Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
COGRASP: Co-Occurrence Graph Based Stock Price Forecasting
Authors: Zhengze Li, Zilin Song, Tingting Yuan, Xiaoming Fu
IJCAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In experiments with real-world open-source stock market data, COGRASP outperforms state-of-the-art methods. We conduct extensive experiments based on the relation graph extracted from social network posts and real-world open-source stock price datasets. Our evaluation shows COGRASP achieves significant performance improvements over the best-performed SOTA methods across four common evaluation metrics: Information Coefficient (IC, up to 39%), the Rank Information Coefficient (Rank IC, up to 140%), the Information Ratio based IC (ICIR, up to 5%), and the Information Ratio based Rank IC (Rank ICIR, up to 94%). Through extensive ablation studies, comparative experiments, and case analyses, we demonstrate that the performance of co-occurrence graphs surpasses that of concept graphs and correlation graphs. |
| Researcher Affiliation | Academia | 1Institute of Computer Science, University of Göttingen, Germany 2Institute of Digitalisation and Informatics, IMC Krems University of Applied Sciences, Austria EMAIL |
| Pseudocode | No | The paper describes the methodology using equations and descriptive text, but does not include a clearly labeled pseudocode or algorithm block. |
| Open Source Code | Yes | *Corresponding authors Code available on https://github.com/Ningbo Song/COGRASP. |
| Open Datasets | Yes | Stock features, including opening price, closing price, highest price, lowest price, trading volume, trading value, amplitude, price change percentage, price change amount, and turnover rate, trading volume, were sourced from the open-source dataset AKShare [King, 2019]. |
| Dataset Splits | Yes | Training and validation used data from January 2015 to February 2024, and testing used data from March to June 2024. |
| Hardware Specification | No | The paper mentions that 'Experiments are based on PyTorch and PyG,' but does not provide specific details about the hardware used, such as GPU/CPU models or memory. |
| Software Dependencies | No | The paper states 'Experiments are based on PyTorch and PyG,' but does not provide specific version numbers for these software dependencies, which is necessary for reproducibility. |
| Experiment Setup | Yes | All models were trained for 500 rounds, and an early stopping mechanism of 20 rounds was introduced to prevent overfitting. The lookback window was set to 15, the learning rate was set to 0.001, and an L2 regularization of 0.001 was applied. During the training process, the Reduce LROn Plateau strategy was used to dynamically adjust the learning rate. In the COGRASP, the time scales were set to 5, 10, 15. The number of layers of the GCN was set to 1 layer and the number of hidden units was 16. For each ALSTM unit, the number of layers was 1, and the number of hidden units was set to 64. |