Model Card for Gemma2-2b-ner

本模型基於 Gemma2:2b 進行微調，目的是讓其依據台灣刑法學中常用的「刑法三階理論」，針對大型語言模型生成的詐欺罪「犯罪事實」段落，依照詐欺罪法條所規定的構成要件進行標註。具備生成詐欺罪「犯罪事實」的模型，可以參考以 BLOOM 560M 為基礎的BLOOM 560M Fraud微調模型，或是以 Gemma2 為基礎的Gemma2:2b Fraud微調模型。

Model Details

Model Description

本模型目前在識別出詐欺罪犯罪事實構成要件要素的平均準確率（percision）及召回率（recall）分別為0.98及0.75。從本模型訓練初期的語料資料錄為 979 筆開始，採用強化學習的流程，將生成的標註資料，採用人工對齊的方式修正後再投入語料庫中進行訓練。最終訓練用的語料計有 2577 筆，經過微調 3 個回合，就完成了本模型。以下是訓練過程各代的準確率及召回率的變化。

版次	資料量	準確率(Precision)	召回率(Recall)
v1	979	0.272727273	0.218623482
v2	1538	0.725888325	0.581300813
v3	1886	0.717277487	0.465986395
v4	2173	0.826086957	0.550724638
v5	2577	0.983606557	0.75

• Developed by: Chun-Hsien Lin
• Funded by [optional]: [More Information Needed]
• Shared by [optional]: [More Information Needed]
• Model type: [More Information Needed]
• Language(s) (NLP): Traditional Chinese
• License: [More Information Needed]
• Finetuned from model [optional]: Gemma2-2b

Model Sources [optional]

• Repository: [More Information Needed]
• Paper [optional]: [More Information Needed]
• Demo [optional]: [More Information Needed]

Uses

目前可以識別出來的標註標籤有以下七種具名實體，無法識別出來的構成要件要素具名實體，則會傳回 None。

  
from colorama import Fore, Back, Style

elements = {'LEO_SOC': ('犯罪主體', 'Subject of Crime'),
            'LEO_VIC': ('客體', 'Victim'),
            'LEO_ACT': ('不法行為', 'Behavior'),
            'LEO_SLE': ('主觀要件', 'Subjective Legal Element of the Offense'),
            'LEO_CAU': ('因果關係', 'Causation'),
            'LEO_ROH': ('危害結果', 'Result of Hazard'),
            'LEO_ATP': ('未遂', 'Attempted')
           }
tag_color = {'LEO_SOC': Fore.BLACK + Back.RED,
             'LEO_VIC': Fore.BLACK + Back.YELLOW,
             'LEO_ACT': Fore.BLACK + Back.GREEN,
             'LEO_SLE': Fore.BLACK + Back.MAGENTA,
             'LEO_CAU': Fore.BLACK + Back.CYAN,
             'LEO_ROH': Fore.BLACK + Back.BLUE,
             'LEO_ATP': Fore.WHITE + Back.BLACK,
            }
  

為了要將本模型標註出來的結果以更明顯的方式識別，可以參考以下的程式碼，將本大型語言模型生成的標註結果以及所標註的標籤，同時送入以下的函數，就可以將結果以 colorama 的方式著色標註。

  
from colorama import Fore, Back, Style
    
def tag_in_color(response_content, tag):
    '''
    說明：
        將標註結果依照標籤進行標色。
    Parameters:
        response_content (str): 已經標註完畢並有標籤的內容。
        tag (str): 標籤名稱，英文，沒有括號。
    Return:
        result (str): 去除標籤並含有 colorama 標色符號的字串。
    '''
    response_head = response_content.split("標註結果:\n")[0]
    response_body = response_content.split("標註結果:\n")[1]
    start_index = 0
    # 使用正規表示式找出所有構成要件要素文字的起始位置
    # 加入 re.escape() 是為了避免處理到有逸脱字元的字串會報錯而中斷程式執行
    findall_open_tags = [m.start() for m in re.finditer(re.escape(f"[{tag}]"), response_body)]
    findall_close_tags = [m.start() for m in re.finditer(re.escape(f"[/{tag}]"), response_body)]
    try:
        parts = [response_body[start_index:findall_open_tags[0]]] # 第一個標籤之前的句子
    except IndexError:
        parts = []
    # 找出每個標籤所在位置，取出標籤文字並加以著色。
    for j, idx in enumerate(findall_open_tags):
        tag_text = response_body[idx + len(tag) + 2:findall_close_tags[j]]
        parts.append(f"{tag_color[tag]}" + tag_text + Style.RESET_ALL) # 標籤內文字著色
        closed_tag = findall_close_tags[j] + len(tag) + 3
        try:
            next_open_tag = findall_open_tags[j+1]
            parts.append(response_body[closed_tag: next_open_tag]) # 結束標籤之後到下一個標籤前的文字
        except IndexError:
            parts.append(response_body[findall_close_tags[-1] + len(tag) + 3 :]) # 加入最後一句
    result = ''
    for _, part in enumerate(parts):
        result = result + part
    if result == '':
        color_result = f"{tag_color[tag]}{tag}" + Fore.RESET + Back.RESET + " " +Fore.YELLOW + Back.RED + "*** 無標註結果 ***" + Fore.RESET + Back.RESET
    else:
        color_result = Fore.RED + Back.YELLOW +  "標註著色結果:\n" + Fore.RESET + Back.RESET + result
    return color_result
  

Direct Use

[More Information Needed]

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

本模型目前僅能標示依據中華民國刑法規定的「詐欺罪」所擬撰（或是語言模型生成）之「犯罪事實」中的構成要件要素，若要具備標註其餘各種不同的犯罪構成要件要素之標註能力，則是後續可以發展以及擴增語料庫的方向。

[More Information Needed]

Bias, Risks, and Limitations

[More Information Needed]

Recommendations

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

How to Get Started with the Model

Use the code below to get started with the model.

[More Information Needed]

Training Details

訓練資料

本模型是以強化學習的方式微調 Gemma2:2b 並經過多回合人工對齊生成資料反覆迭代訓練而成，訓練所需要的資料集是法律要件資料集。使用者可以下載後自己持續迭代後修正及擴充資料集內容。

[More Information Needed]

Training Procedure

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

• Training regime: [More Information Needed]

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

• Hardware Type: [More Information Needed]
• Hours used: [More Information Needed]
• Cloud Provider: [More Information Needed]
• Compute Region: [More Information Needed]
• Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]