Qmsum Cleaned: Meeting Transcripts with Summarization Prompts

A dataset of meeting transcripts curated for summarization tasks, hosted on Hugging Face by pszemraj. The dataset includes documents prefixed with specific prompts instructing a model on the summarization task to perform. The most frequent prefix in the training split is 'Summarize the whole meeting.', appearing 121 times.

Use Cases

• Training models for general meeting summarization based on the 'Summarize the whole meeting' prompt.
• Fine-tuning models for targeted topic summarization based on prompts like 'What did the team discuss about the product cost?'.
• Evaluating model performance on conditional text generation tasks where the output is guided by an input prefix.
• Studying the effect of prompt specificity on summarization quality and model behavior.

Strengths

• The dataset explicitly includes task-specific prefixes for each document, providing clear instruction context.
• The training split contains 121 instances with the 'Summarize the whole meeting.' prefix, indicating a substantial base for a core task.
• The dataset has been cleaned, suggesting some level of preprocessing or curation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face, author pszemraj.

Freshness

Last updated 2025-12-29 03:56:11; freshness should be verified.