LongWanjuan: Towards Systematic Measurement for Long Text Quality
This work addresses the problem of enhancing long-text capabilities in foundation models for AI researchers and practitioners, though it is incremental as it builds on existing data quality efforts.
The authors tackled the lack of systematic quality assessment for long texts in training data by developing metrics based on coherence, cohesion, and complexity, and created LongWanjuan, a bilingual dataset with over 160B tokens that improved model performance on long-text tasks.
The quality of training data are crucial for enhancing the long-text capabilities of foundation models. Despite existing efforts to refine data quality through heuristic rules and evaluations based on data diversity and difficulty, there's a lack of systematic approaches specifically tailored for assessing long texts. Addressing this gap, our work systematically measures the quality of long texts by evaluating three fundamental linguistic dimensions: coherence, cohesion, and complexity. Drawing inspiration from the aforementioned three dimensions, we introduce a suite of metrics designed to evaluate the quality of long texts, encompassing both statistical and pre-trained language model-based ones. Leveraging these metrics, we present LongWanjuan, a bilingual dataset specifically tailored to enhance the training of language models for long-text tasks with over 160B tokens. In LongWanjuan, we categorize long texts into holistic, aggregated, and chaotic types, enabling a detailed analysis of long-text quality. Furthermore, we devise a data mixture recipe that strategically balances different types of long texts within LongWanjuan, leading to significant improvements in model performance on long-text tasks. The code and dataset are available at https://github.com/OpenLMLab/LongWanjuan.