Research And Implementation Of Investment Decision Based On Temporal Knowledge Graph

In the field of investment and financing,various industries and enterprises are complexly related,and it is difficult to mine knowledge from massive information.Knowledge Graph(KG)has become a research hotspot in the field of investment and financing due to its advantages in sorting,summarizing and applying knowledge.However,most events in the financial field are time-sensitive and dynamic,and market information changes with time.Therefore,investment and financing prediction based on knowledge graphs still faces great challenges.As a research hotspot in recent years,KG has not only made key breakthroughs in academia,but also been widely applied in industry.Currently,KG is deeply integrated with various vertical fields such as healthcare,education and finance,and it has become an indispensable technology in more and more application scenarios such as search,recommendation and risk control.However,the current research on KG mainly focuses on static knowledge graph,the research of dynamic Temporal Knowledge Graph(TKG)is deficiency.And the current research on TKG is also less combined with actual domain application scenarios.In view of this,this paper conducts an in-depth research and discussion on how to make investment and financing relationship prediction based on dynamic TKG to provide investment decision advice for investors,as follows:(1)At present,the integration of knowledge graphs and financial fields mainly focuses on static knowledge graphs.In view of the complex and time-sensitive problems of data association in the investment and financing fields,static knowledge graphs can no longer fully meet the application requirements.Therefore,this paper proposes a top-down method for building a knowledge graph of investment and financing time series,from several steps such as ontology design,data collection,data processing,knowledge extraction and knowledge fusion,to form quadruples with knowledge embedded in temporal information and build a temporal knowledge graph of investment and financing,and use the quadruples as a dataset of investment and financing domain(2)Studies have shown that the historical investment behavior of investors has a very large impact on future investment decisions.For example,investors may make additional investments in a company project and follow other investors’ decisions in their investment decision-making.In order to model these phenomena,this paper proposes a multi-task temporal inference model which can make inference predictions from historical repeated entities,all entities and neighboring entities.This inference model can learn from historical information and predict investment events in the future moment,and then provide investment decision recommendations to investors.(3)To verify the effectiveness of the model,this study trained,validated and tested the model on three publicly available generic domain datasets with different temporal granularity and the investment and financing temporal knowledge graph constructed in this paper,compared it with six baseline algorithms for inference and extrapolation to assess the model effectiveness.Otherwise,this study conducted ablation experiments on the model to analyse the effectiveness of each module by adjusting the components of the model.The experiments show that the joint historical cyclic event prediction model is more effective than the existing Temporal Knowledge Graph Completion(TKGC)methods.It can predict events at future moments,which can be combined with the investment and financing domain to provide supplementary opinions for enterprises when making investment or financing decisions.