The paper "Enhancing Complex Question Answering over Knowledge Graphs through Evidence Pattern Retrieval" by Wentao Ding addresses the challenge of improving information retrieval (IR) methods for knowledge graph question answering (KGQA). The authors propose Evidence Pattern Retrieval (EPR) to explicitly model structural dependencies among evidence facts during subgraph extraction, which is a critical step in IR-KGQA. EPR is implemented by indexing atomic adjacency patterns of resource pairs and performing dense retrieval to obtain candidate evidence patterns. These patterns are then scored using a neural model to select the best one for subgraph extraction. Experimental results on datasets such as ComplexWebQuestions and WebQuestionsSP demonstrate that EPR significantly improves the F1 scores of IR-KGQA methods by over 10 points, achieving competitive performance on WebQuestionsSP. The paper also discusses the importance of structural dependencies and the effectiveness of EPR in handling complex questions.The paper "Enhancing Complex Question Answering over Knowledge Graphs through Evidence Pattern Retrieval" by Wentao Ding addresses the challenge of improving information retrieval (IR) methods for knowledge graph question answering (KGQA). The authors propose Evidence Pattern Retrieval (EPR) to explicitly model structural dependencies among evidence facts during subgraph extraction, which is a critical step in IR-KGQA. EPR is implemented by indexing atomic adjacency patterns of resource pairs and performing dense retrieval to obtain candidate evidence patterns. These patterns are then scored using a neural model to select the best one for subgraph extraction. Experimental results on datasets such as ComplexWebQuestions and WebQuestionsSP demonstrate that EPR significantly improves the F1 scores of IR-KGQA methods by over 10 points, achieving competitive performance on WebQuestionsSP. The paper also discusses the importance of structural dependencies and the effectiveness of EPR in handling complex questions.