Lujia Yang, Weicai Yan, Yongbo He, Qifei Zhang, Tao Jin, Jinshan Zhang, Meng Xi, Jianwei Yin. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

A service ecosystem refers to a multilateral network composed of heterogeneous service entities, where the exchange of data, resources, and value through interactions among specific participants forms a service pattern. As service ecosystems like virtual hospital alliance (VHA) evolve towards large-scale, multi-domain integration to meet complex user needs, service pattern fusion has emerged as a fundamental approach to leverage data, resources, and value aggregation. By converging elements from multiple patterns, service pattern fusion enables the fulfillment of composite business objectives with reduced redundancy and lower costs. Existing works primarily address fusion requirements by reorganizing existing services through approaches such as service composition and business process management where only service functions and workflows are considered. However, they lack formalization of pattern fusion constraints and fail to support comprehensive integration of participants, data, resources, and value, let alone identifying optimal fusion solutions that account for participant collaboration and service integration. In this study, we formally define the Service Pattern Fusion Problem (SPFP) as an optimization task aimed at identifying the most efficient and cost-effective pattern by integrating, combining, and pruning elements from multiple patterns while preserving their objectives and meeting business constraints. We adapt traditional heuristic methods to SPFP and propose the Fusion-Oriented Confidence-Aware genetic algorithm (FoCa). FoCa dynamically adjusts the search space and transition probabilities in each iteration, achieving optimal fusion results with a 41.09% reduction in pattern loss and the fastest convergence. In addition, we designed a set of pattern features and conducted random fusion experiments on the public service pattern dataset S-SPD, to explore the correlation between those features and the optimization magnitude across various metrics. The analysis helps identify which types of service patterns benefit most from fusion, providing valuable insights for researchers and practitioners in both academic and engineering contexts.

Portfolio optimization is one of the most important financial problem, suffering from huge computational pressure due to arithmetic complexity. Quantum computing offers polynomial or even exponential speedup that turns out to be a promising approach. However, existing quantum methods is fundamentally limited by either poor scalability or insufficient accuracy. In this paper, we propose SAPO, which formally articulates the quantum circuit that seamlessly integrates financial theory and historical data characteristics with quantum algebra. The circuit design is extended from the HHL algorithm incorporating mean-variance theory, which promotes scalability by equivalent transformation. Then, we present a min-max eigenvalue model that leverages historical financial information to refine parameter settings with high accuracy. Experiments conducted on market data demonstrate that SAPO can effectively reduce the complexity by $\mathbf{3 6. 9 4 \%}$ compared to basic HHL [1], [2] and improve the accuracy by $1.46 \times$ compared to hybrid HHL [3].

Query-based pricing enables personalized data acquisition for data buyers, exhibiting potential in data markets. The state-of-the-art SQL query pricing strategy tackles the #P-hard arbitrage-free pricing task with the quadratic computational complexity, far from promptly fulfilling customer demands. In this paper, we propose a novel arbitrage-free and scalable pricing framework ARIA to calculate the prices for various query types in linear time, including select-project-join and simple aggregate (SPJA) queries. For the first time, we model what the query answer tells about the value of each tuple and formulate the tuple-level information of selection, projection, and simple aggregation queries. We develop several price functions based on the total information gain of all tuples. The containing relationship between the query information prevents possible arbitrage arising from query determinacy. We present efficient price computation algorithms to derive the prices of different types of queries with linear time complexity, which scan the common possible value set of tuples one time. In ARIA, the join query is decomposed as multiple single-relation queries for pricing in linear time. Extensive experiments on real and synthetic datasets demonstrate that, ARIA performs 3x faster than the state of the arts while enjoying desirable pricing characteristics.

Recently, multimodal recommendation (MMRec) has received much attention, which models user preferences based on both user behaviors and modality information. Although current graph neural network based methods yield notable results in MMRec, certain limitations persist among these methods. 1) Most methods rely on pre-trained networks to extract modality features but fail to remove modality noise. 2) Recent methods leverage InfoNCE strategy to align representation, while ignoring the effect of feature redundancy and lacking sufficient alignment between different modality features. Such limitations ultimately harm the recommendation performance. To this end, we propose a Dual Mutual Information-Driven Multimodal Recommendation Model with Denoising Graph Autoencoder (DMIGA). Specifically, to reduce the noise within modality features, we design a denoising graph autoencoder with a cross-modal consistency constraint. Furthermore, we propose a dual mutual information learning mechanism on both feature and instance levels, to reduce the feature redundancy and align different representations. Experimental results on three real-world datasets consistently demonstrate that DMIGA outperforms state-of-the-art methods, with an average of 3.8% improvement.

With the vast range of available services, it has become an important challenge to recommend the optimal service for mashup developer. Recent studies are mainly limited by the service similarity, resulting in challenges such as discrepancy in textual semantics, implicity of inter-service relationships, and the sparsity of historical interactions. Service bundles, which offer a set of services, present a novel approach to address the mashup development problem. In this work, we propose an innovative message-passing model, a Heterogeneous Graph AutoEncoderbased service bundle recommendation model (HGAE), to tackle the issues. Specifically, we introduce the Graph Propagation Module to encode potentially implicit semantic relations in the Mashup-Service-Bundle heterogeneous graph. Furthermore, we build a unified representation for the bundle in the Bundle Prediction Module by combining an autoencoder and spatial attention mechanism, enabling the integration of relationships across different node and edge types. Extensive experiments on real-world datasets demonstrate that HGAE notably outperforms state-of-the-art methods on all metrics, with improvements of 8.69% in NDCG and 9.55% in Recall on the ProgrammableWeb dataset.

Portfolio selection is a critical task in finance, involving the allocation of resources across various assets. However, current methods often struggle to maintain robust performance due to the inherent low signal-to-noise ratio in raw financial data and shifts in data distribution. We propose AlphaGAT, a novel two-stage learning approach for portfolio selection, designed to adapt to different market scenarios. Inspired by the concept of alpha factors, which transform historical market data into actionable signals, the first stage introduces an advanced model named CATimeMixer for alpha factor generation with a novel loss function to improve the effectiveness and robustness. CATimeMixer integrates TimeMixer with Conv1D (C) and cross-asset Attention (A). Specifically, Conv1D enhances TimeMixer by capturing trend and seasonal features across different scales, while cross-asset attention enables TimeMixer to extract interrelationships between different assets. The second stage applies reinforcement learning to dynamically adjust weights, integrating alpha factors into trading signals. Recognizing the varying effectiveness of alpha factors across different periods, our RL agent innovatively transforms the alpha factors into graphs and employs graph attention networks (GAT) to discern the significance of different alpha factors, enhancing policy robustness. Extensive experiments on real-world market data show that our approach outperforms state-of-the-art methods.

Portrait animation aims to transfer the facial expressions and movements of a target character onto a reference character. This task presents two main challenges: accurately transferring motion and expressions while fully preserving the identity features of the reference portrait. We introduce Vividportraits, a diffusion-based model designed to effectively meet these objectives. In contrast to existing methods that rely on sparse representations such as facial landmarks, our approach leverages facial parsing maps for motion guidance, enabling a more precise conveyance of subtle expressions. A random scaling technique is applied during training to prevent the model from internalizing identity-specific features from the driving images. Furthermore, we perform foreground-background segmentation on the reference portrait to reduce data redundancy. The long-video generation process is refined to improve consistency across sequences. Our model, exclusively trained on public datasets, demonstrates superior performance relative to current state-of-the-art methods, achieving a notable 8% improvement in expression metric. More visual results are available on the anonymous website https://www.vividportraits.cn.

Recommendation is a key mechanism for modern users to access items of their interests from massive entities and information. Recently, graph contrastive learning (GCL) has demonstrated satisfactory results on recommendation, due to its ability to enhance representation by integrating graph neural networks (GNNs) with contrastive learning. However, those methods often generate contrastive views by performing random perturbation on edges or embeddings, which is likely to bring noise in representation learning. Besides, in all these methods, the degree of user preference on items is omitted during the representation learning process, which may cause incomplete user/item modeling. To address these limitations, we propose the Adaptive Fusion of Multi-View Graph Contrastive Recommendation (AMGCR) model. Specifically, to generate the informative and less noisy views for better contrastive learning, we design four view generators to learn the edge weights focusing on weight adjustment, feature transformation, neighbor aggregation, and attention mechanism, respectively. Then, we employ an adaptive multi-view fusion module to combine different views from both the view-shared and the view-specific levels. Moreover, to make the model capable of capturing preference information during the learning process, we further adopt a preference refinement strategy on the fused contrastive view. Experimental results on three real-world datasets demonstrate that AMGCR consistently outperforms the state-of-the-art methods, with average improvements of over 10% in terms of Recall and NDCG. Our code is available on https://github.com/Du-danger/AMGCR.

The multi-view data with incomplete information hinder effective data analysis. Existing multi-view imputation methods, which learn the mapping between a complete view and acompletely missingview, are not able to deal with the typical multi-view data withmissing featureinformation. In this paper, we propose a unified generative imputation model named UGit with optimal transport theory to simultaneously impute the missing features/values of all incomplete views. This imputation is conditional onallthe observed values from the multi-view data. UGit consists of two modules, i.e., aunified multi-view generator(UMG) and amasking energy discriminator(MED). To effectively and efficiently impute missing features across all views, the generator UMG employs aunified autoencoderin conjunction with thecross-view attention mechanismto learn the data distribution from all observed multi-view data. The discriminator MED leverages a novelmasking energydivergence function to make UGit differentiable for imputation accuracy enhancement. Extensive experiments on several real-world multi-view data sets demonstrate that, UGit speeds up the model training by 4.28x with more than 41% accuracy gain on average, compared to the state-of-the-art approaches.

Consider an undirected graph G=(V,E) model for a communication network, where each edge is owned by a selfish agent, who reports the cost for offering the use of her edge. Note that each edge agent may misreport her own cost for the use of the edge for her own benefit. In such a non-cooperative setting, we aim at designing an approximately truthful mechanism for establishing a Steiner tree, a minimum cost tree spanning over all the terminals. We present a truthful-in-expectation mechanism that achieves the approximation ratio ln 4 + ε ≈ 1.39, which matches the current best algorithmic ratio for STP.

A/B testing, or controlled experiments, is the gold standard approach to causally compare the performance of algorithms on online platforms. However, conventional Bernoulli randomization in A/B testing faces many challenges such as spillover and carryover effects. Our study focuses on another challenge, especially for A/B testing on two-sided platforms – budget constraints. Buyers on two-sided platforms often have limited budgets, where the conventional A/B testing may be infeasible to be applied, partly because two variants of allocation algorithms may conflict and lead some buyers to exceed their budgets if they are implemented simultaneously. We develop a model to describe two-sided platforms where buyers have limited budgets. We then provide an optimal experimental design that guarantees small bias and minimum variance. Bias is lower when there is more budget and a higher supply-demand rate. We test our experimental design on both synthetic data and real-world data, which verifies the theoretical results and shows our advantage compared to Bernoulli randomization.

The problem of pollution control has been mainly studied in the environmental economics literature where the methodology of game theory is applied for the pollution control. To the best of our knowledge this is the first time this problem is studied from the computational point of view. We introduce a new network model for pollution control and present two applications of this model. On a high level, our model comprises a graph whose nodes represent the agents, which can be thought of as the sources of pollution in the network. The edges between agents represent the effect of spread of pollution. The government who is the regulator, is responsible for the maximization of the social welfare and sets bounds on the levels of emitted pollution in both local areas as well as globally in the whole network. We first prove that the above optimization problem is NP-hard even on some special cases of graphs such as trees. We then turn our attention on the classes of trees and planar graphs which model realistic scenarios of the emitted pollution in water and air, respectively. We derive approximation algorithms for these two kinds of networks and provide deterministic truthful and truthful in expectation mechanisms. In some settings of the problem that we study, we achieve the best possible approximation results under standard complexity theoretic assumptions. Our approximation algorithm on planar graphs is obtained by a novel decomposition technique to deal with constraints on vertices. We note that no known planar decomposition techniques can be used here and our technique can be of independent interest. For trees we design a two level dynamic programming approach to obtain an FPTAS. This approach is crucial to deal with the global pollution quota constraint. It uses a special multiple choice, multi-dimensional knapsack problem where coefficients of all constraints except one are bounded by a polynomial of the input size. We furthermore derive truthful in expectation mechanisms on general networks with bounded degree.

We study the House Allocation problem (also known as the Assignment problem), i.e., the problem of allocating a set of objects among a set of agents, where each agent has ordinal preferences (possibly involving ties) over a subset of the objects. We focus on truthful mechanisms without monetary transfers for finding large Pareto optimal matchings. It is straightforward to show that no deterministic truthful mechanism can approximate a maximum cardinality Pareto optimal matching with ratio better than 2. We thus consider randomised mechanisms. We give a natural and explicit extension of the classical Random Serial Dictatorship Mechanism (RSDM) specifically for the House Allocation problem where preference lists can include ties. We thus obtain a universally truthful randomised mechanism for finding a Pareto optimal matching and show that it achieves an approximation ratio of $$\frac{e}{e-1}$$ . The same bound holds even when agents have priorities (weights) and our goal is to find a maximum weight (as opposed to maximum cardinality) Pareto optimal matching. On the other hand we give a lower bound of $$\frac{18}{13}$$ on the approximation ratio of any universally truthful Pareto optimal mechanism in settings with strict preferences. By using a characterisation result of Bade, we show that any randomised mechanism that is a symmetrisation of a truthful, non-bossy and Pareto optimal mechanism has an improved lower bound of $$\frac{e}{e-1}$$ . Since our new mechanism is a symmetrisation of RSDM for strict preferences, it follows that this lower bound is tight. We moreover interpret our problem in terms of the classical secretary problem and prove that our mechanism provides the best randomised strategy of the administrator who interviews the applicants.

Classical online bipartite matching problem and its generalizations are central algorithmic optimization problems. The second related line of research is in the area of algorithmic mechanism design, referring to the broad class of house allocation or assignment problems. We introduce a single framework that unifies and generalizes these two streams of models. Our generalizations allow for arbitrary matroid constraints or knapsack constraints at every object in the allocation problem. We design and analyze approximation algorithms and truthful mechanisms for this framework. Our algorithms have best possible approximation guarantees for most of the special instantiations of this framework, and are strong generalizations of the previous known results.

We consider the problem of revenue maximization on multi‐unit auctions where items are distinguished by their relative values; any pair of items has the same ratio of values to all buyers. As is common in the study of revenue maximizing problems, we assume that buyers' valuations are drawn from public known distributions and they have additive valuations for multiple items. Our problem is well motivated by sponsored search auctions, which made money for Google and Yahoo! in practice. In this auction, each advertiser bids an amount bi to compete for ad slots on a web page. The value of each ad slot corresponds to its click‐through‐rate, and each buyer has her own per‐click valuations, which is her private information. Obviously, a strategic bidder may bid an amount that is different with her true valuation to improve her utility. Our goal is to design truthful mechanisms avoiding this misreporting. We develop the optimal (with maximum revenue) truthful auction for a relaxed demand model (where each buyer i wants at most di items) and a sharp demand model (where buyer i wants exactly di items). We also find an auction that always guarantees at least half of the revenue of the optimal auction when the buyers are budget constrained. Moreover, all of the auctions we design can be computed efficiently, that is, in polynomial time.

We study the House Allocation problem (also known as the Assignment problem), i.e., the problem of allocating a set of objects among a set of agents, where each agent has ordinal preferences (possibly involving ties) over a subset of the objects. We focus on truthful mechanisms without monetary transfers for finding large Pareto optimal matchings. It is straightforward to show that no deterministic truthful mechanism can approximate a maximum cardinality Pareto optimal matching with ratio better than 2. We thus consider randomized mechanisms. We give a natural and explicit extension of the classical Random Serial Dictatorship Mechanism (RSDM) specifically for the House Allocation problem where preference lists can include ties. We thus obtain a universally truthful randomized mechanism for finding a Pareto optimal matching and show that it achieves an approximation ratio of eovere-1. The same bound holds even when agents have priorities (weights) and our goal is to find a maximum weight (as opposed to maximum cardinality) Pareto optimal matching. On the other hand we give a lower bound of 18 over 13 on the approximation ratio of any universally truthful Pareto optimal mechanism in settings with strict preferences. In the case that the mechanism must additionally be non-bossy, an improved lower bound of eovere-1 holds. This lower bound is tight given that RSDM for strict preference lists is non-bossy. We moreover interpret our problem in terms of the classical secretary problem and prove that our mechanism provides the best randomized strategy of the administrator who interviews the applicants.