids if i]
def neg_reviews(item_id, pages=5):
out = []
for page in range(1, pages + 1):
d = call("taobao_comments", item_id=item_id, page=page)
if not d:
break
batch = d.get("comments") or d.get("list") or []
if not batch:
break
out += [c for c in batch if NEG.search(str(c.get("content") or c.get("text") or ""))]
time.sleep(1.1)
return out
def reviews_1688(item_id, seller_nick, pages=3):
out = []
for page in range(1, pages + 1):
d = call("item_review_1688", item_id=item_id, seller_nick=seller_nick, page=page)
if not d:
break
out += d.get("list") or d.get("comments") or []
time.sleep(1.1)
return out
Gotchas worth knowing
item_review_1688requires bothitem_idandseller_nick(1688 slices reviews by merchant), unlike Taobao/JD where the item ID is enough.- Getting item IDs:
taobao_shop_search/jd_shop_searchby shop,page1–35 at about 60 items per page, so 35 pages = ¥3.5 per shop. That is a shop-level fixed cost that does not grow with the number of products. jd_comments(¥0.3) andjd_comments_v2(¥0.2) are different upstreams: v2 is cheaper, returns 10 per page and usespage. Prefer v2 at volume.taobao_original_detail/jd_item_getexclude coupon price and commission — use them for titles and images only. For pricing usetaobao_item_get(pre-coupon price) and keep the definition consistent.
FAQ
Can I request only 1-star reviews?
The endpoints return pages; filtering by star rating is not guaranteed, so you generally pull pages and filter yourself. A keyword pre-filter saves a lot of storage and downstream compute.
What does mining reviews for 10,000 products cost?
At 13 calls per product and ¥0.2 per call, 10,000 products is about 130,000 calls ≈ ¥26,000. Start with a 100-product sample (about ¥260) to validate your parsing and the pagination depth, then scale.