demo
This commit is contained in:
parent
cb6fa04ca7
commit
9e230bf899
|
@ -0,0 +1,52 @@
|
|||
from sentence_transformers import SentenceTransformer
|
||||
from sklearn.metrics.pairwise import cosine_similarity
|
||||
import numpy as np
|
||||
|
||||
class TextSimilarityScorer:
|
||||
def __init__(self, model_name='bert-base-chinese'):
|
||||
# 初始化 BERT 模型,这里使用中文预训练模型
|
||||
self.model = SentenceTransformer(model_name)
|
||||
|
||||
def calculate_similarity(self, standard_answer, student_answer):
|
||||
# 获取文本嵌入
|
||||
standard_embedding = self.model.encode([standard_answer])
|
||||
student_embedding = self.model.encode([student_answer])
|
||||
|
||||
# 计算余弦相似度
|
||||
similarity = cosine_similarity(standard_embedding, student_embedding)[0][0]
|
||||
return similarity
|
||||
|
||||
def score(self, standard_answer, student_answer, max_score=100):
|
||||
# 计算相似度
|
||||
similarity = self.calculate_similarity(standard_answer, student_answer)
|
||||
|
||||
# 将相似度转换为分数
|
||||
score = similarity * max_score
|
||||
|
||||
# 四舍五入到整数
|
||||
return round(score)
|
||||
|
||||
# 使用示例
|
||||
def main():
|
||||
# 初始化评分器
|
||||
scorer = TextSimilarityScorer()
|
||||
|
||||
# 示例标准答案和学生答案
|
||||
standard_answer = "机器学习是人工智能的一个子领域,它使用统计学方法让计算机系统能够从数据中学习和改进。"
|
||||
student_answers = [
|
||||
"机器学习是AI的分支,通过统计方法让计算机从数据中学习。", # 相似但较简短
|
||||
"哈哈哈哈哈哈。", # 比较相似
|
||||
"人工智能是计算机科学的重要领域。" # 不太相关
|
||||
]
|
||||
|
||||
# 对每个学生答案进行评分
|
||||
for i, student_answer in enumerate(student_answers, 1):
|
||||
similarity = scorer.calculate_similarity(standard_answer, student_answer)
|
||||
score = scorer.score(standard_answer, student_answer)
|
||||
print(f"\n学生答案 {i}:")
|
||||
print(f"答案: {student_answer}")
|
||||
print(f"相似度: {similarity:.2f}")
|
||||
print(f"得分: {score}")
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
Loading…
Reference in New Issue