Convex Aggregation for Opinion Summarization

Recent advances in text autoencoders have significantly improved the quality of the latent space, which enables models to generate grammatical and consistent text from aggregated latent vectors. As a successful application of this property, unsupervised opinion summarization models generate a summary by decoding the aggregated latent vectors of inputs. More specifically, they perform the aggregation via simple average. However, little is known about how the vector aggregation step affects the generation quality. In this study, we revisit the commonly used simple average approach by examining the latent space and generated summaries. We found that text autoencoders tend to generate overly generic summaries from simply averaged latent vectors due to an unexpected $L_2$-norm shrinkage in the aggregated latent vectors, which we refer to as summary vector degeneration. To overcome this issue, we develop a framework Coop, which searches input combinations for the latent vector aggregation using input-output word overlap. Experimental results show that Coop successfully alleviates the summary vector degeneration issue and establishes new state-of-the-art performance on two opinion summarization benchmarks. Code is available at \url{https://github.com/megagonlabs/coop}.

PDF Abstract Findings (EMNLP) 2021 PDF Findings (EMNLP) 2021 Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Unsupervised Opinion Summarization Amazon BiMeanVAE - Coop ROUGE-1 36.57 # 1
ROUGE-2 7.23 # 1
ROUGE-L 21.24 # 1
Unsupervised Opinion Summarization Amazon Optimus - SimpleAvg ROUGE-1 33.54 # 4
ROUGE-2 6.18 # 4
ROUGE-L 19.34 # 4
Unsupervised Opinion Summarization Amazon BiMeanVAE - SimpleAvg ROUGE-1 33.6 # 3
ROUGE-2 6.64 # 2
ROUGE-L 20.87 # 2
Unsupervised Opinion Summarization Amazon Optimus - Coop ROUGE-1 35.32 # 2
ROUGE-2 6.22 # 3
ROUGE-L 19.84 # 3
Unsupervised Opinion Summarization Yelp Optimus - Coop ROUGE-1 33.68 # 2
ROUGE-2 7 # 2
ROUGE-L 18.95 # 3
Unsupervised Opinion Summarization Yelp BiMeanVAE - Coop ROUGE-1 35.37 # 1
ROUGE-2 7.35 # 1
ROUGE-L 19.94 # 1
Unsupervised Opinion Summarization Yelp BiMeanVAE - SimpleAvg ROUGE-1 32.87 # 3
ROUGE-2 6.93 # 3
ROUGE-L 19.89 # 2
Unsupervised Opinion Summarization Yelp Optimus - SimpleAvg ROUGE-1 31.23 # 4
ROUGE-2 6.48 # 4
ROUGE-L 18.27 # 4

Methods