Embed Notice

HTML Code

<blockquote style="position: relative; padding-left: 55px;"><section><a href="https://mastodon.social/users/persagen/statuses/110877502203993245">Victoria Stuart 🇨🇦 🏳️‍⚧️ (persagen@mastodon.social)'s status on Wednesday, 13-Sep-2023 15:58:20 JST</a><a href="https://mastodon.social/@persagen" title="persagen@mastodon.social"><img src="https://gnusocial.jp/avatar/156189-48-20240908055012.webp" width="48" height="48" alt="Victoria Stuart 🇨🇦 🏳️‍⚧️" style="position: absolute; left: 0; top: 0;">Victoria Stuart 🇨🇦 🏳️‍⚧️</a></section><article><p>Addendum 2</p><p>Comments: article placed here due to<br>1. use of prompt engineering + chain of thought (mentioned above)<br>2. application to long documents (here, applied to legal domain, but broadly applicable)<br>3. novelty</p><p>Large Language Model Prompt Chaining for Long Legal Document Classification<br><a href="https://arxiv.org/abs/2308.04138" rel="nofollow noreferrer">https://arxiv.org/abs/2308.04138</a></p><p><a href="https://mastodon.social/tags/LLM" rel="tag">#LLM</a> <a href="https://mastodon.social/tags/LargeLanguageModels" rel="tag">#LargeLanguageModels</a> <a href="https://mastodon.social/tags/ChatGPT" rel="tag">#ChatGPT</a> <a href="https://mastodon.social/tags/PromptEngineering" rel="tag">#PromptEngineering</a> <a href="https://mastodon.social/tags/ChainOfThought" rel="tag">#ChainOfThought</a> <a href="https://mastodon.social/tags/reasoning" rel="tag">#reasoning</a> <a href="https://mastodon.social/tags/classification" rel="tag">#classification</a> <a href="https://mastodon.social/tags/NLP" rel="tag">#NLP</a> <a href="https://mastodon.social/tags/NaturalLanguageProcessing" rel="tag">#NaturalLanguageProcessing</a> <a href="https://mastodon.social/tags/semantics" rel="tag">#semantics</a> <a href="https://mastodon.social/tags/PromptChaining" rel="tag">#PromptChaining</a> <a href="https://mastodon.social/tags/TopicModeling" rel="tag">#TopicModeling</a> <a href="https://mastodon.social/tags/SCOTUS" rel="tag">#SCOTUS</a></p></article><footer><a rel="bookmark" href="https://gnusocial.jp/conversation/2054101#notice-4031216">In conversation</a><time datetime="2023-09-13T15:58:20+09:00" title="Wednesday, 13-Sep-2023 15:58:20 JST">about a year ago</time> <span>from <span><a href="https://mastodon.social/@persagen/110877502203993245" rel="external" title="Sent from mastodon.social via ActivityPub">mastodon.social</a></span></span><a href="https://mastodon.social/@persagen/110877502203993245">permalink</a><h4>Attachments</h4><ol><li><label><a rel="external" href="https://gnusocial.jp/attachment/1514589">Schematic showing the novel "prompt chaining" method for labeling long documents (here U.S. Supreme Court cases) with topic labels.Figure 1 in: Large Language Model Prompt Chaining for Long Legal Document ClassificationLink / URL:
https://arxiv.org/abs/2308.04138</a></label><br><a href="https://files.mastodon.social/media_attachments/files/110/877/587/918/507/105/original/33160b980c3a4030.png" rel="external">https://files.mastodon.social/media_attachments/files/110/877/587/918/507/105/original/33160b980c3a4030.png</a></li><li><article><header><div>Domain not in remote thumbnail source whitelist: static.arxiv.org</div><h5><a href="https://arxiv.org/abs/2308.04138#LLM">Large Language Model Prompt Chaining for Long Legal Document Classification</a></h5><div></div></header><div>Prompting is used to guide or steer a language model in generating an
appropriate response that is consistent with the desired outcome. Chaining is a
strategy used to decompose complex tasks into smaller, manageable components.
In this study, we utilize prompt chaining for extensive legal document
classification tasks, which present difficulties due to their intricate
domain-specific language and considerable length. Our approach begins with the
creation of a concise summary of the original document, followed by a semantic
search for related exemplar texts and their corresponding annotations from a
training corpus. Finally, we prompt for a label - based on the task - to
assign, by leveraging the in-context learning from the few-shot prompt. We
demonstrate that through prompt chaining, we can not only enhance the
performance over zero-shot, but also surpass the micro-F1 score achieved by
larger models, such as ChatGPT zero-shot, using smaller models.</div><footer></footer></article></li></ol></footer></blockquote>

Corresponding Notice

Embed this notice
Victoria Stuart 🇨🇦 🏳️‍⚧️ (persagen@mastodon.social)'s status on Wednesday, 13-Sep-2023 15:58:20 JST Victoria Stuart 🇨🇦 🏳️‍⚧️
Addendum 2
Comments: article placed here due to
1. use of prompt engineering + chain of thought (mentioned above)
2. application to long documents (here, applied to legal domain, but broadly applicable)
3. novelty
Large Language Model Prompt Chaining for Long Legal Document Classification
https://arxiv.org/abs/2308.04138
#LLM #LargeLanguageModels #ChatGPT #PromptEngineering #ChainOfThought #reasoning #classification #NLP #NaturalLanguageProcessing #semantics #PromptChaining #TopicModeling #SCOTUS
In conversationabout a year ago from mastodon.socialpermalink
Attachments
1. Schematic showing the novel "prompt chaining" method for labeling long documents (here U.S. Supreme Court cases) with topic labels. Figure 1 in: Large Language Model Prompt Chaining for Long Legal Document Classification Link / URL: https://arxiv.org/abs/2308.04138
  https://files.mastodon.social/media_attachments/files/110/877/587/918/507/105/original/33160b980c3a4030.png
2. Domain not in remote thumbnail source whitelist: static.arxiv.org
  Large Language Model Prompt Chaining for Long Legal Document Classification
  Prompting is used to guide or steer a language model in generating an appropriate response that is consistent with the desired outcome. Chaining is a strategy used to decompose complex tasks into smaller, manageable components. In this study, we utilize prompt chaining for extensive legal document classification tasks, which present difficulties due to their intricate domain-specific language and considerable length. Our approach begins with the creation of a concise summary of the original document, followed by a semantic search for related exemplar texts and their corresponding annotations from a training corpus. Finally, we prompt for a label - based on the task - to assign, by leveraging the in-context learning from the few-shot prompt. We demonstrate that through prompt chaining, we can not only enhance the performance over zero-shot, but also surpass the micro-F1 score achieved by larger models, such as ChatGPT zero-shot, using smaller models.

Public

Embed Notice

HTML Code

Corresponding Notice