Datasets:
video video 2.93 1.78k |
|---|
YAML Metadata Warning:The task_categories "multimodal-retrieval" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other
CAVFA consumer demand, video supply and product experience
Public research archive for the current CAVFA collection. This repository is the canonical data copy; the companion GitHub repository retains code, lightweight manifests, hashes, and migration notes only.
Contents
data/comments/: 30 de-identified sharing-copy JSONL files, 1,000 sampled shopping-video comments per approved product (30,000 rows total). Stable comment keys are retained; identical text with different comment keys is retained.data/mall_reviews/: 30 separate mall product-review JSONL files, 1,000 main review records per approved product (30,000 total); 29,523 nonempty primary texts, 477 empty primary texts, and 2,081 nested appended reviews. Appends are not extra main-review records.metadata/mall_reviews/: product links and research/native product ID mapping, per-file hashes, counts, time coverage, and collection audit.data/videos/: 450 complete MP4 files, 15 per approved product.metadata/comments/: approved roster, comment-sample acceptance, sharing-copy validation, and redaction receipts.metadata/videos/: per-video metadata, frozen order, collection contract, technical audit, source-page hashes, and the local package manifest.
Research scope
- Shopping-video comments in
data/comments/and their source videos use the approved January–July 2026 Beijing-time window. - Mall reviews were collected on 2026-09-19 without a date filter, in default H5 order. Original primary-review timestamps range from 2021-11-09 to 2026-09-19. These are not the January–July video-comment sample and must not automatically be treated as pre-August evidence.
- Formal videos use the approved August 2026 Beijing-time window.
- This archive contains collection and technical validation only. It does not claim completed semantic video analysis, transcription, or attribute-performance coding. Existing demand-side prototype outputs are documented separately in the companion GitHub repository; this data release does not rerun them.
- Sales, views, GMV, returns, and provider estimates were not used to train, select, or tune the CAVFA measurement model. Any retained provider fields are evidence-only fields.
Mall reviews and product performance
Use this separate corpus to develop attribute-level consumer-experience measures: positive/negative experiences, sentiment, uncertainty, disagreement, and appended-review changes. Product-review source is verified; purchase transactions, stable order/SKU IDs, objective performance truth, attribute labels, and video-claim alignment have not been independently validated. Preserve the original review records and account for default ranking, versions, time, and missing text before analysis.
Products 14 (wide-angle lens) and 18 (rice cooker) have 810 and 713 nonempty primary texts; 190 and 287 primary records are empty. Including append-only text yields 818 and 726 records with any text. Empty text is not neutral or positive evidence. The other 28 products each have 1,000 nonempty primary texts.
Legacy manifest fields comment_files and comment_rows refer only to shopping-video comments. mall_review_* counts describe the separate product-review corpus. No existing video-comment or media file was replaced.
Privacy and use
The comment files are de-identified sharing copies. Direct account/contact fields, credentials, cookies, tokens, browser state, and unrelated identifiers are excluded. De-identification is not a guarantee of anonymity: raw platform media may contain speech, faces, names, or other content visible in the source material. Users must independently verify rights, platform terms, and applicable privacy obligations before reuse.
The seven videos that exceed GitHub's ordinary file limit are included here together with the other 443 videos. Verify all files against the public manifest before analysis.
- Downloads last month
- 237