Scholarlink

The paper presents a simple method for constructing evolutionary trajectories of topic and keyword representations in vector space created using a static language model. Such evolutionary trajectories visualize how the meanings and associations of keywords and topics change over time, which can help in studying scientific progress or social trends at different periods in human history. To construct trajectories, language models Word2Vec, BERT and other modern models can be used without changing their program code. Evolutionary trajectories allow visual comparison of the quality of different language models. To build trajectories, special evolutionary labels are inserted into the analyzed source text next to words from the topic of interest. The case of evolutionary trajectories of keywords in the field of “machine learning” in the vector space of the Word2Vec model is considered. A semantic map based on PCA projection of the evolutionary trajectories and keyword representations onto a 2D plane is presented.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evolutionary Trajectories of Topic and Keyword Representations in Vector Space Created Using Static Language Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • M. Charnine,
  • N. Somin

摘要

The paper presents a simple method for constructing evolutionary trajectories of topic and keyword representations in vector space created using a static language model. Such evolutionary trajectories visualize how the meanings and associations of keywords and topics change over time, which can help in studying scientific progress or social trends at different periods in human history. To construct trajectories, language models Word2Vec, BERT and other modern models can be used without changing their program code. Evolutionary trajectories allow visual comparison of the quality of different language models. To build trajectories, special evolutionary labels are inserted into the analyzed source text next to words from the topic of interest. The case of evolutionary trajectories of keywords in the field of “machine learning” in the vector space of the Word2Vec model is considered. A semantic map based on PCA projection of the evolutionary trajectories and keyword representations onto a 2D plane is presented.