Class	Description
Background	The cited paper provides relevant Background information or is part of the body of literature.
Motivation	The citing paper is directly motivated by the cited paper.
Uses	The citing paper uses the methodology or tools created by the cited paper.
Extends	The citing paper extends the methods, tools or data, etc. of the cited paper.
Comparison or Contrast	The citing paper expresses similarities or differences to, or disagrees with, the cited paper.
Future	*The cited paper may be a potential avenue for future work.

Quickstart

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "sknow-lab/Gemma3-12B-CIC-ACLARC"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

system_prompt = """
# CONTEXT #
You are an expert researcher tasked with classifying the intent of a citation in a scientific publication.

########

# OBJECTIVE # 
You will be given a sentence containing a citation, you must output the appropriate class as an answer.

########

# CLASS DEFINITIONS #

The six (6) possible classes are the following: "BACKGROUND", "MOTIVATION", "USES", "EXTENDS", "COMPARES_CONTRASTS", "FUTURE".

The definitions of the classes are:
1 - BACKGROUND: The cited paper provides relevant Background information or is part of the body of literature.
2 - MOTIVATION: The citing paper is directly motivated by the cited paper.
3 - USES: The citing paper uses the methodology or tools created by the cited paper.
4 - EXTENDS: The citing paper extends the methods, tools or data, etc. of the cited paper.
5 - COMPARES_CONTRASTS: The citing paper expresses similarities or differences to, or disagrees with, the cited paper.
6 - FUTURE: The cited paper may be a potential avenue for future work.

########

# RESPONSE RULES #
- Analyze only the citation marked with the @@CITATION@@ tag.
- Assign exactly one class to each citation.
- Respond only with the exact name of one of the following classes: "BACKGROUND", "MOTIVATION", "USES", "EXTENDS", "COMPARES_CONTRASTS", "FUTURE".
- Do not provide any explanation or elaboration.
"""

test_citing_sentence = "However , the method we are currently using in the ATIS domain ( @@CITATION@@ ) represents our most promising approach to this problem."

user_prompt = f"""
{test_citing_sentence}
### Question: Which is the most likely intent for this citation?
a) BACKGROUND
b) MOTIVATION
c) USES
d) EXTENDS
e) COMPARES_CONTRASTS
f) FUTURE
### Answer:
"""

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": user_prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
# Response: USES

Details about the system prompts and query templates can be found in the paper.

There might be a need for a cleanup function to extract the predicted label from the output. You can find ours on GitHub.

Citation

@misc{koloveas2025llmspredictcitationintent,
      title={Can LLMs Predict Citation Intent? An Experimental Analysis of In-context Learning and Fine-tuning on Open LLMs}, 
      author={Paris Koloveas and Serafeim Chatzopoulos and Thanasis Vergoulis and Christos Tryfonopoulos},
      year={2025},
      eprint={2502.14561},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2502.14561}, 
}