VELDA: Relating an Image Tweet’s Text and Images

Authors: Tao Chen, Hany SalahEldeen, Xiangnan He, Min-Yen Kan, Dongyuan Lu

AAAI 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on real-world image tweets in both English and Chinese and other user generated content, show that VELDA significantly outperforms existing methods on cross-modality image retrieval.Even in other domains where emotion does not factor in image choice directly, our VELDA model demonstrates good generalization ability, achieving higher fidelity modeling of such multimedia documents.
Researcher Affiliation Academia 1School of Computing, National University of Singapore 2Department of Computer Science, Old Dominion University 3NUS Interactive and Digital Media Institute, Singapore
Pseudocode No The information provided is insufficient. The paper describes its model formulation and parameter estimation through text and mathematical equations, but does not present any clearly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The information provided is insufficient. The paper does not contain any statement about releasing its source code, nor does it provide a link to a code repository for the methodology described.
Open Datasets No The information provided is insufficient. The paper describes how the authors collected their own datasets (Weibo, Twitter, Google-Zh, Google-En, POTD) based on queries, but does not provide any specific links, DOIs, repository names, or explicit statements that these collected datasets are publicly available for download or access.
Dataset Splits No The information provided is insufficient. The paper states, 'We randomly split each dataset into 90% as training set and the remaining 10% as testing set,' but does not mention a separate validation split or how validation was handled for hyperparameter tuning.
Hardware Specification No The information provided is insufficient. The paper does not specify any particular hardware components such as GPU models, CPU models, or cloud instance types used for running the experiments.
Software Dependencies No The information provided is insufficient. The paper mentions various software components and methods used (e.g., 'Chinese word segmentation program', 'SIFT', 'k-means', 'Google Translate') but does not provide specific version numbers for any of them.
Experiment Setup Yes Our development testing showed that VELDA operated well over a wide range of hyperparameter settings. As such, we fix the six sets of hyperparameters to relatively standard settings: αV =1, αE=1, βV =0.1, βE=0.1, γ = 0.1, and η = 0.5. We then tune the number of visual topics (K) and emotional topics (E) in a grid search for each dataset (see Table 2 for the detailed settings).