Abstract
Structured representation of product information is a major bottleneck for the efficiency of e-commerce platforms, especially in second-hand ecommerce platforms. Currently, most product information are organized based on manually curated product categories and attributes, which often fail to adequately cover long-tail products and do not align well with buyer preference. To address these problems, we propose \textbf{G}enerative \textbf{S}emantic \textbf{I}n\textbf{D}exings (GSID), a data-driven approach to generate product structured representations. GSID consists of two key components: (1) Pre-training on unstructured product metadata to learn in-domain semantic embeddings, and (2) Generating more effective semantic codes tailored for downstream product-centric applications. Extensive experiments are conducted to validate the effectiveness of GSID, and it has been successfully deployed on the real-world e-commerce platform, achieving promising results on product understanding and other downstream tasks.

Abstract
Identifying attribute values from product profiles is a key task for improving product search, recommendation, and business analytics on e-commerce platforms, which we called Product Attribute Value Identification (PAVI) . However, existing PAVI methods face critical challenges, such as cascading errors, inability to handle out-of-distribution (OOD) attribute values, and lack of generalization capability. To address these limitations, we introduce Multi-Value-Product Retrieval-Augmented Generation (MVP-RAG), combining the strengths of retrieval, generation, and classification paradigms. MVP-RAG defines PAVI as a retrieval-generation task, where the product title description serves as the query, and products and attribute values act as the corpus. It first retrieves similar products of the same category and candidate attribute values, and then generates the standardized attribute values. The key advantages of this work are: (1) the proposal of a multi-level retrieval scheme, with products and attribute values as distinct hierarchical levels in PAVI domain (2) attribute value generation of large language model to significantly alleviate the OOD problem and (3) its successful deployment in a real-world industrial environment. Extensive experimental results demonstrate that MVP-RAG performs better than the state-of-the-art baselines.

Abstract
This paper develops a dynamical-systems framework for modeling influence propagation in product adoption networks, formulated as a positive linear system with Metzler interaction matrices and utility-based decay. Exact solutions are derived for constant, piecewise-constant, and fully time-varying interaction structures using matrix exponentials and the Peano--Baker series. It establishes five results: (i) positive interactions guarantee nonnegative amplification, (ii) perceived utility saturates after $\approx\!3$ complementary additions (Weber--Fechner), (iii) frequency of comparable introductions dominates incremental quality improvements, (iv) reinforcing interactions yields monotone gains while decay control gives ambiguous effects, and (v) long-run retention under SIS-type dynamics is bounded by the inverse spectral radius of the adoption graph. These results extend epidemic-threshold theory and positive-systems analysis to networked adoption, yielding explicit, calibratable expressions for influence dynamics on networks.

Abstract
The random dot product graph is a popular model for network data with extensions that accommodate dynamic (time-varying) networks. However, two significant deficiencies exist in the dynamic random dot product graph literature: (1) no coherent Bayesian way to update one's prior beliefs about the latent positions in dynamic random dot product graphs due to their complicated constraints, and (2) no approach to forecast future networks with meaningful uncertainty quantification. This work proposes a generalized Bayesian framework that addresses these needs using a Gibbs posterior that represents a coherent updating of Bayesian beliefs based on a least-squares loss function. We establish the consistency and contraction rate of this Gibbs posterior under commonly adopted Gaussian random walk priors. For estimation, we develop a fast Gibbs sampler with a time complexity for sampling the latent positions that is linear in the observed edges in the dynamic network, which is substantially faster than existing exact samplers. Simulations and an application to forecasting international conflicts show that the proposed method's in-sample and forecasting performance outperforms competitors.

AI Insights

• GB‑DASE marries Bayesian nonparametrics with optimal transport to capture evolving latent geometry.
• A profile‑likelihood routine selects the embedding dimension, yielding interpretable country trajectories.
• Edge‑probability estimates reveal shifting alliances, matching known historical conflict patterns.
• The Gibbs sampler’s linear‑edge complexity makes it scalable to thousands of nodes over many time steps.
• Posterior contraction rates under Gaussian random‑walk priors guarantee asymptotic consistency.
• Compared to variational and eigenmodel baselines, GB‑DASE achieves superior goodness‑of‑fit to degree distributions.
• Recommended reading: “Bayesian Nonparametrics: Principles and Practice” and “Optimal Bayesian estimation for random dot product graphs” for deeper theory.

Abstract
These notes were originally intended to be complementary material for an introductory course on self-similar graphs and their algebras, presented by the author at the CIMPA School ``K-theory and Operator Algebras'', held in La Plata and Buenos Aires (Argentina) from 28 July to 8 August 2025. In these notes, we introduce the concept of self-similar graph, associated with groups acting on graphs. We define the corresponding $C^*$-algebra using different complementary approaches, to understand its basic properties. We also analyze various generalizations that appear in the literature and, in particular, review the relationship of this construction with Zappa-Sz\'ep products. Finally, we present very recent results on homology and $K$-theory for these algebras.

AI Insights

• Self‑similar actions on groupoids reveal new homology patterns beyond graphs.
• A spectral‑sequence framework computes K‑theory for these C-algebras.
• The Katsura‑Exel‑Pardo groupoid’s homology is linked to classical groupoid cohomology.
• Zappa‑Szép products become twisted groupoid extensions, easing C-algebra analysis.
• Explicit homology calculations for finite‑graph self‑similar actions illustrate the theory.
• Results bridge operator‑algebraic invariants with combinatorial group theory, hinting at new invariants.
• The exposition invites exploration of topology, algebra, and dynamics interplay.

Abstract
Growing concerns surrounding AI safety and data privacy have driven the development of Machine Unlearning as a potential solution. However, current machine unlearning algorithms are designed to complement the offline training paradigm. The emergence of the Continual Learning (CL) paradigm promises incremental model updates, enabling models to learn new tasks sequentially. Naturally, some of those tasks may need to be unlearned to address safety or privacy concerns that might arise. We find that applying conventional unlearning algorithms in continual learning environments creates two critical problems: performance degradation on retained tasks and task relapse, where previously unlearned tasks resurface during subsequent learning. Furthermore, most unlearning algorithms require data to operate, which conflicts with CL's philosophy of discarding past data. A clear need arises for unlearning algorithms that are data-free and mindful of future learning. To that end, we propose UnCLe, an Unlearning framework for Continual Learning. UnCLe employs a hypernetwork that learns to generate task-specific network parameters, using task embeddings. Tasks are unlearned by aligning the corresponding generated network parameters with noise, without requiring any data. Empirical evaluations on several vision data sets demonstrate UnCLe's ability to sequentially perform multiple learning and unlearning operations with minimal disruption to previously acquired knowledge.

AI Insights

• UnCLe consistently beats state‑of‑the‑art baselines on CIFAR‑10, CIFAR‑100, and ImageNet‑subset, lifting accuracy by 3–5 %.
• Its hypernetwork generates task‑specific weights from compact embeddings, enabling zero‑data unlearning via noise alignment.
• The framework sustains performance across diverse architectures (ResNet‑18, VGG‑16) without task‑specific tuning.
• Empirical tests show negligible catastrophic forgetting even after 20 consecutive learn–unlearn cycles.
• Memory overhead stays below 12 % of the base model, though large‑scale nets still demand significant RAM.
• For deeper insights, consult “UnCLe: A Unified Framework for Continual Learning” and the survey on continual learning in deep nets.
• The method’s robustness makes it a promising candidate for privacy‑sensitive deployments in autonomous systems.

Abstract
Proceedings of the Seventh International Conference on Applied Category Theory, held at the University of Oxford on 17 - 21 June 2024. The contributions to ACT 2024 ranged from pure to applied and included contributions in a wide range of disciplines in science and engineering. ACT 2024 included talks in classical mechanics, quantum physics, probability theory, linguistics, decision theory, machine learning, epidemiology, thermodynamics, engineering, and logic.

Abstract
We introduce CoDe-KG, an open-source, end-to-end pipeline for extracting sentence-level knowledge graphs by combining robust coreference resolution with syntactic sentence decomposition. Using our model, we contribute a dataset of over 150,000 knowledge triples, which is open source. We also contribute a training corpus of 7248 rows for sentence complexity, 190 rows of gold human annotations for co-reference resolution using open source lung-cancer abstracts from PubMed, 900 rows of gold human annotations for sentence conversion policies, and 398 triples of gold human annotations. We systematically select optimal prompt-model pairs across five complexity categories, showing that hybrid chain-of-thought and few-shot prompting yields up to 99.8% exact-match accuracy on sentence simplification. On relation extraction (RE), our pipeline achieves 65.8% macro-F1 on REBEL, an 8-point gain over the prior state of the art, and 75.7% micro-F1 on WebNLG2, while matching or exceeding performance on Wiki-NRE and CaRB. Ablation studies demonstrate that integrating coreference and decomposition increases recall on rare relations by over 20%. Code and dataset are available at https://github.com/KaushikMahmud/CoDe-KG_EMNLP_2025

AI Insights

• CoDe‑KG’s clause‑level decomposition isolates independent, dependent, and modifier clauses to enable precise simplification before extraction.
• Hybrid chain‑of‑thought plus few‑shot prompting attains 99.8 % exact‑match accuracy on sentence simplification across five complexity tiers.
• Integrating coreference resolution boosts recall of rare relations by over 20 %, a key advantage over prior pipelines.
• The 150k‑triple corpus, 7248‑row complexity set, and 398 gold triples provide a rich benchmark for future KG research.
• “Language Models as Knowledge Bases” and “Knowledge Graphs: A Survey of the State‑of‑the‑Art” are essential reads for understanding LM‑driven KG construction.
• The GitHub repo (https://github.com/KaushikMahmud/CoDe-KG_EMNLP_2025) hosts both code and data, enabling reproducible experiments.
• A practical tip: use the provided sentence‑conversion policy annotations to fine‑tune LLMs for domain‑specific simplification.

Abstract
Large language models (LLMs) show promise for diagnostic reasoning but often lack reliable, knowledge grounded inference. Knowledge graphs (KGs), such as the Unified Medical Language System (UMLS), offer structured biomedical knowledge that can support trustworthy reasoning. Prior approaches typically integrate KGs via retrieval augmented generation or fine tuning, inserting KG content into prompts rather than enabling structured reasoning. We explore an alternative paradigm: treating the LLM as a reward model of KG reasoning paths, where the model learns to judge whether a candidate path leads to correct diagnosis for a given patient input. This approach is inspired by recent work that leverages reward training to enhance model reasoning abilities, and grounded in computational theory, which suggests that verifying a solution is often easier than generating one from scratch. It also parallels physicians' diagnostic assessment, where they judge which sequences of findings and intermediate conditions most plausibly support a diagnosis. We first systematically evaluate five task formulation for knowledge path judging and eight training paradigm. Second, we test whether the path judging abilities generalize to downstream diagnostic tasks, including diagnosis summarization and medical question answering. Experiments with three open source instruct-tuned LLMs reveal both promise and brittleness: while specific reward optimization and distillation lead to strong path-judging performance, the transferability to downstream tasks remain weak. Our finding provides the first systematic assessment of "reward model style" reasoning over clinical KGs, offering insights into how structured, reward-based supervision influences diagnostic reasoning in GenAI systems for healthcare.

AI Insights

• The study cites foundational models such as BERT, RoBERTa, and DistilBERT to contextualize its reward‑model approach.
• It acknowledges societal risks of LLMs, including bias amplification and misinformation spread.
• The authors discuss interpretability tools like saliency maps to audit diagnostic reasoning paths.
• A systematic review of five task formulations and eight training paradigms underpins the experimental design.
• The paper highlights computational theory that verification is easier than generation, inspiring the reward‑model paradigm.
• It recommends key resources: “Deep Learning” by Goodfellow et al., “Attention Is All You Need” by Vaswani et al., and the BERT paper.
• The authors provide clear definitions of “large‑scale language model” and “model interpretability” to ground the discussion.

Abstract
Personal service robots are increasingly used in domestic settings to assist older adults and people requiring support. Effective operation involves not only physical interaction but also the ability to interpret dynamic environments, understand tasks, and choose appropriate actions based on context. This requires integrating both hardware components (e.g. sensors, actuators) and software systems capable of reasoning about tasks, environments, and robot capabilities. Frameworks such as the Robot Operating System (ROS) provide open-source tools that help connect low-level hardware with higher-level functionalities. However, real-world deployments remain tightly coupled to specific platforms. As a result, solutions are often isolated and hard-coded, limiting interoperability, reusability, and knowledge sharing. Ontologies and knowledge graphs offer a structured way to represent tasks, environments, and robot capabilities. Existing ontologies, such as the Socio-physical Model of Activities (SOMA) and the Descriptive Ontology for Linguistic and Cognitive Engineering (DOLCE), provide models for activities, spatial relationships, and reasoning structures. However, they often focus on specific domains and do not fully capture the connection between environment, action, robot capabilities, and system-level integration. In this work, we propose the Ontology for roBOts and acTions (OntoBOT), which extends existing ontologies to provide a unified representation of tasks, actions, environments, and capabilities. Our contributions are twofold: (1) we unify these aspects into a cohesive ontology to support formal reasoning about task execution, and (2) we demonstrate its generalizability by evaluating competency questions across four embodied agents - TIAGo, HSR, UR3, and Stretch - showing how OntoBOT enables context-aware reasoning, task-oriented execution, and knowledge sharing in service robotics.

Abstract
People face overwhelming information during work activities, necessitating effective organization and management strategies. Even in personal lives, individuals must keep, annotate, organize, and retrieve knowledge from daily routines. The collection of records for future reference is known as a personal knowledge base. Note-taking applications are valuable tools for building and maintaining these bases, often called a ''second brain''. This paper presents a case study on how people build and explore personal knowledge bases for various purposes. We selected the note-taking tool Obsidian and researchers from a Brazilian lab for an in-depth investigation. Our investigation reveals interesting findings about how researchers build and explore their personal knowledge bases. A key finding is that participants' knowledge retrieval strategy influences how they build and maintain their content. We suggest potential features for an AI system to support this process.

AI Insights

• Participants mixed search, tags, and folders to surface notes, revealing a hybrid retrieval ecology.
• The study stresses that a PKB’s creation plan must align with its retrieval logic from day one.
• A flexible framework should give a minimal scaffold yet allow on‑the‑fly re‑structuring.
• Mental‑model alignment proved key for experts to encode, organize, and recall knowledge.
• Findings echo the DIKW hierarchy, mapping data‑to‑wisdom flows in personal bases.
• Generalizability is limited by the narrow Brazilian CS researcher sample, hinting at nuance.
• The study omitted Obsidian’s AI plugins, opening a fertile gap for future work.

Abstract
Knowledge production is often viewed as an endogenous process in which discovery arises through the recombination of existing theories, findings, and concepts. Yet given the vast space of potential recombinations, not all are equally valuable, and identifying those that may prove most generative remains challenging. We argue that a crucial form of recombination occurs when linking concepts creates knowledge gaps-empty regions in the conceptual landscape that focus scientific attention on proximal, unexplored connections and signal promising directions for future research. Using computational topology, we develop a method to systematically identify knowledge gaps in science at scale. Applying this approach to millions of articles from Microsoft Academic Graph (n = 34,363,623) over a 120-year period (1900-2020), we uncover papers that create topological gaps in concept networks, tracking how these gap-opening works reshape the scientific knowledge landscape. Our results indicate that gap-opening papers are more likely to rank among the most highly cited works (top 1-20%) compared with papers that do not introduce novel concept pairings. In contrast, papers that introduce novel combinations without opening gaps are not more likely to rank in the top 1% for citation counts, and are even less likely than baseline papers to appear in the top 5% to 20%. Our findings also suggest that gap-opening papers are more disruptive, highlighting their generative role in stimulating new directions for scientific inquiry.