User Review Writing via Interview with Dialogue Systems

The repository contains the code that accompanies our SIGDIAL 2024 paper.

Note

A Japanese version of this README is available here.

📝 Abstract

User reviews on e-commerce and review sites are crucial for making purchase decisions, although creating detailed reviews is time-consuming and labor-intensive. In this study, we propose a novel use of dialogue systems to facilitate user review creation by generating reviews from information gathered during interview dialogues with users. To validate our approach, we implemented our system using GPT-4 and conducted comparative experiments from the perspectives of system users and review readers. The results indicate that participants who used our system rated their interactions positively. Additionally, reviews generated by our system required less editing to achieve user satisfaction compared to those by the baseline. We also evaluated the reviews from the readers’ perspective and found that our system-generated reviews are more helpful than those written by humans. Despite challenges with the fluency of the generated reviews, our method offers a promising new approach to review writing.

🖥️ System Overview

Our system is designed to facilitate user review writing via an interview dialogue. It first interviews the user to gather detailed product information, then generates a review based on the dialogue history. The application is developed using Flask and utilizes OpenAI models at each stage. Customization options allow users to adjust the model, language, and other parameters via settings/config.py.

🧱 Project Structure

├── app.py               # Flask application entry point
├── modules/             # Core modules for generation and utilities
├── resources/
│   ├── questionnaire/   # Predefined questions for baseline system
│   └── guidance/        # Guidance text for users
├── settings/            # Configuration file (`config.py`)
├── templates/           # HTML templates
├── static/              # Static assets (CSS, JavaScript, images)
├── data/                # Logs and completed session outputs
├── requirements.txt     # Python dependencies
├── run.sh               # Shell script to start the application
├── README.md            # English README file
└── README_ja.md         # Japanese README file

🚀 Quick Start

Follow these steps to get up and running quickly:

1. Clone the repository

   git clone https://github.com/UEC-InabaLab/InterviewToReview.git
   cd InterviewToReview

2. Install dependencies

   pip install -r requirements.txt

3. Configure your API key

   Create a .env file in the project root and add your OpenAI API key:

   API_KEY="{YOUR_OPENAI_API_KEY}"

   You can create .env file by copying the provided .env.example file.

4. Start the application

   bash run.sh

This will launch a local Flask server (default at http://127.0.0.1:8000).

⚙️ Configuration

System settings are defined in settings/config.py. You can customize:

• MODEL: The OpenAI model to use (e.g., gpt-4).
• BOT_TYPE: Dialogue mode (gpt or rule-based).
  • If rule-based, the system will use fixed questions from ./resources/questionnaire/. For more details, please see Section 4.1.2 of our paper.
  • If BOT_TYPE is set to gpt, the system uses the OpenAI model specified in MODEL to generate questions for each turn of the dialogue.
• LANG : Language code (ja for Japanese, en for English).
• MIN_QUESTIONS, MAX_QUESTIONS: Range for the number of interview questions. (i.e., the number of dialogue turns)
• TEMPERATURE_INTERVIEW, TEMPERATURE_REVIEW, TEMPERATURE_RATING: Temperature settings for the LLM responses during the interview, review text generation, and rating prediction phases.
• SAVE_COMPLETED: Enable or disable saving of completed sessions.

🧪 Tested Environment

• Confirmed to work on Python 3.10.12.

📚 Citation

@inproceedings{tanaka2024user,
  title     = {User Review Writing via Interview with Dialogue Systems},
  author    = {Yoshiki Tanaka and Michimasa Inaba},
  booktitle = {Proceedings of the 25th Annual Meeting of the Special Interest Group on Discourse and Dialogue},
  year      = {2024},
  url       = {https://aclanthology.org/2024.sigdial-1.37/}
}