To address these issues（解决）
These issues largely diminish the performance and trustworthiness of LLMs （降低）
Our model is mainly divided into three parts, namely Topic Entity Recog-
nition, Relation Recognition and Answer Selection. The overall model is
shown in Figure 1 （划分，命名）
planning optimization, which distills the knowledge from KGs into LLMs to generate faithful
relation paths as plans;（提炼）
where E and R denote the set of entities and relations, respectively.（各自代表）
We simply fine-tune bert model to measure how well an entity matches a question.（微调）
By treating relation paths as plans, we can make sure the plans are grounded by KGs（基于）
Can the planning module of RoG be integrated with other LLMs to improve their performance?（融为一体）
After calculating the above scores, we normalize and add them, the entity
with the highest score is selected as the topic entity.（归一化）
in the latter task our results significantly outperform earlier methods.（优于）
In this paper, we propose a novel method called reasoning on graphs (RoG) that *synergizes LLMs with KGs to conduct faithful and interpretable reasoning.（统筹）
we can find that ChatGPT+CoT suffers from the lack of knowledge issue（遭受）
To tackle the issues,（解决）
To further unleash LLMs’ reasoning ability （释放）
To utilize the instruction-following ability of LLMs（利用）
In the former task, our results are comparable to state-of-the-art; （相当的）
What’s especially striking is their ability to handle complex tasks （突破性的）
DECAF (Yu et al., 2022a) combines semantic parsing and LLMs reasoning to jointly generate answers, which also reach salient performance on KGQA tasks （突出的）
The underlying technology is intended to enhance decision-making.（基础的）
Finally, we build a classier to judge whether a question is simple or complicated （分类器）
Topic entity is the core entity of the corresponding KB query（核心）
To address the issues of hallucinations and lack of knowledge, we present a planning-retrieval-reasoning framework （框架）
plan-and-solve paradigm （范例）
Moreover, the planning module of RoG can be plug-and-play with different LLMs during inference to improve their performance. （即插即用）
These two models have their own preponderance in entity recognition, so we decide to combine the results of both（优势）
Firstly, we construct a scoring mechanism to select the core entity of the question （分数选择器）
We also evaluate the transferability of RoG to other KGs. （迁移性）
as shown in Figure
On the contrary, RoG can generate faithful relation paths and retrieve valid reasoning paths from KGs for reasoning.
Previous works that jointly use KGs and LLMs for KGQA reasoning can be broadly divided into two categories:
In this paper, we propose a novel method called reasoning on graphs (RoG) that synergizes LLMs with KGs to conduct faithful and interpretable reasoning.
We conduct extensive experiments on two benchmark KGQA datasets, and the results demonstrate that RoG achieves state-of-the-art performance on KG reasoning tasks
In this section, we introduce our method: reasoning on graphs (RoG).
We will discuss the implementation details of these two tasks in the following subsections.
We evaluate the reasoning ability of RoG on two benchmark KGQA datasets:
We compare RoG with 21 baselines grouping into 5 categories: 1) Embedding-
based methods, 2) Retrieval-augmented methods, 3) Semantic parsing methods, 4) LLMs, and 5) LLMs+KGs methods. The details of each baseline are described in Appendix A.4
Following previous works, we use Hits@1 and F1 as the evaluation metrics
Since UniKGQA (Jiang et al., 2022) and DECAF (Yu et al., 2022a) are state-of-the-art methods, we directly refer their results and those of the other baselines reported in their papers for comparison
From the results, it is evident that
We conduct an ablation study to analyze the effectiveness of the planning module
and reasoning module in our method (RoG). We compare four variant