LCSTS: A Large Scale Chinese Short Text Summarization Dataset

TL;DR

The paper tackles the lack of large-scale datasets for Chinese short text summarization by constructing LCSTS from Sina Weibo, yielding 2.4 million training pairs plus a manually labeled subset and an evaluation set. It adopts a neural encoder-decoder approach, exploring both character-based and word-based inputs and two decoding contexts, with GRU-based RNNs and ADADELTA optimization, evaluated via ROUGE. Key contributions include the dataset itself, standard data splits, a human-labeled subset for quality analysis, and a baseline neural model demonstrating promising results with context-aware decoding. The work enables supervised learning for Chinese short text summarization at scale and points to avenues for improvement, such as hierarchical architectures and rare-word handling, to further close the gap to human performance.

Abstract

Automatic text summarization is widely regarded as the highly difficult problem, partially because of the lack of large text summarization data set. Due to the great challenge of constructing the large scale summaries for full text, in this paper, we introduce a large corpus of Chinese short text summarization dataset constructed from the Chinese microblogging website Sina Weibo, which is released to the public {http://icrc.hitsz.edu.cn/Article/show/139.html}. This corpus consists of over 2 million real Chinese short texts with short summaries given by the author of each text. We also manually tagged the relevance of 10,666 short summaries with their corresponding short texts. Based on the corpus, we introduce recurrent neural network for the summary generation and achieve promising results, which not only shows the usefulness of the proposed corpus for short text summarization research, but also provides a baseline for further research on this topic.

Figures (8)

• Figure 1: A Weibo Posted by People's Daily.
• Figure 2: Diagram of the process for creating the dataset.
• Figure 3: Box plot of lengths for short text(ST), segmented short text(Segmented ST), summary(SUM) and segmented summary(Segmented SUM). The red line denotes the median, and the edges of the box the quartiles.
• Figure 4: Five examples of different scores.
• Figure 5: The graphical depiction of RNN encoder and decoder framework without context.