针对获取碎片化纺纱工艺信息导致的生产效率低下、资源浪费及决策失误等问题,文章提出了一种基于双向长短期记忆网络的纺纱工艺重用知识图谱构建方法.首先,自上而下定义纺纱工艺相关概念、术语和关系,完成对知识图谱模式层的构建;其次,根据模式层规则来构建数据层,采用双向长短期记忆网络(Bi-directional Long Short-Term Memory,BiLSTM)模型捕捉输入序列的上下文信息作为条件随机场(Conditional Random Fields,CRF)的输入,对标签序列进行建模标注以提取关键知识信息,并通过词向量模型(Word2Vec)来计算纺纱相关的文本数据之间的相似度来实现知识融合,从而提升分词准确率;最后通过Neo4j图数据库存储抽取到的纺纱工艺知识,并可视化展示原料、工艺等复杂关系网络,可帮助纺织企业优化生产、提升决策效率.实例分析结果表明,该知识抽取方法具有较高的召回率(88.7％)、准确率(89.9％)和F1值(89.3％),优于BiLSTM-CRF和LSTM-CRF模型,抽取效果有了显著提升.
Construction of a knowledge graph for reusable spinning processes based on bidirectional long short-term memory networks
Textile spinning process knowledge is a vital technical heritage in the textile industry,and its systematic organization and efficient reuse are crucial for driving industrial technological innovation and preserving cultural diversity.However,spinning process knowledge is often scattered across various documents,practical experiences,and master-apprentice transmissions,exhibiting a high degree of fragmentation and unstructuredness,and posing significant challenges to knowledge integration,retrieval,and application.With the rapid development of the textile industry,the updating and iteration of spinning process knowledge have accelerated,rendering traditional manual recording and inheritance methods inadequate for meeting the demands of modem industries for efficient and precise knowledge management.Particularly when dealing with complex and varied spinning processes,traditional methods often fail to fully capture process details,leading to information loss or distortion during transmission,thereby affecting the integrity and accuracy of spinning process knowledge.To address this challenge,research on the reuse of spinning process knowledge based on knowledge graphs has emerged.As a structured knowledge representation method,knowledge graphs can formally describe entities,attributes,and relationships within spinning process knowledge,forming an interconnected knowledge network.By constructing a spinning process knowledge graph,systematic organization,standardized expression,and intelligent management of spinning process knowledge can be achieved,so as to provide textile practitioners with a comprehensive,accurate,and easily understandable knowledge resource platform.To address issues such as low production efficiency,resource waste,and decision-making errors caused by fragmented spinning process information,a method for constructing a spinning process reuse knowledge graph based on a bidirectional long short-term memory network(BiLSTM)was proposed.Firstly,spinning process-related concepts,terminologies,and relationships were defined top-down to complete the construction of the knowledge graph's schema layer.Secondly,the data layer was constructed according to the schema layer rules.The BiLSTM model was employed to capture contextual information from the input sequence,which serves as input for the Conditional Random Fields(CRF)model to model and annotate tag sequences for extracting key knowledge information.Furthermore,the Word2Vec model was utilized to calculate the similarity between textile-related text data,facilitating knowledge fusion and enhancing tokenization accuracy.Finally,the extracted spinning process knowledge was stored in a Neo4j graph database and visually presented,showcasing complex relationships among raw materials,processes,etc.,aiding textile enterprises in optimizing production and enhancing decision-making efficiency.This paper integrates the BiLSTM and CRF models,leveraging BiLSTM's contextual capture capabilities and CRF's sequence labeling advantages to achieve precise extraction of key knowledge information from spinning process texts.Additionally,the introduction of a word embedding model for knowledge fusion and the utilization of Word2Vec to calculate text similarity promote the effective knowledge integration and improve the accuracy of tokenization and entity recognition.The results of case studies indicate that the proposed knowledge extraction method achieves high recall(88.7％),precision(89.9％),andF1score(89.3％),significantly outperforming the BiLSTM-CRF and LSTM-CRF models.This demonstrates the method's remarkable effectiveness in spinning process knowledge extraction,effectively enhancing production and decision-making efficiency in textile enterprises.In the future,further optimization and refinement of this method will be undertaken,and more advanced natural language processing and knowledge graph construction technologies will be explored to further enhance the accuracy and efficiency of spinning process knowledge extraction.Additionally,strengthened cooperation with textile enterprises will facilitate the application of this method in actual production,so as to generate greater economic and social benefits for enterprises.
knowledge graphspinning process knowledgebidirectional long short-term memory networksknowledge extractionknowledge fusionentity relationship
万方数据
维普
