Influencer: Empowering Everyday Users in Creating Promotional Posts via AI-infused Exploration and Customization

Research Background and Problem

• What problems or challenges did the authors identify?
  Creating engaging promotional posts on social platforms is an essential tool for ordinary users to share creativity, enhance community engagement, or generate revenue through small businesses. However, ordinary users often lack design skills, making it difficult and time-consuming to produce promotional posts with high-quality images and effective copywriting. Additionally, during the creation phase, methods for drawing inspiration are inefficient and lack flexibility in modifying recommended design examples.

• Why is this problem important?
  Designing high-quality promotional posts is crucial for increasing brand exposure, attracting customers, and ultimately achieving higher conversion rates. Failing to address these issues may reduce users' creative capabilities, limit their economic opportunities, and hinder their ability to express ideas.

• Research Motivation and Related Work
  The authors point out that existing tools (e.g., Google Image Search, ChatGPT, Adobe Photoshop) have shortcomings such as uneditable search results and reliance on other tools to refine designs. While previous research has proposed some design tools to support example exploration, these tools are typically targeted at professional users and lack an integrated workflow. Furthermore, the generated designs cannot dynamically match users' brand or product information. These issues motivate the development of an integrated tool that combines inspiration generation with promotional post creation.

Solution

• What methods or solutions did the authors propose?
  The authors proposed an interactive tool called "Infuencer" to help non-professional design users quickly generate creative ideas and efficiently produce high-quality promotional posts. Its main features include:

  1. Multidimensional Recommendation System: Supports multidimensional exploration based on images and copywriting, recommending design materials from aspects such as storylines, colors, and objects.
  2. Context-Aware Exploration Module: Allows users to drag brand information, product images, etc., into the recommendation module to update content.
  3. Flexible Fusion Mechanism: Supported by machine learning, it enables seamless integration of different content (images and copywriting).
  4. Mind Map Interface: Organizes ideas, tracks design paths, and generates multiple design alternatives.

• What are the innovative aspects of this solution?

  1. Compared to traditional systems, it integrates all steps from inspiration generation to promotional post creation into a single workflow.
  2. Provides multidimensional recommendation mechanisms and dynamic context-based recommendation features.
  3. Enhances user interactivity in the design process, such as quick creation through dragging and merging design materials.
  4. Introduces a mind map interface to track users' thought processes during design.

• What are the implementation steps and key technologies used?

  1. Build a database containing associations between images and keywords to support recommendation data.
  2. Use backend systems incorporating GPT-4 and other AI models (e.g., OFA framework, DALL·E) to generate copywriting and image recommendations.
  3. Combine user interactions (dragging images, creating text boxes, etc.) with AI recommendations to dynamically update recommended content.
  4. Provide flexible content fusion options, allowing users to customize designs via text prompts or image uploads.
  5. Design a visual canvas-based interface to facilitate users in managing the design process and tracking exploration paths.

Research Outcomes

• What specific outcomes were achieved?

  1. Experimental Validation: Compared to industry-standard tools (Google Search + ChatGPT + PowerPoint), Infuencer significantly improved design efficiency. In experiments, users saved design time on average and produced higher-quality promotional posts.
  2. User Satisfaction: The Creative Support Index (CSI) score for Infuencer was significantly higher than baseline tools, with users enjoying the exploration and creation process more.
  3. Expert Evaluation: Design experts found that promotional posts generated using Infuencer had advantages in content clarity, consistency, and other aspects.
  4. Third-Party User Evaluation: Consumer surveys showed that promotional posts created with Infuencer had higher visual appeal and interactivity.

• What advantages does it have compared to existing solutions?

  1. Automated and user-friendly design workflow, reducing reliance on other complex tools.
  2. Provides multidimensional inspiration exploration and context-sensitive dynamic recommendation features, significantly improving design efficiency and outcomes.
  3. Supports generating multiple design alternatives, enabling users to compare and choose conveniently.

• What were the experimental or evaluation results?
  Task completion success rate: Users achieved a 100% task completion rate using Infuencer, while baseline tools (Google Search + ChatGPT + PowerPoint) showed significantly lower completion rates across multiple tasks. Design outcomes were also rated higher in quality by design experts and users.

• Limitations and Future Directions

  1. The current context-aware recommendation module has limitations in color extraction and object detection accuracy. Future improvements could involve optimizing recommendation algorithms (e.g., incorporating advanced vision models) to enhance recommendation quality.
  2. The mind map interface may become cluttered when handling large amounts of design materials, necessitating improvements in layout management functionality.
  3. The system does not yet address complex typography and font design areas, which could be enhanced by expanding datasets and integrating other professional design tools (e.g., Photoshop).
  4. Experimental validation covered only limited task scenarios; future research could extend to more real-world application scenarios to further evaluate its generalizability.

Through these innovative features, Infuencer demonstrates significant potential in empowering ordinary users to easily create high-quality promotional posts.