Real-time Semantic Data Stream Reasoning Base On Knowledge Representation Learning

In recent years,knowledge graph has become an important way of knowledge representation.A triple formed by subject predicate object is the basic component of knowledge graph.With subject and object as labeled nodes and predicate as labeled directed edges,a set of triples can be represented as a networked graph,that is,a knowledge graph.Knowledge graph brings better data interactivity and strong knowledge reasoning ability.However,it is necessary to mine the hidden relationships in the dynamic knowledge stream by efficient real-time semantic reasoning.Traditional semantic data stream reasoning uses forward or backward chained reasoning to generate deterministic answers,but it is inefficient in complex transfer rule reasoning,which can not meet the timeliness requirements of real-time data stream processing scenarios.Therefore,in view of the above problems,this thesis studies the semantic data stream reasoning based on the deep learning model.This thesis extends a knowledge representation method based on the joint embedding model and applies it to semantic data stream processing.The algorithm first embeds the transfer rules and the fact triples,and uses the deep learning model for training.In the reasoning stage,the inference template is built according to the rules involved in the query,and the triples generated by the inference template are predicted and classified by the deep learning model,and then the results are output as the query and reasoning answers.Furthermore,aiming at the problem of dynamic increase of model data set,this thesis proposes a multi vector space training method,which puts the increased data set into a new vector space for training,and sets different training parameters for each space according to the data characteristics,so as to avoid the problem of consuming a lot of time to retrain the whole data set when the data is updated.Finally,this thesie applies the multi space dynamic training model to the data stream reasoning platform,and proposes the corresponding index and triple prediction mechanism to realize the efficient semantic data stream reasoning based on deep learning.Experiments show that for complex transfer rule reasoning,the real-time semantic data flow reasoning based on knowledge representation learning can effectively reduce the delay under the premise of ensuring better reasoning accuracy and hit rate.