| --- |
| base_model: roberta-base |
| license: mit |
| metrics: |
| - accuracy |
| - f1 |
| tags: |
| - generated_from_trainer |
| - unsloth |
| model-index: |
| - name: OpenSesame |
| results: [] |
| --- |
| |
| <!-- This model card has been generated automatically according to the information the Trainer had access to. You |
| should probably proofread and complete it, then remove this comment. --> |
|
|
| # How to interpretate the output |
|
|
| LABEL 0 = User hasn't buying intentions. |
|
|
| LABEL 1 = User has buying intentions. |
|
|
| # Word Of Prompt |
|
|
| **Overview:** |
| "Word Of Prompt" redefines advertising by integrating it seamlessly into natural language conversations. |
| Utilizing fine-tuned RoBERTa and Llama3, "Word Of Prompt" detects user intent to purchase and responds with contextually relevant product suggestions as if coming from a trusted friend. |
|
|
| **Core Features:** |
| - **Intent Recognition:** Harnesses a fine-tuned RoBERTa model to accurately interpret buying signals within textual conversations: the model is OpenSesame and you can find it [here](https://huggingface.co/PiGrieco/OpenSesame/). |
| - **Intelligent Response Generation:** Employs an Agentic Retrieval-Augmented Generation (RAG) mechanism built on Llama3, dynamically setting and manipulating API parameters to fetch the most suitable products: the technology is called "OpenTheVault" and you can find it [here](https://colab.research.google.com/drive/1ydT7cvNn0FhnAj8ZhPojToOBsiC5Djom?usp=sharing). |
| - **Seamless Integration:** Designed to be integrated easily into any existing LLM or AI agent, enhancing their functionality with minimal setup: find the SDK [here](https://github.com/PiGrieco/WordOfPrompt-Integration). |
|
|
| NB, IMPORTANT: OpenTheVault and SDK will be uploaded soon! |
|
|
| ### Vision & Mission |
|
|
| **Vision:** |
| To transform advertising into a helpful, integral part of the conversational experience, mirroring the trust and personal relevance of advice from a friend. |
| "Word Of Prompt" envisions a world where ads are not just tolerated but valued components of our digital interactions. |
|
|
| **Mission:** |
| Our mission is to provide AI developers and marketers with powerful tools that enhance user engagement without disrupting the natural flow of conversation. |
| By doing so, we aim to foster a more sustainable and user-centric advertising landscape that aligns advertisers' goals with consumer satisfaction and help AI Agents and LLMs democratization helping AI developers to earn from their developing efforts. |
|
|
| ### Join Us! |
|
|
| We're looking for AI developers which want to join our team: contact Piermatteo Grieco on [LinkedIn](https://www.linkedin.com/in/piermatteo-grieco/) if you're interested in knowing more about the project. |
|
|
| ## How to Use "Word Of Prompt" |
|
|
| **Integration Steps:** |
| 1. **Incorporate the Library:** |
| Download and integrate the "Word Of Prompt" library into your LLM or AI agent's development environment. |
|
|
| The library is open-source, allowing for custom modifications if needed. |
|
|
| 3. **Configure the API:** |
| Set up the necessary API credentials and configure the settings to connect with product databases like Amazon’s Product API, ensuring that your agent can retrieve product information in real time. |
|
|
| 4. **Activate in Your Application:** |
| Implement "Word Of Prompt" within your conversational models or customer service bots. |
|
|
| Configure the system to detect purchase-related queries and trigger the product recommendation features. |
|
|
| 6. **Customize Responses:** |
| Tailor the response format to fit the tone and style of your AI agent, ensuring that the product recommendations appear as natural and organic parts of the conversation. |
|
|
|
|
|
|
| # OpenSesame |
|
|
| This model is a fine-tuned version of [roberta-base](https://huggingface.co/roberta-base) on the [this](https://www.researchgate.net/publication/372788974_Purchase_Intention_and_Sentiment_Analysis_on_Twitter_Related_to_Social_Commerce) dataset. |
| It achieves the following results on the evaluation set: |
| - Loss: 0.0903 |
| - Accuracy: 0.9825 |
| - F1: 0.9826 |
|
|
| ## Model description |
|
|
| **Overview:** |
| "Open Sesame" is an advanced open-source model designed to detect users' buying intentions from textual data. |
|
|
| **Core Features:** |
| - **Intent Detection:** Utilizes a fine-tuned version of RoBERTa to analyze text and identify potential buying signals, enhancing the accuracy and relevance of generated insights. |
| - **Integration Capability:** Engineered to be seamlessly integrated into any LLM or AI agent, "Open Sesame" offers a plug-and-play solution for developers looking to enhance e-commerce and retail applications. |
| - **Customizable:** While pre-trained to detect purchasing intentions, "Open Sesame" can be further adapted or fine-tuned to meet specific industry needs or to cover additional conversational scenarios. |
|
|
| **Use Cases:** |
| - **E-commerce Platforms:** Improve product recommendation systems by understanding user intent in real-time. |
| - **Customer Service Automation:** Equip chatbots and virtual assistants to better respond to customer inquiries with purchase intent detection. |
| - **Marketing and Sales:** Enable more targeted and personalized marketing campaigns based on detected user interests and needs. |
|
|
| **Getting Started:** |
| To start using "Open Sesame" in your projects, simply load the model from the Hugging Face Model Hub using the following commands: |
|
|
| ```python |
| from transformers import AutoModelForSequenceClassification, AutoTokenizer |
| |
| model_name = "PiGrieco/OpenSesame" |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForSequenceClassification.from_pretrained(model_name) |
| ``` |
|
|
| **Contribute:** |
| "Open Sesame" is open-source and we welcome contributions from the community! Whether it's improving the model, expanding the dataset, or refining the documentation, your input helps make "Open Sesame" better for everyone. |
|
|
| ## Training procedure |
|
|
| ### Training hyperparameters |
|
|
| The following hyperparameters were used during training: |
| - learning_rate: 5e-05 |
| - train_batch_size: 8 |
| - eval_batch_size: 8 |
| - seed: 42 |
| - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
| - lr_scheduler_type: linear |
| - num_epochs: 8 |
|
|
| ### Training results |
|
|
| | Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 | |
| |:-------------:|:-----:|:----:|:---------------:|:--------:|:------:| |
| | 0.4003 | 1.0 | 129 | 0.1545 | 0.9649 | 0.9659 | |
| | 0.4802 | 2.0 | 258 | 0.1453 | 0.9708 | 0.9714 | |
| | 0.1132 | 3.0 | 387 | 0.1655 | 0.9678 | 0.9688 | |
| | 0.0753 | 4.0 | 516 | 0.1038 | 0.9825 | 0.9826 | |
| | 0.1563 | 5.0 | 645 | 0.1078 | 0.9766 | 0.9769 | |
| | 0.0665 | 6.0 | 774 | 0.0914 | 0.9825 | 0.9826 | |
| | 0.0677 | 7.0 | 903 | 0.0909 | 0.9825 | 0.9826 | |
| | 0.0659 | 8.0 | 1032 | 0.0903 | 0.9825 | 0.9826 | |
|
|
|
|
| ### Framework versions |
|
|
| - Transformers 4.41.2 |
| - Pytorch 2.3.0+cu121 |
| - Datasets 2.19.2 |
| - Tokenizers 0.19.1 |
|
|
