Knowledge graphs(KGs)express relationships between entity pairs,and many real-life problems can be formulated as knowledge graph reasoning(KGR).Conventional approaches to KGR have achieved promising performance but st...Knowledge graphs(KGs)express relationships between entity pairs,and many real-life problems can be formulated as knowledge graph reasoning(KGR).Conventional approaches to KGR have achieved promising performance but still have some drawbacks.On the one hand,most KGR methods focus only on one phase of the KG lifecycle,such as KG completion or refinement,while ignoring reasoning over other stages,such as KG extraction.On the other hand,traditional KGR methods,broadly categorized as symbolic and neural,are unable to balance both scalability and interpretability.To resolve these two problems,we take a more comprehensive perspective of KGR with regard to the whole KG lifecycle,including KG extraction,completion,and refinement,which correspond to three subtasks:knowledge extraction,relational reasoning,and inconsistency checking.In addition,we propose the implementation of KGR using a novel neural symbolic framework,with regard to both scalability and interpretability.Experimental results demonstrate that our proposed methods outperform traditional neural symbolic models.展开更多
Objective To establish the knowledge graph of“disease-syndrome-symptom-method-formula”in Treatise on Febrile Diseases(Shang Han Lun,《伤寒论》)for reducing the fuzziness and uncertainty of data,and for laying a foun...Objective To establish the knowledge graph of“disease-syndrome-symptom-method-formula”in Treatise on Febrile Diseases(Shang Han Lun,《伤寒论》)for reducing the fuzziness and uncertainty of data,and for laying a foundation for later knowledge reasoning and its application.Methods Under the guidance of experts in the classical formula of traditional Chinese medicine(TCM),the method of“top-down as the main,bottom-up as the auxiliary”was adopted to carry out knowledge extraction,knowledge fusion,and knowledge storage from the five aspects of the disease,syndrome,symptom,method,and formula for the original text of Treatise on Febrile Diseases,and so the knowledge graph of Treatise on Febrile Diseases was constructed.On this basis,the knowledge structure query and the knowledge relevance query were realized in a visual manner.Results The knowledge graph of“disease-syndrome-symptom-method-formula”in the Treatise on Febrile Diseases was constructed,containing 6469 entities and 10911 relational triples,on which the query of entities and their relationships can be carried out and the query result can be visualized.Conclusion The knowledge graph of Treatise on Febrile Diseases systematically realizes its digitization of the knowledge system,and improves the completeness and accuracy of the knowledge representation,and the connection between“disease-syndrome-symptom-treatment-formula”,which is conducive to the sharing and reuse of knowledge can be obtained in a clear and efficient way.展开更多
The effectiveness of the logic mining approach is strongly correlated to the quality of the induced logical representation that represent the behaviour of the data.Specifically,the optimum induced logical representati...The effectiveness of the logic mining approach is strongly correlated to the quality of the induced logical representation that represent the behaviour of the data.Specifically,the optimum induced logical representation indicates the capability of the logic mining approach in generalizing the real datasets of different variants and dimensions.The main issues with the logic extracted by the standard logic mining techniques are lack of interpretability and the weakness in terms of the structural and arrangement of the 2 Satisfiability logic causing lower accuracy.To address the issues,the logical permutation serves as an alternative mechanism that can enhance the probability of the 2 Satisfiability logical rule becoming true by utilizing the definitive finite arrangement of attributes.This work aims to examine and analyze the significant effect of logical permutation on the performance of data extraction ability of the logic mining approach incorporated with the recurrent discrete Hopfield Neural Network.Based on the theory,the effect of permutation and associate memories in recurrent Hopfield Neural Network will potentially improve the accuracy of the existing logic mining approach.To validate the impact of the logical permutation on the retrieval phase of the logic mining model,the proposed work is experimentally tested on a different class of the benchmark real datasets ranging from the multivariate and timeseries datasets.The experimental results show the significant improvement in the proposed logical permutation-based logic mining according to the domains such as compatibility,accuracy,and competitiveness as opposed to the plethora of standard 2 Satisfiability Reverse Analysis methods.展开更多
The fight against fraud and trafficking is a fundamental mission of customs. The conditions for carrying out this mission depend both on the evolution of economic issues and on the behaviour of the actors in charge of...The fight against fraud and trafficking is a fundamental mission of customs. The conditions for carrying out this mission depend both on the evolution of economic issues and on the behaviour of the actors in charge of its implementation. As part of the customs clearance process, customs are nowadays confronted with an increasing volume of goods in connection with the development of international trade. Automated risk management is therefore required to limit intrusive control. In this article, we propose an unsupervised classification method to extract knowledge rules from a database of customs offences in order to identify abnormal behaviour resulting from customs control. The idea is to apply the Apriori principle on the basis of frequent grounds on a database relating to customs offences in customs procedures to uncover potential rules of association between a customs operation and an offence for the purpose of extracting knowledge governing the occurrence of fraud. This mass of often heterogeneous and complex data thus generates new needs that knowledge extraction methods must be able to meet. The assessment of infringements inevitably requires a proper identification of the risks. It is an original approach based on data mining or data mining to build association rules in two steps: first, search for frequent patterns (support >= minimum support) then from the frequent patterns, produce association rules (Trust >= Minimum Trust). The simulations carried out highlighted three main association rules: forecasting rules, targeting rules and neutral rules with the introduction of a third indicator of rule relevance which is the Lift measure. Confidence in the first two rules has been set at least 50%.展开更多
Online Social Networks(OSNs)are based on the sharing of different types of information and on various interactions(comments,reactions,and sharing).One of these important actions is the emotional reaction to the conten...Online Social Networks(OSNs)are based on the sharing of different types of information and on various interactions(comments,reactions,and sharing).One of these important actions is the emotional reaction to the content.The diversity of reaction types available on Facebook(namely FB)enables users to express their feelings,and its traceability creates and enriches the users’emotional identity in the virtual world.This paper is based on the analysis of 119875012 FB reactions(Like,Love,Haha,Wow,Sad,Angry,Thankful,and Pride)made at multiple levels(publications,comments,and sub-comments)to study and classify the users’emotional behavior,visualize the distribution of different types of reactions,and analyze the gender impact on emotion generation.All of these can be achieved by addressing these research questions:who reacts the most?Which emotion is the most expressed?展开更多
Assembly process documents record the designers'intention or knowledge.However,common knowl-edge extraction methods are not well suitable for assembly process documents,because of its tabular form and unstructured...Assembly process documents record the designers'intention or knowledge.However,common knowl-edge extraction methods are not well suitable for assembly process documents,because of its tabular form and unstructured natural language texts.In this paper,an assembly semantic entity recognition and relation con-struction method oriented to assembly process documents is proposed.First,the assembly process sentences are extracted from the table through concerned region recognition and cell division,and they will be stored as a key-value object file.Then,the semantic entities in the sentence are identified through the sequence tagging model based on the specific attention mechanism for assembly operation type.The syntactic rules are designed for realizing automatic construction of relation between entities.Finally,by using the self-constructed corpus,it is proved that the sequence tagging model in the proposed method performs better than the mainstream named entity recognition model when handling assembly process design language.The effectiveness of the proposed method is also analyzed through the simulation experiment in the small-scale real scene,compared with manual method.The results show that the proposed method can help designers accumulate knowledge automatically and efficiently.展开更多
Knowledge mining is a widely active research area across disciplines such as natural language processing(NLP), data mining(DM), and machine learning(ML). The overall objective of extracting knowledge from data source ...Knowledge mining is a widely active research area across disciplines such as natural language processing(NLP), data mining(DM), and machine learning(ML). The overall objective of extracting knowledge from data source is to create a structured representation that allows researchers to better understand such data and operate upon it to build applications. Each mentioned discipline has come up with an ample body of research, proposing different methods that can be applied to different data types. A significant number of surveys have been carried out to summarize research works in each discipline. However, no survey has presented a cross-disciplinary review where traits from different fields were exposed to further stimulate research ideas and to try to build bridges among these fields.In this work, we present such a survey.展开更多
Traditional experimental economics methods often consume enormous resources of qualified human participants,and the inconsistence of a participant’s decisions among repeated trials prevents investigation from sensiti...Traditional experimental economics methods often consume enormous resources of qualified human participants,and the inconsistence of a participant’s decisions among repeated trials prevents investigation from sensitivity analyses.The problem can be solved if computer agents are capable of generating similar behaviors as the given participants in experiments.An experimental economics based analysis method is presented to extract deep information from questionnaire data and emulate any number of participants.Taking the customers’willingness to purchase electric vehicles(EVs)as an example,multi-layer correlation information is extracted from a limited number of questionnaires.Multiagents mimicking the inquired potential customers are modelled through matching the probabilistic distributions of their willingness embedded in the questionnaires.The authenticity of both the model and the algorithmis validated by comparing the agent-based Monte Carlo simulation results with the questionnaire-based deduction results.With the aid of agent models,the effects of minority agents with specific preferences on the results are also discussed.展开更多
Knowlege is important for text-related applications.In this paper,we introduce Microsoft Concept Graph,a knowledge graph engine that provides concept tagging APIs to facilitate the understanding of human languages.Mic...Knowlege is important for text-related applications.In this paper,we introduce Microsoft Concept Graph,a knowledge graph engine that provides concept tagging APIs to facilitate the understanding of human languages.Microsoft Concept Graph is built upon Probase,a universal probabilistic taxonomy consisting of instances and concepts mined from the Web.We start by introducing the construction of the knowledge graph through iterative semantic extraction and taxonomy construction procedures,which extract 2.7 million concepts from 1.68 billion Web pages.We then use conceptualization models to represent text in the concept space to empower text-related applications,such as topic search,query recommendation,Web table understanding and Ads relevance.Since the release in 2016,Microsoft Concept Graph has received more than 100,000 pageviews,2 million API calls and 3,000 registered downloads from 50,000 visitors over 64 countries.展开更多
A main focus of machine learning research has been improving the generalization accuracy and efficiency of prediction models. However, what emerges as missing in many applications is actionability, i.e., the ability t...A main focus of machine learning research has been improving the generalization accuracy and efficiency of prediction models. However, what emerges as missing in many applications is actionability, i.e., the ability to turn prediction results into actions. Existing effort in deriving such actionable knowledge is few and limited to simple action models while in many real applications those models are often more complex and harder to extract an optimal solution. In this paper, we propose a novel approach that achieves actionability by combining learning with planning, two core areas of AI. In particular, we propose a framework to extract actionable knowledge from random forest, one of the most widely used and best off-the-shelf classifiers. We formulate the actionability problem to a sub-optimal action planning （SOAP） problem, which is to find a plan to alter certain features of a given input so that the random forest would yield a desirable output, while minimizing the total costs of actions. Technically, the SOAP problem is formulated in the SAS＋ planning formalism, and solved using a Max-SAT based ap- proach. Our experimental results demonstrate the effectiveness and efficiency of the proposed approach on a personal credit dataset and other benchmarks. Our work represents a new application of automated planning on an emerging and challenging machine learning paradigm.展开更多
Although amazing progress has been made in ma- chine learning to achieve high generalization accuracy and ef- ficiency, there is still very limited work on deriving meaning- ful decision-making actions from the result...Although amazing progress has been made in ma- chine learning to achieve high generalization accuracy and ef- ficiency, there is still very limited work on deriving meaning- ful decision-making actions from the resulting models. How- ever, in many applications such as advertisement, recommen- dation systems, social networks, customer relationship man- agement, and clinical prediction, the users need not only ac- curate prediction, but also suggestions on actions to achieve a desirable goal （e.g., high ads hit rates） or avert an unde- sirable predicted result （e.g., clinical deterioration）. Existing works for extracting such actionability are few and limited to simple models such as a decision tree. The dilemma is that those models with high accuracy are often more complex and harder to extract actionability from. In this paper, we propose an effective method to extract ac- tionable knowledge from additive tree models （ATMs）, one of the most widely used and best off-the-shelf classifiers. We rigorously formulate the optimal actionable planning （OAP） problem for a given ATM, which is to extract an action- able plan for a given input so that it can achieve a desirable output while maximizing the net profit. Based on a state space graph formulation, we first propose an optimal heuris- tic search method which intends to find an optimal solution. Then, we also present a sub-optimal heuristic search with an admissible and consistent heuristic function which can re- markably improve the efficiency of the algorithm. Our exper- imental results demonstrate the effectiveness and efficiency of the proposed algorithms on several real datasets in the application domain of personal credit and banking.展开更多
As one of the most important components in knowledge graph construction,entity linking has been drawing more and more attention in the last decade.In this paper,we propose two improvements towards better entity linkin...As one of the most important components in knowledge graph construction,entity linking has been drawing more and more attention in the last decade.In this paper,we propose two improvements towards better entity linking.On one hand,we propose a simple but effective coarse-to-fine unsupervised knowledge base(KB)extraction approach to improve the quality of KB,through which we can conduct entity linking more efficiently.On the other hand,we propose a highway network framework to bridge key words and sequential information captured with a self-attention mechanism to better represent both local and global information.Detailed experimentation on six public entity linking datasets verifies the great effectiveness of both our approaches.展开更多
基金funded by National Natural Science Foundation of China(Grant no.91846204 and U19B2027)National Key Research and Development Program of China(Grant no.2018YFB1402800).
文摘Knowledge graphs(KGs)express relationships between entity pairs,and many real-life problems can be formulated as knowledge graph reasoning(KGR).Conventional approaches to KGR have achieved promising performance but still have some drawbacks.On the one hand,most KGR methods focus only on one phase of the KG lifecycle,such as KG completion or refinement,while ignoring reasoning over other stages,such as KG extraction.On the other hand,traditional KGR methods,broadly categorized as symbolic and neural,are unable to balance both scalability and interpretability.To resolve these two problems,we take a more comprehensive perspective of KGR with regard to the whole KG lifecycle,including KG extraction,completion,and refinement,which correspond to three subtasks:knowledge extraction,relational reasoning,and inconsistency checking.In addition,we propose the implementation of KGR using a novel neural symbolic framework,with regard to both scalability and interpretability.Experimental results demonstrate that our proposed methods outperform traditional neural symbolic models.
基金The Open Fund of Hunan University of Traditional Chinese Medicine for the First-Class Discipline of Traditional Chinese Medicine(2018ZYX66)the Science Research Project of Hunan Provincial Department of Education(20C1391)the Natural Science Foundation of Hunan Province(2020JJ4461)。
文摘Objective To establish the knowledge graph of“disease-syndrome-symptom-method-formula”in Treatise on Febrile Diseases(Shang Han Lun,《伤寒论》)for reducing the fuzziness and uncertainty of data,and for laying a foundation for later knowledge reasoning and its application.Methods Under the guidance of experts in the classical formula of traditional Chinese medicine(TCM),the method of“top-down as the main,bottom-up as the auxiliary”was adopted to carry out knowledge extraction,knowledge fusion,and knowledge storage from the five aspects of the disease,syndrome,symptom,method,and formula for the original text of Treatise on Febrile Diseases,and so the knowledge graph of Treatise on Febrile Diseases was constructed.On this basis,the knowledge structure query and the knowledge relevance query were realized in a visual manner.Results The knowledge graph of“disease-syndrome-symptom-method-formula”in the Treatise on Febrile Diseases was constructed,containing 6469 entities and 10911 relational triples,on which the query of entities and their relationships can be carried out and the query result can be visualized.Conclusion The knowledge graph of Treatise on Febrile Diseases systematically realizes its digitization of the knowledge system,and improves the completeness and accuracy of the knowledge representation,and the connection between“disease-syndrome-symptom-treatment-formula”,which is conducive to the sharing and reuse of knowledge can be obtained in a clear and efficient way.
基金Universiti Sains Malaysia for Short Term Grant with Grant Number 304/PMATHS/6315390.
文摘The effectiveness of the logic mining approach is strongly correlated to the quality of the induced logical representation that represent the behaviour of the data.Specifically,the optimum induced logical representation indicates the capability of the logic mining approach in generalizing the real datasets of different variants and dimensions.The main issues with the logic extracted by the standard logic mining techniques are lack of interpretability and the weakness in terms of the structural and arrangement of the 2 Satisfiability logic causing lower accuracy.To address the issues,the logical permutation serves as an alternative mechanism that can enhance the probability of the 2 Satisfiability logical rule becoming true by utilizing the definitive finite arrangement of attributes.This work aims to examine and analyze the significant effect of logical permutation on the performance of data extraction ability of the logic mining approach incorporated with the recurrent discrete Hopfield Neural Network.Based on the theory,the effect of permutation and associate memories in recurrent Hopfield Neural Network will potentially improve the accuracy of the existing logic mining approach.To validate the impact of the logical permutation on the retrieval phase of the logic mining model,the proposed work is experimentally tested on a different class of the benchmark real datasets ranging from the multivariate and timeseries datasets.The experimental results show the significant improvement in the proposed logical permutation-based logic mining according to the domains such as compatibility,accuracy,and competitiveness as opposed to the plethora of standard 2 Satisfiability Reverse Analysis methods.
文摘The fight against fraud and trafficking is a fundamental mission of customs. The conditions for carrying out this mission depend both on the evolution of economic issues and on the behaviour of the actors in charge of its implementation. As part of the customs clearance process, customs are nowadays confronted with an increasing volume of goods in connection with the development of international trade. Automated risk management is therefore required to limit intrusive control. In this article, we propose an unsupervised classification method to extract knowledge rules from a database of customs offences in order to identify abnormal behaviour resulting from customs control. The idea is to apply the Apriori principle on the basis of frequent grounds on a database relating to customs offences in customs procedures to uncover potential rules of association between a customs operation and an offence for the purpose of extracting knowledge governing the occurrence of fraud. This mass of often heterogeneous and complex data thus generates new needs that knowledge extraction methods must be able to meet. The assessment of infringements inevitably requires a proper identification of the risks. It is an original approach based on data mining or data mining to build association rules in two steps: first, search for frequent patterns (support >= minimum support) then from the frequent patterns, produce association rules (Trust >= Minimum Trust). The simulations carried out highlighted three main association rules: forecasting rules, targeting rules and neutral rules with the introduction of a third indicator of rule relevance which is the Lift measure. Confidence in the first two rules has been set at least 50%.
文摘Online Social Networks(OSNs)are based on the sharing of different types of information and on various interactions(comments,reactions,and sharing).One of these important actions is the emotional reaction to the content.The diversity of reaction types available on Facebook(namely FB)enables users to express their feelings,and its traceability creates and enriches the users’emotional identity in the virtual world.This paper is based on the analysis of 119875012 FB reactions(Like,Love,Haha,Wow,Sad,Angry,Thankful,and Pride)made at multiple levels(publications,comments,and sub-comments)to study and classify the users’emotional behavior,visualize the distribution of different types of reactions,and analyze the gender impact on emotion generation.All of these can be achieved by addressing these research questions:who reacts the most?Which emotion is the most expressed?
文摘Assembly process documents record the designers'intention or knowledge.However,common knowl-edge extraction methods are not well suitable for assembly process documents,because of its tabular form and unstructured natural language texts.In this paper,an assembly semantic entity recognition and relation con-struction method oriented to assembly process documents is proposed.First,the assembly process sentences are extracted from the table through concerned region recognition and cell division,and they will be stored as a key-value object file.Then,the semantic entities in the sentence are identified through the sequence tagging model based on the specific attention mechanism for assembly operation type.The syntactic rules are designed for realizing automatic construction of relation between entities.Finally,by using the self-constructed corpus,it is proved that the sequence tagging model in the proposed method performs better than the mainstream named entity recognition model when handling assembly process design language.The effectiveness of the proposed method is also analyzed through the simulation experiment in the small-scale real scene,compared with manual method.The results show that the proposed method can help designers accumulate knowledge automatically and efficiently.
文摘Knowledge mining is a widely active research area across disciplines such as natural language processing(NLP), data mining(DM), and machine learning(ML). The overall objective of extracting knowledge from data source is to create a structured representation that allows researchers to better understand such data and operate upon it to build applications. Each mentioned discipline has come up with an ample body of research, proposing different methods that can be applied to different data types. A significant number of surveys have been carried out to summarize research works in each discipline. However, no survey has presented a cross-disciplinary review where traits from different fields were exposed to further stimulate research ideas and to try to build bridges among these fields.In this work, we present such a survey.
基金This work is supported by NSFC-EPSRC Collaborative Project(NSFC-No.51361130153,EPSRC-EP/L001063/1),State Grid Corporation of China.
文摘Traditional experimental economics methods often consume enormous resources of qualified human participants,and the inconsistence of a participant’s decisions among repeated trials prevents investigation from sensitivity analyses.The problem can be solved if computer agents are capable of generating similar behaviors as the given participants in experiments.An experimental economics based analysis method is presented to extract deep information from questionnaire data and emulate any number of participants.Taking the customers’willingness to purchase electric vehicles(EVs)as an example,multi-layer correlation information is extracted from a limited number of questionnaires.Multiagents mimicking the inquired potential customers are modelled through matching the probabilistic distributions of their willingness embedded in the questionnaires.The authenticity of both the model and the algorithmis validated by comparing the agent-based Monte Carlo simulation results with the questionnaire-based deduction results.With the aid of agent models,the effects of minority agents with specific preferences on the results are also discussed.
文摘Knowlege is important for text-related applications.In this paper,we introduce Microsoft Concept Graph,a knowledge graph engine that provides concept tagging APIs to facilitate the understanding of human languages.Microsoft Concept Graph is built upon Probase,a universal probabilistic taxonomy consisting of instances and concepts mined from the Web.We start by introducing the construction of the knowledge graph through iterative semantic extraction and taxonomy construction procedures,which extract 2.7 million concepts from 1.68 billion Web pages.We then use conceptualization models to represent text in the concept space to empower text-related applications,such as topic search,query recommendation,Web table understanding and Ads relevance.Since the release in 2016,Microsoft Concept Graph has received more than 100,000 pageviews,2 million API calls and 3,000 registered downloads from 50,000 visitors over 64 countries.
基金This work was supported in part by the National Natural Science Foundation of China （Grant Nos. 61502412, 61379066, and 61402395）, Natural Science Foundation of the Jiangsu Province （BK20150459, BK20151314, and BK20140492）, Natural Science Foundation of the Jiangsu Higher Education Institutions （15KJB520036）, United States NSF grants （IIS-0534699, IIS-0713109, CNS-1017701）, Microsoft Research New Faculty Fellowship, and the Research Innovation Program for Graduate Student in Jiangsu Province （KYLX16 1390）.
文摘A main focus of machine learning research has been improving the generalization accuracy and efficiency of prediction models. However, what emerges as missing in many applications is actionability, i.e., the ability to turn prediction results into actions. Existing effort in deriving such actionable knowledge is few and limited to simple action models while in many real applications those models are often more complex and harder to extract an optimal solution. In this paper, we propose a novel approach that achieves actionability by combining learning with planning, two core areas of AI. In particular, we propose a framework to extract actionable knowledge from random forest, one of the most widely used and best off-the-shelf classifiers. We formulate the actionability problem to a sub-optimal action planning （SOAP） problem, which is to find a plan to alter certain features of a given input so that the random forest would yield a desirable output, while minimizing the total costs of actions. Technically, the SOAP problem is formulated in the SAS＋ planning formalism, and solved using a Max-SAT based ap- proach. Our experimental results demonstrate the effectiveness and efficiency of the proposed approach on a personal credit dataset and other benchmarks. Our work represents a new application of automated planning on an emerging and challenging machine learning paradigm.
基金This work was supported in part by China Postdoctoral Science Foundation （2013M531527）, the Fundamental Research Funds for the Central Universities （0110000037）, the National Natural Science Foun- dation of China （Grant Nos. 61502412, 61033009, and 61175057）, Natural Science Foundation of the Jiangsu Province （BK20150459）, Natural Science Foundation of the Jiangsu Higher Education Institutions （15KJB520036）, National Science Foundation, United States （IIS-0534699, IIS-0713109, CNS-1017701）, and a Microsoft Research New Faculty Fellowship.
文摘Although amazing progress has been made in ma- chine learning to achieve high generalization accuracy and ef- ficiency, there is still very limited work on deriving meaning- ful decision-making actions from the resulting models. How- ever, in many applications such as advertisement, recommen- dation systems, social networks, customer relationship man- agement, and clinical prediction, the users need not only ac- curate prediction, but also suggestions on actions to achieve a desirable goal （e.g., high ads hit rates） or avert an unde- sirable predicted result （e.g., clinical deterioration）. Existing works for extracting such actionability are few and limited to simple models such as a decision tree. The dilemma is that those models with high accuracy are often more complex and harder to extract actionability from. In this paper, we propose an effective method to extract ac- tionable knowledge from additive tree models （ATMs）, one of the most widely used and best off-the-shelf classifiers. We rigorously formulate the optimal actionable planning （OAP） problem for a given ATM, which is to extract an action- able plan for a given input so that it can achieve a desirable output while maximizing the net profit. Based on a state space graph formulation, we first propose an optimal heuris- tic search method which intends to find an optimal solution. Then, we also present a sub-optimal heuristic search with an admissible and consistent heuristic function which can re- markably improve the efficiency of the algorithm. Our exper- imental results demonstrate the effectiveness and efficiency of the proposed algorithms on several real datasets in the application domain of personal credit and banking.
基金This work was supported by the key project of the National Natural Science Foundation of China(Grant No.61836007)the normal project of the National Natural Science Foundation of China(Grant No.61876118)the project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions.
文摘As one of the most important components in knowledge graph construction,entity linking has been drawing more and more attention in the last decade.In this paper,we propose two improvements towards better entity linking.On one hand,we propose a simple but effective coarse-to-fine unsupervised knowledge base(KB)extraction approach to improve the quality of KB,through which we can conduct entity linking more efficiently.On the other hand,we propose a highway network framework to bridge key words and sequential information captured with a self-attention mechanism to better represent both local and global information.Detailed experimentation on six public entity linking datasets verifies the great effectiveness of both our approaches.
