优化BERTopic主题模型：有效减少-1异常主题文档的方法

import pandas as pd
from sentence_transformers import SentenceTransformer
from bertopic import BERTopic

# 示例文档数据 (实际应用中请替换为您的真实数据)
skills_augmented = [
    "Python programming skills and data analysis",
    "Data analysis with R and statistical modeling",
    "Machine learning algorithms and deep learning",
    "Project management techniques and agile methodologies",
    "Effective communication strategies and public speaking",
    "Advanced Excel skills for data manipulation and reporting",
    "Team collaboration tools and remote work strategies",
    "Statistical modeling with Python and data visualization",
    "Database administration with SQL and NoSQL",
    "Cloud computing fundamentals and AWS services",
    "Web development with JavaScript and React",
    "Agile methodology in software development and Scrum",
    "Financial accounting principles and auditing",
    "Digital marketing strategies and SEO",
    "Network security protocols and cybersecurity",
    "User experience design and UI prototyping",
    "Big data technologies like Hadoop and Spark",
    "Customer relationship management and sales automation",
    "Content writing and SEO optimization",
    "Leadership and negotiation skills",
    "Data visualization with Tableau and Power BI",
    "Cybersecurity awareness and threat intelligence",
    "Business intelligence tools and dashboards",
    "Supply chain optimization and logistics",
    "Artificial intelligence concepts and applications",
    "Mobile app development for iOS and Android",
    "Risk management in finance and investment",
    "Brand building and marketing campaigns",
    "Blockchain technology basics and cryptocurrency",
    "Customer service and support skills",
    "Technical writing and documentation",
    "Human resources management and talent acquisition",
    "Environmental sustainability and green technology",
    "Medical research and clinical trials",
    "Legal compliance and regulatory affairs",
    "Product management and lifecycle",
    "Sales forecasting and market analysis",
    "Quality assurance and testing",
    "Graphic design and multimedia production",
    "Event planning and coordination"
]

# 1. 准备嵌入模型
llm_mod = "all-MiniLM-L6-v2"
model = SentenceTransformer(llm_mod)

# 2. 训练BERTopic模型
# 如果您已经预先计算了embeddings，可以直接传入 embeddings=embeddings
bertopic_model = BERTopic(verbose=True)
topics, probs = bertopic_model.fit_transform(skills_augmented)

print("原始主题分布（前5个主题和-1）：")
# 打印原始主题分布，包括-1主题
original_topic_counts = pd.Series(topics).value_counts().sort_index()
print(original_topic_counts.head(6) if -1 in original_topic_counts.index else original_topic_counts.head(5))

# 3. 减少异常文档
new_topics = bertopic_model.reduce_outliers(skills_augmented, topics)

print("\n减少异常文档后的主题分布（前5个主题和-1）：")
# 打印减少异常文档后的主题分布
new_topic_counts = pd.Series(new_topics).value_counts().sort_index()
print(new_topic_counts.head(6) if -1 in new_topic_counts.index else new_topic_counts.head(5))

# 您现在可以使用 new_topics 进行后续分析，例如更新主题表示
# bertopic_model.update_topics(skills_augmented, new_topics)