Topic generation method based on pre-training model

• Main Authors: WANG LEI, WANG YUQI, CAI CHANGYAN, GUO HONGFEI, WANG JUNYAN, XU CAI, CAO JIA, LUO YIN, WANG KEQI
• Format: Patent
• Language: Chinese; English
• Published: 30.06.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a topic generation method based on a pre-training model, and the method comprises the steps: obtaining a feature vector and a keyword of each text in to-be-clustered texts, each text comprising h keywords; clustering the to-be-clustered text by using a set clustering algorithm to obtain a plurality of topics; performing cleaning and merging processing on the plurality of topics to obtain n processed topics; for any topic in the n topics, generating a corresponding topic description based on a pre-training generation model; and outputting the topic descriptions of the n topics and the corresponding texts. According to the method, the topic description is generated by adopting the pre-training generation model, so that the obtained topic description is smooth and high in readability, and the clustering result is more accurate due to the fact that the topics are cleaned and combined. 本发明提供了一种基于预训练模型的话题生成方法，包括：获取待聚类文本中的每个文本的特征向量和关键词，每个文本包括h个关键词；利用设定聚类算法对待聚类文本进行聚类，得到多个话题；对多个话题进行清洗和合并处理，得到处理后