Task 2

Introduction

The bursty event cascade classification task aims to classify social network structures as either the cascade of bursty events or normal events. Efficient classification of bursty event cascade enables the early detection of bursty events, safeguarding the integrity of online discourse and public perception.

Dataset

For the bursty event cascade classification challenge, we collect a dataset from Weibo with 6,886 cascades of bursty events and 13,762 cascades of normal events. The dataset comprises posts made during the event, along with interactions such as comments and shares, forming a cascade network. Each instance in the dataset contains information about the content and timestamp of the share. The dataset is labeled with the event class for each cascade network.

📥 Dataset Download

Dataset Description

The dataset comprises two files: train.json and test.json. The test.json file does not contain labels.

Field	Meaning
edge_index	This field represents the edges of the cascade network, denoting the Weibo share interactions. It is structured as a 2D array with a shape of `[2, edge_num]`, where `edge_num` signifies the number of interactions within the cascade.
x	This field contains the feature embeddings of each Weibo post in the cascade network. These embeddings represent the content of the Weibo posts processed by BERT, a language model known for its contextual representation learning.
label	This field indicates the label of the event class associated with the cascade network. It serves as the ground truth for classification tasks, distinguishing between bursty events and normal events.

Metric

The evaluation metric is the F1 score, the harmonic mean of the precision and recall.

Submission Rules

There is no limit to the number of submissions per team. The final assessment will be based on the team's most recent submission.

Submission Format

The submission should be a .zip file with the following structure:

{team_id}.zip
    ├── source_code
    │   ├── evaluation.py
    │   ├── ...
    ├── result.json
    ├── introduction.pdf

Result File Format

[0, 1, 0, 1, 0, ..., 1, 0, 0]

Source Code

The source_code folder must contain an executable evaluation.py script that can test the provided data in test.json and generate the corresponding result.json file.

Model Introduction

The introduction.pdf should describe the method and its innovative aspects. The innovation of the method will contribute to 20% of the total score.

Submission Method

Upload your submission files to a secure platform like Google Drive, generate a shareable link, and submit via the form below.

Only submissions that fully comply with the rules will be accepted.

📝 Submission Form

⚔️ Task 2: Bursty Event Cascade Classification