WZDavid commited on
Commit
92e3750
·
verified ·
1 Parent(s): 25a40cf

Expose sampled negative item metadata and expand dataset documentation

Browse files
README.md CHANGED
@@ -22,6 +22,10 @@ configs:
22
  data_files:
23
  - split: test
24
  path: tables/products/*.parquet
 
 
 
 
25
  - config_name: users
26
  data_files:
27
  - split: test
@@ -40,66 +44,109 @@ configs:
40
  path: tables/product_images/*.parquet
41
  ---
42
 
43
- # MemoryVCD — Full edition
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
 
45
- **323 users · 12 domains · 535,698 interaction records · 276,418 history products**
 
 
 
 
46
 
47
- **Full edition:** includes 243,394 verified cached product images embedded in Parquet, with native image previews. Full means all matched valid images available in the local cache at release time; it does **not** mean 100% image coverage. No missing images were synthesized or fetched to fill gaps.
48
 
49
- Release: **2026-11-12**. The [URL edition](https://huggingface.co/datasets/WZDavid/MemoryVCD) and [Full edition](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full) share identical histories, product tables, evaluation windows, and candidates. Full additionally contains the `product_images` subset. Use file hashes in `manifest.json` to verify alignment.
 
 
 
 
 
 
 
 
 
 
 
50
 
51
- ## Browse the data
52
 
53
- The default **interactions** subset shows one interaction per row: domain, user, product, rating, UTC date, review title/body, and image URL. It is the complete corpus, not a preview sample. Use the subset selector or the links below.
54
 
55
- | Subset | Rows | Purpose |
56
  |---|---:|---|
57
- | [interactions](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/interactions/test) | 535,698 | Read, search and filter individual reviews |
58
- | [products](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/products/test) | 481,333 | Product metadata for histories and saved evaluation candidates |
59
- | [users](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/users/test) | 323 | User activity counts and original ordering |
60
- | [evaluation_windows](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/evaluation_windows/test) | 15,504 | Time-axis and domain-axis test windows and memory boundaries |
61
- | [evaluation_candidates](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/evaluation_candidates/test) | 19,751 | Saved ranking/choice candidates, with source provenance |
62
- | [product_images](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/product_images/test) | 243,394 | Actual cached images with product title and identifiers |
 
63
 
64
- The `test` split label denotes an **evaluation corpus**, not that every interaction is a held-out test target. Memory and test targets are selected **within each user's history** using `evaluation_windows`. There is no random train/validation/test split.
65
 
66
- ## Load
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
67
 
68
  ```python
69
  from datasets import load_dataset
70
 
71
  repo = "WZDavid/MemoryVCD-Full"
72
  reviews = load_dataset(repo, "interactions", split="test")
73
- products = load_dataset(repo, "products", split="test")
74
- windows = load_dataset(repo, "evaluation_windows", split="test")
75
- print(reviews[0])
76
- ```
77
 
78
- ```python
79
- # Stream one cached product image without downloading all image shards.
80
- images = load_dataset(repo, "product_images", split="test", streaming=True)
81
- example = next(iter(images))
82
- display(example["image"])
83
  ```
84
 
85
- ### Original user histories
86
-
87
- `users/users_12domains.jsonl.gz` preserves the original nested histories, one user per line (323 lines). The `users` preview subset is a compact index, not the nested histories themselves. The archive is provided separately to keep the browser readable.
88
 
89
  ```python
90
- import gzip, json
91
- from huggingface_hub import hf_hub_download
92
-
93
- path = hf_hub_download(repo, "users/users_12domains.jsonl.gz", repo_type="dataset")
94
- with gzip.open(path, "rt") as f:
95
- users = [json.loads(line) for line in f if line.strip()]
96
  ```
97
 
98
- `tools/release_dataset_helpers.py` provides `load_users`, `split_window` and `restore_ranking_input`. These reconstruct data inputs without invoking any models. This is a **data release**, not a bundle of model outputs, API credentials, or experiment caches.
99
-
100
- ## Corpus distribution
101
 
102
- The first four domains are evaluation targets; the remaining eight provide additional cross-domain memory. The primary four contain 314,116 interactions; all twelve contain 535,698. Users are shared across domains, so domain user counts must not be summed.
103
 
104
  | Domain | Users | Interactions | History products |
105
  |---|---:|---:|---:|
@@ -115,14 +162,11 @@ The first four domains are evaluation targets; the remaining eight provide addit
115
  | Movies_and_TV | 278 | 9,395 | 8,433 |
116
  | Office_Products | 323 | 28,325 | 14,032 |
117
  | Sports_and_Outdoors | 323 | 26,847 | 15,194 |
 
118
 
119
- ## Image and metadata coverage
120
-
121
- All **481,333 required product/domain records** have metadata. This includes the random ranking candidates and all saved time-cutoff hard candidates, beyond the 276,418 history products.
122
-
123
- There are **456,761 product image URLs** and **243,394 valid cached product images** (50.57% of required product/domain rows). Among history products, **83,565/276,418** have a cached image (30.23%). URL availability is not a live HTTP success guarantee.
124
 
125
- | Domain | Required products | Image URLs | Cached images |
126
  |---|---:|---:|---:|
127
  | Beauty_and_Personal_Care | 82,891 | 82,890 | 49,225 |
128
  | Books | 100,625 | 78,730 | 49,539 |
@@ -136,38 +180,57 @@ There are **456,761 product image URLs** and **243,394 valid cached product imag
136
  | Movies_and_TV | 8,433 | 5,840 | 244 |
137
  | Office_Products | 14,032 | 13,977 | 3,936 |
138
  | Sports_and_Outdoors | 15,194 | 15,194 | 5,485 |
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
 
140
- `products.cached_image_available` describes availability in the companion Full release, including when read from the URL edition. `image_shard` is relative to the Full repository. Images are embedded as bytes in that shard; the image field's `path` is only a filename label, not an additional loose file to download. Join using `(domain, parent_asin)`. Missing-cache entries remain in the product and interaction tables; only valid cached images enter `product_images`. `image_sha256` checks image payload integrity.
141
-
142
- Product images differ from user-uploaded **review images**. The primary-domain histories preserve their review-image fields; the eight added domains' extraction did not retain these fields. A missing field there must not be interpreted as proof that the original review had no image. `review_images_field_present` makes this distinction explicit. Review image bytes are not bundled.
143
-
144
- ## Fields and ordering
145
-
146
- - `interaction_id`: stable identifier formed from source user index, domain, and original position within that domain.
147
- - `interaction_index`: zero-based index after a stable timestamp sort within one user/domain; used by evaluation windows.
148
- - `source_user_index` and `source_interaction_index`: preserve source ordering and allow reconstruction.
149
- - `timestamp`: original Unix milliseconds; `date` is its UTC ISO representation.
150
- - `review_title` and `review_text`: complete review text. `rating` preserves the original numeric score.
151
- - `image_url`: selected MAIN product image (large, then hi_res, then thumb; fallback to another variant).
152
- - `review_images_json`: JSON representation of original review images; `null` if absent.
153
- - Product metadata mirrors the evaluator's slim representation: title limited to 200 characters and description to 400. No `price` field is promised in the new product tables. Per-domain slim JSONL files are in `metadata/`.
154
-
155
- ## Evaluation protocol and actual candidate coverage
156
-
157
- All 323 users have windows for the four primary target domains. Time cutoffs are 0, .25, .5, .75 and 1. A three-interaction window moves along the timestamp-sorted target history; preceding interactions are the initial memory. Interior centers use Python `round(c*n)`, with indices `[center-1, center+2)`.
158
-
159
- Domain windows hold the target's final three interactions fixed. D0 has no initial memory. D1–D3 add non-target primary domains in the order Personal Care, Books, Electronics, Home & Kitchen, skipping the target. D4 adds the target; D5 adds the eight additional domains; Dtx uses only the additional eight. Memory must precede the earliest test timestamp **strictly**, and source-domain segments remain in their specified order. For Image Prediction, earlier test interactions are revealed before later predictions; other tasks retain the same initial memory throughout the window.
160
-
161
- **Saved candidates have different coverage from windows.** The `data/v8e/...` source files cover the original first 100 users at each cutoff. The base `data/cross_domain.jsonl` ranking candidates and `hard_negatives_*_v3.jsonl` choice candidates are separately identified by `source_file`. Missing saved candidates are not fabricated for the other users. Filter by `task`, `source_file`, and `cutoff` before evaluation; do not mix candidate versions.
162
-
163
- The stored hard-negative records contain **3 negatives**, although a paper description says 4. This release preserves the actual files rather than silently adding a fourth candidate. Ranking pools retain the saved negatives. `evaluation/source_candidates/` also contains unchanged hard-negative JSONL files. The six tasks reuse these histories/windows; they are not six independent corpora.
164
-
165
- ## Integrity and limitations
166
-
167
- The original histories contain **10,404 excess records sharing a user/domain/product/timestamp key**. They are retained to preserve the original evaluation timelines; this is not a claim that all 535,698 rows are unique events. Stable `interaction_id` values distinguish the retained rows. Users were selected for substantial overlap/activity across the four primary domains; they are not a representative random sample of Amazon users. Ratings are strongly skewed toward 4–5 stars.
168
-
169
- See `reports/` for corpus, metadata, image and release validation results. `manifest.json` lists SHA-256 hashes and sizes for release files. Image caches were checked for valid decoding before export. No external image downloads were performed during packaging.
170
-
171
- ## Source and attribution
172
 
173
- Derived from [Amazon Reviews 2023](https://huggingface.co/datasets/McAuley-Lab/Amazon-Reviews-2023). Original reviews and product images retain their applicable source terms and third-party rights; this packaging does not grant a new license to those materials. Refer to the upstream dataset for source attribution and usage terms.
 
22
  data_files:
23
  - split: test
24
  path: tables/products/*.parquet
25
+ - config_name: negative_items
26
+ data_files:
27
+ - split: test
28
+ path: tables/negative_items/*.parquet
29
  - config_name: users
30
  data_files:
31
  - split: test
 
44
  path: tables/product_images/*.parquet
45
  ---
46
 
47
+ # 🧠 MemoryVCD · Full Edition
48
+
49
+ **323 users · 12 domains · 535,698 interactions**
50
+
51
+ [📄 Paper](https://aclanthology.org/2026.findings-emnlp.1804/) · [📥 PDF](https://aclanthology.org/2026.findings-emnlp.1804.pdf) · [💻 MemoryCD Code](https://github.com/AgentMemoryWorld/MemoryCD) · [🔗 URL Edition](https://huggingface.co/datasets/WZDavid/MemoryVCD) · [🖼️ Full Edition](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full)
52
+
53
+ Real user histories for evaluating long-term, cross-domain personalization. MemoryVCD extends the MemoryCD data release with product image references, complete metadata for the saved candidate pools, and explicit evaluation inputs. Full adds the available cached product images.
54
+
55
+ ## 📄 Paper and release relationship
56
+
57
+ **MemoryCD: Benchmarking Long-Context User Memory of LLM Agents for Lifelong Cross-Domain Personalization** — Zhang et al., *Findings of EMNLP 2026*.
58
+
59
+ The paper studies memory grounded in real Amazon review histories, with four tasks: rating prediction, item ranking, review summarization, and review generation. It evaluates single-domain and cross-domain memory. See the [paper](https://aclanthology.org/2026.findings-emnlp.1804/) for methods and results.
60
+
61
+ This repository is the **MemoryVCD extended release**, not a claim that every added visual feature or candidate file was evaluated in the MemoryCD paper. Statistics below describe the released files.
62
 
63
+ | Repository | Purpose | Product images |
64
+ |---|---|---|
65
+ | [MemoryCD](https://huggingface.co/datasets/WZDavid/MemoryCD) | Original release; unchanged | Original item metadata |
66
+ | [MemoryVCD](https://huggingface.co/datasets/WZDavid/MemoryVCD) | Extended histories, candidate metadata and evaluation inputs | URL strings |
67
+ | [MemoryVCD-Full](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full) | Same data plus existing image cache | 243,394 embedded image records |
68
 
69
+ ## 📊 At a glance
70
 
71
+ | Statistic | Count |
72
+ |---|---:|
73
+ | Users | 323 |
74
+ | Domains | 12: 4 evaluation targets + 8 memory domains |
75
+ | Interaction records | 535,698 |
76
+ | History product/domain pairs | 276,418 |
77
+ | All product/domain metadata records | 481,333 |
78
+ | Unique sampled negative product/domain pairs | 211,364 |
79
+ | Negative items outside the history product set | 204,915 |
80
+ | Missing metadata for saved negatives | **0** |
81
+ | Product image URLs | 456,761 |
82
+ | Valid cached image records (Full) | 243,394 |
83
 
84
+ Counts use `(domain, parent_asin)` as the item key. A negative for one user can be a historical item for another; history and negative counts therefore overlap.
85
 
86
+ ## 🗂️ Browse the data
87
 
88
+ | Subset | Rows | Contents |
89
  |---|---:|---|
90
+ | [interactions](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/interactions/test) | 535,698 | One review per row |
91
+ | [products](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/products/test) | 481,333 | History and candidate item metadata |
92
+ | [negative_items](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/negative_items/test) | 211,364 | Metadata for all saved sampled negatives |
93
+ | [users](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/users/test) | 323 | User index and activity counts |
94
+ | [evaluation_windows](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/evaluation_windows/test) | 15,504 | Memory boundaries and test targets |
95
+ | [evaluation_candidates](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/evaluation_candidates/test) | 19,751 | Positive and negative item IDs with source provenance |
96
+ | [product_images](https://huggingface.co/datasets/WZDavid/MemoryVCD-Full/viewer/product_images/test) | 243,394 | Cached product images with native image fields |
97
 
98
+ All subsets use `test` to label the evaluation corpus. This is **not** a random train/test split: `evaluation_windows` defines memory and test targets within each history.
99
 
100
+ ## 🎯 Negative items and sampling
101
+
102
+ `products` includes both historical items and all items required by the saved ranking/choice candidates. `negative_items` is a convenient filtered view of those same metadata records; it does not add new interactions or change the candidate pools.
103
+
104
+ | Domain | Unique negative items | Outside history | Missing metadata |
105
+ |---|---:|---:|---:|
106
+ | Beauty_and_Personal_Care | 51,966 | 49,416 | 0 |
107
+ | Books | 53,433 | 52,571 | 0 |
108
+ | Electronics | 52,641 | 51,075 | 0 |
109
+ | Home_and_Kitchen | 53,324 | 51,853 | 0 |
110
+ | **Total** | **211,364** | **204,915** | **0** |
111
+
112
+ | Candidate task | Saved rows | Negatives per row | Source |
113
+ |---|---:|---:|---|
114
+ | Item ranking | 9,876 | 19 | Saved random candidate pools |
115
+ | Item preference choice | 9,875 | 3 | Saved hard-negative pools |
116
+
117
+ **No resampling was performed.** Candidate order and IDs are preserved. The saved hard-negative files contain 3 negatives; any description expecting 4 does not match these files. Base and time-cutoff sources are distinct: filter `task`, `source_file`, and `cutoff` before evaluation. Time-axis source files cover the original first 100 users, while evaluation windows cover all 323 users. Missing candidate rows are not fabricated.
118
+
119
+ Join `evaluation_candidates.neg_asins` to `negative_items.parent_asin` **within the same domain**. `candidate_reference_count` counts occurrences across all saved candidate rows; `candidate_tasks` lists the tasks using each item. The coverage audit is in `reports/negative_metadata_coverage.json`.
120
+
121
+ ## 🚀 Load and join
122
 
123
  ```python
124
  from datasets import load_dataset
125
 
126
  repo = "WZDavid/MemoryVCD-Full"
127
  reviews = load_dataset(repo, "interactions", split="test")
128
+ candidates = load_dataset(repo, "evaluation_candidates", split="test")
129
+ negatives = load_dataset(repo, "negative_items", split="test")
 
 
130
 
131
+ meta = {(r["domain"], r["parent_asin"]): r for r in negatives}
132
+ example = candidates[0] # Select your intended task/source/cutoff for evaluation.
133
+ negative_metadata = [meta[(example["domain"], a)] for a in example["neg_asins"]]
134
+ print([(r["parent_asin"], r["title"]) for r in negative_metadata])
 
135
  ```
136
 
137
+ For positive item metadata, use the `products` subset. To read cached images:
 
 
138
 
139
  ```python
140
+ images = load_dataset("WZDavid/MemoryVCD-Full", "product_images",
141
+ split="test", streaming=True)
142
+ example_image = next(iter(images))
143
+ print(example_image["domain"], example_image["parent_asin"])
144
+ image = example_image["image"] # PIL image; cached bytes are embedded in Parquet.
 
145
  ```
146
 
147
+ ## 🌍 Domain statistics
 
 
148
 
149
+ The first four domains are evaluation targets (314,116 interactions). The remaining eight provide additional memory (221,582 interactions). Users overlap across domains.
150
 
151
  | Domain | Users | Interactions | History products |
152
  |---|---:|---:|---:|
 
162
  | Movies_and_TV | 278 | 9,395 | 8,433 |
163
  | Office_Products | 323 | 28,325 | 14,032 |
164
  | Sports_and_Outdoors | 323 | 26,847 | 15,194 |
165
+ | **Total (unique users)** | **323** | **535,698** | **276,418** |
166
 
167
+ ## 🖼️ Images and metadata
 
 
 
 
168
 
169
+ | Domain | Required metadata | Image URLs | Cached image records |
170
  |---|---:|---:|---:|
171
  | Beauty_and_Personal_Care | 82,891 | 82,890 | 49,225 |
172
  | Books | 100,625 | 78,730 | 49,539 |
 
180
  | Movies_and_TV | 8,433 | 5,840 | 244 |
181
  | Office_Products | 14,032 | 13,977 | 3,936 |
182
  | Sports_and_Outdoors | 15,194 | 15,194 | 5,485 |
183
+ | **Total** | **481,333** | **456,761** | **243,394** |
184
+
185
+ Full includes all matched valid images in the release cache, **not 100% image coverage**: 50.57% of required metadata rows and 30.23% of history products have cached images. One invalid cached image was excluded. External URLs may expire. Product photos are separate from user-uploaded review photos; review image bytes are not bundled.
186
+
187
+ ## 🧾 Key fields and files
188
+
189
+ | Table / file | Important fields or contents |
190
+ |---|---|
191
+ | `interactions` | `user_id`, `domain`, `parent_asin`, `rating`, `timestamp`, review title/text, `image_url` |
192
+ | `products` / `negative_items` | `title`, `description`, category, store, average rating, image URL; `in_user_history`, cache availability |
193
+ | `evaluation_candidates` | `task`, `source_file`, `cutoff`, `pos_asin`, `neg_asins` |
194
+ | `evaluation_windows` | Target indices, memory prefix/boundary, source domains |
195
+ | `product_images` (Full) | `(domain, parent_asin)`, embedded `image`, source URL and image hash |
196
+ | `users/users_12domains.jsonl.gz` | Original nested histories, one user per line |
197
+ | `metadata/*.jsonl.gz` | Per-domain metadata including required candidate items |
198
+ | `evaluation/source_candidates/` | Original saved hard-negative files |
199
+ | `tools/release_dataset_helpers.py` | Load nested histories and reconstruct evaluation inputs |
200
+ | `reports/`, `manifest.json` | Coverage checks, validation, file sizes and SHA-256 hashes |
201
+
202
+ Metadata uses the evaluator's slim fields: title up to 200 characters and description up to 400; price is not included. `image_shard` paths refer to the Full repository. Join images with `(domain, parent_asin)`. The `users` subset is an index; use the compressed histories for the complete nested records.
203
+
204
+ ## ⚙️ Evaluation and limitations
205
+
206
+ | Setting | Released definition |
207
+ |---|---|
208
+ | Time axis | Five cutoffs: 0, .25, .5, .75, 1; a three-interaction target window |
209
+ | Domain axis | D0–D5 and Dtx; final three target-domain interactions |
210
+ | Chronology | Memory precedes the target boundary; see helper functions for exact reconstruction |
211
+ | Ordering | `source_user_index`, `source_interaction_index` and stable interaction IDs preserve provenance |
212
+ | Repeated keys | 10,404 excess records share user/domain/item/timestamp; retained to preserve timelines |
213
+ | Selection bias | Users have substantial cross-domain activity; not a representative random sample |
214
+ | Review images | Extra-eight-domain extraction did not retain review-image fields; absence is not proof of no original image |
215
+
216
+ The same histories/windows support multiple tasks; task counts must not be summed as independent users. Exact source ordering and candidate versions matter for reproducibility.
217
+
218
+ ## 📚 Citation and source
219
+
220
+ Please cite the foundational MemoryCD paper and identify the MemoryVCD edition/revision used:
221
+
222
+ ```bibtex
223
+ @inproceedings{zhang-etal-2026-memorycd,
224
+ title = "{M}emory{CD}: Benchmarking Long-Context User Memory of {LLM} Agents for Lifelong Cross-Domain Personalization",
225
+ author = "Zhang, Weizhi and Wei, Xiaokai and Huang, Wei-Chieh and Hui, Zheng and Wang, Chen and Gong, Michelle and Yu, Philip S.",
226
+ booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2026",
227
+ year = "2026",
228
+ pages = "38554--38575",
229
+ url = "https://aclanthology.org/2026.findings-emnlp.1804/",
230
+ doi = "10.18653/2026.findings-emnlp.1804"
231
+ }
232
+ ```
233
 
234
+ Derived from [Amazon Reviews 2023](https://huggingface.co/datasets/McAuley-Lab/Amazon-Reviews-2023). Original reviews and images retain applicable source terms and third-party rights. This packaging does not grant new rights to those materials.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
235
 
236
+ Data release: **2026-11-12** · Negative-item view and documentation: **2026-11-18**. Shared tables are identical across URL and Full editions; Full additionally includes cached image bytes.
manifest.json CHANGED
@@ -4,8 +4,8 @@
4
  "files": [
5
  {
6
  "path": "README.md",
7
- "bytes": 11295,
8
- "sha256": "6601ba8e8d8cdb002303d2104f06970004fa42abd0a2bd743b6b096de80cf12b"
9
  },
10
  {
11
  "path": "evaluation/source_candidates/hard_negatives_Beauty_and_Personal_Care_v3.jsonl",
@@ -207,6 +207,11 @@
207
  "bytes": 1743,
208
  "sha256": "b33d349b41046d6e9e508733b3614b3ceb681d7693bdc07ffb53963cfee5b02d"
209
  },
 
 
 
 
 
210
  {
211
  "path": "reports/release_validation.json",
212
  "bytes": 3223,
@@ -332,6 +337,11 @@
332
  "bytes": 5014826,
333
  "sha256": "623d74e2f3259a822f2acd381fcd7d005baafb47ab57325baecee7e2acc6567b"
334
  },
 
 
 
 
 
335
  {
336
  "path": "tables/product_images/test-00000.parquet",
337
  "bytes": 61728063,
 
4
  "files": [
5
  {
6
  "path": "README.md",
7
+ "bytes": 12702,
8
+ "sha256": "0fd176a1f9d0878be91758313fdba7cc4e32d7eeec6631e65bb75988a37681ce"
9
  },
10
  {
11
  "path": "evaluation/source_candidates/hard_negatives_Beauty_and_Personal_Care_v3.jsonl",
 
207
  "bytes": 1743,
208
  "sha256": "b33d349b41046d6e9e508733b3614b3ceb681d7693bdc07ffb53963cfee5b02d"
209
  },
210
+ {
211
+ "path": "reports/negative_metadata_coverage.json",
212
+ "bytes": 763,
213
+ "sha256": "27a35444a286714b2cb49ba69bf05f5bad1424b7e0fa7ba0512e22d337ea2e8a"
214
+ },
215
  {
216
  "path": "reports/release_validation.json",
217
  "bytes": 3223,
 
337
  "bytes": 5014826,
338
  "sha256": "623d74e2f3259a822f2acd381fcd7d005baafb47ab57325baecee7e2acc6567b"
339
  },
340
+ {
341
+ "path": "tables/negative_items/test.parquet",
342
+ "bytes": 36411001,
343
+ "sha256": "d706bcf9c2e1e55b963f8fb11600f21d3b3fae7d4e0713d45fc7c77ca9a14fea"
344
+ },
345
  {
346
  "path": "tables/product_images/test-00000.parquet",
347
  "bytes": 61728063,
reports/negative_metadata_coverage.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "status": "passed",
3
+ "sampling": "Existing saved candidates; no resampling",
4
+ "candidate_rows": 19751,
5
+ "unique_negative_product_domain_pairs": 211364,
6
+ "negative_items_outside_history": 204915,
7
+ "missing_metadata": [],
8
+ "domains": [
9
+ {
10
+ "domain": "Beauty_and_Personal_Care",
11
+ "negative_items": 51966,
12
+ "outside_history": 49416
13
+ },
14
+ {
15
+ "domain": "Books",
16
+ "negative_items": 53433,
17
+ "outside_history": 52571
18
+ },
19
+ {
20
+ "domain": "Electronics",
21
+ "negative_items": 52641,
22
+ "outside_history": 51075
23
+ },
24
+ {
25
+ "domain": "Home_and_Kitchen",
26
+ "negative_items": 53324,
27
+ "outside_history": 51853
28
+ }
29
+ ],
30
+ "tasks": {
31
+ "item_ranking": 9876,
32
+ "item_preference_choice": 9875
33
+ }
34
+ }
tables/negative_items/test.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d706bcf9c2e1e55b963f8fb11600f21d3b3fae7d4e0713d45fc7c77ca9a14fea
3
+ size 36411001