A Tree-shaped Fuzzy Clustering Answer Retrieval Model Based on Question Alignment

Open domain question answering (QA) refers to the model that can retrieve multiple supporting documents related to the answer from comprehensive knowledge bases. The difficulties lie in discovering the semantic logic relationship between supporting documents and providing explainable reasoning processes. Current retrieval models rely heavily on the scale of parameters in text embedding and iterative optimization, resulting in high training costs and a lack of interpretability. Given the challenges outlined above, this article proposes an unsupervised tree-shaped axiomatic fuzzy set (AFS) clustering model with semantic framework, tailored for open-domain answer prediction. Unlike traditional rule-based methods that require repeated iterations leading to complex computational overhead, the hierarchical single-feature clustering model proposed in this article achieves high retrieval efficiency and semantic interpretability. Moreover, a retrieval strategy based on tree-structured clustering semantic descriptions for unsupervised question alignment reasoning paths is introduced, effectively enhancing the capacity for intricate reasoning of the model. The application of the AFS clustering theory to information retrieval with the single feature tree clustering method is original. Experimental results on three open domain QA datasets show the superiority of the proposed model.