| 摘 要: 针对目前绘画领域缺乏标准的价值评估指标体系,提出了基于BERT-LDA和K-means聚类的绘画作品价值要素挖掘方法。运用超平面法对绘画文献进行了停用词筛选,基于BERT-LDA模型构建了包含文本语义信息的融合特征向量,运用K-means算法对融合特征向量进行降维可视化,随之构建了绘画作品价值评估指标体系。结果表明,基于BERT-LDA模型和K-means算法识别的主题及主题词相比传统LDA模型的查准率、查全率和F值分别提升了28.5%、10%和21.5%。通过随机森林等算法对指标体系进行验证,验证了构建的绘画作品价值评估指标体系的科学性。 |
| 关键词: BERT-LDA;融合特征向量;K-means聚类;绘画;指标体系 |
|
中图分类号: TP18
文献标识码: A
|
|
| Construction of a Painting Works Valuation Index System Based on BERT-LDA and K-means Clustering |
|
LI Tianyi, LIU Qinming
|
(Business School, University of Shanghai for Science and Technology, Shanghai 200093, China)
18437773865@163.com; lqm0531@163.com
|
| Abstract: This paper proposes a method for mining the value elements of painting works based on BERT-LDA ((Bidirectional Encoder Representations from Transformers-Latent Dirichlet Allocation) and K-means clustering in view of the lack of standard value evaluation index system in the current painting field. The hyperplane method is applied to filter out stop words in painting literature, fusion feature vectors containing textual semantic information are constructed by using the BERT-LDA model, and the K-means algorithm is employed for dimensionality reduction and visualization of the fusion feature vectors. Subsequently, a valuation index system for painting works is established. The results indicate that, compared to the traditional LDA model, the precision, recall, and F-value of the themes and keywords identified by the BERT-LDA model and K-means algorithm increase by 28. 5% , 10% , and 21. 5% , respectively. The index system is validated by algorithm such as Random forest, verifying the scientific nature of the constructed valuation index system for painting works. |
| Keywords: BERT-LDA; fusion feature vectors; K-means clustering; painting; index system |
