Composed Image Retrieval (CIR) enables image retrieval by combining multiple query modalities, but existing benchmarks predominantly focus on general-domain imagery and rely on reference images with short textual modifications. As a result, they provide limited support for retrieval scenarios that require fine-grained semantic reasoning, structured visual understanding, and domain-specific knowledge. In this work, we introduce CIRThan, a sketch+text composed image retrieval dataset for Thangka imagery, a culturally grounded and knowledge-specific visual domain characterized by complex structures, dense symbolic elements, and domain-dependent semantic conventions. CIRThan contains 2,287 high-quality Thangka images, each paired with a human-drawn sketch and hierarchical textual descriptions at three semantic levels, enabling composed queries that jointly express structural intent and multi-level semantic specification. We provide standardized data splits, comprehensive dataset analysis, and benchmark evaluations of representative supervised and zero-shot CIR methods. Experimental results reveal that existing CIR approaches, largely developed for general-domain imagery, struggle to effectively align sketch-based abstractions and hierarchical textual semantics with fine-grained Thangka images, particularly without in-domain supervision. We believe CIRThan offers a valuable benchmark for advancing sketch+text CIR, hierarchical semantic modeling, and multimodal retrieval in cultural heritage and other knowledge-specific visual domains. The dataset is publicly available at https://github.com/jinyuxu-whut/CIRThan.

Line charts are a valuable tool for data analysis and exploration, distilling essential insights from a dataset. However, access to the underlying data used to create a line chart is rarely readily available. In this paper, we explore a novel dataset discovery problem, dataset discovery via line charts, focusing on the use of line charts as queries to discover datasets within a large data repository that are capable of generating similar line charts. To solve this problem, we propose a novel approach called Fine-grained Cross-modal Relevance Learning Model (FCM), which aims to estimate the relevance between a line chart and raw data from a candidate dataset. To achieve this goal, FCM first applies a visual element extractor to extract visual elements, i.e., lines and y-axis ticks, from a line chart. Then, two novel segment-level encoders are applied to learn representations for a line chart and a candidate dataset, preserving fine-grained information, followed by a cross-modal matcher that matchs the learned representations in a fine-grained manner. Furthermore, we extend FCM to support line chart query generated based on data aggregation. Last, we provide a benchmark tailored for this problem since no such dataset exists. Extensive evaluation on the new benchmark verifies the effectiveness of our proposed method. Specifically, our proposed approach surpasses the best baseline by 30.1% and 41.0% in terms of prec@50 and ndcg@50, respectively.

Abstract Table integration aims to create a comprehensive table by consolidating tuples containing relevant information. In this work, we investigate the challenge of integrating multiple tables from a data lake, focusing on three core tasks: (1) pairwise integrability judgment , which determines whether a tuple pair is integrable, accounting for any occurrences of semantic equivalence or typographical errors; (2) integrable set discovery , which identifies all integrable sets in a table based on pairwise integrability judgments established in the first task; (3) multi-tuple conflict resolution , which resolves conflicts between multiple tuples during integration. To this end, we train a binary classifier to address the task of pairwise integrability judgment. Given the scarcity of labeled data in data lakes, we propose a self-supervised adversarial contrastive learning algorithm to perform classification, which incorporates data augmentation methods and adversarial examples to autonomously generate new training data. Upon the output of pairwise integrability judgment, each integrable set can be considered as a community—a densely connected sub-graph where nodes and edges correspond to tuples in the table and their pairwise integrability, respectively—we proceed to investigate various community detection algorithms to address the integrable set discovery objective. Moving forward to tackle multi-tuple conflict resolution , we introduce an innovative in-context learning methodology. This approach capitalizes on the knowledge embedded within large language models to effectively resolve conflicts that arise when integrating multiple tuples. Notably, our method minimizes the need for annotated data, making it particularly suited for scenarios where labeled datasets are scarce. Since no suitable test collections are available for our tasks, we develop our own benchmarks using two real-world dataset repositories: Real and Join . We conduct extensive experiments on these benchmarks to validate the robustness and applicability of our methodologies in the context of integrating tables within data lakes.

Line charts are fundamental to data analysis and exploration, offering concise visual representations of trends. However, gaining access to the underlying data used to construct these charts is often challenging. In this paper, we describe DDLC (short for Dataset discovery via line charts), an automatic dataset discovery tool that is able to not only identify datasets (from a dataset repository) that are "relevant" to the information depicted from a line chart provided by the users, but also empower users to refine search results based on specific visual elements extracted from the line chart. Moreover, DDLC offers multiple avenues for users to validate search outcomes: 1) Providing explanations on how a similar line chart could be generated from the identified dataset; 2) enabling comparison of line charts generated from different datasets via different ways (e.g., the aggregation vs. non-aggregation operator); 3) facilitating fine-grained examination of the correspondence between the line chart and the identified dataset. By seamlessly combining dataset retrieval with visual refinement and validation mechanisms, DDLC offers a comprehensive solution for the data-driven exploration and analysis.

Visualization recommendation (VisRec) is to automatically generate the most relevant visualization for a table of interest to a user. In this paper, we present a novel machine learning-based VisRec method, VisFormer, which solves VisRec in three stages: 1) Table representation learning, which is to learn accurate column-level representations for a table. To achieve it, we resort to Transformer, a powerful language model that can learn accurate word embeddings by modeling context. Specifically, we propose a hierarchical Transformer-based architecture to learn expressive column representations by capturing two types of context, intra-column context and cross-column context; 2) Visual Relation Learning, which is to capture column relations. To achieve it, we regard each visualization as a relation tuple with a special relation, visual relation, between the columns. Then for each visual relation, we use a neural network to evaluate the corresponding visualizations; 3) Visual Preference Learning, which is to extract visual preference features that can affect users’ decision from a visualization. To achieve so, we use a Convolution Neural Network to extract such features and explore how to use them to refine the recommendation results. We conduct experiments to compare with three state-of-the-art ML-based methods on a large real-world dataset, Plotly community feed. The experimental results show that compared with the most competitive baseline, the relative improvements of VisFormer on Recall@1, Recall@2, and Recall@3 are 8.8%, 20.6%, and 21.0%, respectively.