Inference Completion Algorithms Based On Knowledge Graph Construction

After entering the 21 st century,with the revolution of information technology,computer-related technology has developed rapidly.At the same time,various information data in the Internet world has also shown exponential growth,which is bringing me a more convenient lifestyle.At the same time,it also allows us to think about how to deal with such a huge amount of data.Faced with such a huge amount of data.Google company proposed the emerging concept of Knowledge Grahp in 2012,and set out to study new forms of data and more advanced data processing methods.Knowledge graph technology is simply structured information based on entities and relationships with a large number of entities and relationships set.Knowledge graph is also widely used,which is the basis for the realization of technologies such as big data,artificial intelligence,question and answer systems.The knowledge graph construction process includes several key technologies such as knowledge extraction,entity linking,relational reasoning,and disambiguation.Among them,relational reasoning supplementary technology is one of the key steps.Technology of relational reasoning gives the computer a certain "thinking" ability.It makes possible for the computer to improve the knowledge graph independently.It is a key research technology of this article.The function of relational reasoning completion technology is to analyze,summarize and summarize the three tuples through statistical learning or vector mapping.Find and discover the potential facts or new triple information in the knowledge map to make the original knowledge map data more perfect.This thesis first makes a detailed investigation on the development of reasoning completion algorithm,and finally proposes an algorithm in the direction of representation learning and logic rule walking.The main works is summarized as follows:1.This thesis investigated the background of the existing relational reasoning technology,analyzed the advantages and disadvantages of the common mainstream reasoning algorithms,and introduced the common technical background of relational reasoning.2.Improved representation-based learning method: An algorithm is proposed which is based on the Trans E reasoning algorithm.It is a factual description as a supplementary reasoning algorithm model,and the core is to use TF-IWF-based weighting technology.The supplementary fact description is used to increase the semantic space distance between the three entities in the reasoning,and the accuracy of reasoning is improved by increasing the fault-tolerant space.Finally,the relevant results in the experimental part verified the effectiveness of the algorithm proposed in this chapter.3.Optimized logic-based reasoning algorithm: Analyzed the relational reasoning algorithm model of logic rules,and proposed a new undirected graph random walk reasoning fusion algorithm UGRW based on this.This algorithm is divided into undirected graph logic walks reasoning and relational subgraph inference,using the triple data in the knowledge graph to learn and train the inference model corresponding to a specific relation,and divide the undirected graph topology into a global graph and relational sub-graphs,which are processed separately,through feature modeling and calculation of adjacency matrix to complete the task of relational reasoning.In the final part of the experiment,the fusion inference performance is compared and analyzed on the two commonly used data,the different performance of the relational sub-graphs reasoning part under four different relationship types is discussed.