Currently,　most　knowledge　graph　question　answering　(KGQA)　systems　need　to　retrieve　all　entities　when　dealing　with　complex　problems　involving　multiple　entities　and　relationships,　which　results　in　high　time　complexity　and　resource　consumption.　The　response　time　of　KGQA　systems　is　an　important　indicator　for　evaluating　their　performance.　To　address　this　issue,　this　paper　proposes　a　Chinese　KGQA　system　based　on　dense　relationship　retrieval.　The　system　uses　the　Faiss　index　mechanism　for　vector　similarity　retrieval,　quickly　extracts　the　top　K　relationships　from　the　pre-built　vector　relationship　library,　and　constructs　paths　to　reduce　a　large　number　of　irrelevant　semantic　paths　in　most　KGQA　tasks,　thereby　improving　the　time　cost　and　computational　resource　overhead　caused　by　the　exponential　growth　of　path　numbers　in　second-order　and　higher-order　questions.　By　using　this　method,　the　response　time　of　the　QA　system　can　be　shortened　to　within　1　second　with　minimal　loss　of　accuracy.　This　paper　combines　pre-training　models　to　complete　tasks　such　as　text　semantic　similarity　and　entity　mention　recognition,　and　achieves　an　average　F1　value　of　71.3%　on　the　CCKS2019-CKBQA　test　set.　Comparing　with　other　systems　demonstrates　the　superiority　of　our　method　in　terms　of　accuracy　and　efficiency.　©　2023　IEEE.