Accurate stock prediction is critical for portfolio management, where learning to adapt to market changes is the key to sustainable profitability. Financial markets, as complex interactive systems, exhibit evolution in both temporal dynamics and relational structures. While current temporal-relational models have achieved remarkable success in stock prediction, they face fundamental challenges in learning and adapting to market changes, particularly the systematic shifts in temporal and relational distributions that challenge the i.i.d. assumption underlying model training. In this study, we pioneer the research of temporal-relational distribution shifts in stock prediction and introduce Enhancer, a model-agnostic framework that can be applied to any downstream predictor. Enhancer adopts a meta-learning architecture featuring both a Temporal Meta-Learner (TML) and a Relational Meta-Learner (RML). Specifically, we introduce Reactive Point Processes Attention (RPPsAtt) within TML to overcome the limitations of missing fine-grained temporal point information, a common issue with prior methods that rely on distribution inference for mitigating temporal distribution shift. To enhance relational generalization, we introduce the Approximation-Intervention (Ant) mechanism within RML, marking the first method to mitigate relational distribution shift for quantitative investment. We conduct experiments on four long-term stock datasets, categorizing them into two tasks: stock trend prediction and stock investment recommendation. Our experimental results show that Enhancer achieves an average increase of 29.3% in profit ratio and 18.54% in the Sharpe ratio compared to the baselines across two tasks.

Visual grounding refers to the process of associating natural language expressions with corresponding regions within an image. Existing benchmarks for visual grounding primarily operate within small-scale scenes with a few objects. Nevertheless, recent advances in imaging technology have enabled the acquisition of gigapixel-level images, providing high-resolution details in large-scale scenes containing numerous objects. To bridge this gap between imaging and computer vision benchmarks and make grounding more practically valuable, we introduce a novel dataset, named GigaGrounding, designed to challenge visual grounding models in gigapixel-level large-scale scenes. We extensively analyze and compare the dataset with existing benchmarks, demonstrating that GigaGrounding presents unique challenges such as large-scale scene understanding, gigapixel-level resolution, significant variations in object scales, and the “multi-hop expressions”. Furthermore, we introduced a simple yet effective grounding approach, which employs a “glance-to-zoom-in” paradigm and exhibits enhanced capabilities for addressing the GigaGrounding task. The dataset is available at www.gigavision.ai.

Stock forecast is a crucial yet challenging task in modern quantitative trading. Given theoretical and investment merits, recently a variety of deep learning methods have been proposed for automatically simulating stock movements from historical time series. However, these methods typically follow the i.i.d. assumption that actually contradicts the complex trading environment. In reality, individual stocks often exhibit diverse volatility patterns, while macro market scenarios may also change over time, jointly resulting in distribution shifts and weak generalization. To combat these bottlenecks, in this paper we propose a new learning architecture calledPersonalized Temporal Embedding and Routing(PTER) to improve stock forecast by forming a relaxed weight-sharing paradigm. The key of PTER is introducing hypernetworks to guide tailoring target network parameters, such that stock time series are embedded adapting to multi-object multi-scenario data disparities. Specifically, in the encoding stage, PTER first captures hyper-knowledge characterizing the similarity and peculiarity of different stocks and market scenarios. The knowledge space is then projected onto the temporal parameter space, enabling the customization of protruded features from chaotic observation signals. In the inference stage, each sample is dispatched to orthogonal predictor heads to dynamically output expected returns based on market conditions. Through experiments on benchmark datasets spanning over five years on four of the world's largest exchange markets, we show that PTER improves the cumulative and risk-adjusted revenue performance by a significant margin.

Out-of-distribution (OOD) graph generalization are critical for many real-world applications. Existing methods neglect to discard spurious or noisy features of inputs, which are irrelevant to the label. Besides, they mainly conduct instance-level class-invariant graph learning and fail to utilize the structural class relationships between graph instances. In this work, we endeavor to address these issues in a unified framework, dubbedIndividual andStructuralGraphInformationBottlenecks (IS-GIB). To remove class spurious feature caused by distribution shifts, we propose Individual Graph Information Bottleneck (I-GIB) which discards irrelevant information by minimizing the mutual information between the input graph and its embeddings. To leverage the structural intra- and inter-domain correlations, we propose Structural Graph Information Bottleneck (S-GIB). Specifically for a batch of graphs with multiple domains, S-GIB first computes the pair-wise input-input, embedding-embedding, and label-label correlations. Then it minimizes the mutual information between input graph and embedding pairs while maximizing the mutual information between embedding and label pairs. The critical insight of S-GIB is to simultaneously discard spurious features and learn invariant features from a high-order perspective by maintaining class relationships under multiple distributional shifts. Notably, we unify the proposed I-GIB and S-GIB to form our complementary framework IS-GIB. Extensive experiments conducted on both node- and graph-level tasks consistently demonstrate the superior generalization ability of IS-GIB. The code is available athttps://github.com/YangLing0818/GraphOOD.

With the development of event knowledge graph technology, researchers have solved the singleness problem of event graph based on temporal relationship by constructing event logic graph, but have not integrated the multiple relationships among events with time series data for trend prediction.In addition, due to the impact of COVID-19, corporate credit risks have been gradually exposed in recent years, and defaults have occurred frequently.The technology of event graph and event logic graph is mostly used for event schema induction, script induction, etc., but abundant graph knowledge is not well exploited for forecast task.To fill this gap, we construct an event logic graph by extracting various types of event relationships, such as causal relationship, sequential relationship, parallel relationship, and reversal relationship.Different types of edges among events are used to represent different relationships.Combined with the time series of corporate credit bonds, a temporal convolutional network driven by event logic graph is built, and applied to forecast corporate credit risk.We extract structured events from financial news, construct event logic graph and learn the graph knowledge.Then, the event logic graph embedding is combined with time series of bonds to forecast whether the corporate will default.Experiments show that the proposed method outperforms baseline methods in forecasting credit risk.

Stock trend prediction is a hot issue in theFintechfield. Effective stock profiling is challenging due to highly non-stationary dynamics and complex interplays. Existing methods usually regard each stock independently or detect simplistic homogeneous structures. Practically, stock correlation originates from diverse aspects and underlying relationship signals are implicit in comprehensive graphs. Besides, RNNs are extensively used to simulate stock volatility while inadequate in capturing fine-granular patterns across local time snippets. To this end, in this paper we propose HATR-I, a Hierarchical Adaptive Temporal-Relational Interaction model to characterize and predict stock evolutions. Specifically, we grasp short- and long-term transition regularities of stock dynamics based on cascaded dilated convolutions and gating paths. By formulating different views of domain adjacency graphs into a unified multiplex network with edge attributes, we inject node- and semantic-level dual attention to refine the propagation of inter-stock collaborative information. Particularly, the stock pair matching is proceeding along each time-stage rather than until final compressed representations, meanwhile significant feature points and scales are identified considering the effect of time attenuation. Finally, we deduce latent shared clusters as global regularization to optimize the stock representations. Experiments on three real-world stock market datasets demonstrate the effectiveness of our proposed model.

Stock recommendation plays a critical role in modern quantitative trading. The large volumes of social media information such as investment reviews that delegate emotion-driven factors, together with price technical indicators formulate a “snapshot” of the evolving stock market profile. However, previous studies usually model the temporal trajectories of price and media modalities separately while losing their interrelated influences. Moreover, they mainly extract review semantics via sequential or attentive models, whereas the rich text associated knowledge is largely neglected. In this paper, we propose a novel heterogeneous interactive snapshot network for stock profiling and recommendation. We model investment reviews in each snapshot as a heterogeneous document graph, and develop a flexible hierarchical attentive propagation framework to capture fine-grained proximity features. Further, to learn stock embedding for ranking, we introduce a novel twins-GRU method, which tightly couples the media and price parallel sequences in a cross-interactive fashion to catch dynamic dependencies between successive snapshots. Our approach excels state-of-the-arts over 7.6% in terms of cumulative and risk-adjusted returns in trading simulations on both English and Chinese benchmarks.

Stock investment selection is a hard issue in the Fintech field due to non-stationary dynamics and complex market interdependencies. Existing studies are mostly based on RNNs, which struggle to capture interactive information among fine granular volatility patterns. Besides, they either treat stocks as isolated, or presuppose a fixed graph structure heavily relying on prior domain knowledge. In this paper, we propose a novel Adaptive Long-Short Pattern Transformer (ALSP-TF) for stock ranking in terms of expected returns. Specifically, we overcome the limitations of canonical self-attention including context and position agnostic, with two additional capacities: (i) fine-grained pattern distiller to contextualize queries and keys based on localized feature scales, and (ii) time-adaptive modulator to let the dependency modeling among pattern pairs sensitive to different time intervals. Attention heads in stacked layers gradually harvest short- and long-term transition traits, spontaneously boosting the diversity of representations. Moreover, we devise a graph self-supervised regularization, which helps automatically assimilate the collective synergy of stocks and improve the generalization ability of overall model. Experiments on three exchange market datasets show ALSP-TF’s superiority over state-of-the-art stock forecast methods.

Stock trend prediction is a challenging task due to the non-stationary dynamics and complex market dependencies. Existing methods usually regard each stock as isolated for prediction, or simply detect their correlations based on a fixed predefined graph structure. Genuinely, stock associations stem from diverse aspects, the underlying relation signals should be implicit in comprehensive graphs. On the other hand, the RNN network is mainly used to model stock historical data, while is hard to capture fine-granular volatility patterns implied in different time spans. In this paper, we propose a novel Hierarchical Adaptive Temporal-Relational Network (HATR) to characterize and predict stock evolutions. By stacking dilated causal convolutions and gating paths, short- and long-term transition features are gradually grasped from multi-scale local compositions of stock trading sequences. Particularly, a dual attention mechanism with Hawkes process and target-specific query is proposed to detect significant temporal points and scales conditioned on individual stock traits. Furthermore, we develop a multi-graph interaction module which consolidates prior domain knowledge and data-driven adaptive learning to capture interdependencies among stocks. All components are integrated seamlessly in a unified end-to-end framework. Experiments on three real-world stock market datasets validate the effectiveness of our model.

Investment messages published on social media platforms are highly valuable for stock prediction. Most previous work regards overall message sentiments as forecast indicators and relies on shallow features (bag-of-words, noun phrases, etc.) to determine the investment opinion signals. These methods neither capture the time-sensitive and target-aware characteristics of stock investment reviews, nor consider the impact of investor's reliability. In this study, we provide an in-depth analysis of public stock reviews and their application in stock movement prediction. Specifically, we propose a novel framework which includes the following three key components: time-sensitive and target-aware investment stance detection, expert-based dynamic stance aggregation, and stock movement prediction. We first introduce our stance detection model named MFN, which learns the representation of each review by integrating multi-view textual features and extended knowledge in financial domain to distill bullish/bearish investment opinions. Then we show how to identify the validity of each review, and enhance stock movement prediction by incorporating expert-based aggregated opinion signals. Experiments on real datasets show our framework can effectively improve the performance of both investment opinion mining and individual stock forecasting.

Personalized news recommendation is a critical technology to improve users' online news reading experience.The core of news recommendation is accurate matching between user's interests and candidate news.The same user usually has diverse interests that are reflected in different news she has browsed.Meanwhile, important semantic features of news are implied in text segments of different granularities.Existing studies generally represent each user as a single vector and then match the candidate news vector, which may lose fine-grained information for recommendation.In this paper, we propose FIM, a Finegrained Interest Matching method for neural news recommendation.Instead of aggregating user's all historical browsed news into a unified vector, we hierarchically construct multilevel representations for each news via stacked dilated convolutions.Then we perform finegrained matching between segment pairs of each browsed news and the candidate news at each semantic level.High-order salient signals are then identified by resembling the hierarchy of image recognition for final click prediction.Extensive experiments on a real-world dataset from MSN news validate the effectiveness of our model on news recommendation.

Building an intelligent chatbot with multi-turn dialogue ability is a major challenge, which requires understanding the multi-view semantic and dependency correlation among words, n-grams and sub-sequences. In this paper, we investigate selecting the proper response for a context through multi-grained representation and interactive matching. To construct hierarchical representation types of text segments, we propose a refined architecture which exclusively consists of gated dilated-convolution and self-attention. Compared with the recurrent-based sentence modeling methods, this architecture provides more flexibility and a speedup. The matching signals of each utterance-response pair are extracted by integrating the interactive information from different views. Then a turns-aware attention mechanism is utilized to aggregate the matching sequence, so as to identify important utterances and capture the implicit relationship of the whole context. Experiments on two large-scale public data sets show that our model significantly outperforms the state-of-the-art methods in terms of all metrics. We empirically provide a thorough ablation test, as well as the comparison of different representation and matching strategies, for a better insight into how each component affects the performance of the model.