基于知识图谱的服装问答系统  

Clothing Question Answering System Based on Knowledge Graph

作　　者：赖佳扬 张晓滨[1] 马瑛超 LAI Jia-yang;ZHANG Xiao-bin;MA Ying-chao(School of Computer Science,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048

出　　处：《计算机技术与发展》2023年第2期99-104,共6页Computer Technology and Development

基　　金：西安工程大学研究生创新基金(chx2021028);国家级大学生创新创业项目(202110709013)。

摘　　要：随着电商产业的不断发展,消费者希望在网上购物时能够和商家进行更好的沟通,而人工客服需要浪费大量的人力物力。为合理、有效地利用服装资源,文章通过构建服装知识图谱,并基于知识图谱实现服装知识自动问答。该问答系统利用目标实体的多跳关系与问句进行匹配从而完成答案生成,知识问答模型采用BERT作为编码层,使用LSTM网络对知识库的多跳关系进行学习,利用自注意力机制对知识库特征和问句特征进行计算,最终通过二分类的输出将问句和知识库的匹配结果作为评分,并根据评分给出答案。文章对问答系统的准确性和运行效率在自建的服装数据上进了实验。实验表明,该问答方法相对于传统的答案匹配方法效果更好,同时在运行效率的实验上验证了方法在实际中的可行性。基于知识图谱进行问答系统的搭建可以有效地解答消费者在服装知识和服装搭配推荐上的问题,在提高用户体验的同时节约了人力资源。With the continuous development of the e-commerce industry,consumers hope to have better communication with merchants when shopping online,and manual customer service requires a lot of waste of manpower and material resources.In order to make reasonable and effective use of clothing resources,we construct clothing knowledge graph and realize automatic question answering of clothing knowledge based on the knowledge graph.The question answering system uses the multi-hop relationship of the target entity to match the question sentence to complete the answer generation.The knowledge question answering model uses BERT as the encoding layer,uses the LSTM network to learn the multi-hop relationship of the knowledge base,and uses the self-attention mechanism to calculate the feature of knowledge base and the feature of the question base.Finally,the matching result of the question base and the knowledge base is scored through the binary output,and the answer is given according to the score.The article conducts experiments on the accuracy and operation efficiency of the question answering system on the self-built clothing data.Experiments show that the proposed question answering method has better effect than the traditional answer matching method,and its feasibility in practice is verified in the experiment of operation efficiency.The construction of question answering system based on knowledge graph can effectively answer consumers’questions about clothing knowledge and clothing matching recommendation,which can improve user experience and save human resources.

关 键 词：服装 知识图谱 知识问答 知识抽取 Bert预训练模型 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]