To alleviate the hallucination problem of large language model (LLM), retrieval-augmented generation (RAG) has been proposed and widely adopted.Due to the limitations in cross-chunk summarization task of naive RAG, graph-based RAG has emerged as a promising solution.However, a close study reveals several flaws in these works.First, most graphbased RAGs suffer from less efficient indexing process, which leads to information loss and expensive costs.Second, they heavily rely on LLM for retrieval thus inference slowly, which hinders their application in industry.To build a more efficient and effective RAG, we propose the multi-semantic RAG (MS-RAG).In this work, we combine knowledge graphs with dense vector to build a multi-semantic RAG.To be specific, (i) at indexing stage, we create multiple semantic-level indexes, including chunk-level, relation-level, and entity-level, to leverage the merits of dense vector and knowledge graph.(ii) at retrieval stage, unlike the previous LLM-empowered entity extraction, we propose a novel mix recall algorithm.Finally, we employ a multi-semantic rerank module to purify the results.Extensive experiments show that MS-RAG achieves superior performance.In terms of retrieval effect, MS-RAG achieves state-of-the-art performance, which is about 10%-30% improvement than the existing methods.In terms of question-answering effect, MS-RAG still achieves promising results with faster inference speed.More analysis and experiments are provided in Appendix.

In recent years, blockchain technology has received widespread attention. Smart contracts are programs that run on the blockchain, and their security faces serious challenges for blockchain applications. Inspired by the success of artificial intelligence technology, some smart contract vulnerability detection methods based on deep learning have been proposed and achieved meaningful progress. However, a closer look reveals three flaws in these works. First, some works simply use LSTM modules to perform sequence learning on pre-processing smart contracts, which lacks the ability to extract long-range features and does not support parallel processing of inputs. Secondly, smart contract vulnerability detection methods based on deep learning lack interpretability of results. From the perspective of system design, in order to solve these problems, we propose SmartDT, a more effective, faster, and interpretable smart contract vulnerability detection system. Specifically, (i) we propose an attention-based deep learning smart contract detection module, which is able to learn long-range dependencies in inputs and supports parallel processing of inputs. (ii) After the deep learning module, we stack an optional symbolic execution module for enhancing the interpretability of the classification results. Compared with general symbolic execution detection, our method can achieve faster detection because we can call a specific symbolic analysis model to detect input based on the classification results of the deep learning module. Extensive experiments demonstrate the effectiveness of our proposed method. Compared with other machine learningbased methods, our method achieves better performance and better interpretability, which demonstrates SmartDT’s outstanding feature learning capabilities. Compared with symbolic execution methods, our method achieves faster and more effective detection. In addition, we conducted ablation experiments to verify the effectiveness of each module.