Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Instruction-Guided Bullet Point Summarization of Long Financial Earnings Call Transcripts

Subhendu Khatuya, Koushiki Sinha, Niloy Ganguly, Saptarshi Ghosh, Pawan Goyal

TL;DR

This work tackles bullet-point summarization of long Earnings Call Transcripts (ECTs) using the ECTSum dataset. It introduces FLAN-FinBPS, a two-stage pipeline that first builds an extractive context with an unsupervised question-based method and topic modeling, then generates abstractive bullet points via a parameter-efficient, instruction-tuned FLAN-T5 model with LoRA. The approach achieves state-of-the-art results across ROUGE, BERTScore, and factual-consistency metrics, significantly outperforming the previous baseline (ECTBPS) and reducing training requirements. The method enables accurate, concise, and numerically faithful summaries of complex financial transcripts, with potential applicability to other long, domain-specific documents.

Abstract

While automatic summarization techniques have made significant advancements, their primary focus has been on summarizing short news articles or documents that have clear structural patterns like scientific articles or government reports. There has not been much exploration into developing efficient methods for summarizing financial documents, which often contain complex facts and figures. Here, we study the problem of bullet point summarization of long Earning Call Transcripts (ECTs) using the recently released ECTSum dataset. We leverage an unsupervised question-based extractive module followed by a parameter efficient instruction-tuned abstractive module to solve this task. Our proposed model FLAN-FinBPS achieves new state-of-the-art performances outperforming the strongest baseline with 14.88% average ROUGE score gain, and is capable of generating factually consistent bullet point summaries that capture the important facts discussed in the ECTs.

Instruction-Guided Bullet Point Summarization of Long Financial Earnings Call Transcripts

TL;DR

This work tackles bullet-point summarization of long Earnings Call Transcripts (ECTs) using the ECTSum dataset. It introduces FLAN-FinBPS, a two-stage pipeline that first builds an extractive context with an unsupervised question-based method and topic modeling, then generates abstractive bullet points via a parameter-efficient, instruction-tuned FLAN-T5 model with LoRA. The approach achieves state-of-the-art results across ROUGE, BERTScore, and factual-consistency metrics, significantly outperforming the previous baseline (ECTBPS) and reducing training requirements. The method enables accurate, concise, and numerically faithful summaries of complex financial transcripts, with potential applicability to other long, domain-specific documents.

Abstract

While automatic summarization techniques have made significant advancements, their primary focus has been on summarizing short news articles or documents that have clear structural patterns like scientific articles or government reports. There has not been much exploration into developing efficient methods for summarizing financial documents, which often contain complex facts and figures. Here, we study the problem of bullet point summarization of long Earning Call Transcripts (ECTs) using the recently released ECTSum dataset. We leverage an unsupervised question-based extractive module followed by a parameter efficient instruction-tuned abstractive module to solve this task. Our proposed model FLAN-FinBPS achieves new state-of-the-art performances outperforming the strongest baseline with 14.88% average ROUGE score gain, and is capable of generating factually consistent bullet point summaries that capture the important facts discussed in the ECTs.
Paper Structure (7 sections, 2 figures, 3 tables)

This paper contains 7 sections, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Extractive Phase
  4. Abstractive Phase
  5. Testing Phase
  6. Results and Analysis
  7. Conclusion

Figures (2)

  • Figure 1: FLAN-FinBPS Architecture - Unsupervised Extractive module takes ECT document and a list of generated questions from reference summary as input to generate the extractive context. FLAN-T5 based Abstractive module takes task specific instruction and the extractive context to generate the bullet point summary
  • Figure 2: Distribution of questions across key topics. Revenue, earnings per Share, and sales together encompass nearly 45% of all topics.