In contrast, the stepwise reasoning strategy performs reasoning path searching according to the similarity matching of relation embeddings during the answer retrieval process. Moreover, the incomplete answer set also results in inaccurate performance evaluation. (2) For questions with multiple answer entities, random selection of the target entity causes oscillation and slow convergence during the training phase. A false reasoning path may be generated even when the model reaches the correct answer entity. It is unclear whether the model has correctly interpreted the semantic meaning of the question. (1) The reasoning process is not interpretable. Many existing models , use the final answer entity as supervision for the training of representation and embeddings matching, which has the following two major drawbacks. Compared with single-hop KGQA that only performs triplet matching once, the multi-hop KGQA task is much more challenging due to its larger search space and implicit order of multi-relation reasoning.Īmong various multi-hop KGQA methods, embedding-based methods have become the mainstream solution due to their powerful representation learning ability. Reasoning from the topic entity Seylvia Brett (i.e., the red node) to the answer entity Greenwich (i.e., the green node) requires reasoning through two fact triplets, which are ( Seylvia Brett, spouse, Charles Vyner Brooke) and then ( Charles Vyner Brooke, place_of_birth, Greenwich). Depending on the length of the reasoning path, KGQA tasks are divided into two categories: (1) single-hop KGQA, which retrieves the answer entity in only one hop of reasoning (2) multi-hop KGQA, in which multiple hops of reasoning are required to reach the answer entity.Īn example of a knowledge graph with a 2-hop question is presented in Fig. 1. The traversed reasoning path may consist of one or multiple fact triplets. The answer entity to the question can be retrieved by searching a reasoning path from the topic entity in the knowledge graph. Generally speaking, the KGQA process begins with identifying the topic entity of the question. The head_entity and tail_entity are two entity nodes, while relation is a directed edge representing the relation type from head_entity to tail_entity. In knowledge graphs, a fact triplet is often represented as ( head_entity, relation, tail_entity). Its core concern lies in understanding natural language problems and searching for corresponding factual answer entities through semantic similarity evaluation. The KGQA task aims to retrieve the answer entity to a natural language question from the knowledge graph, enabling users to conveniently access the knowledge without any concern for the internal graph structure. Compared with traditional question answering tasks, KGQA has the advantages of better semantic understanding and higher answer accuracy. With the construction of large-scale knowledge graphs, such as Freebase and DBpedia , question answering over knowledge graph (KGQA) has attracted widespread attention ,. High-quality knowledge graphs facilitate the storage, organization, and access processes of human knowledge and intelligence. Knowledge graph organizes various concepts in the real world as entities along with their intricate relation. The results demonstrate that the major modules of UMRNet are effective, and UMRNet outperforms the state-of-the-art methods regarding both accuracy and efficiency. Extensive ablation studies and comparative experiments have been conducted on four KGQA benchmark datasets. ![]() Furthermore, to avoid the need for an extra hop of reasoning, we propose a non-delayed termination detection mechanism that performs effective evaluation of the remaining reasoning information based on history attention. Specifically, the proposed dynamic update strategy of question embeddings based on our attention redistribution mechanism is capable of handling the mapping problems. Therefore, in this paper, we address these two issues of delayed determination and mapping problems by proposing an Unrestricted Multi-Hop Reasoning Network for Interpretable KGQA named UMRNet. In addition, they suffer from two mapping problems between the question and relations: (1) one-to-many mapping when an individual question word corresponds to multiple hops of reasoning (2) many-to-one mapping when a single hop of reasoning corresponds to multiple question words. For the more challenging task of multi-hop KGQA, existing methods either address fixed-length multi-hop reasoning, or perform a delayed detection of termination that requires an extra hop of reasoning. The knowledge graph based question answering (KGQA) task returns accurate answer entities instead of keyword matches. Knowledge graphs significantly boost the answer retrieval quality for natural language questions.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |