README.md

metadata

license: cc-by-nc-4.0
viewer: false

Baidu ULTR Dataset - UvA BERT-12l-12h

Query-document vectors and clicks for a subset of the Baidu Unbiased Learning to Rank dataset. This dataset uses a BERT cross-encoder with 12 layers trained on a Masked Language Modeling (MLM) and click-through-rate (CTR) prediction task to compute query-document vectors (768 dims). The model is available under model/.

Setup

1. Install huggingface datasets
2. Install pandas and pyarrow: pip install pandas pyarrow
3. Optionally, you might need to install a pyarrow-hotfix if you cannot install pyarrow >= 14.0.1
4. You can now use the dataset as described below.

Load train / test click dataset:

from datasets import load_dataset

dataset = load_dataset(
    "philipphager/baidu-ultr_uva-mlm-ctr",
    name="clicks",
    split="train", # ["train", "test"]
    cache_dir="~/.cache/huggingface",
)

dataset.set_format("torch") #  [None, "numpy", "torch", "tensorflow", "pandas", "arrow"]

Load expert annotations:

from datasets import load_dataset

dataset = load_dataset(
    "philipphager/baidu-ultr_uva-mlm-ctr",
    name="annotations",
    split="test",
    cache_dir="~/.cache/huggingface",
)

dataset.set_format("torch") #  [None, "numpy", "torch", "tensorflow", "pandas", "arrow"]

Available features

Each row of the click / annotation dataset contains the following attributes. Use a custom collate_fn to select specific features (see below):

Click dataset

name	dtype	description
query_id	string	Baidu query_id
query_md5	string	MD5 hash of query text
url_md5	List[string]	MD5 hash of document url, most reliable document identifier
text_md5	List[string]	MD5 hash of document title and abstract
query_document_embedding	Tensor[float16]	BERT CLS token
click	Tensor[int32]	Click / no click on a document
n	int32	Number of documents for current query, useful for padding
position	Tensor[int32]	Position in ranking (does not always match original item position)
media_type	Tensor[int32]	Document type (label encoding recommended as ids do not occupy a continous integer range)
displayed_time	Tensor[float32]	Seconds a document was displayed on screen
serp_height	Tensor[int32]	Pixel height of a document on screen
slipoff_count_after_click	Tensor[int32]	Number of times a document was scrolled off screen after previously clicking on it

Expert annotation dataset

name	dtype	description
query_id	string	Baidu query_id
query_md5	string	MD5 hash of query text
text_md5	List[string]	MD5 hash of document title and abstract
query_document_embedding	Tensor[float16]	BERT CLS token
label	Tensor[int32]	Relevance judgment on a scale from 0 (bad) to 4 (excellent)
n	int32	Number of documents for current query, useful for padding
frequency_bucket	int32	Monthly frequency of query (bucket) from 0 (high frequency) to 9 (low frequency)

Example PyTorch collate function

Each sample in the dataset is a single query with multiple documents. The following example demonstrates how to create a batch containing multiple queries with varying numbers of documents by applying padding:

import torch
from typing import List
from collections import defaultdict
from torch.nn.utils.rnn import pad_sequence
from torch.utils.data import DataLoader


def collate_clicks(samples: List):
    batch = defaultdict(lambda: [])

    for sample in samples:
        batch["query_document_embedding"].append(sample["query_document_embedding"])
        batch["position"].append(sample["position"])
        batch["click"].append(sample["click"])
        batch["n"].append(sample["n"])

    return {
        "query_document_embedding": pad_sequence(batch["query_document_embedding"], batch_first=True),
        "position": pad_sequence(batch["position"], batch_first=True),
        "click": pad_sequence(batch["click"], batch_first=True),
        "n": torch.tensor(batch["n"]),
    }

loader = DataLoader(dataset, collate_fn=collate_clicks, batch_size=16)