Hugging Face's logo

Datasets:
t-tech
/
T-ECD

Languages:
Russian
English
Size:
100B<n<1T
Tags:
recsys
e-commerce
retrieval
dataset
ranking
cross-domain
License:
Dataset card Files
xet
Community
kliakhnovich commited on
Commit
9c08a47
·
verified ·
1 Parent(s): 3b944d4

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -8
README.md CHANGED
@@ -19,7 +19,7 @@ pretty_name: T-ECD
19
 
20
  # T-ECD: T-Tech E-commerce Cross-Domain Dataset
21
 
22
- ![Habr_cover_cat@1x](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/WTyxU8ugrf-XzHBjpoGq-.jpeg)
23
 
24
  ⭐️ **T-ECD** is a large-scale synthetic cross-domain dataset for recommender systems research, created by T-Bank's RecSys R&D team.
25
  It captures real-world e-commerce interaction patterns across multiple domains while ensuring complete anonymity through synthetic generation.
@@ -45,8 +45,11 @@ Cross-domain consistency is achieved by aligning identifiers across all domains:
45
 
46
  This alignment allows researchers to seamlessly link interactions from different services, enabling studies in transfer learning, cross-domain personalization, and multi-task modeling.
47
 
48
- 📊 *[Graph 1: Distribution of interactions per user (heavy tail)]*
49
- 📊 *[Graph 2: Overlap of users across domains]*
 
 
 
50
 
51
  ---
52
 
@@ -95,8 +98,10 @@ t-ecd/
95
  - category information (if available)
96
  - pretrained embedding (if available)
97
 
98
- 📊 *[Graph 3: Distribution of event types per domain]*
99
- 📊 *[Graph 4: Distribution of subdomains (e.g., recommendations vs catalog)]*
 
 
100
 
101
  - **User catalog (`users.pq`)**: Contains anonymized user attributes such as region and socio-demographic cluster.
102
 
@@ -113,8 +118,6 @@ t-ecd/
113
  - **Reviews (`reviews/{day}.pq`)**:
114
  Provide explicit ratings per brand.
115
  Raw text reviews are not included; instead, we release pretrained text embeddings to preserve privacy while enabling multimodal research.
116
-
117
-
118
  ---
119
 
120
  ### 🛠️ Data Collection
@@ -126,7 +129,8 @@ T-ECD was generated through a multi-step process:
126
 
127
  This process ensures that the dataset is privacy-preserving while remaining representative of industrial recommender system data.
128
 
129
- 📊 *[Graph 5: Temporal coverage and dataset scale]*
 
130
 
131
  ---
132
 
 
19
 
20
  # T-ECD: T-Tech E-commerce Cross-Domain Dataset
21
 
22
+ ![image (2)](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/t8w2QTQbnH0DhEIbJYCnT.png)
23
 
24
  ⭐️ **T-ECD** is a large-scale synthetic cross-domain dataset for recommender systems research, created by T-Bank's RecSys R&D team.
25
  It captures real-world e-commerce interaction patterns across multiple domains while ensuring complete anonymity through synthetic generation.
 
45
 
46
  This alignment allows researchers to seamlessly link interactions from different services, enabling studies in transfer learning, cross-domain personalization, and multi-task modeling.
47
 
48
+
49
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/s8a8iC4RmUjsD_hzOVPvD.png)
50
+
51
+
52
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/QG0DavvcvccN1GcN_gRL6.png)
53
 
54
  ---
55
 
 
98
  - category information (if available)
99
  - pretrained embedding (if available)
100
 
101
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/Q7aeb_I-Yf-rcqyPDTOLa.png)
102
+
103
+ ![image](https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/W1248ufT1-x3GUzoziuRM.png)
104
+
105
 
106
  - **User catalog (`users.pq`)**: Contains anonymized user attributes such as region and socio-demographic cluster.
107
 
 
118
  - **Reviews (`reviews/{day}.pq`)**:
119
  Provide explicit ratings per brand.
120
  Raw text reviews are not included; instead, we release pretrained text embeddings to preserve privacy while enabling multimodal research.
 
 
121
  ---
122
 
123
  ### 🛠️ Data Collection
 
129
 
130
  This process ensures that the dataset is privacy-preserving while remaining representative of industrial recommender system data.
131
 
132
+ <img src="https://cdn-uploads.huggingface.co/production/uploads/645d4947f5760d1530d55023/c2Clc9bNxL9i7jgGBfBq2.png" style="max-height: 500px; width: auto;" alt="Temporal distribution of events over domains">
133
+ *Temporal distribution of events over domains*
134
 
135
  ---
136