DescribeEvents/amazon_review_events

Amazon Product Review Events

Curated per-user product review sequences from Amazon Reviews 2023 (McAuley Lab), designed for temporal point process (TPP) and marked temporal point process (MTPP) modeling. Each sequence captures a single user's chronologically-ordered reviews within one product category.

Sequences are aggressively filtered to prevent pattern exploitation — uninformative "other" events are removed, and sequences with dominant-type or consecutive-repeat shortcuts are dropped to ensure models must perform genuine reasoning.

Schema

Each record is a dictionary with 10 fields:

Field	Type	Description
seq_idx	int	Sequence index
seq_len	int	Number of events in the sequence
type_category	str	Parent product category (e.g., "Electronics")
span_weeks	float	Total time span of the sequence (weeks)
description	str	Category, user activity window, and review period
metadata	str (JSON)	user_id, parent_category, num_reviews, num_sub_categories, date_range_start, date_range_end, span_weeks, time_unit
time_since_start	list[float]	Time since the first event (in weeks)
time_since_last_event	list[float]	Time since the previous event (in weeks)
type_event	list[str]	Product sub-category slug (see below)
type_text	list[str]	Natural language description with first sub-category name, product title, star rating, and review title. HTML tags, entities, embedded media tags, and URLs are stripped.

Event Types (40 sub-categories)

Event types are the 2nd-level product sub-categories, normalized to lowercase slugs. Rare sub-categories are mapped to "other" during initial curation, and all "other" events are then stripped from the final sequences. Products without a proper category hierarchy (e.g., generic "All Electronics") are excluded.

Home & Kitchen (159 sequences, 10 types): kitchen_dining (2,131), home_d_cor_products (1,888), bedding (1,520), furniture (1,018), storage_organization (837), bath (814), heating_cooling_air_quality (272), wall_art (254), seasonal_d_cor (228), event_party_supplies (170)

Beauty & Personal Care (57 sequences, 10 types): hair_care (696), skin_care (693), makeup (642), tools_accessories (483), foot_hand_nail_care (437), personal_care (163), shave_hair_removal (127), fragrance (108), salon_spa (80), men_s_grooming (3)

Electronics (10 sequences, 10 types): computers_accessories (119), television_video (98), home_audio (84), camera_photo (71), portable_audio_video (58), accessories_supplies (39), security_surveillance (38), headphones_earbuds_accessories (37), car_vehicle_electronics (27), gps_finders_accessories (5)

Books (3 sequences, 10 types): literature_fiction (40), history (25), biographies_memoirs (20), children_s_books (17), arts_photography (17), teen_young_adult (15), mystery_thriller_suspense (11), christian_books_bibles (7), politics_social_sciences (5), crafts_hobbies_home (2)

Curation Filters

Sequences are selected to represent moderately prolific reviewers with diverse, non-trivial sub-category distributions:

Filter	Value	Description
min-events	50	Min reviews per user sequence
max-events	80	Max reviews per user sequence
min-types	3	At least 3 distinct sub-category types
min-span	3 months	Exclude sequences spanning less than 3 months
max-span	60 months	Exclude sequences spanning 60+ months
max-sub-categories	10	Top 10 sub-categories kept per parent category; rest mapped to "other"
min-subcat-count	50	Sub-categories with fewer than 50 products mapped to "other"
Verified purchase only	—	Only verified purchase reviews are included
Exclude catch-all categories	—	Products without proper sub-category hierarchy (e.g., "All Electronics") are excluded
Remove "other" events	—	All "other" events stripped; time_since_last_event recomputed
max-dominant-ratio	0.30	Drop sequences where any single type exceeds 30% of events
max-repeat-ratio	0.35	Drop sequences where consecutive same-type rate exceeds 35%

Event Text

Each event's type_text is a natural language sentence containing the first sub-category, product title, star rating, and review title:

The user reviewed "Instant Pot Duo 7-in-1 Electric Pressure Cooker" under Kitchen & Dining, rating it 5 out of 5 stars. Their review says: "Best kitchen purchase ever".

The model must learn to predict the product sub-category from product titles, sub-category context, rating patterns, and review titles.

Example

{
  "seq_idx": 0,
  "seq_len": 61,
  "type_category": "Electronics",
  "span_weeks": 269.34,
  "description": "Amazon Electronics review timeline for a user spanning May 2015 to Jul 2020. This sequence tracks the user's product review activity within the Electronics category on Amazon.",
  "metadata": "{\"user_id\": \"AF7S6K2L...\", \"parent_category\": \"Electronics\", \"num_reviews\": 76, \"num_sub_categories\": 10, ...}",
  "time_since_start": [0.0, 0.0016, 0.0024, ...],
  "time_since_last_event": [0.0, 0.0016, 0.0008, ...],
  "type_event": ["camera_photo", "computers_accessories", "accessories_supplies", ...],
  "type_text": ["The user reviewed \"GE 26571 Line Cord with Coupler (25 Feet, White)\" under Accessories & Supplies, rating it 3 out of 5 stars. Their review says: \"Like quality\".", ...]
}

Intended Use

• Training and evaluating temporal point process models
• Studying consumer purchasing patterns across product sub-categories
• Benchmarking next-event prediction and event forecasting
• Modeling marked temporal point processes with rich text marks

Citation

If you use this dataset, please cite:

@dataset{amazon_review_events_2025,
  title={Amazon Product Review Events},
  author={XiaoBB},
  year={2025},
  url={https://huggingface.co/datasets/XiaoBB/amazon_review_events},
  note={Curated from Amazon Reviews 2023 (McAuley Lab)}
}