PaperMatch is an AI-driven solution designed to streamline the process of discovering academic research and interpreting media narratives around key topics.
With the exponential rise in scholarly publications and fragmented media coverage, researchers face information overload and lack tools that connect academic rigor with real-world discourse. PaperMatch bridges that gap by offering:
- 🧠 Summarization of academic papers using LLMs (Llama3 via Ollama)
- 🌍 News analysis and clustering by topic
- 🤖 A single interface (via Telegram) to query and receive structured reports
Backed by insights from academic research on information behavior and media framing (e.g., Bawden & Robinson, 2009; Tenopir et al., 2012; Entman, 1993), PaperMatch offers both relevance and context in one unified platform.
| Name | Student ID | Contributions |
|---|---|---|
| Alvin Wong Ann Ying | A0266486M | Project lead, academic scraping, graph/embedding pipeline, LLM summarization, Telegram API, code integration, evaluation workflow, all README documentation, system demo, report and pptx proposal, LaTex integration |
| Bertrand Tan Yu-Jin | A0292314J | News scraping, clustering for media content, evaluation workflow, README documentation, pptx proposal |
🛠️ Supported OS: Ubuntu 24.04
Install Ollama and pull the Llama3.3 model:
git clone https://github.com/MRAWAY77/PaperMatch.git
cd PaperMatch/query/
ollama pull llama3.3 # depending on what models you want to useYou can also choose from other available models you’ve already pulled:
llama3:latest(4.7 GB)qwen3:latest(5.2 GB)llama3.3:latest(42 GB)qwen3:30b-a3b(18 GB)mistral-small3.1:latest(15 GB)gemma3:latest(3.3 GB)
More models are available at https://ollama.com/search.
Set up the environment:
conda create -n papermatch python=3.10.16 -y
conda activate papermatch
pip install -r requirements.txtNext, download and extract the required folders into the project root directory:
📁 Download from Google Drive 🔗 https://drive.google.com/drive/folders/1dBYjVw8gvEo9h2YO9v3y-TY6y9i8uvvH?usp=drive_link
Download and unzip:
- datasets.zip
- cluster_embeddings.zip
- EVALUATION/ (Only needed for reviewing the evaluation artifacts in detail)
Make sure the extracted folders are placed according to the directory structure shown in Section 3.
The directory tree should look like this:
papermatch/
├── academic_papers.csv
├── assets/
│ ├── architecture.png
│ ├── MDDI_blend.jpg
│ ├── MDDI.jpg
│ ├── sample_report.pdf
│ └── telegram.png
├── datasets/
│ ├── Academic_Clusters/
│ ├── News_Clusters/
│ └── academic_metadata.csv
├── graph_network/
│ ├── cluster_embeddings/
│ ├── graph_network.py
│ ├── cluster_embeddings.py
│ ├── output_graphs/
│ └── sim_score_graph_network.txt
├── project_report/
│ ├── PLP_Project_Proposal_Presentation.pdf
│ ├── PaperMatch.pdf
├── eval/
│ ├── config.py
│ ├── eval_summaries.py
│ ├── query_eval.py
├── query/
│ ├── config.py
│ ├── llm.py
│ ├── query.py
│ ├── main.py
│ ├── eval_logs/
│ └── ...
├── scrapers/
├── news_webscraper/
│ ├── Arachne.py
│ ├── assets/
│ │ ├── arachne_interface_01.jpg
│ │ ├── arachne_interface_02.jpg
│ │ └── arachne_interface_03.jpg
│ ├── cluster_organizer.py
│ ├── MetaCluster-I.py
│ ├── README.md
│ ├── Tapestry-Ik.py
│ └── Weaver.py
├── paperwithcode_chrome.py
└── paperwithcode_firefox.py
├── utils/
│ ├── categorise_papers.py
│ ├── helper.ipynb
│ ├── sorting.py
│ ├── topic_modeling.py
│ └── topic_model_results/
├── LICENSE
├── README.md
├── requirements.txt
└── setup.sh
To connect PaperMatch to a Telegram group:
- Visit https://my.telegram.org and create an app.
- Retrieve your api_id and api_hash.
- Create a group chat and copy the group link.
Update query/config.py:
api_id = <insert your api_id>
api_hash = '<insert your api_hash>'
phone_number = '<insert your phone number>'
TARGET_CHANNEL = '<insert telegram group link>'Start the Telegram-integrated query interface:
python query/main.pyOnce the script is running and connected to your Telegram group, interact with the bot using the following command:
📩 Send this in your Telegram group:
/ask [your query here]
This command triggers the AI pipeline, which retrieves relevant academic papers, summarizes them, matches media coverage, and generates a contextualized PDF report directly in your Telegram chat.
Example:
/ask What are the latest Deepfake trends?
This will generate an AI-enhanced academic-media insight report based on your query.
Example report generated by the system:
📄 Sample Report: View PDF
See the full project report in the project_report folder.
For any issues or inquiries, please contact the support personnel listed below:
| Name | Role | Contact Email |
|---|---|---|
| Alvin Wong | Team Lead/AI & SW Eng | alvinwongannying@gmail.com |
| Bertrand Tan | Member/AI Eng | bertrand.tanyj@gmail.com |