结合词向量和自注意力机制的设计素材推荐系统  

Design Resources Recommendation Based on Word Vectors and Self-Attention Mechanisms

作　　者：孙清华 邓程 顾振宇 Sun Qinghua;Deng Cheng;Gu Zhenyu(School of Design,Shanghai Jiao Tong University,Shanghai 2002402;Department of Computer Science and Engineering,Shanghai Jiao Tong University,Shanghai 200240)

机构地区：[1]上海交通大学设计学院,上海200240 [2]上海交通大学计算机科学与工程系,上海200240

出　　处：《计算机辅助设计与图形学学报》2024年第1期63-72,共10页Journal of Computer-Aided Design & Computer Graphics

摘　　要：当设计师使用文本在设计库中检索素材时,基于词匹配的搜索无法有效地识别文本中的设计意图并推荐合适的素材.对此,提出一个结合词向量和自注意力机制的设计素材推荐系统.首先从预训练词向量模型中获取文本的词向量表征;然后利用自注意力机制学习文本中不同词的权重以及文本与设计语义间的映射关系;最后基于模型输出与向量相似,检索并重排序设计素材形成推荐.所提系统还能检索带有不同语义标签的素材数据,通过计算其语义与设计语义的词相似作为权重,转换其数据维度与模型输出维度一致.基于1300个电商设计案例和200种字体进行实验的结果表明,用户评估推荐结果的有效率分别为55%和57.3%,处于同类工作中上水平;验证了该系统的可行性及有效性.When designers utilize text prompts to retrieve resources in a design repository,the keyword-matching approach usually fails to capture the real design semantics in the text and recommends irrelevant design re-sources.To address this problem,this paper proposes a design material recommendation system exploiting word vectors and self-attention mechanisms.First,the model segments the text into tokens and takes the token word vectors from the pre-trained word vector model as input.Then,we exploit the self-attention mechanism to learn the weight of tokens and the mapping function between tokens and design semantics.Finally,we retrieve and sort the design resources by computing the vector similarity between the model output and annotated resource data.The system can retrieve other design resources with different semantic tags as well by unifying the similarity be-tween the different semantics into weight vectors and converting the data dimension to be consistent with the model output.With 1300 e-commerce design cases and 200 fonts,the efficiency of user-assessed recommendation results is 55%and 57.3%,respectively,which are above average for similar works and validate the feasibility and effectiveness of the proposed system.

关 键 词：设计素材推荐 语义提取 自注意力机制 词向量 设计工具 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]