---
language:
- zh
- en
pretty_name: TopicVid
---
# TopicVid Dataset

This dataset provides structured metadata, content features, and a heterogeneous graph related to short-video topics and subtopics. It is designed for tasks such as topic analysis, audience interaction modeling, peak prediction, and research on graph neural networks or graph retrieval.

---

## Contents

- `available_dataset_with_subtopic.json` — Processed structured raw data of short video content and interaction statistics about topics.
- `comment.npy` — Comment features.
- `content.npy` — Content features.
- `desc.npy` — Description features.
- `heterogeneous_graph.pkl` — Heterogeneous graph file.
- `title.npy` — Title features.
- `topic.npy` — Topic embeddings.
- `video.npy` — Video features.

---

## Data Structure

### 1) `available_dataset_with_subtopic.json`
This file contains the raw data of short video content and interaction statistics.

Fields:

- `url` (string) — Direct link to the video on the platform.
- `desc` (string) — Description text of the video content.
- `title` (string) — Title of the video post.
- `content` (string) — Additional text content; may be empty.
- `user_id` (string) — Unique identifier of the publishing user.
- `duration` (integer) — Video duration in seconds.
- `platform` (string) — Source platform name (e.g., Douyin, Kuaishou).
- `post_create_time` (string) — Time of publication in "YYYY-MM-DD HH:MM:SS" format.
- `topic` (string) — Main topic associated with the video.
- `subtopic` (string) — Numbered subcategory under the main topic.
- `time_frames` (dict) — Interaction statistics recorded at different dates.
  - Key: Date in "YYYY-MM-DD" format
  - Value: Dictionary with fields:
    - `fans_count` — Number of followers
    - `like_count` — Number of likes
    - `view_count` — Number of views
    - `share_count` — Number of shares
    - `collect_count` — Number of collections
    - `comment_count` — Number of comments
- `comments` (dict) — Collection of user comments.
  - Key: Comment index (string)
  - Value: Dictionary with fields:
    - `comment_user_id` — Commenting user ID
    - `comment_nickname` — Commenting user's display name
    - `comment_content` — Comment text
    - `comment_time` — Time of comment
    - `ip_address` — IP location of the commenting user

### 2) `*.npy`
Numpy arrays containing preprocessed embeddings or feature vectors.

### 3) `heterogeneous_graph.pkl`
A serialized Python object containing:
- Node types and indices
- Edge types and lists
- Labels information is available at [link](https://github.com/chensh911/TLGM/blob/main/data/subtopic_label.csv)