File size: 12,117 Bytes
1c2019e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
# Explorer_LLM_Rec_Competition

This dataset contains user historical behaviors and content metadata, constructed from real user interaction histories. It covers behavior sequences across multiple domains for a single user and supports cross-domain recommendation, semantic-ID retrieval / generation, content understanding, and related tasks.

## Files

| File | Description |
|------|-------------|
| `OneReason_UserProfile/` | Per-user multi-domain behavior records (~500k rows) |
| `OneReason_Pid2Sid/` | Mapping from content PID to three-segment semantic ID |
| `OneReason_Pid2Caption/` | Mapping from content PID to text caption |
| `OneReason_Pid2Tag/` | Mapping from content PID to level-3 category tag |
| `OneReason_General/` | General knowledge data |

---

## Anonymization & Join Rules

- All `pid` / item id / author id / live id are hashed `int64` values.
- The user table contains no `uid`; each row is an anonymous user sample.
- Hashed item ids join the three item-metadata tables via `(domain, pid)`.
- Regular videos and ads both come from the video system, distinguished by `domain`:
  - Regular video: `video/video`
  - Ad: `video/ad`
  - Some users' video-watch sequences also contain ads.

## Domain Values

| Domain | Meaning | Associated user fields | Token prefix |
|--------|---------|------------------------|--------------|
| `video/video` | Regular short videos | `video_*`, `video_history_*` | `<\|video_begin\|>` |
| `video/ad` | Ads / outer-loop content | `outer_loop_*` | `<\|ad_begin\|>` |
| `goods` | E-commerce products | `ec_*` | `<\|prod_begin\|>` |
| `live` | Live-streaming hosts | `live_hist_author_id_list` | `<\|living_begin\|>` |

---

## OneReason_Pid2Sid

Semantic-ID mapping table. Converts a content ID into a three-segment semantic ID.

| Field | Type | Description |
|-------|------|-------------|
| `pid` | int64 | Hashed content / product / host ID |
| `domain` | string | Domain of the content (see Domain Values) |
| `sid_three` | list[float] | Three-segment semantic ID `[s_a, s_b, s_c]` |

Token concatenation example:

```text
video/video: <|video_begin|><s_a_123><s_b_456><s_c_789>
video/ad:    <|ad_begin|><s_a_123><s_b_456><s_c_789>
goods:       <|prod_begin|><s_a_123><s_b_456><s_c_789>
live:        <|living_begin|><s_a_123><s_b_456><s_c_789>
```

## OneReason_Pid2Caption

Lookup of a content's text description by hashed PID.

| Field | Type | Description |
|-------|------|-------------|
| `pid` | int64 | Hashed content / product / host ID |
| `domain` | string | Domain of the content (see Domain Values) |
| `caption` | string | Content description, or stringified list of description tags |

## OneReason_Pid2Tag

Level-3 category tag for video, ad, and live content.

| Field | Type | Description |
|-------|------|-------------|
| `pid` | int64 | Hashed content / host ID |
| `domain` | string | Domain; currently `video/video`, `video/ad`, `live` |
| `tag_lv3` | string | Level-3 category tag |

---

## OneReason_UserProfile

Each row is an anonymous user-profile sample containing behavior sequences from e-commerce, short-video, live-streaming, and ad domains.

### Sequence Alignment Rules

`user_history` fields are grouped into 4 domains. Each domain has one or more **primary sequences (pid / ID sequences)**, and every primary sequence is paired with a set of **aligned feature sequences** of identical length. If both a primary sequence and its aligned feature sequences are `null`, the user has no behavior in that domain.

#### `video/video` (short video)

| Primary sequence | Aligned feature sequences |
|------------------|---------------------------|
| `video_sampled_pid_list` | `video_neg_feedback_list`, `video_like_list`, `video_comment_list`, `video_forward_list`, `video_collect_list`, `video_watch_time_list`, `video_play_done_list`, `video_duration_list`, `video_ts_list` |
| `video_history_sampled_pid_list` | `video_history_neg_feedback_list`, `video_history_like_list`, `video_history_comment_list`, `video_history_forward_list`, `video_history_collect_list`, `video_history_watch_time_list`, `video_history_play_done_list`, `video_history_duration_list`, `video_history_ts_list` |

#### `video/ad` (ads / outer-loop)

| Primary sequence | Aligned feature sequences |
|------------------|---------------------------|
| `outer_loop_history_action_pid_list_pos` | `outer_loop_history_action_pid_list_pos_ts` |
| `outer_loop_history_action_pid_list_click` | `outer_loop_history_action_pid_list_click_ts`, `outer_loop_history_action_pid_list_click_type`, `outer_loop_history_action_pid_list_click_industry` |
| `outer_loop_deep_target_pid` | `outer_loop_deep_target_pid_ts` |

#### `goods` (e-commerce)

| Primary sequence | Aligned feature sequences |
|------------------|---------------------------|
| `ec_item_id_list` | `ec_cvr_label_list`, `ec_time_ms_list` |
| `ec_good_click_item_id_list_extend` | `ec_trunc_clk_lag` |
| `ec_good_order_item_id_list_extend` | `ec_trunc_buy_lag` |
| `ec_colossus_rs_item_id_list` | `ec_colossus_rs_lagv1_list`, `ec_colossus_rs_lagv2_list`, `ec_colossus_rs_is_click_list`, `ec_colossus_rs_is_cart_list`, `ec_colossus_rs_is_buy_list` |

#### `live` (live-streaming)

| Primary sequence | Aligned feature sequences |
|------------------|---------------------------|
| `live_hist_author_id_list` | `live_hist_timestamp_list`, `live_hist_live_id_list`, `live_hist_show_cnt_list`, `live_hist_play_cnt_list`, `live_hist_valid_play_cnt_list`, `live_hist_play_duration_list`, `live_hist_valid_play_duration_list`, `live_hist_like_cnt_list`, `live_hist_comment_cnt_list`, `live_hist_reduce_similar_cnt_list`, `live_hist_report_live_cnt_list`, `live_hist_author_category_type_list`, `live_hist_author_type_list`, `live_hist_is_interactive_mp_live_list`, `live_hist_is_building_live_list`, `live_hist_is_local_life_live_list`, `live_hist_is_detect_game_live_list`, `live_hist_is_recruit_live_list`, `live_hist_follow_author_cnt_list` |

---

### E-commerce Fields

| Field | Type | Description |
|-------|------|-------------|
| `ec_item_id_list` | list[int64] | E-commerce item ID sequence |
| `ec_cvr_label_list` | list[int64] | Conversion label aligned with `ec_item_id_list` |
| `ec_time_ms_list` | list[int64] | Absolute timestamp (ms) aligned with `ec_item_id_list` |
| `ec_colossus_rs_item_id_list` | list[int64] | Recently shown e-commerce item ID sequence |
| `ec_colossus_rs_lagv1_list` | list[int64] | Hour-level lag of `ec_colossus_rs_item_id_list` relative to `ec_time_ms` |
| `ec_colossus_rs_lagv2_list` | list[int64] | Supplementary hour-level lag of `ec_colossus_rs_item_id_list` relative to `ec_time_ms` |
| `ec_colossus_rs_is_click_list` | list[int64] | Click flag for each recently shown item (0/1) |
| `ec_colossus_rs_is_cart_list` | list[int64] | Add-to-cart flag for each recently shown item (0/1) |
| `ec_colossus_rs_is_buy_list` | list[int64] | Purchase flag for each recently shown item (0/1) |
| `ec_good_click_item_id_list_extend` | list[int64] | Recently clicked e-commerce item ID sequence |
| `ec_trunc_clk_lag` | list[int64] | Hour-level lag of `ec_good_click_item_id_list_extend` relative to `ec_time_ms` |
| `ec_good_order_item_id_list_extend` | list[int64] | Recently purchased e-commerce item ID sequence |
| `ec_trunc_buy_lag` | list[int64] | Hour-level lag of `ec_good_order_item_id_list_extend` relative to `ec_time_ms` |
| `ec_time_ms` | int64 | Snapshot time when e-commerce features were captured |

### Short Video Fields

| Field | Type | Description |
|-------|------|-------------|
| `video_sampled_pid_list` | list[int64] | Currently clicked short-video ID sequence |
| `video_neg_feedback_list` | list[float] | Negative-feedback flag for currently clicked videos (0/1) |
| `video_like_list` | list[float] | Like flag for currently clicked videos (0/1) |
| `video_comment_list` | list[float] | Comment flag for currently clicked videos (0/1) |
| `video_forward_list` | list[float] | Forward / share flag for currently clicked videos (0/1) |
| `video_collect_list` | list[float] | Collect flag for currently clicked videos (0/1) |
| `video_watch_time_list` | list[float] | Watch time for currently clicked videos |
| `video_play_done_list` | list[float] | Play-done (full-watch) flag for currently clicked videos (0/1) |
| `video_duration_list` | list[float] | Duration of currently clicked videos |
| `video_ts_list` | list[int64] | Absolute timestamp of currently clicked videos |
| `video_history_sampled_pid_list` | list[int64] | Historical shown short-video ID sequence |
| `video_history_neg_feedback_list` | list[float] | Negative-feedback flag for historical videos (0/1) |
| `video_history_like_list` | list[float] | Like flag for historical videos (0/1) |
| `video_history_comment_list` | list[float] | Comment flag for historical videos (0/1) |
| `video_history_forward_list` | list[float] | Forward / share flag for historical videos (0/1) |
| `video_history_collect_list` | list[float] | Collect flag for historical videos (0/1) |
| `video_history_watch_time_list` | list[float] | Watch time for historical videos |
| `video_history_play_done_list` | list[float] | Play-done flag for historical videos (0/1) |
| `video_history_duration_list` | list[float] | Duration of historical videos |
| `video_history_ts_list` | list[int64] | Absolute timestamp of historical videos |
| `video_history_seq_len` | int64 | Length of the historical-video sequence |

### Live-streaming Fields

| Field | Type | Description |
|-------|------|-------------|
| `live_hist_timestamp_list` | list[string] | Date of watched live streams|
| `live_hist_author_id_list` | list[int64] | Host ID of watched live streams |
| `live_hist_live_id_list` | list[int64] | Live-room ID of watched live streams |
| `live_hist_show_cnt_list` | list[int64] | Impression count |
| `live_hist_play_cnt_list` | list[int64] | Play count |
| `live_hist_valid_play_cnt_list` | list[int64] | Valid-play count |
| `live_hist_play_duration_list` | list[int64] | Total play duration |
| `live_hist_valid_play_duration_list` | list[int64] | Valid-play duration |
| `live_hist_like_cnt_list` | list[int64] | Like count |
| `live_hist_comment_cnt_list` | list[int64] | Comment count |
| `live_hist_reduce_similar_cnt_list` | list[int64] | "Reduce similar" / "not interested" count |
| `live_hist_report_live_cnt_list` | list[int64] | Report count |
| `live_hist_author_category_type_list` | list[string] | Host category type |
| `live_hist_author_type_list` | list[string] | Host type |
| `live_hist_is_interactive_mp_live_list` | list[int64] | Interactive mini-program live flag (0/1) |
| `live_hist_is_building_live_list` | list[int64] | Real-estate live flag (0/1) |
| `live_hist_is_local_life_live_list` | list[int64] | Local-life live flag (0/1) |
| `live_hist_is_detect_game_live_list` | list[int64] | Gaming live flag (0/1) |
| `live_hist_is_recruit_live_list` | list[int64] | Recruitment live flag (0/1) |
| `live_hist_follow_author_cnt_list` | list[int64] | List of follow-author action counts |

### Ad / Outer-loop Fields

| Field | Type | Description |
|-------|------|-------------|
| `outer_loop_history_action_pid_list_pos` | list[int64] | Recently deep-converted outer-loop ad ID sequence |
| `outer_loop_history_action_pid_list_pos_ts` | list[int64] | Absolute timestamp aligned with `outer_loop_history_action_pid_list_pos` |
| `outer_loop_history_action_pid_list_click` | list[int64] | Recently clicked outer-loop ad ID sequence |
| `outer_loop_history_action_pid_list_click_ts` | list[int64] | Absolute timestamp aligned with `outer_loop_history_action_pid_list_click` |
| `outer_loop_history_action_pid_list_click_type` | list[string] | Click type aligned with `outer_loop_history_action_pid_list_click` |
| `outer_loop_history_action_pid_list_click_industry` | list[string] | Industry aligned with `outer_loop_history_action_pid_list_click` |
| `outer_loop_deep_target_pid` | list[int64] | Deep-conversion target ID sequence of recent outer-loop ads |
| `outer_loop_deep_target_pid_ts` | list[int64] | Absolute timestamp aligned with `outer_loop_deep_target_pid` |