OperA
A textual dataset describing user behaviors during online shopping sessions, including observations, actions, persona profiles, and annotated reasons.
692 sessions, 28,904 observation-action pairs, 604 JSON/structured text annotations
CC-BY 4.0
Description
The dataset OperA combines web browsing data collected from real users during their online shopping sessions. It includes observation-action pairs synchronized over time, detailed persona profiles, and human justifications for the choices made. Data is de-identified to ensure confidentiality.
What is this dataset for?
- Evaluate and train language models (LLM) on the simulation of personalized user behaviors.
- Study the decision-making processes in online business.
- Develop intelligent agents capable of reproducing and anticipating human actions in a purchasing context.
Can it be enriched or improved?
Yes, this dataset can be enriched by additional annotations, in particular on the motivations of users or their demographic context. It is also possible to complete it with data from other platforms or integrating multimodal modalities (e.g. images of the products consulted).
🔎 In summary
🧠 Recommended for
- Behavioral AI researchers
- LLM Agent Developers
- UX specialists
🔧 Compatible tools
- Classic NLP frameworks
- JSON annotation tools
- Fine-tuning platforms like Hugging Face
💡 Tip
To make the most of this dataset, combine log analysis with persona profiles to model behavioral variations.
Frequently Asked Questions
What is the size and diversity of users in this dataset?
The dataset includes 692 sessions from 51 unique users, with detailed persona profiles but limited cultural diversity.
Can this dataset be used to train custom chatbots?
Yes, OperA is specifically designed for fine-tuning and evaluating LLM agents that simulate personalized user behavior.
What are the guarantees in terms of data confidentiality?
The data has been rigorously filtered and de-identified to ensure the protection of users' privacy.
