---

---

LLMRS: Unlocking Potentials of LLM-Based Recommender Systems for Software Purchase [WITS 2023]

Angela John*, Theophilus Aidoo*, Hamayoon Behmanush*, Irem B. Gunduz**, Hewan Shrestha**, Maxx Richard Rahman and Wolfgang Maaß

*Equally contributing first authors
**Equally contributing second authors

---

Table of Contents

1. Project Structure
2. Pipeline
3. Installation
4. Running the Code
5. Citation
6. License

---

Project Structure

├── data
├── figures
├── output
├── src
|   ├── notebooks
|   ├── static
|   ├── web_interface
|   ├── __init__.py
|   ├── data_preprocessing.py
|   ├── logger.py
|   ├── recommendation_api.py
|   ├── software_data_processor.py
|   └── software_recommender.py
├── .gitignore 
├── LICENSE
├── README.md
└── requirements.txt

---

Pipeline

The pipeline contains 3 steps as follows:

1. Data preprocessing
2. Process Software Data
3. Recommender

Downloading data

The dataset is in .json format here. After downloading Software reviews and metadata, we ran this script to get data ready for preprocessing.

Generating additional features

1. Licensing Fee is set to 80% of the minimum price in the software category. Licensing fees could be similar in a particular software category and country.

2. Implementation Cost is set to 50% of the cost of the software.

3. Maintenance cost is assumed to be a monthly service so it was set to 1% of the price of the product.

Installation

Setup environment

pip install --upgrade pip

python -m venv .llmrs

source .llmrs/bin/activate

conda deactivate

pip install -r requirements.txt

Note: All monetary values are in USD($) * 100

Running the Code

To run recommendation:

1. Run python src/recommendation_api.py

2. Visit 127.0.0.1:500

   a. Enter Software description with price, license, maintenace and implementation costs in the respective boxes.

   b. When you click Get Recommendation, this would load pre-processed data/softwares_with_scores.csv and compute similarity with input software specification from user input.

   c. Output is then ranked with our ranking algorithm and parsed to the web interface

Sample Input:
"Software for managing employee files"
price = 0,10
license cost = 0, 10
maintenance cost = 0, 10
implementation cost= 0, 10

Citation

If you're using LLMRS in your research or applications, please cite using this BibTeX:

@misc{john2024llmrsunlockingpotentialsllmbased,
      title={LLMRS: Unlocking Potentials of LLM-Based Recommender Systems for Software Purchase}, 
      author={Angela John and Theophilus Aidoo and Hamayoon Behmanush and Irem B. Gunduz and Hewan Shrestha and Maxx Richard Rahman and Wolfgang Maaß},
      year={2024},
      eprint={2401.06676},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2401.06676}, 
}

License

This work is licensed under a GNU General Public License v3.