Share E-Book
Scan to open this page

Scan with your phone to open this page

Author: Enamul Hoque Prince

Rating 3.5 AI initial rating

Over the past few years, the rapid evolution of large language models (LLMs) has transformed the landscape of artificial intelligence and rekindled old dreams in visualization research — of making data communication conversational, adaptive, and accessible. Yet, despite the explosion of research at this intersection, there has been no comprehensive book dedicated to this timely and rapidly developing field. This volume aims to fill that gap, representing the synthesis of those developments: an attempt to articulate how language and visualization, when combined, can reshape how we reason with data, tell stories, and design tools that augment human intelligence. The book demonstrates a coherent account of the emerging landscape of LLM-driven visualization systems.

AI Reading Assistant

Whole-book reading guide from stratified index samples; jump to passages in the text

AI guide
【One-Line Pitch】 A research-grounded synthesis of how large language models are reshaping data visualization—from chart question answering to automated storytelling and accessibility—written for visualization researchers, HCI practitioners, and data scientists who want a structured map of a fast-moving field rather than a scattered pile of papers. 【Book Arc】 - **Opening (~0%–10%)**: Frames the core premise—natural language and visualization are complementary communication modes—and previews the book's scope, explicitly excluding visualization-for-AI (explaining model internals) in favor of LLM-for-visualization. - **Early (~10%–30%)**: Lays the conceptual groundwork: why text and charts work better together, the NLP/NLG split (understanding queries vs. generating explanations), and the book's three thematic phases—foundations, building and using, responsible and future. - **Middle (~30%–50%)**: Technical foundations of language models—self-attention, Transformer encoder/decoder variants (encoder–decoder, encoder-only, decoder-only), training pipelines, fine-tuning vs. parameter-efficient fine-tuning, and instruction tuning—plus multimodal LLM architectures and their implications for visualization. - **Late (~50%–75%)**: Application space: understanding and interacting with visualizations (deconstruction, chart QA, benchmarks), authoring and communicating with charts and text, and making visualizations accessible and inclusive. *(Excerpts only sketch these via the table of contents; detailed content is not covered.)* - **Ending (~75%–100%)**: Forward-looking themes—responsible AI, multimodality, and evaluation—positioned as the closing phase of the book's thematic arc. *(Excerpts do not cover the specific arguments of these chapters.)* 【Key Takeaways】 - **Language and visualization are complementary, not competing** (Early): Charts exploit preattentive processing to surface patterns; text supplies context and explanation. Research cited in the book shows users prefer annotated charts over bare charts or text alone—this pairing is the book's foundational argument. - **The field splits into two directions—understanding (NLP) and generation (NLG)** (Early): Systems must both interpret user queries about data and produce readable explanations or summaries. The book treats this input/output duality as the organizing logic for applications. - **Transformer architecture is the technical bedrock** (Middle): Self-attention, encoder/decoder variants, and cross-attention are explained with concrete analogies (e.g., pronoun resolution) before the book moves to visualization applications—so readers can connect model mechanics to downstream capabilities. - **Fine-tuning strategy matters for practical deployment** (Middle): The book contrasts full fine-tuning with parameter-efficient fine-tuning (PEFT) and instruction tuning, framing them as cost/adaptability trade-offs—useful for anyone building visualization tools on top of LLMs. - **Multimodal LLMs are the enabling technology for chart understanding** (Middle): Model architectures and training stages for multimodal models are covered specifically for their implications for visualization, bridging text-only LLMs to chart comprehension. - **Applications cluster into three areas** (Late): Understanding/interacting with visualizations, authoring/communicating with charts and text, and accessibility/inclusiveness—a practical taxonomy for scoping projects or literature reviews. - **Scope is deliberately bounded** (Early): The book excludes using visualization to explain LLM internals; it focuses on LLMs supporting visualization, which helps readers calibrate expectations before diving in. 【Reading Tips】 - **Skim Chapter 2 if you already know Transformers**; deep-read the sections on PEFT, instruction tuning, and multimodal implications, which are the bridge to visualization applications. - **Use the book preview and thematic-phase figure (Ch. 1) as a map**: the three phases—foundations, building and using, responsible and future—tell you which chapters are conceptual vs. applied. - **Treat the application chapters as a taxonomy, not a tutorial**: the excerpts show a problem-space structure (understand, author, accessibility); use it to organize your own reading of primary papers. - **Watch for benchmark references** (ChartQA, Chart-to-text, OpenCQA): these recur and are useful entry points if you want to reproduce or evaluate systems. - **Don't expect implementation code**: this is a synthesis lecture volume; pair it with primary papers for hands-on work. 【Coverage Limits】 The excerpts are heavily front-loaded—title pages, acknowledgements, table of contents, and early foundational chapters—so the guide's Late and Ending sections rely on chapter titles and the book preview rather than substantive content. Specific arguments, case studies, and conclusions in the application and future-work chapters are not covered by the available excerpts.
Page 9
slam, Shadikur Rahman, Aaryaman Kartha, and Puja Das. Their creativity, dedication, and enthusiasm have been at the heart of many ideas that shaped this book...
View in text
Page 14
e language models connect to external tools or data sources for tasks like calculations or information retrieval. ARIA Accessible Rich Internet Applications....
View in text
Excerpt 3
ences 11 putational Linguistics: ACL 2024, pp. 13921–13937. https://aclanthology.org/2024.findings-acl. 828/ (2024) 2. Center, P.R.: Positive Economic Views...
View in text
Excerpt 4
during pre-training, where parts of the input text are cor- rupted, and the model must reconstruct the original. This makes BART highly effective for text ge...
View in text
Excerpt 5
C., Chen, X., Yang, X., Wang, Z., Yu, G., Fu, B., Zhang, H.: ChartLlama: A multimodal LLM for Chart Understanding and Generation (2023) 5. Kantharaj, S., Leo...
View in text
Excerpt 6
Models 3.6.1 Model Architectures As illustrated in Fig. 3.10, most MLLMs combine three main components: (i) a visual encoder that converts an image into feat...
View in text
Excerpt 7
t text is presented separately from the visualization (e.g., a paragraph below or alongside a chart) without explicit links between the two [9]. In other cas...
View in text
Excerpt 8
oppose, neutral) and provides linked visual sum- maries.. ● Literary Analysis: Beyond news or social media, LLMs can support the study of narrative texts suc...
View in text
Tags
AI categories
Artificial IntelligenceDataTechnology
ISBN: 3032378737
Publisher: Springer
Publish Year: 2027
Language: English
Pages: 138
File Format: PDF
File Size: 15.0 MB
Text Preview (First 20 pages)
Registered users can read the full content for free

Register as a Gaohf Library member to read the complete e-book online for free and enjoy a better reading experience.

Generating text preview…