Freight Question Answer System Based On Embedded Knowledge Graph

A question answering system for transportation can provide value-added suggestions,such as the location of transportation facilities,route recommendation,and freight transportation decision-making.Due to the large scale of freight data and various types,it is difficult for people to directly observe the correlation between the data from the massive traffic data and form valuable information only by retrieving in traditional databases.Thus,this paper proposes a question answering framework,called as KEFQA,based on the knowledge graphs for freight(Knowledge Graph of Freight,KG-Fre).The data information is stored into KG-Fre,and then the problem of QA is transformed into finding a question that matches answers in the embedded space of the f KG-Fre.Specifically,we first preprocess the freight data,then build an ontology-based library to form the KG-Fre which is stored using the Neo4j’s graph storage method.Then,we adopt the Trans E knowledge representation learning model to represent each predicate and entity in the graph as a low-dimensional vector,and encode the predicate and head entity in the user’s question via a proposed predicate learning model and a head entity learning model.Finally,a similarity measurement method based on joint distance is proposed,which searches for the most relevant facts to the question in KG-Fre and returns it to the user as the answer.This paper validates the accuracy of KEFQA using the massive heavy truck freight data set in Shaanxi Province.It shows that the accuracy reaches 92.56%,satisfying the need of QA.At the same time,the FB2 M and FB5 M open-source data sets and Simple Questions simple question sets are selected to compare the performance with the existing QA methods based on knowledge graphs.The experimental results show that the accuracy of the KEFQA framework reaches 81.9%,which is superior to other QA methods.