Datasets:
ArXiv:
License:
Update README.md
Browse files
README.md
CHANGED
@@ -19,7 +19,7 @@ The model is available under `model/`.
|
|
19 |
from datasets import load_dataset
|
20 |
|
21 |
dataset = load_dataset(
|
22 |
-
"philipphager/baidu-
|
23 |
name="clicks",
|
24 |
split="train", # ["train", "test"]
|
25 |
cache_dir="~/.cache/huggingface",
|
@@ -33,7 +33,7 @@ dataset.set_format("torch") # [None, "numpy", "torch", "tensorflow", "pandas",
|
|
33 |
from datasets import load_dataset
|
34 |
|
35 |
dataset = load_dataset(
|
36 |
-
"philipphager/baidu-
|
37 |
name="annotations",
|
38 |
split="test",
|
39 |
cache_dir="~/.cache/huggingface",
|
@@ -50,16 +50,32 @@ Each row of the click / annotation dataset contains the following attributes. Us
|
|
50 |
|------------------------------|----------------|-------------|
|
51 |
| query_id | string | Baidu query_id |
|
52 |
| query_md5 | string | MD5 hash of query text |
|
53 |
-
|
|
|
|
|
|
|
|
54 |
| text_md5 | List[string] | MD5 hash of document title and abstract |
|
55 |
-
|
|
|
|
|
|
56 |
| click | Tensor[int32] | Click / no click on a document |
|
57 |
-
| n | int32 | Number of documents for current query, useful for padding |
|
58 |
| position | Tensor[int32] | Position in ranking (does not always match original item position) |
|
59 |
-
| media_type | Tensor[int32] | Document type (label encoding recommended as
|
60 |
-
| displayed_time | Tensor[float32]| Seconds a document was displayed on screen |
|
61 |
-
| serp_height | Tensor[int32] | Pixel height of a document on screen |
|
62 |
-
| slipoff_count_after_click | Tensor[int32] | Number of times a document was scrolled off screen after previously clicking on it |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
|
64 |
|
65 |
### Expert annotation dataset
|
@@ -67,11 +83,28 @@ Each row of the click / annotation dataset contains the following attributes. Us
|
|
67 |
|------------------------------|----------------|-------------|
|
68 |
| query_id | string | Baidu query_id |
|
69 |
| query_md5 | string | MD5 hash of query text |
|
70 |
-
|
|
71 |
-
|
|
72 |
-
| label | Tensor[int32] | Relevance judgment on a scale from 0 (bad) to 4 (excellent) |
|
73 |
-
| n | int32 | Number of documents for current query, useful for padding |
|
74 |
| frequency_bucket | int32 | Monthly frequency of query (bucket) from 0 (high frequency) to 9 (low frequency) |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
76 |
## Example PyTorch collate function
|
77 |
Each sample in the dataset is a single query with multiple documents.
|
@@ -95,7 +128,9 @@ def collate_clicks(samples: List):
|
|
95 |
batch["n"].append(sample["n"])
|
96 |
|
97 |
return {
|
98 |
-
"query_document_embedding": pad_sequence(
|
|
|
|
|
99 |
"position": pad_sequence(batch["position"], batch_first=True),
|
100 |
"click": pad_sequence(batch["click"], batch_first=True),
|
101 |
"n": torch.tensor(batch["n"]),
|
@@ -103,3 +138,4 @@ def collate_clicks(samples: List):
|
|
103 |
|
104 |
loader = DataLoader(dataset, collate_fn=collate_clicks, batch_size=16)
|
105 |
```
|
|
|
|
19 |
from datasets import load_dataset
|
20 |
|
21 |
dataset = load_dataset(
|
22 |
+
"philipphager/baidu-ultr_baidu-mlm-ctr",
|
23 |
name="clicks",
|
24 |
split="train", # ["train", "test"]
|
25 |
cache_dir="~/.cache/huggingface",
|
|
|
33 |
from datasets import load_dataset
|
34 |
|
35 |
dataset = load_dataset(
|
36 |
+
"philipphager/baidu-ultr_baidu-mlm-ctr",
|
37 |
name="annotations",
|
38 |
split="test",
|
39 |
cache_dir="~/.cache/huggingface",
|
|
|
50 |
|------------------------------|----------------|-------------|
|
51 |
| query_id | string | Baidu query_id |
|
52 |
| query_md5 | string | MD5 hash of query text |
|
53 |
+
| query | List[int32] | List of query tokens |
|
54 |
+
| query_length | int32 | Number of query tokens |
|
55 |
+
| n | int32 | Number of documents for current query, useful for padding |
|
56 |
+
| url_md5 | List[string] | MD5 hash of document URL, most reliable document identifier |
|
57 |
| text_md5 | List[string] | MD5 hash of document title and abstract |
|
58 |
+
| title | List[List[int32]] | List of tokens for document titles |
|
59 |
+
| abstract | List[List[int32]] | List of tokens for document abstracts |
|
60 |
+
| query_document_embedding | Tensor[Tensor[float16]]| BERT CLS token |
|
61 |
| click | Tensor[int32] | Click / no click on a document |
|
|
|
62 |
| position | Tensor[int32] | Position in ranking (does not always match original item position) |
|
63 |
+
| media_type | Tensor[int32] | Document type (label encoding recommended as IDs do not occupy a continuous integer range) |
|
64 |
+
| displayed_time | Tensor[float32]| Seconds a document was displayed on the screen |
|
65 |
+
| serp_height | Tensor[int32] | Pixel height of a document on the screen |
|
66 |
+
| slipoff_count_after_click | Tensor[int32] | Number of times a document was scrolled off the screen after previously clicking on it |
|
67 |
+
| bm25 | Tensor[float32] | BM25 score for documents |
|
68 |
+
| bm25_title | Tensor[float32] | BM25 score for document titles |
|
69 |
+
| bm25_abstract | Tensor[float32] | BM25 score for document abstracts |
|
70 |
+
| tf_idf | Tensor[float32] | TF-IDF score for documents |
|
71 |
+
| tf | Tensor[float32] | Term frequency for documents |
|
72 |
+
| idf | Tensor[float32] | Inverse document frequency for documents |
|
73 |
+
| ql_jelinek_mercer_short | Tensor[float32] | Query likelihood score for documents using Jelinek-Mercer smoothing (alpha = 0.1) |
|
74 |
+
| ql_jelinek_mercer_long | Tensor[float32] | Query likelihood score for documents using Jelinek-Mercer smoothing (alpha = 0.7) |
|
75 |
+
| ql_dirichlet | Tensor[float32] | Query likelihood score for documents using Dirichlet smoothing (lambda = 128) |
|
76 |
+
| document_length | Tensor[int32] | Length of documents |
|
77 |
+
| title_length | Tensor[int32] | Length of document titles |
|
78 |
+
| abstract_length | Tensor[int32] | Length of document abstracts |
|
79 |
|
80 |
|
81 |
### Expert annotation dataset
|
|
|
83 |
|------------------------------|----------------|-------------|
|
84 |
| query_id | string | Baidu query_id |
|
85 |
| query_md5 | string | MD5 hash of query text |
|
86 |
+
| query | List[int32] | List of query tokens |
|
87 |
+
| query_length | int32 | Number of query tokens |
|
|
|
|
|
88 |
| frequency_bucket | int32 | Monthly frequency of query (bucket) from 0 (high frequency) to 9 (low frequency) |
|
89 |
+
| n | int32 | Number of documents for current query, useful for padding |
|
90 |
+
| url_md5 | List[string] | MD5 hash of document URL, most reliable document identifier |
|
91 |
+
| text_md5 | List[string] | MD5 hash of document title and abstract |
|
92 |
+
| title | List[List[int32]] | List of tokens for document titles |
|
93 |
+
| abstract | List[List[int32]] | List of tokens for document abstracts |
|
94 |
+
| query_document_embedding | Tensor[Tensor[float16]] | BERT CLS token |
|
95 |
+
| label | Tensor[int32] | Relevance judgments on a scale from 0 (bad) to 4 (excellent) |
|
96 |
+
| bm25 | Tensor[float32] | BM25 score for documents |
|
97 |
+
| bm25_title | Tensor[float32] | BM25 score for document titles |
|
98 |
+
| bm25_abstract | Tensor[float32] | BM25 score for document abstracts |
|
99 |
+
| tf_idf | Tensor[float32] | TF-IDF score for documents |
|
100 |
+
| tf | Tensor[float32] | Term frequency for documents |
|
101 |
+
| idf | Tensor[float32] | Inverse document frequency for documents |
|
102 |
+
| ql_jelinek_mercer_short | Tensor[float32] | Query likelihood score for documents using Jelinek-Mercer smoothing (alpha = 0.1) |
|
103 |
+
| ql_jelinek_mercer_long | Tensor[float32] | Query likelihood score for documents using Jelinek-Mercer smoothing (alpha = 0.7) |
|
104 |
+
| ql_dirichlet | Tensor[float32] | Query likelihood score for documents using Dirichlet smoothing (lambda = 128) |
|
105 |
+
| document_length | Tensor[int32] | Length of documents |
|
106 |
+
| title_length | Tensor[int32] | Length of document titles |
|
107 |
+
| abstract_length | Tensor[int32] | Length of document abstracts |
|
108 |
|
109 |
## Example PyTorch collate function
|
110 |
Each sample in the dataset is a single query with multiple documents.
|
|
|
128 |
batch["n"].append(sample["n"])
|
129 |
|
130 |
return {
|
131 |
+
"query_document_embedding": pad_sequence(
|
132 |
+
batch["query_document_embedding"], batch_first=True
|
133 |
+
),
|
134 |
"position": pad_sequence(batch["position"], batch_first=True),
|
135 |
"click": pad_sequence(batch["click"], batch_first=True),
|
136 |
"n": torch.tensor(batch["n"]),
|
|
|
138 |
|
139 |
loader = DataLoader(dataset, collate_fn=collate_clicks, batch_size=16)
|
140 |
```
|
141 |
+
|