p988744/eland-stance-zh

Eland Stance - Chinese Stance Detection LoRA Adapter

LoRA adapter for Chinese stance detection with sarcasm recognition, fine-tuned on Qwen3-4B.

Task Overview

Stance Detection (觀點支持度分析): Determines whether a response supports, opposes, or is neutral toward the main text's viewpoint.

Label	Description
支持 (Support)	Response genuinely agrees with main text
反對 (Oppose)	Response genuinely disagrees with main text
中立 (Neutral)	Unclear or balanced position

Performance

Overall Benchmark (368 samples)

Metric	Score
Accuracy	82.34%
Macro-F1	81.55%

By Category

Category	Accuracy
Standard Samples	85.5%
Sarcasm Detection	90.0%

Quick Start

Installation

pip install transformers peft torch

Load Model

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen3-4B",
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-4B")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "p988744/eland-stance-zh")

Inference

# System prompt with few-shot examples for best sarcasm detection
system_prompt = """你是一位專業的觀點支持度分析助手，擅長識別諷刺和反話。

## 任務
分析「回應」對「主文觀點」的立場，判斷回應是支持、反對、還是中立於主文觀點。

## 標籤定義
- 支持：回應真正認同主文觀點
- 反對：回應真正反對主文觀點
- 中立：態度不明確

## 範例

### 範例 1（諷刺 → 反對）
主文：這家餐廳的服務非常好
回應：是啊，特別是等了一小時才有人來點餐的時候
分析：回應使用諷刺，字面附和但實際批評服務差
答案：反對

### 範例 2（真正支持）
主文：這個政策有助於經濟發展
回應：確實，從最近的數據來看效果很明顯
分析：回應提供正面證據，真心認同
答案：支持

## 判斷要點
1. 識別諷刺：「字面肯定 + 負面細節」= 諷刺 = 實際反對
2. 諷刺式附和（是啊、沒錯、對啊）+ 批評性內容 = 反對
3. 按真實意圖判斷，不要被字面意思誤導"""

def analyze_stance(main_text: str, response: str) -> str:
    messages = [
        {"role": "system", "content": system_prompt},
        {"role": "user", "content": f"主文：{main_text}\n回應：{response}"}
    ]

    text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
        enable_thinking=False  # Important for Qwen3
    )

    inputs = tokenizer(text, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=20,
        temperature=0,
        do_sample=False
    )

    result = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
    return result.strip()

# Test examples
print(analyze_stance(
    "這家餐廳的服務非常好",
    "是啊，特別是等了一小時才有人來點餐的時候"
))  # Expected: 反對 (Sarcasm detected)

print(analyze_stance(
    "這個政策有助於經濟發展",
    "確實，從最近的數據來看效果很明顯"
))  # Expected: 支持

Training Configuration

Parameter	Value
Base Model	Qwen/Qwen3-4B
Method	SFT (Supervised Fine-Tuning)
LoRA Rank	32
LoRA Alpha	64
Target Modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Learning Rate	1e-5
Epochs	8
Batch Size	4
Trainable Parameters	66M (1.62%)

Dataset

Trained on p988744/eland-stance-zh-data (520 samples):

Domain	Train	Test
Financial (金融)	100	30
Product (產品)	100	30
Brand (品牌)	100	30
Social (社群)	100	30
Total	400	120

Challenge types:

• Implicit stance (隱含立場): Position not directly stated
• Sarcasm detection (諷刺反諷): Surface meaning opposite to true intent
• Mixed stance (混合立場): Contains both supporting and opposing elements
• Emotional separation (情感與立場分離): Emotion doesn't match stance
• Complex reasoning (複雜推理): Requires multi-step inference

Model Variants

Version	Repository	Use Case
LoRA Adapter	p988744/eland-stance-zh	HuggingFace + PEFT (this repo)
GGUF	p988744/eland-stance-zh-gguf	Ollama / llama.cpp
Full Merged	p988744/eland-stance-zh-vllm	vLLM

Input Format

主文：[main text with viewpoint]
回應：[response to analyze]

Example Inputs & Expected Outputs

Main Text	Response	Expected	Type
這家餐廳的服務非常好	是啊，特別是等了一小時才有人來點餐的時候	反對	Sarcasm
這款手機的電池續航力很強	沒錯，強到每天要充三次電	反對	Sarcasm
這個政策有助於經濟發展	確實，從最近的數據來看效果很明顯	支持	Direct
遠端工作會降低生產力	我不同意，研究顯示遠端工作反而提高效率	反對	Direct
AI 將取代大部分工作	這需要看具體的產業和職位類型	中立	Neutral

Difference from Sentiment Analysis

Task	Sentiment Analysis	Stance Detection
Question	What is the emotional tone?	Does response agree with main text?
Labels	正面/負面/中立	支持/反對/中立
Input	Single text	Main text + Response pair
Focus	Emotion polarity	Agreement/disagreement

Important: A negative sentiment doesn't always mean opposition. Sarcastic agreement ("是啊，特別是...") expresses opposition despite surface agreement.

License

Apache 2.0