Qimin Zhong, Hao Liao, Haiming Qin, Mingyang Zhou, Rui Mao, Wei Chen, Naipeng Chao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Knowledge tracing aims to dynamically track and assess learners’ mastery of specific knowledge concepts (e.g., item taxonomies, skill hierarchies). Integrating it into recommender systems greatly enhances model explainability, yet current models either overemphasize temporal dynamics or overlook inter-concept spatial correlations, resulting in suboptimal performance in integrated spatiotemporal modeling. Such disjointed designs also lead to limited adaptability in long learning sequences, poor compatibility with real-world dynamic scenarios, and restricted model interpretability. To solve these problems, we propose an A-SGNN framework fusing higher-order path spiking and graph neural networks, which leverages Graph Convolutional Network (GCN) and Spiking Neural Network (SNN) modules to make up for existing shortcomings. Specifically, a Bidirectional GCN module fully captures bidirectional high-order spatial graph structures between concepts, accurately modeling multi-knowledge relationships and reducing information loss. An SNN module with adaptive path-finding strategy dynamically optimizes individual learning trajectories, overcoming traditional fixed-path rigidity, while its membrane potential decay simulates human forgetting. A time decay factor is integrated to better capture memory effects in learning. Experiments demonstrate that our method outperforms state-of-the-art approaches in predicting learners’ future academic performance, with this advantage deriving from the combined synergistic benefits of the GCN and SNN modules. Our implementation is publicly available at: https://github.com/jianruichen/A-SGNN .

Jianlyu Chen, Nan Wang, Chaofan Li, Bo Wang, Shitao Xiao, Han Xiao, Hao Liao, Defu Lian, Zheng Liu. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Yuhong Dai, Jianxun Lian, Yitian Huang, Wei Zhang, Mingyang Zhou, Mingqi Wu, Xing Xie, Hao Liao. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Role-playing capabilities in large language models (LLMs) often lack cognitive consistency in complex scenarios that require deep understanding and coherent reasoning.While recent reasoning models excel in math and coding tasks, they show limited effectiveness in open-ended role-playing scenarios.We introduce R-CHAR (Role-Consistent Hierarchical Adaptive Reasoning), a metacognition-driven framework that enhances role-playing performance through guided thinking trajectories synthesis and adaptive evaluation.Our approach demonstrates that concise thinking processes can achieve superior performance efficiently compared to elaborate reasoning chains in roleplaying social intelligence tasks, outperforming existing specialized models.Experimental results on the SocialBench benchmark show significant and stable performance improvements across varying scenario complexities, showing particular strength in long-context comprehension (from 34.64% to 68.59%) and grouplevel social interactions.Our work advances the development of cognitively consistent roleplaying systems, bridging the gap between surface-level mimicry and authentic character simulation.

A novel dynamic network model is proposed to capture evolving latent communities within temporal networks. To achieve this, we decompose each observed dynamic edge between vertices using a Poisson-gamma edge partition model, assigning each vertex to one or more latent communities through nonnegative vertex-community memberships. Specifically, hierarchical transition kernels are employed to model the interactions between these latent communities. A hierarchical graph prior is placed on the transition structure of the latent communities, allowing us to model how they evolve and interact over time. Consequently, our dynamic network model enables the inferred community structure to merge, split, and interact with one another, providing a comprehensive understanding of complex network dynamics. Experiments on various real-world network datasets demonstrate that the proposed model not only effectively uncovers interpretable latent structures but also surpasses other state-of-the-art dynamic network models in the tasks of link prediction and community detection.

Network connectivity minimization is a fundamental problem in controlling the spread of viruses in the Internet and facilitating information propagation in online social networks. The problem aims to identify a budget number of key nodes whose removal would minimize the connectivity of a network. However, the existing solutions heavily rely on the number of edges, making it challenging to handle large and densely connected social networks. In this study, we present a fast algorithm that is independent of the number of edges. To achieve this, we first introduce a surrogate matrix that approximates the residual adjacency matrix with arbitrary small predefined error. We then devise an efficient approach for inferring k influential nodes by optimizing the eigenvalues of the surrogate matrix. Remarkably, the algorithm has a small time complexity of O(knr3), with r being a small tunable number. Our algorithm thereby maintains a linear scalability in terms of the number of nodes and is unaffected by the number of edges. Hence, it has the capability to efficiently handle large and dense social networks. At last, we evaluate its performance against state-of-the-art techniques using diverse real-world datasets. The experimental results demonstrate the superiority of our proposed method in terms of both solution quality and computational efficiency.

Large language models (LLMs) provide powerful foundations to perform fine-grained text re-ranking. However, they are often prohibitive in reality due to constraints on computation bandwidth. In this work, we propose a flexible architecture called Matroyshka Re-Ranker, which is designed to facilitate runtime customization of model layers and sequence lengths at each layer based on users' configurations. Consequently, the LLM-based re-rankers can be made applicable across various real-world situations. The increased flexibility may come at the cost of precision loss. To address this problem, we introduce a suite of techniques to optimize the performance. First, we propose cascaded self-distillation, where each sub-architecture learns to preserve a precise re-ranking performance from its super components, whose predictions can be exploited as smooth and informative teacher signals. Second, we design a factorized compensation mechanism, where two collaborative LoRA modules, vertical and horizontal, are jointly employed to compensate for the precision loss resulted from arbitrary combinations of layer and sequence compression. We perform comprehensive experiments using passage and document retrieval datasets from MSMARCO, along with all public datasets from BEIR. In our experiments, Matryoshka Re-Ranker substantially outperforms existing methods, while effectively preserving its superior performance across various compression forms and application scenarios. We have publicly released our method at this https://github.com/FlagOpen/FlagEmbedding repo.

Explainable recommender systems ( ERS ) aim to enhance users’ trust in the systems by offering personalized recommendations with transparent explanations. This transparency provides users with a clear understanding of the rationale behind the recommendations, fostering a sense of confidence and reliability in the system’s outputs. Generally, the explanations are presented in a familiar and intuitive way, which is in the form of natural language, thus enhancing their accessibility to users. Recently, there has been an increasing focus on leveraging reviews as a valuable source of rich information in both modeling user-item preferences and generating textual interpretations, which can be performed simultaneously in a multi-task framework. Despite the progress made in these review-based recommendation systems, the integration of implicit feedback derived from user-item interactions and user-written text reviews has yet to be fully explored. To fill this gap, we propose a model named SERMON (A s pect-enhanced E xplainable R ecommendation with M ulti-modal C o ntrast Lear n ing). Our model explores the application of multimodal contrastive learning to facilitate reciprocal learning across two modalities, thereby enhancing the modeling of user preferences. Moreover, our model incorporates the aspect information extracted from the review, which provides two significant enhancements to our tasks. Firstly, the quality of the generated explanations is improved by incorporating the aspect characteristics into the explanations generated by a pre-trained model with controlled textual generation ability. Secondly, the commonly used user-item interactions are transformed into user-item-aspect interactions, which we refer to as interaction triple, resulting in a more nuanced representation of user preference. To validate the effectiveness of our model, we conduct extensive experiments on three real-world datasets. The experimental results show that our model outperforms state-of-the-art baselines, with a 2.0% improvement in prediction accuracy and a substantial 24.5% enhancement in explanation quality for the TripAdvisor dataset.

Wensheng Lu, Jianxun Lian, Wei Zhang, Guanghua Li, Mingyang Zhou, Hao Liao, Xing Xie. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

The influence maximization (IM) problem aims to identify a budgeted set of nodes with the highest potential to influence the largest number of users in a cascade model, a key challenge in viral marketing. Traditional \emph{IM} approaches consider each user/node independently as a potential target customer. However, in many scenarios, the target customers comprise motifs, where activating only one or a few users within a motif is insufficient for effective viral marketing, which, nevertheless, receives little attention. For instance, if a motif of three friends planning to dine together, targeting all three simultaneously is crucial for a restaurant advertisement to succeed. In this paper, we address the motif-oriented influence maximization problem under the linear threshold model. We prove that the motif-oriented IM problem is NP-hard and that the influence function is neither supermodular nor submodular, in contrast to the classical \emph{IM} setting. To simplify the problem, we establish the submodular upper and lower bounds for the influence function. By leveraging the submodular property, we propose a natural greedy strategy that simultaneously maximizes both bounds. Our algorithm has an approximation ratio of $\tau\cdot (1-1/e-\varepsilon)$ and a near-linear time complexity of $O((k+l)(m+\eta)\log \eta/\varepsilon^2)$. Experimental results on diverse datasets confirm the effectiveness of our approach in motif maximization.

Federated learning enables collaborative AI training across organizations without compromising data privacy. Decentralized federated learning (DFL) improves this by offering enhanced reliability and security through peer-to-peer (P2P) model sharing. However, DFL faces challenges in terms of slow convergence rate due to complex P2P graphs. To address this issue, we propose an efficient algorithm to accelerate DFL by introducing a limited number of k of edges into the P2P graphs. Specifically, we establish a connection between the convergence rate and the second smallest eigenvalue of the laplacian matrix of the P2P graph. We prove that finding the optimal set of edges to maximize this eigenvalue is an NP-complete problem. Our quantitative analysis shows the positive effect of strategic edge additions on improving this eigenvalue. Based on the analysis, we then propose an efficient algorithm to compute the best set of candidate edges to maximize the second smallest eigenvalue, and consequently the convergence rate is maximized. Our algorithm has a low time complexity of O(krn^2). Experimental results on diverse datasets validate the effectiveness of our proposed algorithms in accelerating DFL convergence.

Recommendation systems guide users in locating their desired information within extensive content repositories. Usually, a recommendation model is optimized to enhance accuracy metrics from a user utility standpoint, such as click-through rate or matching relevance. However, a responsible industrial recommendation model must address not only user utility (responsibility to users) but also other objectives, including increasing platform revenue (responsibility to platforms), ensuring fairness (responsibility to content creators), and maintaining unbiasedness (responsibility to long-term healthy development). Multi-objective learning is a promising approach for achieving responsible recommendation models. Nevertheless, current methods encounter two challenges: difficulty in scaling to heterogeneous objectives within a unified framework, and inadequate controllability over objective priority during optimization, leading to uncontrollable solutions.

Hao Cheng, Shuo Wang, Wensheng Lu, Wei Zhang, Mingyang Zhou, Kezhong Lu, Hao Liao. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

In this paper, we propose a novel approach to enhance user and restaurant representations in the context of predicting the closure of a restaurant and give an explanation based on data generated from user-restaurant interactions. In order to accurately predict the operating status of a restaurant and give a reasonable explanation, we need rich relevant information of user-restaurant interaction and reviews to model their representations. However, the interaction information between the user and the restaurant is usually sparse. To address this issue, we propose a new model, which is called the Co-Attentive Contrastive Learning (CACL) model. Our model employs a contrastive learning algorithm to deal with data sparsity and leverages co-attention mechanism to select the most relevant review information assisting the former to obtain a more accurate and granular representation. By fusing these module, we obtain rich information to perform two tasks with better quality. To demonstrate the effectiveness of our model, extensive experiments was conducted in six different cities, and the results showed that CACL was superior to the previous method in terms of prediction accuracy and explainable ability (average 4.8% improvement in prediction and 35.4% improvement in explanation).

The ease of spreading false information online enables individuals with malicious intent to manipulate public opinion and destabilize social stability. Recently, fake news detection based on evidence retrieval has gained popularity in an effort to identify fake news reliably and reduce its impact. Evidence retrieval-based methods can improve the reliability of fake news detection by computing the textual consistency between the evidence and the claim in the news. In this paper, we propose a framework for fake news detection based on MUlti- Step Evidence Retrieval enhancement (MUSER), which simulates the steps of human beings in the process of reading news, summarizing, consulting materials, and inferring whether the news is true or fake. Our model can explicitly model dependencies among multiple pieces of evidence, and perform multi-step associations for the evidence required for news verification through multi-step retrieval. In addition, our model is able to automatically collect existing evidence through paragraph retrieval and key evidence selection, which can save the tedious process of manual evidence collection. We conducted extensive experiments on real-world datasets in different languages, and the results demonstrate that our proposed model outperforms state-of-the-art baseline methods for detecting fake news by at least 3% in F1-Macro and 4% in F1-Micro. Furthermore, it provides interpretable evidence for end users.

In this paper, we present an algorithmic study on how to surpass competitors in popularity by strategic promotions in social networks. We first propose a novel model, in which we integrate the Preferential Attachment (PA) model for popularity growth with the Independent Cascade (IC) model for influence propagation in social networks called PA-IC model. In PA-IC, a popular item and a novice item grab shares of popularity from the natural popularity growth via the PA model, while the novice item tries to gain extra popularity via influence cascade in a social network. The popularity ratio is defined as the ratio of the popularity measure between the novice item and the popular item. We formulate Popularity Ratio Maximization (PRM) as the problem of selecting seeds in multiple rounds to maximize the popularity ratio in the end. We analyze the popularity ratio and show that it is monotone but not submodular. To provide an effective solution, we devise a surrogate objective function and show that empirically it is very close to the original objective function while theoretically, it is monotone and submodular. We design two efficient algorithms, one for the overlapping influence and non-overlapping seeds (across rounds) setting and the other for the non-overlapping influence and overlapping seed setting, and further discuss how to deal with other models and problem variants. Our empirical evaluation further demonstrates that our proposed method consistently achieves the best popularity promotion compared to other methods. Our theoretical and empirical analyses shed light on the interplay between influence maximization and preferential attachment in social networks.

The detection of fake news often requires sophisticated reasoning skills, such as logically combining information by considering word-level subtle clues. In this paper, we move towards fine-grained reasoning for fake news detection by better reflecting the logical processes of human thinking and enabling the modeling of subtle clues. In particular, we propose a fine-grained reasoning framework by following the human’s information-processing model, introduce a mutual-reinforcement-based method for incorporating human knowledge about which evidence is more important, and design a prior-aware bi-channel kernel graph network to model subtle differences between pieces of evidence. Extensive experiments show that our model outperforms the state-of-the-art methods and demonstrate the explainability of our approach.

Networks are one of the most powerful structures for modeling problems in the real world. Many machine learning algorithms, however, require that each input example is a real vector. Network embedding learns from feature representations of nodes and links in a network, and converts it to vectors. Community structure is an important feature of the network, which represents the relationship among nodes and attracts the attention of relevant researchers. Many algorithms have been developed to identify the community structure. These algorithms usually identify different communities in the network, generating different types of information. In this paper, we propose a "Community Splitter" model based on random walk and RNN (Recurrent Neural Networks) that combines the node information generated by multiple community detection algorithms to improve node representation and link prediction. Extensive experiments on nine real datasets demonstrate that our proposed Community Splitter model has a significant prediction power compared to state-of-the-art link prediction models.

The bloom of the Internet and the recent breakthroughs in deep learning techniques open a new door to AI for E-commence, with a trend evolved from using a few financial factors such as liquidity and profitability to using more advanced AI techniques to process complex and multi-modal data.In this paper, we tackle the practical problem of restaurant survival prediction.We argue that traditional methods ignore two essential aspects, which are very helpful for the task: 1) modeling customer reviews and 2) jointly considering status prediction and result explanation.Thus, we propose a novel joint learning framework for explainable restaurant survival prediction based on the multi-modal data of user-restaurant interactions and users' textual reviews.Moreover, we design a graph neural network to capture the high-order interactions and design a co-attention mechanism to capture the most informative and meaningful signal from noisy textual reviews.Our results on two datasets show a significant and consistent improvement over the SOTA techniques (average 6.8% improvement in prediction and 45.3% improvement in explanation).

Vertical federated learning (VFL) is a privacy-preserving machine learning paradigm that can learn models from features distributed on different platforms in a privacy-preserving way. Since in real-world applications the data may contain bias on fairness-sensitive features (e.g., gender), VFL models may inherit bias from training data and become unfair for some user groups. However, existing fair machine learning methods usually rely on the centralized storage of fairness-sensitive features to achieve model fairness, which are usually inapplicable in federated scenarios. In this paper, we propose a fair vertical federated learning framework (FairVFL), which can improve the fairness of VFL models. The core idea of FairVFL is to learn unified and fair representations of samples based on the decentralized feature fields in a privacy-preserving way. Specifically, each platform with fairness-insensitive features first learns local data representations from local features. Then, these local representations are uploaded to a server and aggregated into a unified representation for the target task. In order to learn a fair unified representation, we send it to each platform storing fairness-sensitive features and apply adversarial learning to remove bias from the unified representation inherited from the biased data. Moreover, for protecting user privacy, we further propose a contrastive adversarial learning method to remove private information from the unified representation in server before sending it to the platforms keeping fairness-sensitive features. Experiments on three real-world datasets validate that our method can effectively improve model fairness with user privacy well-protected.

Nowadays, with the development of web services, the survival of online service has captured public attention significantly. Future prediction of business is crucial to the shopkeepers. Predicting how the business will develop and explaining what key factors are leading to it are two most important tasks. Existing literatures usually tackle only one of these two tasks, ignoring that they are two closely related tasks and complement each other. In this paper, we propose a review-based neural model, named Deep Concept-aware Model (DCA), to predict restaurants' future status and provide explainable sentences simultaneously in an end2end framework. Specifically, we use co-attention to select concepts and implement prediction and explanation tasks through Factorization Machine and Gated Recurrent Unit, respectively. We conduct extensive experiments on three Chinese cities' datasets. The proposed joint model outperforms the state-of-the-art baseline methods for both prediction (average 40.96% improvement in AUC) and explanation (average 9.72% improvement in BLEU and 86.05% in Precision metric of ROUGE).

In the past two decades, analyzing the information network has been intensively studied from various disciplines. Small world property and scale-free property prevail in network science research. The comparison and classification of different kinds of graphs are extremely important. However, how to design a robust and accurate classification with deep learning techniques for network property still lack enough attention, which is a vital task in various application scenarios. In this paper, we proposed the reliable network property classifier based on graph embedding(RNC) to classify the network property (scale free or small world property). In order to process non-euclidean data, we embedded each network into an image and use dimensional reduction, rasterization, and convolutional neural networks to complete the classification problem. The method can effectively accomplish classification tasks in not only artificial networks but also real networks. Besides, RNC wins greatly in terms of robustness on real networks, showing the robustness of RNC against the incomplete structure of the network.

With the notion of virtual servers, peers participating in a heterogeneous, structured peer-to-peer (P2P) network may host different numbers of virtual servers, and by migrating virtual servers, peers can balance their loads proportional to their capacities. The existing and decentralized load balance algorithms designed for the heterogeneous, structured P2P networks either explicitly construct auxiliary networks to manipulate global information or implicitly demand the P2P substrates organized in a hierarchical fashion. Without relying on any auxiliary networks and independent of the geometry of the P2P substrates, we present, in this paper, a novel load balancing algorithm that is unique in that each participating peer is based on the partial knowledge of the system to estimate the probability distributions of the capacities of peers and the loads of virtual servers, resulting in imperfect knowledge of the system state. With the imperfect system state, peers can compute their expected loads and reallocate their loads in parallel. Through extensive simulations, we compare our proposal to prior load balancing algorithms.

Unstructured peer-to-peer (P2P) file-sharing networks are popular in the mass market. As the peers participating in unstructured networks interconnect randomly, they rely on flooding query messages to discover objects of interest. Empirical measurement studies indicate that the peers in P2P networks have similar preferences, and recently proposed unstructured P2P networks intend to organize the participating peers in a small-world (SW) fashion by exploiting the knowledge of contents stored in peers. As existing algorithms for constructing SW-based unstructured P2P networks may not precisely reveal the object sharing patterns, the resultant networks thus may not perform searches efficiently and effectively by exploiting the common interests among peers. In this paper, we suggest a novel P2P network formation algorithm to construct SW-based unstructured networks. We validate our proposal in simulations with an empirical data set, and the simulation results prove that our proposal greatly outperforms existing algorithms in terms of search efficiency and effectiveness.

In an unstructured peer-to-peer (P2P) network (e.g., Gnutella), participating peers choose their neighbors randomly such that the resultant P2P network mismatches its underlying physical network, resulting in the lengthy communication between the peers and redundant network traffics generated in the underlying network. Previous solutions to the topology mismatch problem in the literature either have no performance guarantees or are far from the optimum. In this paper, we propose a novel topology matching algorithm based on the Metropolis-Hastings method. Our proposal is guided by our insight analytical model and is close to the optimal design. Specifically, we show that our proposal constructs an unstructured P2P network in which a broadcast message, originated by any node v, reaches any other node u by taking approximately the only physical end-to-end delay between v and u. In addition, our design guarantees the exponential broadcast scope. We verify our solution through extensive simulations and show that our proposal considerably outperforms state-of-the-art solutions.

In an unstructured peer-to-peer (P2P) network (e.g., Gnutella), participating peers choose their neighbors randomly such that the resultant P2P network mismatches its underlying physical network, resulting in the lengthy communication between the peers and redundant network traffics generated in the underlying network. Previous solutions to the topology-mismatch problem in the literature are far from the optimum. In this paper, we propose a novel topology-matching algorithm based on the Metropolis-Hastings method. Through extensive simulations, we show that our proposal constructs an unstructured P2P network where a broadcast message, originated by any node v, reaches any other node u by taking approximately the only physical end-to-end delay between v and u.

Prior studies show that more than 70 percent of communication paths in a popular unstructured peer-to-peer (P2P) system (i.e., Gnutella) do not exploit the physical network topology, leading to the topology mismatch problem, and thus, lengthen communication between participating peers. While previous efforts in solving overlay topology matching problems do not guarantee the bounds of performance metrics (e.g., the communication delay between any two overlay peers and the broadcasting scope of any participating peer), in this paper, we present a novel topology matching algorithm that has provable performance qualities. In our proposal, each participating node creates and manages a constant number of overlay connections to other peers in a distributed manner. In rigorous performance analysis, we show that 1) the expected overlay communication delay between any two nodes in our P2P network is a constant; 2) in addition, any joining node has the exponential broadcasting scope in expectation; 3) furthermore, a participating node takes a polylogarithmic overhead to exploit the physical network locality and maintain its flooding scope. Together with extensive simulations, we present our proposal that significantly outperforms two recent solutions, i.e., THANCS and mOverlay, in terms of overlay communication latency and/or broadcasting scope.

Small-world (SW) networks possess two properties, namely low diameter and high clustering coefficient, which are often desired by large-scale peer-to-peer networks. Prior studies have shown that the construction of an SW network can be based on a d-regular graph, and each node in the graph maintains d local neighbors and a small constant number of long-distance contacts. However, it is commonly understood that it is difficult to construct a short route in an SW network, given source (s) and target (t) nodes, though an SW network guarantees that a short route from s to t exists. Prior work in [1] proposed a navigable SW network for a d-dimensional lattice such that a simple localized routing algorithm can be devised to route a message from s to t using O(log2,X) hops, where X is the number of nodes in the network. In this paper, we present a novel navigable SW network based on a hierarchical model. Compared to previous efforts, the novelty of our study presents 1) that our network construction based on a hierarchical model is decentralized, 2) that routing a message between any two nodes in our SW network takes logarithmic hopcount in expectation, 3) that our SW network has high cluster coefficient, and 4) that the performance of our proposal is mathematically provable. We support the performance of our proposal in this study through rigorous, thorough performance analysis and extensive simulations.

Virtual servers provide an abstraction layer to exploit the heterogeneity of peers (or nodes) that participate in a peer-to-peer (P2P) network based on the distributed hash table (DHT). In this paper, we present a load-balanced DHT network based on virtual servers. In our proposal, participating peers accept loads proportional to their capacities. By sampling the capacities of peers and workloads of virtual servers in the system, our design approximates the probability distributions for the capacities and workloads. Each node is based on the probability distributions to match their local virtual servers and other peers. Our proposal is developed through rigorous performance analysis, and is evaluated in extensive simulations.