Knowledge Graph Question Answering (KGQA) aims to answer natural language questions by reasoning over structured knowledge graphs (KGs). While large language models (LLMs) have advanced KGQA through their strong reasoning capabilities, existing methods continue to struggle to fully exploit both the rich knowledge encoded in KGs and the reasoning capabilities of LLMs, particularly in complex scenarios. They often assume complete KG coverage and lack mechanisms to judge when external information is needed, and their reasoning remains locally myopic, failing to maintain coherent multi-step planning, leading to reasoning failures even when relevant knowledge exists. We propose Graph-RFT, a novel two-stage reinforcement fine-tuning KGQA framework with a ''plan–KGsearch–and–Websearch–during–think'' paradigm, that enables LLMs to perform autonomous planning and adaptive retrieval scheduling across KG and web sources under incomplete knowledge conditions. Graph-RFT introduces a chain-of-thought (CoT) fine-tuning method with a customized plan–retrieval dataset activates structured reasoning and resolves the GRPO cold-start problem. It then introduces a novel plan–retrieval guided reinforcement learning process integrates explicit planning and retrieval actions with a multi-reward design, enabling coverage-aware retrieval scheduling. It employs a Cartesian-inspired planning module to decompose complex questions into ordered sub-questions, and logical expression to guide tool invocation for globally consistent multi-step reasoning. This reasoning–retrieval process is optimized with a multi-reward combining outcome and retrieval-specific signals, enabling the model to learn when and how to combine KG and web retrieval effectively. Experiments on multiple KGQA benchmarks demonstrate that Graph-RFT achieves superior performance over strong baselines, even with smaller LLM backbones, and substantially improves complex question decomposition, factual coverage, and tool coordination.

Recent advancements have highlighted that Large Language Models (LLMs) are prone to hallucinations when solving complex reasoning problems, leading to erroneous results. To tackle this issue, researchers incorporate Knowledge Graphs (KGs) to improve the reasoning ability of LLMs. However, existing methods face two limitations: 1) they typically assume that all answers to the questions are contained in KGs, neglecting the incompleteness issue of KGs, and 2) they treat the KG as a static repository and overlook the implicit logical reasoning structures inherent in KGs. In this paper, we introduce SymAgent, an innovative neural-symbolic agent framework that achieves collaborative augmentation between KGs and LLMs. We conceptualize KGs as dynamic environments and transform complex reasoning tasks into a multi-step interactive process, enabling KGs to participate deeply in the reasoning process. SymAgent consists of two modules: Agent-Planner and Agent-Executor. The Agent-Planner leverages LLM's inductive reasoning capability to extract symbolic rules from KGs, guiding efficient question decomposition. The Agent-Executor autonomously invokes predefined action tools to integrate information from KGs and external documents, addressing the issues of KG incompleteness. Furthermore, we design a self-learning framework comprising online exploration and offline iterative policy updating phases, enabling the agent to automatically synthesize reasoning trajectories and improve performance. Experimental results demonstrate that SymAgent with weak LLM backbones (i.e., 7B series) yields better or comparable performance compared to various strong baselines. Further analysis reveals that our agent can identify missing triples, facilitating automatic KG updates.

Despite Spanish's pivotal role in the global finance industry, a pronounced gap exists in Spanish financial natural language processing (NLP) and application studies compared to English, especially in the era of large language models (LLMs).To bridge this gap, we unveil Toisón de Oro, the first bilingual framework that establishes instruction datasets, finetuned LLMs, and evaluation benchmark for financial LLMs in Spanish joint with English.We construct a rigorously curated bilingual instruction dataset including over 144K Spanish and English samples from 15 datasets covering 7 tasks.Harnessing this, we introduce FinMA-ES, an LLM designed for bilingual financial applications.We evaluate our model and existing LLMs using FLARE-ES, the first comprehensive bilingual evaluation benchmark with 21 datasets covering 9 tasks.The FLARE-ES benchmark results

Link prediction (LP) in knowledge graph (KG) is a crucial task that has received increasing attention recently. Due to the heterogeneous structures of KGs, various application scenarios, and demand-specific downstream objectives, there exist multiple subtasks in LP. Most studies only focus on designing a dedicated architecture for a specific subtask, which results in various complicated LP models. The isolated architectures and chaotic situations make it significant to construct a unified model that can handle multiple LP subtasks simultaneously. However, unifying all subtasks in LP presents numerous challenges, including unified input forms, task-specific context modeling, and topological information encoding. To address these challenges, we propose a topology-aware generative framework, namely UniLP, which utilizes a generative pre-trained language model to accomplish different LP subtasks universally. Specifically, we introduce a context demonstration template to convert task-specific context into a unified generative formulation. Based on the unified formulation, to address the limitation of transformer architecture that may overlook important structural signals in KGs, we design novel topology-aware soft prompts to deeply couple topology and text information in a contextualized manner. Extensive experiment results demonstrate that our framework achieves substantial performance gain and provides a real unified end-to-end solution for the whole LP subtasks. We also perform comprehensive ablation studies to support in-depth analysis of each component in UniLP.

Graph Neural Networks (GNNs) such as GCN can effectively learn document representations via the semantic relation graph among documents and words. However, despite a few exceptions, most of the previous work in this line of research does not consider the underlying topical semantics inherited in document contents and the relation graph, making the representations less effective and hard to interpret. In a few recent studies trying to incorporate latent topics into GNNs, the topics have been learned independently from the relation graph modeling. Intuitively, topic extraction can benefit much from the information propagation of the relation graph structure - directly and indirectly connected documents and words have similar topics. In this paper, we propose a novel Graph Topic Neural Network (GTNN) model to mine latent topic semantics for interpretable document representation learning, taking into account the document-document, document-word, and word-word relationships in the graph. We also show that our model can be viewed as semi-amortized inference for relational topic model based on Poisson distribution, with high order correlations. We test our model in several settings: unsupervised, semi-supervised, and supervised representation learning, for both connected and unconnected documents. In all the cases, our model outperforms the state-of-the-art models for these tasks.

Deep generative models (DGMs) have shown strong performance in semi-supervised learning (SSL), which incorporate discrete class information into the learning process. Yet existing methods generally overfit to the given labeled data, for only considering the conditional probability of labels. In this paper, we propose a novel discriminative regularized deep generative method for SSL, which fully exploits the discriminative and geometric information of data to address the aforementioned issue. Our method introduces the cluster and manifold assumption that maximizes the classification margin between clusters and simultaneously smooths the predictions of the data which is close in the sub-manifold of each cluster, to regularize the learning of the classifier in DGMs. To derive the regularization based on introduced assumptions, we adopt the generated data of DGMs along with labelled and unlabelled data, to model the data manifold and yield clusters based on the Gumbel-softmax distribution. Experimental results on both text and image datasets demonstrate the effectiveness and flexibility of our method, and prove that two introduced assumptions are complementary in guiding the classification boundary, thus improving the discriminative ability of the classifier.

Sparse topic models (STMs) are widely used for learning a semantically rich latent sparse representation of short texts in large scale, mainly by imposing sparse priors or appropriate regularizers on topic models. However, it is difficult for these STMs to model the sparse structure and pattern of the corpora accurately, since their sparse priors always fail to achieve real sparseness, and their regularizers bypass the prior information of the relevance between sparse coefficients. In this paper, we propose a novel Bayesian hierarchical topic models called Bayesian Sparse Topical Coding with Poisson Distribution (BSTC-P) on the basis of Sparse Topical Coding with Sparse Groups (STCSG). Different from traditional STMs, it focuses on imposing hierarchical sparse prior to leverage the prior information of relevance between sparse coefficients. Furthermore, we propose a sparsity-enhanced BSTC, Bayesian Sparse Topical Coding with Normal Distribution (BSTC-N), via mathematic approximation. We adopt superior hierarchical sparse inducing prior, with the purpose of achieving the sparsest optimal solution. Experimental results on datasets of Newsgroups and Twitter show that both BSTC-P and BSTC-N have better performance on finding clear latent semantic representations. Therefore, they yield better performance than existing works on document classification tasks.

This article presents an unsupervised multi-view hierarchical embedding (UMHE) framework to sufficiently reveal the intrinsic topical knowledge in social events. Event-oriented topics are highly related to such events as it can provide explicit descriptions of what have happened in social community. In many real-world cases, however, it is difficult to include all attributes of microblogs, more often, textual aspects only are available. Traditional topic modelling methods have failed to generate event-oriented topics with the textual aspects, since the inherent relations between topics are often overlooked in these methods. Meanwhile, the metrics in original word vocabulary space might not effectively capture semantic distances. Our UMHE framework overcomes the severe information deficiency and poor feature representation. The UMHE first develops a multi-view Bayesian rose tree to preliminarily generate prior knowledge for latent topics and their relations. With such prior knowledge, we design an unsupervised translation-based hierarchical embedding method to make a better representation of these latent topics. By applying self-adaptive spectral clustering on the embedding space and the original space concomitantly, we eventually extract event-oriented topics in word distributions to express social events. Our framework is purely data-driven and unsupervised, without any external knowledge. Experimental results on TREC Tweets2011 dataset and Sina Weibo dataset demonstrate that the UMHE framework can construct hierarchical structure with high fitness, but also yield topic embeddings with salient semantics; therefore, it can derive event-oriented topics with meaningful descriptions.

Compressing textstreams generated by social networks can both reduce storage consumption and improve efficiency such as fast searching. However, the compression process is a challenge due to the large scale of textstreams. In this article, we propose a textstream compression framework based on compressed sensing theory and design a series of matching parallel procedures. The new approach uses a linear projection technique in the textstream compression process, achieving fast compression speed and low compression ratio. Two processes are executed by designing elaborated parallel procedures for efficient compressing and decompressing of large-scale textstreams. The decompression process is implemented for approximate solutions of underdetermined linear systems. Experimental results show that the new method can efficiently achieve the compression and decompression tasks on a large amount of text generated by social networks.

To date, data generates and arrives in the form of stream to propagate discussions of public events in microblog services. Discovering event-oriented topics from the stream will lead to a better understanding of the change of public concern. However, as the massive scale of the data stream, traditional static topic models, such as LDA, are no longer fit for topic detection and tracking tasks. In this paper, we propose a central topic model (CenTM), where a Multi-view Clustering algorithm with Two-phase Random Walk (MC-TRW) is devised to aggregate the LDA's latent topics into central topics. Furthermore, we leverage the aggregation of central topics alternately with MC-TRW and sequential topic inference to improve the scalability in the stream fashion, so as to derive the dynamic central topic model (DCenTM). Specifically, our model is able to uncover the intrinsic characteristics of the central topics and predict the trend of their intensity along a life cycle. Experimental results demonstrate that the proposed central topic model is event-oriented and of high generalization, it therefore can dispose the topic trend prediction effectively and precisely in massive data stream.