Predictive AI with External Knowledge Infusion: Datasets and Benchmarks for Stock Markets
Ambedkar Dukkipati, Kawin Mayilvaghanan, Naveen Kumar Pallekonda, Sai Prakash Hadnoor, Ranga Shaarad Ayyagari · 2025
Read Predictive AI with External Knowledge Infusion: Datasets and Benchmarks for Stock Markets by Ambedkar Dukkipati, Kawin Mayilvaghanan, Naveen Kumar Pallekonda, Sai Prakash Hadnoor, Ranga Shaarad Ayyagari (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-20058.txt.
Introduction
Ambedkar Dukkipati and four co-authors build data sets and tests for stock forecasting that uses outside knowledge, not prices alone. Their work appeared in twenty twenty-five. Prices move with company links, policy shifts, wars, and the wider economy, and all of those change over time.
No public data captured that web, so the team built time-stamped knowledge graphs for the stock market. Their method treats events on those graphs as sparks in a Hawkes process, a model where one event raises the odds of others. The learned patterns rank stocks by expected returns over several holding spans. In wide tests, the rankings beat rival baselines. The authors offer the data and benchmarks for others to build on.
Catalog record
| CREATOR | Ambedkar Dukkipati, Kawin Mayilvaghanan, Naveen Kumar Pallekonda, Sai Prakash Hadnoor, Ranga Shaarad Ayyagari |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-20058
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Predictive AI with External Knowledge Infusion: Datasets and Benchmarks for Stock Markets Ambedkar Dukkipati, Kawin Mayilvaghanan, Naveen Kumar Pallekonda, Sai Prakash Hadnoor, Ranga Shaarad Ayyagari Abstract Fluctuations in stock prices are influenced by a complex interplay of factors that go beyond mere historical data. These factors, themselves influenced by external forces, encompass inter-stock dynamics, broader economic factors, various government policy decisions, outbreaks of wars, etc. Furthermore, all of these factors are dynamic and exhibit changes over time. In this paper, for the first time, we tackle the forecasting problem under external influence by proposing learning mechanisms that not only learn from historical trends but also incorporate external knowledge from temporal knowledge graphs. Since there are no such datasets or temporal knowledge graphs available, we study this problem with stock market data, and we construct comprehensive temporal knowledge graph datasets. In our proposed approach, we model relations on external temporal knowledge graphs as events of a Hawkes process on graphs. With extensive experiments, we show that learned dynamic representations effectively rank stocks based on returns across multiple holding periods, outperforming related baselines on relevant metrics. 1 Introduction The development of learning algorithms for forecasting events is an important component in building AI systems with a wide range of applications, including in financial markets and healthcare. Figure 1 : Interdependence of stock trends between stocks in the same class during COVID-19 (January–May 2020, NSE). Pharmaceutical stocks (CIPLA, DIVISLAB, DRREDDY) trended positively, whereas banking stocks (PNB, INDUSINDBANK, ICICIBANK) declined. Recent progress in these areas has been notable, and methods involving a combination of power sequence models, such as transformers, and elegant probabilistic models, such as temporal point processes, have made enormous progress ( Mei and Eisner, 2017 ; Gracious et al., 2021 ; Gracious and Dukkipati, 2023 ; Gracious et al., 2025 ; Gracious and Dukkipati, 2025 ) . However, nearly all existing methods fall short of adequately accounting for or integrating time-varying external factors that influence the outcomes. For instance, stock price variations are affected by a complex interplay of factors that extend beyond mere historical patterns of stock prices. These elements can be shaped by external influences, such as prevailing economic conditions, shifts in governmental policies, and incidents of conflict. Additionally, these factors can be dynamic, with some changing rapidly, others very slowly, some very frequently, and others infrequently. To study this problem, we develop an approach that captures external influences via temporal Knowledge Graphs (KG) and learns historical patterns via temporal point processes. Considering that such specific temporal KGs are not available in the literature, we develop temporal KGs for financial markets for the machine learning community and propose representation learning techniques to establish benchmark results. The temporal KGs we constructed integrated diverse components, including first- and second-order stock relationships, corporate events, macroeconomic indicators, events derived from financial statements, financial ratios disclosed during earnings calls, and analyst sentiments. Notably, all these elements are equipped with time-aware attributes. The capacity to incorporate entities of varying categories, such as Person, Stock Assets, Products, and Alliance, results in an intrinsically heterogeneous knowledge graph. We refer to these KGs as \mathsf{STOCKnowledge} aims to encapsulate all pertinent information related to stock assets within a given share market. Contributions (1) We propose a time-series forecasting algorithm that incorporates knowledge of external events influencing the evolution of the series. (2) For the task of stock prediction, we construct temporal knowledge graphs that capture relevant macroeconomic and financial information related to the NSE and NASDAQ stock markets and use them to enhance sequential forecasting models. (3) To this end, we propose a time-aware knowledge graph representation technique that leverages a heterogeneous graph convolution operator, along with a temporal point process-based method to model external events. We integrate these embeddings with transformer-based sequential embeddings to obtain accurate rankings of future stock performance over different time horizons. (4) We empirically demonstrate that our method outperforms related baselines in ranking stocks across three different holding periods. Additionally, we conduct ablation studies and further experiments to analyze the effectiveness of the different components of our model. 2 Related work Forecasting has traditionally used time-series models that predict future values solely from historical numerical data, assuming that the target variable evolves as a function of its own past. Deep learning techniques using models such as RNNs ( Hewamalage et al., 2021 ) , LSTMs ( Elsworth and Güttel, 2020 ; Kong et al., 2025 ) , and transformer-based architectures ( Zhou et al., 2021 ; Liu et al., 2024 ) enable the capture of long-range dependencies and complex non-linear patterns that classical models struggle to represent. However, these approaches often assume that all relevant signals are contained within the target series and a limited set of covariates, overlooking the broader context in which the data arise. This limits their ability to handle situations in which external events (e.g., pandemics, geopolitical shifts) or inter-entity interactions (e.g., firm-to-firm relationships) cause abrupt or correlated changes. While recent work has begun to incorporate covariates, there remains a gap in systematically modeling rich, temporally dynamic external influences. To address this gap, recent studies have incorporated unstructured data sources into forecasting models. These include news articles ( Wang et al., 2024a ) , social media posts ( Karlemstrand and Leckström, 2021 ) , financial reports ( Zheng et al., 2019 ) , and product reviews ( Liu et al., 2023 ) , which offer valuable contextual signals that complement numerical time series. Text-Guided Time Series Forecasting (TGTSF) ( Xu et al., 2024 ) uses textual cues, such as channel metadata or news content, to guide predictions. Con ( Chattopadhyay et al., 2025 ) combines temporal and textual modalities using cross-attention to achieve improved results. In a multimodal direction, MM-Forecast ( Li et al., 2024 ) integrates images and other modalities for temporal-event prediction. These approaches typically treat unstructured signals as flat texts, failing to capture the relationships between entities or how they evolve over time. This limits their ability to reason about structured causal chains or model how influences unfold temporally. Incorporating structural knowledge into learning problems has been studied for natural language problems ( Annervaz et al., 2018 ) . Recently, this technique has been applied to forecasting problems. This includes time-varying categorical variables, continuous external signals, and relational knowledge captured using graphs and knowledge graphs. For example, TimeXer ( Wang et al., 2024b ) enhances transformers with patch-wise and variate-wise attention to handle structured exogenous inputs in multivariate time-series. ChronosX ( Arango et al., 2025 ) fuses historical and future auxiliary inputs using dedicated networks to adapt pretrained models without retraining the core models. Beyond tabular features, recent studies have constructed graphs ( Cini et al., 2023 ) and knowledge graphs ( Jain et al., 2023 ) to model prior knowledge and event dynamics that influence targets. This structured context supports better generalization across domains and improves interpretability by linking the forecasts to causally relevant external signals. However, these efforts are often static or limited to coarse-grained relationships and do not explicitly model the evolving temporal interactions between entities. Our work addresses this gap by constructing temporal knowledge graphs that encode fine-grained, time-indexed external events and using point-process modeling to dynamically represent their influence on forecasting targets.