Qinghua Zhao

Qinghua Zhao, Shuai Ma, Yuxuan Lei

External knowledge,e.g., entities and entity descriptions, can help humans understand texts. Many works have been explored to include external knowledge in the pre-trained models. These methods, generally, design pre-training tasks and implicitly introduce knowledge by updating model weights, alternatively, use it straightforwardly together with the original text. Though effective, there are some limitations. On the one hand, it is implicit and only model weights are paid attention to, the pre-trained entity embeddings are ignored. On the other hand, entity descriptions may be lengthy, and inputting into the model together with the original text may distract the model's attention. This paper aims to explicitly include both entities and entity descriptions in the fine-tuning stage. First, the pre-trained entity embeddings are fused with the original text representation and updated by the backbone model layer by layer. Second, descriptions are represented by the knowledge module outside the backbone model, and each knowledge layer is selectively connected to one backbone layer for fusing. Third, two knowledge-related auxiliary tasks, i.e., entity/description enhancement and entity enhancement/pollution task, are designed to smooth the semantic gaps among evolved representations. We conducted experiments on four knowledge-oriented tasks and two common tasks, and the results achieved new state-of-the-art on several datasets. Besides, we conduct an ablation study to show that each module in our method is necessary. The code is available at https://github.com/lshowway/Ered.

Bo Li, Qinghua Zhao, Lijie Wen

Previous works have evaluated memorization by comparing model outputs with training corpora, examining how factors such as data duplication, model size, and prompt length influence memorization. However, analyzing these extensive training corpora is highly time-consuming. To address this challenge, this paper proposes an innovative approach named ROME that bypasses direct processing of the training data. Specifically, we select datasets categorized into three distinct types -- context-independent, conventional, and factual -- and redefine memorization as the ability to produce correct answers under these conditions. Our analysis then focuses on disparities between memorized and non-memorized samples by examining the logits and representations of generated texts. Experimental findings reveal that longer words are less likely to be memorized, higher confidence correlates with greater memorization, and representations of the same concepts are more similar across different contexts. Our code and data will be publicly available when the paper is accepted.

Qinghua Zhao, Xueling Gong, Xinyu Chen et al.

While critical for alignment, Supervised Fine-Tuning (SFT) incurs the risk of catastrophic forgetting, yet the layer-wise emergence of instruction-following capabilities remains elusive. We investigate this mechanism via a comprehensive analysis utilizing information-theoretic, geometric, and optimization metrics across model scales (1B-32B). Our experiments reveal a distinct depth-dependent pattern: middle layers (20\%-80\%) are stable, whereas final layers exhibit high sensitivity. Leveraging this insight, we propose Mid-Block Efficient Tuning, which selectively updates these critical intermediate layers. Empirically, our method outperforms standard LoRA up to 10.2\% on GSM8K (OLMo2-7B) with reduced parameter overhead, demonstrating that effective alignment is architecturally localized rather than distributed. The code is publicly available at https://anonymous.4open.science/r/base_sft.

Hao Yang, Qinghua Zhao, Lei Li

Chain-of-Thought (CoT) prompting significantly enhances model reasoning, yet its internal mechanisms remain poorly understood. We analyze CoT's operational principles by reversely tracing information flow across decoding, projection, and activation phases. Our quantitative analysis suggests that CoT may serve as a decoding space pruner, leveraging answer templates to guide output generation, with higher template adherence strongly correlating with improved performance. Furthermore, we surprisingly find that CoT modulates neuron engagement in a task-dependent manner: reducing neuron activation in open-domain tasks, yet increasing it in closed-domain scenarios. These findings offer a novel mechanistic interpretability framework and critical insights for enabling targeted CoT interventions to design more efficient and robust prompts. We released our code and data at https://anonymous.4open.science/r/cot-D247.

Qinghua Zhao, Vinit Ravishankar, Nicolas Garneau et al.

Word order is an important concept in natural language, and in this work, we study how word order affects the induction of world knowledge from raw text using language models. We use word analogies to probe for such knowledge. Specifically, in addition to the natural word order, we first respectively extract texts of six fixed word orders from five languages and then pretrain the language models on these texts. Finally, we analyze the experimental results of the fixed word orders on word analogies and show that i) certain fixed word orders consistently outperform or underperform others, though the specifics vary across languages, and ii) the Wov2Lex hypothesis is not hold in pre-trained language models, and the natural word order typically yields mediocre results. The source code will be made publicly available at https://github.com/lshowway/probing_by_analogy.

Qinghua Zhao, Shuai Ma, Shuo Ren

Though some recent works focus on injecting sentiment knowledge into pre-trained language models, they usually design mask and reconstruction tasks in the post-training phase. In this paper, we aim to benefit from sentiment knowledge in a lighter way. To achieve this goal, we study sentence-level sentiment analysis and, correspondingly, propose two sentiment-aware auxiliary tasks named sentiment word cloze and conditional sentiment prediction. The first task learns to select the correct sentiment words within the input, given the overall sentiment polarity as prior knowledge. On the contrary, the second task predicts the overall sentiment polarity given the sentiment polarity of the word as prior knowledge. In addition, two kinds of label combination methods are investigated to unify multiple types of labels in each task. We argue that more information can promote the models to learn more profound semantic representation. We implement it in a straightforward way to verify this hypothesis. The experimental results demonstrate that our approach consistently outperforms pre-trained models and is additive to existing knowledge-enhanced post-trained models. The code and data are released at https://github.com/lshowway/KESA.

Shan Yang, Yongfei Zhang, Guanglin Niu et al.

Few-shot relation extraction (FSRE) is of great importance in long-tail distribution problem, especially in special domain with low-resource data. Most existing FSRE algorithms fail to accurately classify the relations merely based on the information of the sentences together with the recognized entity pairs, due to limited samples and lack of knowledge. To address this problem, in this paper, we proposed a novel entity CONCEPT-enhanced FEw-shot Relation Extraction scheme (ConceptFERE), which introduces the inherent concepts of entities to provide clues for relation prediction and boost the relations classification performance. Firstly, a concept-sentence attention module is developed to select the most appropriate concept from multiple concepts of each entity by calculating the semantic similarity between sentences and concepts. Secondly, a self-attention based fusion module is presented to bridge the gap of concept embedding and sentence embedding from different semantic spaces. Extensive experiments on the FSRE benchmark dataset FewRel have demonstrated the effectiveness and the superiority of the proposed ConceptFERE scheme as compared to the state-of-the-art baselines. Code is available at https://github.com/LittleGuoKe/ConceptFERE.

Kaiyuan Zheng, Qinghua Zhao, Lei Li

The relationship between language and thought remains an unresolved philosophical issue. Existing viewpoints can be broadly categorized into two schools: one asserting their independence, and another arguing that language constrains thought. In the context of large language models, this debate raises a crucial question: Does a language model's grasp of semantic meaning depend on thought processes? To explore this issue, we investigate whether reasoning techniques can facilitate semantic understanding. Specifically, we conceptualize thought as reasoning, employ chain-of-thought prompting as a reasoning technique, and examine its impact on sentiment analysis tasks. The experiments show that chain-of-thought has a minimal impact on sentiment analysis tasks. Both the standard and chain-of-thought prompts focus on aspect terms rather than sentiment in the generated content. Furthermore, counterfactual experiments reveal that the model's handling of sentiment tasks primarily depends on information from demonstrations. The experimental results support the first viewpoint.

Qinghua Zhao, Jiaang Li, Lei Li et al.

Existing works have studied the impacts of the order of words within natural text. They usually analyze it by destroying the original order of words to create a scrambled sequence, and then comparing the models' performance between the original and scrambled sequences. The experimental results demonstrate marginal drops. Considering this findings, different hypothesis about word order is proposed, including ``the order of words is redundant with lexical semantics'', and ``models do not rely on word order''. In this paper, we revisit the aforementioned hypotheses by adding a order reconstruction perspective, and selecting datasets of different spectrum. Specifically, we first select four different datasets, and then design order reconstruction and continuing generation tasks. Empirical findings support that ChatGPT relies on word order to infer, but cannot support or negate the redundancy relations between word order lexical semantics.

Bo Li, Guanzhi Deng, Ronghao Chen et al.

Understanding how Large Language Models (LLMs) perform complex reasoning and their failure mechanisms is a challenge in interpretability research. To provide a measurable geometric analysis perspective, we define the concept of the Reasoning Manifold, a latent low-dimensional geometric structure formed by the internal representations corresponding to all correctly reasoned generations. This structure can be conceptualized as the embodiment of the effective thinking paths that the model has learned to successfully solve a given task. Based on this concept, we build REMA, a framework that explains the origins of failures by quantitatively comparing the spatial relationships of internal model representations corresponding to both erroneous and correct reasoning samples. Specifically, REMA first quantifies the geometric deviation of each erroneous representation by calculating its k-nearest neighbors distance to the approximated manifold formed by correct representations, thereby providing a unified failure signal. It then localizes the divergence points where these deviations first become significant by tracking this deviation metric across the model's layers and comparing it against a baseline of internal fluctuations from correct representations, thus identifying where the reasoning chain begins to go off-track. Our extensive experiments on diverse language and multimodal models and tasks demonstrate the low-dimensional nature of the reasoning manifold and the high separability between erroneous and correct reasoning representations. The results also validate the effectiveness of the REMA framework in analyzing the origins of reasoning failures. This research connects abstract reasoning failures to measurable geometric deviations in representations, providing new avenues for in-depth understanding and diagnosis of the internal computational processes of black-box models.

Jie Zhang, Qinghua Zhao, Chi-ho Lin et al.

Memorization in large language models poses critical risks for privacy and fairness as these systems scale to billions of parameters. While previous studies established correlations between memorization and factors like token frequency and repetition patterns, we revealed distinct response patterns: frequency increases minimally impact memorized samples (e.g. 0.09) while substantially affecting non-memorized samples (e.g., 0.25), with consistency observed across model scales. Through counterfactual analysis by perturbing sample prefixes and quantifying perturbation strength through token positional changes, we demonstrate that redundancy correlates with memorization patterns. Our findings establish that: about 79% of memorized samples are low-redundancy, these low-redundancy samples exhibit 2-fold higher vulnerability than high-redundancy ones, and consequently memorized samples drop by 0.6 under perturbation while non-memorized samples drop by only 0.01, indicating that more redundant content becomes both more memorable and more fragile. These findings suggest potential redundancy-guided approaches for data preprocessing, thereby reducing privacy risks and mitigating bias to ensure fairness in model deployments.

Yuki Hou, Haruki Tamoto, Qinghua Zhao et al.

Existing retrieval methods in Large Language Models show degradation in accuracy when handling temporally distributed conversations, primarily due to their reliance on simple similarity-based retrieval. Unlike existing memory retrieval methods that rely solely on semantic similarity, we propose SynapticRAG, which uniquely combines temporal association triggers with biologically-inspired synaptic propagation mechanisms. Our approach uses temporal association triggers and synaptic-like stimulus propagation to identify relevant dialogue histories. A dynamic leaky integrate-and-fire mechanism then selects the most contextually appropriate memories. Experiments on four datasets of English, Chinese and Japanese show that compared to state-of-the-art memory retrieval methods, SynapticRAG achieves consistent improvements across multiple metrics up to 14.66% points. This work bridges the gap between cognitive science and language model development, providing a new framework for memory management in conversational systems.

Qinghua Zhao, Shuai Ma

In this paper, we study sentiment analysis task where the outcomes are mainly contributed by a few key elements of the inputs. Motivated by the two-streams hypothesis, we propose a neural architecture, named TraceNet, to address this type of task. It not only learns discriminative representations for the target task via its encoders, but also traces key elements at the same time via its locators. In TraceNet, both encoders and locators are organized in a layer-wise manner, and a smoothness regularization is employed between adjacent encoder-locator combinations. Moreover, a sparsity constraints are enforced on locators for tracing purposes and items are proactively masked according to the item weights output by locators.A major advantage of TraceNet is that the outcomes are easier to understand, since the most responsible parts of inputs are identified. Also, under the guidance of locators, it is more robust to attacks due to its focus on key elements and the proactive masking training strategy. Experimental results show its effectiveness for sentiment classification. Moreover, we provide several case studies to demonstrate its robustness and interpretability.

Qinghua Zhao

News recommendation models often fall short in capturing users' preferences due to their static approach to user-news interactions. To address this limitation, we present a novel dynamic news recommender model that seamlessly integrates continuous time information to a hierarchical attention network that effectively represents news information at the sentence, element, and sequence levels. Moreover, we introduce a dynamic negative sampling method to optimize users' implicit feedback. To validate our model's effectiveness, we conduct extensive experiments on three real-world datasets. The results demonstrate the effectiveness of our proposed approach.