PPTAgent - A framework developed by the Chinese Academy of Sciences for automatically generating high-quality presentations. - AiBoss

What is PPTAgent?

PPTAgent is an innovative framework developed by the Chinese Information Processing Laboratory of the Institute of Software, Chinese Academy of Sciences. Based on a two-stage editing method that mimics human workflows, it automatically generates high-quality presentations from documents. PPTAgent analyzes reference presentations, extracts structural and content patterns, drafts outlines based on code actions, and generates slides, ensuring content consistency and alignment. Leveraging its Large Language Model (LLM) capabilities, PPTAgent decomposes the presentation generation process into an iterative editing workflow, improving the coherence and adaptability of generated presentations and better handling complex formatting issues. PPTAgent introduces the PPT Eval evaluation framework, comprehensively assessing the quality of generated presentations across three dimensions: content, design, and coherence, providing valuable resources and insights for future research on presentation generation.

PPTAgent's main functions

Analysis Reference PresentationUnderstand its structural and content patterns.
Drafting an outlineBased on the analysis results, create a detailed presentation outline.
Generate slidesBased on code actions, the outline is transformed into specific slide content, ensuring consistency and alignment of the content.
Editing and correctionIt provides a feedback mechanism to edit and self-correct the generated slides, thereby improving the quality of the generated documents.
Comprehensive assessmentBased on the PPT Eval framework, the quality of the generated presentation is evaluated from three dimensions: content, design, and coherence.

PPTAgent's technical principles

Phase 1: Presentation Analysis
- Slide clustering:
  - Slides can be divided into two categories: slides that support the presentation structure (such as opening slides) and slides that convey specific content (such as bulleted slides).
  - Different clustering algorithms were used to cluster the slides based on textual or visual features. For structured slides, LLM was used to infer the functional role of each slide; for content slides, hierarchical clustering was used based on image similarity.
- Pattern Extraction:
  - Further analyze the content pattern of the slides to ensure the editing is purposeful.
  - Leveraging the context-aware capabilities of LLM, diverse content patterns are extracted. Each element is represented by category, modality, and content, and the pattern of each slide is extracted based on LLM's instructional compliance and structured output capabilities.
Phase Two: Presentation Generation
- Outline generation:
  - Guide LLM creators to create a structured outline containing multiple entries. Each entry specifies a reference slide, an index of related document sections, and a title and description for the new slide.
  - Leveraging the planning and summarizing capabilities of LLM, and combining the semantic information of document content and reference presentations, a coherent and engaging outline can be generated to guide the subsequent generation process.
- Slide generation:
  - Guided by the outline, new slides are generated by iteratively editing reference slides.
  - Implement five specialized APIs to support LLM editing, deleting, and copying of text elements, as well as editing and deleting of visual elements.
  - Converting the slides from their original XML format to an HTML representation makes LLM easier to understand and manipulate.
  - LLM accepts two types of input: source document text based on partial indexes and titles of available images. New slide content is generated based on content patterns.
  - LLM uses the generated content, the HTML representation of the reference slides, and the API documentation to produce executable editing actions. Executed in a REPL environment, the system detects errors during execution and provides real-time feedback. LLM iteratively optimizes editing actions based on intermediate results, enhancing the robustness of the generation process.

PPTAgent project address

GitHub repository:https://github.com/icip-cas/PPTAgent
arXiv technical paper:https://arxiv.org/pdf/2501.03936

Application scenarios of PPTAgent

EducationTeachers can quickly generate presentation slides for course explanations, including key knowledge points, charts, and examples, improving teaching efficiency and students' learning interest.
Corporate TrainingThis tool generates training presentations for new employees, introducing company culture, rules and regulations, business processes, and other content to help new employees quickly understand the company environment.
MarketingThe marketing team generates product promotion presentations, including product features, market analysis, user cases, etc., for use in client meetings or marketing events.
project managementThe project team generates a presentation document to report on project progress, including project goals, progress status, problems encountered and solutions, for use in reporting to management or clients.
Personal useGenerate your own presentation slides, including the presentation topic, key points, supporting materials, etc., to improve the effectiveness of your presentation.