---
license: apache-2.0
dataset_info:
  config_name: data
  features:
  - name: event
    dtype: int64
  - name: word
    dtype: string
  - name: topic
    dtype: string
  - name: selected_topic
    dtype: string
  - name: semantic_relevance
    dtype: int64
  - name: interestingness
    dtype: int64
  - name: pre-knowledge
    dtype: int64
  - name: sentence_number
    dtype: int64
  - name: participant
    dtype: string
  - name: eeg
    dtype:
      array2_d:
        shape:
        - 32
        - 2001
        dtype: float64
  splits:
  - name: train
    num_bytes: 11925180913
    num_examples: 23270
  download_size: 11927979870
  dataset_size: 11925180913
configs:
- config_name: data
  data_files:
  - split: train
    path: data/train-*
  default: true
task_categories:
- text-classification
- token-classification
language:
- en
size_categories:
- 10K<n<100K
---

We release a novel dataset containing 23,270 time-locked (0.7s) word-level EEG recordings acquired from participants who read both text that was semantically relevant and irrelevant to self-selected topics.

Submitted to ICLR 2025. The raw EEG data and the datasheet will be avaialble after acceptance to avoid disclosure of the authors' identity.

See [code repository][1] for benchmark results.


EEG data acquisition:
![Data acquisition](data-acquisition.PNG)

Explanations of the variables:

- **event** corresponds to a specific point in time during EEG data collection and represents the onset of an event (presentation of a word)
- **word** is a word read by the participant
- **topic** is the topic of the document to which the **word** belongs to
- **selected topic** indicates the topic the participant has selected
- **semantic relevance** indicates whether the word is semantically relevant (expressed as 1) or semantically irrelevant (expressed as 0) to the topic selected by the participant
- **interestingness** indicates the participant's interest in the topic of a document
- **pre-knowledge** indicates the participant's previous knowledge about the topic of the document
- **sentence number** represents the sentence number to which the word belongs
- **eeg** - brain recordings having a shape of 32 x 2001 for each word 

The dataset can be downloaded and used as follows:

```py
import numpy as np
from datasets import load_dataset

# Load the cleaned version of the dataset
d = load_dataset("Quoron/EEG-semantic-text-relevance", "data")

# See the structure of the dataset
print(d)

# Get the first entry in the dataset
first_entry = d['train'][0]

# Get EEG data as numpy array in the first entry
eeg = np.array(first_entry['eeg'])

# Get a word in the first entry
word = first_entry['word']

```

We recommend using the Croissant metadata to explore the dataset.


  [1]: https://anonymous.4open.science/r/EEG-semantic-text-relevance-651D