DynamicTOC: Persona-based Table of Contents for Consumption of Long Documents

Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 5133 - 5143 July 10-15, 2022 ©2022 Association for Computational Linguistics DYNAMICTOC: Persona-based Table of Contents for Consumption of Long Documents Himanshu Maheshwari, Shelly Jain, Tanvi Karandikar IIIT Hyderabad him.mj26,shelly.jain.viii,tanvi.karandikar141@gmail.com Nethraa Sivakumar SSNCE, Chennai nethraa18096@ece.ssn.edu.in Navita Goyal UMD, College Park navita@umd.edu Vinay Aggarwal Adobe Research vinayagg@adobe.com Sumit Shekhar Adobe Research sushekha@adobe.com Abstract Long documents like contracts, financial doc- uments, etc., are often tedious to read through. Linearly consuming (via scrolling or naviga- tion through default table of content) these documents is time-consuming and challenging. These documents are also authored to be con- sumed by varied entities (referred to as persona in the paper) interested in only certain parts of the document. In this work, we describe DYNAMICTOC, a dynamic table of content- based navigator, to aid in the task of non-linear, persona-based document consumption. DY- NAMICTOC highlights sections of interest in the document as per the aspects relevant to dif- ferent personas. DYNAMICTOC is augmented with short questions to assist the users in un- derstanding underlying content. This uses a novel deep-reinforcement learning technique to generate questions on these persona-clustered paragraphs. Human and automatic evaluations suggest the efficacy of both end-to-end pipeline and different components of DYNAMICTOC. 1 Introduction Documents such as financial statements, reports and contracts are often long and comprehensive, replete with domain-specific description and infor- mation. They are meant to be consumed by several entities or personas, e.g. legal department of com- panies, customers or financial organizations such as banks. As these documents contain vital infor- mation about the business, the business personas are often required to read through and analyze the documents in details. These personas are often interested in different sections of the document, based on the business requirements. For example, employees might be interested in the stock pro- grams of the company, whereas the lenders and investors would like to read through profit state- ments. The traditional technology to navigate long documents is through a Table of Contents (ToC) populated with the heading of each section and chapters. However, the Table of Contents does not show the information present in the underlying paragraphs of a section, and there is no way to high- light information relevant to different personas. To this effect, we propose DYNAMICTOC, an intelligent table of contents-based navigator. DY- NAMICTOC provides user the flexibility to choose the persona and read the document from its lens. For the current work, we focus on the finance and legal domain, and hence, personas are taken as com- monplace entities like investors, lenders, financial bodies, etc. DYNAMICTOC highlights the rele- vant sections of the document as per the persona. For this, the input finance or contract document is segmented at the paragraph level and a cluster of “aspects or topics” is inferred for each para. These are then mapped to the interest topics of the personas. Further, DYNAMICTOC has a novel question-based guided experience, to …

Content truncated to 500 words