TIEM, a timestamp-gated framework with three coordinated components: an Event-Evidence Hypergraph (EEH) for timestamp-filtered multi-tier retrieval; a Case-based Skill Memory (CSM) for source-tagged temporal skills; and Heterogeneous Evidence-Experience Fusion Reasoning (HEFR) for evidence-experience fusion and prediction.
Abstract
Event-driven catalyst-outcome forecasting increasingly uses retrieval- and memory-augmented large language model agents for prediction. However, training-data contamination and temporal leakage can create an Evidence Chasm between reported accuracy and true predictive ability. We propose TIEM, a timestamp-gated framework with three coordinated components: an Event-Evidence Hypergraph (EEH) for timestamp-filtered multi-tier retrieval; a Case-based Skill Memory (CSM) for source-tagged temporal skills; and Heterogeneous Evidence-Experience Fusion Reasoning (HEFR) for evidence-experience fusion and prediction. We also introduce FinPURE, a recent-period A-share holdout benchmark, and use a Name-Date Probe to assess per-model name-date sensitivity rather than assuming training cutoffs. Results on five financial forecasting benchmarks show TIEM outperforms current baselines. Our project is available at https://github.com/QwenQKing/Fin_TIEM.
This work investigates LLM-based forecasting agents, meaning systems in which a language model contributes to a scored prediction about a future or currently unobserved target, and organizes architectures into three groups.
Xiaogang Xu, Jiaqi Tang, Jianmin Chen et al.· 0 citations
REATS is proposed, which leverages LLM reasoning capabilities as an intelligent ensemble router that jointly processes textual temporal pattern descriptions and numerical features to produce interpretable, sample-adaptive ensemble weights through chain-of-thought reasoning.
Large language models (LLMs) can summarize heterogeneous user evidence in natural language, but current LLM recommenders often collapse enduring preferences, transient intent, and exposure-induced behavior into one profile. This makes recommendation vulnerable to feedback loops: repeated exposure is mistaken for preference, immediate clicks dominate delayed satisfaction, and fluent explanations need not reflect the ranking decision. We propose our method, a model-agnostic framework for long-horizon recommendation. Our method uses a frozen multimodal language model to convert item content and feedback into evidence-grounded semantic atoms, then maintains separate short-term, long-term, and exposure memories. Propensity-weighted updates reduce policy-induced exposure bias, while a conservative offline critic reranks candidates for delayed satisfaction under a behavior-support constraint. Explanations use only influential evidence atoms and are checked by counterfactual deletion. We provide an identification result and evaluate the framework in e-commerce-like, news-like, and short-video-like environments. Across ten seeds, our method improves discounted long-term value over the strongest alternative by 6.1%, 7.6%, and 6.7%, respectively. Twenty-seed paired ablations show significant value drops after removing propensity correction (0.739 +/- 0.191) or conservative support regularization (0.523 +/- 0.234). A frozen instruction language model also more than doubles semantic-atom NDCG over TF-IDF on a held-out paraphrase benchmark.
We present a diffusion based model for asynchronous time series prediction, where the goal is to predict the next inter event time and event type. To address the inherent uncertainty of future events, we introduce ReDiTT, a retrieval augmented conditional diffusion transformer that operates in latent space. ReDiTT retrieves structurally similar latent sequences from a memory bank during both training and inference and incorporates them as reference conditions through cross attention. This retrieval based conditioning allows the model to attend to relevant temporal dynamics and provides global structural guidance for generation. As a result, ReDiTT stabilizes long horizon forecasting and improves sample diversity. Experiments on seven real world datasets demonstrate state of the art performance on next event prediction and long horizon forecasting. Our code is available at https://github.com/BorealisAI/ReDiTT.
Saiyue Lyu, Zhitian Zhang, Ruizhi Deng et al.· Trans. Mach. Learn. Res.· 0 citations
FraLLM is a novel LLM fine-tuning framework that seamlessly internalizes transaction-oriented knowledge for FRA and introduces the Memory Token Mechanism, which recurrently aggregates historical text prototypes into a compact, continuously updated memory token that allows LLMs to effectively synthesize long-term transaction history while ensuring cost-efficiency.
Siwei Zhang, Yun Xiong, Xi Chen et al.· Proceedings of the 32nd ACM...· 0 citations
Risk prediction in conflict-affected environments requires models that can learn from evolving data streams. This study develops an incremental recurrent learning framework for adverse-event count prediction and evaluates it in a field study of human-social-infrastructure projects in Afghanistan. The framework implements GRU-Linear (GRU-L) and LSTM-Linear (LSTM-L) predictors that encode short historical sequences with a recurrent layer and map the learned representation to nonnegative count predictions through a fully connected regression head. For each arriving instance, the model first predicts the target and then updates its parameters after the true value becomes available, following a prequential (test-then-train) evaluation protocol. We compare these predictors with statistical, ensemble, gradient-boosted, recurrent batch, and rule-based stream baselines under a unified chronological protocol. Batch models use expanding-window refitting, whereas stream models use prequential test-then-train evaluation, ensuring identical information availability constraints across all model classes. The experiments report MAE, RMSE, paired 400-instance block-level significance tests, sliding-window error trajectories, Page-Hinkley prediction-error drift alarms, and runtime. The results show that LSTM-L obtains the lowest MAE for Events and Wounded, and GRU-L obtains the lowest MAE for Hijacked and Dead. Linear Regression (LR) and TargetMean provide the lowest runtimes, while the recurrent stream models offer stronger MAE performance, particularly when recent temporal patterns are predictive. The study demonstrates that incremental recurrent learning can support adaptive risk monitoring when adverse-event distributions change over time.
Donghui Shi, Jifei Cheng, Yakun Sun et al.· IEEE Access· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
