Automatically Summarizing Evidence from Clinical Trials: A Prototype Highlighting Current Challenges
Automatically Summarizing Evidence from Clinical Trials: A Prototype Highlighting Current Challenges
复制标题
DOI：
10.48550/arxiv.2303.05392
复制
发表时间：
2023-03
期刊：
影响因子：
--
通讯作者：
S. Ramprasad;Denis Jered McInerney;Iain J. Marshal;Byron Wallace
中科院分区：
文献类型：
--
作者：
S. Ramprasad;Denis Jered McInerney;Iain J. Marshal;Byron Wallace
在这项工作中，我们提出了TrialsSummarizer，一个系统，其目的是自动总结的随机对照试验中最相关的一个给定的查询的证据。在先前工作的基础上，该系统检索与指定条件、干预和结果的组合的查询相匹配的试验出版物，并根据样本量和估计的研究质量对它们进行排名。前k个此类研究通过神经多文档摘要系统，产生这些试验的概要。我们考虑两种架构：基于BART的标准序列到序列模型，以及旨在为最终用户提供更大透明度和可控性的多头架构。这两种模型都可以为查询检索到的证据生成流畅和相关的摘要，但是它们引入不受支持的语句的倾向使得它们目前不适合在该领域中使用。所提出的体系结构可以帮助用户验证输出，允许用户将生成的令牌追溯到输入。演示视频可以在https://vimeo.com/735605060The上找到原型，源代码和模型重量可以在https://sanjanaramprasad.github.io/trials-summarizer/上找到
In this work we present TrialsSummarizer, a system that aims to automatically summarize evidence presented in the set of randomized controlled trials most relevant to a given query. Building on prior work, the system retrieves trial publications matching a query specifying a combination of condition, intervention(s), and outcome(s), and ranks these according to sample size and estimated study quality.The top-k such studies are passed through a neural multi-document summarization system, yielding a synopsis of these trials. We consider two architectures: A standard sequence-to-sequence model based on BART, and a multi-headed architecture intended to provide greater transparency and controllability to end-users.Both models produce fluent and relevant summaries of evidence retrieved for queries, but their tendency to introduce unsupported statements render them inappropriate for use in this domain at present.The proposed architecture may help users verify outputs allowing users to trace generated tokens back to inputs. The demonstration video can be found at https://vimeo.com/735605060The prototype, source code, and model weights are available at: https://sanjanaramprasad.github.io/trials-summarizer/