Bertopic ArXiv

🚀 BERTopic_ArXiv

BERTopic_ArXiv是一个基于BERTopic的模型。BERTopic是一个灵活且模块化的主题建模框架，可从大型数据集中生成易于解释的主题。此预训练模型展示了BERTopic中可使用的几种表示模型，它在约30000篇ArXiv摘要上进行训练，采用了多种主题表示方法。

🚀 快速开始

本模型的使用步骤如下：

1. 安装BERTopic：

pip install -U bertopic
pip install -U safetensors

2. 使用模型：

from bertopic import BERTopic
topic_model = BERTopic.load("MaartenGr/BERTopic_ArXiv")

topic_model.get_topic_info()

3. 查看不同的主题表示：

>>> topic_model.get_topic(0, full=True)
{'Main': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['oriented', 0.009217253131615378],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181]],
 'POS': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181],
  ['user', 0.008753551043296965]],
 'KeyBERTInspired': [['task oriented dialogue', 0.6559894680976868],
  ['dialogue systems', 0.6249060034751892],
  ['oriented dialogue', 0.5788208246231079],
  ['dialog systems', 0.530449628829956],
  ['dialogue state', 0.5167528390884399],
  ['response generation', 0.5143576860427856],
  ['spoken language understanding', 0.46739083528518677],
  ['oriented dialog', 0.4600704610347748],
  ['dialog', 0.4534587264060974],
  ['dialogues', 0.44082391262054443]],
 'MMR': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['oriented', 0.009217253131615378],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181]],
 'KeyBERT + MMR': [['task oriented dialogue', 0.6559894680976868],
  ['dialogue systems', 0.6249060034751892],
  ['oriented dialogue', 0.5788208246231079],
  ['dialog systems', 0.530449628829956],
  ['dialogue state', 0.5167528390884399],
  ['response generation', 0.5143576860427856],
  ['spoken language understanding', 0.46739083528518677],
  ['oriented dialog', 0.4600704610347748],
  ['dialog', 0.4534587264060974],
  ['dialogues', 0.44082391262054443]],
 'OpenAI_Label': [['Challenges and Approaches in Developing Task-oriented Dialogue Systems',
   1]],
 'OpenAI_Summary': [['Task-oriented dialogue systems and their components, such as dialogue policy, natural language understanding, dialogue state tracking, response generation, and end-to-end training using neural networks. These components are crucial in assisting users to complete various activities such as booking tickets and restaurant reservations through spoken language understanding dialogue. The challenge lies in tracking dialogue states of multiple domains and obtaining annotations for training. Effective SLU is achieved by utilizing context from the prior dialogue history.',
   1]]}

✨ 主要特性

• 多种主题表示方法：使用了POS、KeyBERTInspired、MaximalMarginalRelevance、KeyBERT + MaximalMarginalRelevance、ChatGPT labels、ChatGPT summaries等多种主题表示方法。
• 可视化示例：提供了默认c-TF-IDF表示和ChatGPT生成标签的可视化示例。
• 详细的主题信息：可以查看每个主题的关键词、频率、标签等信息。

📦 安装指南

要使用此模型，请安装BERTopic：

pip install -U bertopic
pip install -U safetensors

💻 使用示例

基础用法

from bertopic import BERTopic
topic_model = BERTopic.load("MaartenGr/BERTopic_ArXiv")

topic_model.get_topic_info()

高级用法

查看所有不同的主题表示（关键词、标签、摘要等）：

>>> topic_model.get_topic(0, full=True)
{'Main': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['oriented', 0.009217253131615378],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181]],
 'POS': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181],
  ['user', 0.008753551043296965]],
 'KeyBERTInspired': [['task oriented dialogue', 0.6559894680976868],
  ['dialogue systems', 0.6249060034751892],
  ['oriented dialogue', 0.5788208246231079],
  ['dialog systems', 0.530449628829956],
  ['dialogue state', 0.5167528390884399],
  ['response generation', 0.5143576860427856],
  ['spoken language understanding', 0.46739083528518677],
  ['oriented dialog', 0.4600704610347748],
  ['dialog', 0.4534587264060974],
  ['dialogues', 0.44082391262054443]],
 'MMR': [['dialogue', 0.02704485163341523],
  ['dialog', 0.01677038224466311],
  ['response', 0.011692640237477233],
  ['responses', 0.01002788412923778],
  ['intent', 0.00990720856306287],
  ['oriented', 0.009217253131615378],
  ['slot', 0.009177118721490055],
  ['conversational', 0.009129311385144046],
  ['systems', 0.009101146153425574],
  ['conversation', 0.008845392252307181]],
 'KeyBERT + MMR': [['task oriented dialogue', 0.6559894680976868],
  ['dialogue systems', 0.6249060034751892],
  ['oriented dialogue', 0.5788208246231079],
  ['dialog systems', 0.530449628829956],
  ['dialogue state', 0.5167528390884399],
  ['response generation', 0.5143576860427856],
  ['spoken language understanding', 0.46739083528518677],
  ['oriented dialog', 0.4600704610347748],
  ['dialog', 0.4534587264060974],
  ['dialogues', 0.44082391262054443]],
 'OpenAI_Label': [['Challenges and Approaches in Developing Task-oriented Dialogue Systems',
   1]],
 'OpenAI_Summary': [['Task-oriented dialogue systems and their components, such as dialogue policy, natural language understanding, dialogue state tracking, response generation, and end-to-end training using neural networks. These components are crucial in assisting users to complete various activities such as booking tickets and restaurant reservations through spoken language understanding dialogue. The challenge lies in tracking dialogue states of multiple domains and obtaining annotations for training. Effective SLU is achieved by utilizing context from the prior dialogue history.',
   1]]}

📚 详细文档

主题概述

• 主题数量：107
• 训练文档数量：33189

训练过程

模型的训练过程如下：

from cuml.manifold import UMAP
from cuml.cluster import HDBSCAN
from bertopic import BERTopic
from sklearn.feature_extraction.text import CountVectorizer
from bertopic.representation import PartOfSpeech, KeyBERTInspired, MaximalMarginalRelevance, OpenAI

# 准备子模型
embedding_model = SentenceTransformer('sentence-transformers/all-mpnet-base-v2')
umap_model = UMAP(n_components=5, n_neighbors=50, random_state=42, metric="cosine", verbose=True)
hdbscan_model = HDBSCAN(min_samples=20, gen_min_span_tree=True, prediction_data=False, min_cluster_size=20, verbose=True)
vectorizer_model = CountVectorizer(stop_words="english", ngram_range=(1, 3), min_df=5)

# 使用ChatGPT进行摘要
summarization_prompt = """
I have a topic that is described by the following keywords: [KEYWORDS]
In this topic, the following documents are a small but representative subset of all documents in the topic:
[DOCUMENTS]

Based on the information above, please give a description of this topic in the following format:
topic: <description>
"""
summarization_model = OpenAI(model="gpt-3.5-turbo", chat=True, prompt=summarization_prompt, nr_docs=5, exponential_backoff=True, diversity=0.1)

# 表示模型
representation_models = {
    "POS": PartOfSpeech("en_core_web_lg"),
    "KeyBERTInspired": KeyBERTInspired(),
    "MMR": MaximalMarginalRelevance(diversity=0.3),
    "KeyBERT + MMR": [KeyBERTInspired(), MaximalMarginalRelevance(diversity=0.3)],
    "OpenAI_Label": OpenAI(model="gpt-3.5-turbo", exponential_backoff=True, chat=True, diversity=0.1),
    "OpenAI_Summary": [KeyBERTInspired(), summarization_model],
}

# 拟合BERTopic
topic_model= BERTopic(
        embedding_model=embedding_model,
        umap_model=umap_model,
        hdbscan_model=hdbscan_model,
        vectorizer_model=vectorizer_model,
        representation_model=representation_models,
        verbose=True
).fit(docs)

训练超参数

• calculate_probabilities：False
• language：None
• low_memory：False
• min_topic_size：10
• n_gram_range：(1, 1)
• nr_topics：None
• seed_topic_list：None
• top_n_words：10
• verbose：True

框架版本