Recent advancements in knowledge graph question answering (KGQA) have shown promise, yet existing methods often fail to align with human reasoning patterns that involve continuous reflection and refinement. This paper proposes FD-PORT (flow-guided direct preference optimization for knowledge graph reasoning with trees), a novel approach that combines Monte Carlo Tree Search (MCTS) with flow-guided direct preference optimization (FDPO) for KGQA tasks. MCTS simulates human-like reasoning by systematically exploring multiple inference paths in knowledge graphs, while FDPO transforms the search feedback into fine-grained training signals through flow balance conditions. Unlike traditional methods focusing on end-to-end training or sequence-level preferences, FD-PORT establishes flow consistency between any states along the reasoning chain, enabling robust multi-hop reasoning that adapts to local decisions and long-range dependencies. Experimental results on three benchmark datasets demonstrate that FD-PORT significantly outperforms state-of-the-art methods, achieving up to 50.6% improvements over GPT-4 on complex multi-hop reasoning tasks with a smaller open-source language model. The framework is advanced in maintaining diverse reasoning paths while ensuring answer quality, closely mirroring human problem-solving strategies.

Dimensional sentiment analysis has received considerable attention because it can represent affective states as continuous numerical values on multiple dimensions such as valence (positive–negative) and arousal (excited–calm). Compared to the categorical approach, which represents affective states as several discrete classes (e.g., positive and negative), the dimensional approach can provide more fine-grained (real-valued) sentiment analysis. Traditional approaches to predicting dimensional sentiment scores typically treat each dimension independently without consideration of relations between dimensions. In fact, different dimensions may correlate with each other. For example, expressions with a higher valence score usually have a higher arousal score, And higher irony expressions usually have a lower valence score. Such relations between dimensions are useful for dimension score prediction. To this end, this study proposes a multi-dimensional relation model to incorporate relations between dimensions into deep neural networks for dimension score prediction. The proposed method has two modes: internal and external. The internal mode incorporates the relations between dimensions into sentence representations before prediction, whereas the external mode builds a linear regression model that can capture the relations between dimensions to refine the predicted scores after prediction. To evaluate the proposed method, we created a Chinese three-dimensional corpus with valence-arousal-irony (VAI) ratings. Experiments using various neural network architectures demonstrate that the proposed multi-dimensional relation model outperformed those that treat each dimension independently. In addition, the internal mode outperformed the external mode, and a combination of the two modes achieved the best performance.