Knowledge-based Decision-making and Services

Abstract

Knowledge can be divided into two types: tacit knowledge and explicit knowledge. The tacit knowledge is about the experiences which are in disorder and can not be transferred to others. The genre that studies tacit knowledge-based decision-making and services focuses on knowledge-based systems, expert systems, rough set, fuzzy theory, MCDA (Multi-criteria decision analysis), and GDM (Group decision-making). The explicit knowledge is about the rules which are streamlined in order and can be transferred to others. The genre that studies explicit knowledge to support decision-making and services includes first-order logic, ontology, semantic web, and knowledge graph.

In the first part of this thesis, we modify the ANP method, the most popular method in determining the weights of criteria, whose limitation is the un-convergence problem. we propose a hybrid ANP (H-ANP) method, which aims to improve the ANP by combining the weights obtained from the analytic hierarchy process (AHP). The proposed method is proved to be convergent since the network of the H-ANP is strongly connected. According to the simulation experiments, H-ANP is more robust than ANP under different settings of parameters. It also shows a higher Kendall cor-relationship and lower MSE concerning AHP, compared with the existing method (e.g., the averagely connected ANP method). An empirical example is also provided, which uses H-ANP to evaluate the government data sustainability of a city.

In the second part of this thesis, we discuss a certain method of explicit knowledge-based decision-making and services called the Tucker-reasoning learning method. This method aims to build a learning model to teach the knowledge graph to make binary decisions like "Yes" or "No". This study first embeds a knowledge graph into a 3-order tensor. The first slice of tensor represents the weak connection relationships. The second slice represents the mid connection relationships. The third slice represents the strong connection relationships like "is-a". Each slice is an adjacent matrix representing the existence of a relationship between head nodes and tail nodes. Based on Tucker Decomposition, this study invents a new score function that can measure the score of each edge in the knowledge graph. An objective function is created to compare the weighted scores of paths supporting the first decision and the weighted scores supporting the second decision. The numerical experiments show that the method's accuracy is 84% on the condition of no conflict. The proposed method is the first to train knowledge graphs to make binary decisions directly. Thus, this method conserves the transparency of reasoning paths and is suitable for application scenarios like financial, justice, and medical decision-making, which require clear reasoning paths. Compared to Trans-related methods used for one-step reasoning and link prediction, the proposed method greatly contributes to the multiple-step reasoning in the knowledge graph domain.

Date of Award	7 Aug 2023
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Duan LI (Supervisor), Xiaomi AN (External Supervisor) & Qi WU (Supervisor)