Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture
Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it...
Ausführliche Beschreibung
Gespeichert in:
| Autor*in: |
Wang, Qi [verfasserIn] |
|---|
| Format: |
E-Artikel |
|---|---|
| Sprache: |
Englisch |
| Erschienen: |
2021transfer abstract |
|---|
| Schlagwörter: |
|---|
| Umfang: |
9 |
|---|
| Übergeordnetes Werk: |
Enthalten in: The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast - Liu, Yang ELSEVIER, 2018, an international journal, Amsterdam |
|---|---|
| Übergeordnetes Werk: |
volume:429 ; year:2021 ; day:14 ; month:03 ; pages:101-109 ; extent:9 |
| Links: |
|---|
| DOI / URN: |
10.1016/j.neucom.2020.12.040 |
|---|
| Katalog-ID: |
ELV052908410 |
|---|
| LEADER | 01000caa a22002652 4500 | ||
|---|---|---|---|
| 001 | ELV052908410 | ||
| 003 | DE-627 | ||
| 005 | 20230626033938.0 | ||
| 007 | cr uuu---uuuuu | ||
| 008 | 210910s2021 xx |||||o 00| ||eng c | ||
| 024 | 7 | |a 10.1016/j.neucom.2020.12.040 |2 doi | |
| 028 | 5 | 2 | |a /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica |
| 035 | |a (DE-627)ELV052908410 | ||
| 035 | |a (ELSEVIER)S0925-2312(20)31940-8 | ||
| 040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
| 041 | |a eng | ||
| 082 | 0 | 4 | |a 570 |q VZ |
| 084 | |a BIODIV |q DE-30 |2 fid | ||
| 084 | |a 35.70 |2 bkl | ||
| 084 | |a 42.12 |2 bkl | ||
| 100 | 1 | |a Wang, Qi |e verfasserin |4 aut | |
| 245 | 1 | 0 | |a Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| 264 | 1 | |c 2021transfer abstract | |
| 300 | |a 9 | ||
| 336 | |a nicht spezifiziert |b zzz |2 rdacontent | ||
| 337 | |a nicht spezifiziert |b z |2 rdamedia | ||
| 338 | |a nicht spezifiziert |b zu |2 rdacarrier | ||
| 520 | |a Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. | ||
| 520 | |a Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. | ||
| 650 | 7 | |a Reinforcement learning |2 Elsevier | |
| 650 | 7 | |a IDA* algorithm |2 Elsevier | |
| 650 | 7 | |a Markov decision process |2 Elsevier | |
| 650 | 7 | |a Neural network |2 Elsevier | |
| 650 | 7 | |a Knowledge graph |2 Elsevier | |
| 700 | 1 | |a Hao, Yongsheng |4 oth | |
| 700 | 1 | |a Chen, Feng |4 oth | |
| 773 | 0 | 8 | |i Enthalten in |n Elsevier |a Liu, Yang ELSEVIER |t The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |d 2018 |d an international journal |g Amsterdam |w (DE-627)ELV002603926 |
| 773 | 1 | 8 | |g volume:429 |g year:2021 |g day:14 |g month:03 |g pages:101-109 |g extent:9 |
| 856 | 4 | 0 | |u https://doi.org/10.1016/j.neucom.2020.12.040 |3 Volltext |
| 912 | |a GBV_USEFLAG_U | ||
| 912 | |a GBV_ELV | ||
| 912 | |a SYSFLAG_U | ||
| 912 | |a FID-BIODIV | ||
| 912 | |a SSG-OLC-PHA | ||
| 936 | b | k | |a 35.70 |j Biochemie: Allgemeines |q VZ |
| 936 | b | k | |a 42.12 |j Biophysik |q VZ |
| 951 | |a AR | ||
| 952 | |d 429 |j 2021 |b 14 |c 0314 |h 101-109 |g 9 | ||
| author_variant |
q w qw |
|---|---|
| matchkey_str |
wangqihaoyongshengchenfeng:2021----:epnntedagrtmokolderpraoighognu |
| hierarchy_sort_str |
2021transfer abstract |
| bklnumber |
35.70 42.12 |
| publishDate |
2021 |
| allfields |
10.1016/j.neucom.2020.12.040 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica (DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 DE-627 ger DE-627 rakwb eng 570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Wang, Qi verfasserin aut Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture 2021transfer abstract 9 nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier Hao, Yongsheng oth Chen, Feng oth Enthalten in Elsevier Liu, Yang ELSEVIER The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast 2018 an international journal Amsterdam (DE-627)ELV002603926 volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 https://doi.org/10.1016/j.neucom.2020.12.040 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA 35.70 Biochemie: Allgemeines VZ 42.12 Biophysik VZ AR 429 2021 14 0314 101-109 9 |
| spelling |
10.1016/j.neucom.2020.12.040 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica (DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 DE-627 ger DE-627 rakwb eng 570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Wang, Qi verfasserin aut Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture 2021transfer abstract 9 nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier Hao, Yongsheng oth Chen, Feng oth Enthalten in Elsevier Liu, Yang ELSEVIER The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast 2018 an international journal Amsterdam (DE-627)ELV002603926 volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 https://doi.org/10.1016/j.neucom.2020.12.040 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA 35.70 Biochemie: Allgemeines VZ 42.12 Biophysik VZ AR 429 2021 14 0314 101-109 9 |
| allfields_unstemmed |
10.1016/j.neucom.2020.12.040 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica (DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 DE-627 ger DE-627 rakwb eng 570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Wang, Qi verfasserin aut Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture 2021transfer abstract 9 nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier Hao, Yongsheng oth Chen, Feng oth Enthalten in Elsevier Liu, Yang ELSEVIER The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast 2018 an international journal Amsterdam (DE-627)ELV002603926 volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 https://doi.org/10.1016/j.neucom.2020.12.040 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA 35.70 Biochemie: Allgemeines VZ 42.12 Biophysik VZ AR 429 2021 14 0314 101-109 9 |
| allfieldsGer |
10.1016/j.neucom.2020.12.040 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica (DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 DE-627 ger DE-627 rakwb eng 570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Wang, Qi verfasserin aut Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture 2021transfer abstract 9 nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier Hao, Yongsheng oth Chen, Feng oth Enthalten in Elsevier Liu, Yang ELSEVIER The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast 2018 an international journal Amsterdam (DE-627)ELV002603926 volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 https://doi.org/10.1016/j.neucom.2020.12.040 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA 35.70 Biochemie: Allgemeines VZ 42.12 Biophysik VZ AR 429 2021 14 0314 101-109 9 |
| allfieldsSound |
10.1016/j.neucom.2020.12.040 doi /cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica (DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 DE-627 ger DE-627 rakwb eng 570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Wang, Qi verfasserin aut Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture 2021transfer abstract 9 nicht spezifiziert zzz rdacontent nicht spezifiziert z rdamedia nicht spezifiziert zu rdacarrier Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier Hao, Yongsheng oth Chen, Feng oth Enthalten in Elsevier Liu, Yang ELSEVIER The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast 2018 an international journal Amsterdam (DE-627)ELV002603926 volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 https://doi.org/10.1016/j.neucom.2020.12.040 Volltext GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA 35.70 Biochemie: Allgemeines VZ 42.12 Biophysik VZ AR 429 2021 14 0314 101-109 9 |
| language |
English |
| source |
Enthalten in The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast Amsterdam volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 |
| sourceStr |
Enthalten in The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast Amsterdam volume:429 year:2021 day:14 month:03 pages:101-109 extent:9 |
| format_phy_str_mv |
Article |
| bklname |
Biochemie: Allgemeines Biophysik |
| institution |
findex.gbv.de |
| topic_facet |
Reinforcement learning IDA* algorithm Markov decision process Neural network Knowledge graph |
| dewey-raw |
570 |
| isfreeaccess_bool |
false |
| container_title |
The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |
| authorswithroles_txt_mv |
Wang, Qi @@aut@@ Hao, Yongsheng @@oth@@ Chen, Feng @@oth@@ |
| publishDateDaySort_date |
2021-01-14T00:00:00Z |
| hierarchy_top_id |
ELV002603926 |
| dewey-sort |
3570 |
| id |
ELV052908410 |
| language_de |
englisch |
| fullrecord |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">ELV052908410</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230626033938.0</controlfield><controlfield tag="007">cr uuu---uuuuu</controlfield><controlfield tag="008">210910s2021 xx |||||o 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1016/j.neucom.2020.12.040</subfield><subfield code="2">doi</subfield></datafield><datafield tag="028" ind1="5" ind2="2"><subfield code="a">/cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)ELV052908410</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(ELSEVIER)S0925-2312(20)31940-8</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">570</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">BIODIV</subfield><subfield code="q">DE-30</subfield><subfield code="2">fid</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">35.70</subfield><subfield code="2">bkl</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">42.12</subfield><subfield code="2">bkl</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Wang, Qi</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2021transfer abstract</subfield></datafield><datafield tag="300" ind1=" " ind2=" "><subfield code="a">9</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zzz</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">z</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zu</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework.</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework.</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Reinforcement learning</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">IDA* algorithm</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Markov decision process</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Neural network</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Knowledge graph</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Hao, Yongsheng</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Chen, Feng</subfield><subfield code="4">oth</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="n">Elsevier</subfield><subfield code="a">Liu, Yang ELSEVIER</subfield><subfield code="t">The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast</subfield><subfield code="d">2018</subfield><subfield code="d">an international journal</subfield><subfield code="g">Amsterdam</subfield><subfield code="w">(DE-627)ELV002603926</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:429</subfield><subfield code="g">year:2021</subfield><subfield code="g">day:14</subfield><subfield code="g">month:03</subfield><subfield code="g">pages:101-109</subfield><subfield code="g">extent:9</subfield></datafield><datafield tag="856" ind1="4" ind2="0"><subfield code="u">https://doi.org/10.1016/j.neucom.2020.12.040</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ELV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">FID-BIODIV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PHA</subfield></datafield><datafield tag="936" ind1="b" ind2="k"><subfield code="a">35.70</subfield><subfield code="j">Biochemie: Allgemeines</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="936" ind1="b" ind2="k"><subfield code="a">42.12</subfield><subfield code="j">Biophysik</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">429</subfield><subfield code="j">2021</subfield><subfield code="b">14</subfield><subfield code="c">0314</subfield><subfield code="h">101-109</subfield><subfield code="g">9</subfield></datafield></record></collection>
|
| author |
Wang, Qi |
| spellingShingle |
Wang, Qi ddc 570 fid BIODIV bkl 35.70 bkl 42.12 Elsevier Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| authorStr |
Wang, Qi |
| ppnlink_with_tag_str_mv |
@@773@@(DE-627)ELV002603926 |
| format |
electronic Article |
| dewey-ones |
570 - Life sciences; biology |
| delete_txt_mv |
keep |
| author_role |
aut |
| collection |
elsevier |
| remote_str |
true |
| illustrated |
Not Illustrated |
| topic_title |
570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph Elsevier |
| topic |
ddc 570 fid BIODIV bkl 35.70 bkl 42.12 Elsevier Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph |
| topic_unstemmed |
ddc 570 fid BIODIV bkl 35.70 bkl 42.12 Elsevier Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph |
| topic_browse |
ddc 570 fid BIODIV bkl 35.70 bkl 42.12 Elsevier Reinforcement learning Elsevier IDA* algorithm Elsevier Markov decision process Elsevier Neural network Elsevier Knowledge graph |
| format_facet |
Elektronische Aufsätze Aufsätze Elektronische Ressource |
| format_main_str_mv |
Text Zeitschrift/Artikel |
| carriertype_str_mv |
zu |
| author2_variant |
y h yh f c fc |
| hierarchy_parent_title |
The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |
| hierarchy_parent_id |
ELV002603926 |
| dewey-tens |
570 - Life sciences; biology |
| hierarchy_top_title |
The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |
| isfreeaccess_txt |
false |
| familylinks_str_mv |
(DE-627)ELV002603926 |
| title |
Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| ctrlnum |
(DE-627)ELV052908410 (ELSEVIER)S0925-2312(20)31940-8 |
| title_full |
Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| author_sort |
Wang, Qi |
| journal |
The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |
| journalStr |
The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast |
| lang_code |
eng |
| isOA_bool |
false |
| dewey-hundreds |
500 - Science |
| recordtype |
marc |
| publishDateSort |
2021 |
| contenttype_str_mv |
zzz |
| container_start_page |
101 |
| author_browse |
Wang, Qi |
| container_volume |
429 |
| physical |
9 |
| class |
570 VZ BIODIV DE-30 fid 35.70 bkl 42.12 bkl |
| format_se |
Elektronische Aufsätze |
| author-letter |
Wang, Qi |
| doi_str_mv |
10.1016/j.neucom.2020.12.040 |
| dewey-full |
570 |
| title_sort |
deepening the ida* algorithm for knowledge graph reasoning through neural network architecture |
| title_auth |
Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| abstract |
Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. |
| abstractGer |
Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. |
| abstract_unstemmed |
Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework. |
| collection_details |
GBV_USEFLAG_U GBV_ELV SYSFLAG_U FID-BIODIV SSG-OLC-PHA |
| title_short |
Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture |
| url |
https://doi.org/10.1016/j.neucom.2020.12.040 |
| remote_bool |
true |
| author2 |
Hao, Yongsheng Chen, Feng |
| author2Str |
Hao, Yongsheng Chen, Feng |
| ppnlink |
ELV002603926 |
| mediatype_str_mv |
z |
| isOA_txt |
false |
| hochschulschrift_bool |
false |
| author2_role |
oth oth |
| doi_str |
10.1016/j.neucom.2020.12.040 |
| up_date |
2024-07-06T17:29:49.636Z |
| _version_ |
1803851650445606912 |
| fullrecord_marcxml |
<?xml version="1.0" encoding="UTF-8"?><collection xmlns="http://www.loc.gov/MARC21/slim"><record><leader>01000caa a22002652 4500</leader><controlfield tag="001">ELV052908410</controlfield><controlfield tag="003">DE-627</controlfield><controlfield tag="005">20230626033938.0</controlfield><controlfield tag="007">cr uuu---uuuuu</controlfield><controlfield tag="008">210910s2021 xx |||||o 00| ||eng c</controlfield><datafield tag="024" ind1="7" ind2=" "><subfield code="a">10.1016/j.neucom.2020.12.040</subfield><subfield code="2">doi</subfield></datafield><datafield tag="028" ind1="5" ind2="2"><subfield code="a">/cbs_pica/cbs_olc/import_discovery/elsevier/einzuspielen/GBV00000000001282.pica</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(DE-627)ELV052908410</subfield></datafield><datafield tag="035" ind1=" " ind2=" "><subfield code="a">(ELSEVIER)S0925-2312(20)31940-8</subfield></datafield><datafield tag="040" ind1=" " ind2=" "><subfield code="a">DE-627</subfield><subfield code="b">ger</subfield><subfield code="c">DE-627</subfield><subfield code="e">rakwb</subfield></datafield><datafield tag="041" ind1=" " ind2=" "><subfield code="a">eng</subfield></datafield><datafield tag="082" ind1="0" ind2="4"><subfield code="a">570</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">BIODIV</subfield><subfield code="q">DE-30</subfield><subfield code="2">fid</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">35.70</subfield><subfield code="2">bkl</subfield></datafield><datafield tag="084" ind1=" " ind2=" "><subfield code="a">42.12</subfield><subfield code="2">bkl</subfield></datafield><datafield tag="100" ind1="1" ind2=" "><subfield code="a">Wang, Qi</subfield><subfield code="e">verfasserin</subfield><subfield code="4">aut</subfield></datafield><datafield tag="245" ind1="1" ind2="0"><subfield code="a">Deepening the IDA* algorithm for knowledge graph reasoning through neural network architecture</subfield></datafield><datafield tag="264" ind1=" " ind2="1"><subfield code="c">2021transfer abstract</subfield></datafield><datafield tag="300" ind1=" " ind2=" "><subfield code="a">9</subfield></datafield><datafield tag="336" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zzz</subfield><subfield code="2">rdacontent</subfield></datafield><datafield tag="337" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">z</subfield><subfield code="2">rdamedia</subfield></datafield><datafield tag="338" ind1=" " ind2=" "><subfield code="a">nicht spezifiziert</subfield><subfield code="b">zu</subfield><subfield code="2">rdacarrier</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework.</subfield></datafield><datafield tag="520" ind1=" " ind2=" "><subfield code="a">Inferring missing links in Knowledge Graphs (KGs) is a key evaluation task for KG reasoning, which aims to find relations for a given entity pair. Existing research often employs the IDA* (Iterative Deepening A*) algorithm for the path discovery task owing to its efficiency and accuracy. However, it relies on heuristics to set cost functions and is also difficult to utilize useful context information in the search process. In this paper, we propose the Deep-IDA* framework which applies neural networks and reinforcement learning (RL) to empower the IDA* algorithm to tackle the path discovery problem in KG reasoning. We model KG reasoning as a Markov Decision Process (MDP) and divide our Deep-IDA* framework and the resulting path into two parts: path-finding and path-reasoning. For path-finding, we propose a policy network to model the cost from the source to a candidate location. In this process, we employ the GCN (Graph Convolutional Network) to embed the observable sub-track, then employ the LSTM (Long Short-Term Memory) to record the historical trajectory, and introduce the attention to utilize the context information, and finally form policy. For path-reasoning with the searched candidate paths passed from the former process, we employ a value network to estimate the cost from the candidate to the destination entity, using the GNN (Graph Neural Networks) to learn a message-passing algorithm that solves the path inference problem, and using the GRU (Gated Recurrent Unit) to update the historical information. Finally, the actor-learner algorithm is utilized to minimize the sum of the losses of the two parts. Experiment results on three datasets demonstrate the effectiveness and efficiency of our framework.</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Reinforcement learning</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">IDA* algorithm</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Markov decision process</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Neural network</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="650" ind1=" " ind2="7"><subfield code="a">Knowledge graph</subfield><subfield code="2">Elsevier</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Hao, Yongsheng</subfield><subfield code="4">oth</subfield></datafield><datafield tag="700" ind1="1" ind2=" "><subfield code="a">Chen, Feng</subfield><subfield code="4">oth</subfield></datafield><datafield tag="773" ind1="0" ind2="8"><subfield code="i">Enthalten in</subfield><subfield code="n">Elsevier</subfield><subfield code="a">Liu, Yang ELSEVIER</subfield><subfield code="t">The TORC1 signaling pathway regulates respiration-induced mitophagy in yeast</subfield><subfield code="d">2018</subfield><subfield code="d">an international journal</subfield><subfield code="g">Amsterdam</subfield><subfield code="w">(DE-627)ELV002603926</subfield></datafield><datafield tag="773" ind1="1" ind2="8"><subfield code="g">volume:429</subfield><subfield code="g">year:2021</subfield><subfield code="g">day:14</subfield><subfield code="g">month:03</subfield><subfield code="g">pages:101-109</subfield><subfield code="g">extent:9</subfield></datafield><datafield tag="856" ind1="4" ind2="0"><subfield code="u">https://doi.org/10.1016/j.neucom.2020.12.040</subfield><subfield code="3">Volltext</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_USEFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">GBV_ELV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SYSFLAG_U</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">FID-BIODIV</subfield></datafield><datafield tag="912" ind1=" " ind2=" "><subfield code="a">SSG-OLC-PHA</subfield></datafield><datafield tag="936" ind1="b" ind2="k"><subfield code="a">35.70</subfield><subfield code="j">Biochemie: Allgemeines</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="936" ind1="b" ind2="k"><subfield code="a">42.12</subfield><subfield code="j">Biophysik</subfield><subfield code="q">VZ</subfield></datafield><datafield tag="951" ind1=" " ind2=" "><subfield code="a">AR</subfield></datafield><datafield tag="952" ind1=" " ind2=" "><subfield code="d">429</subfield><subfield code="j">2021</subfield><subfield code="b">14</subfield><subfield code="c">0314</subfield><subfield code="h">101-109</subfield><subfield code="g">9</subfield></datafield></record></collection>
|
| score |
7.3982077 |