Natural Language Generation (NLG) is the main use of NLP in information visualization, used to supplement or summarize visual data displays.
NLP visualization
Visualization is a broad NLP method that aids in understanding text data, language models, and algorithms outside of this particular application field. Making intricate linguistic patterns, semantic links, and textual material easier to understand and interpret is the aim. The following are some instances of visualization strategies that are discussed:
- Word clouds, where word frequency is shown by font size.
- Word frequencies or other data categories can be shown using bar charts.
- The distribution of text metrics, such as word or sentence lengths, is displayed via histograms.
- Geographic mapping of textual data, including sentiment analysis findings from various places.
- Contextualizing visualizations by adding labels, tooltips, or annotations.
- Showing a hierarchical representation of related phrases according to their embeddings with the use of clustering techniques.
- Use techniques such as t-SNE to project high-dimensional word embeddings into two dimensions.
- Interactive graphics for examining intricate systems or huge text databases.
Explainable AI (XAI) for Natural Language Processing(NLP) models uses visualization approaches such feature importance methods that highlight important characteristics and attention maps that display word contributions.
Information Visualization with NLG
- A software program can use natural language to convey information to the user to Natural Language Generation.
- The goal of NLG applications in information visualization is to assist users in obtaining valuable information, frequently in fields like healthcare and education.
- These systems produce spoken and written content, usually to supplement visual displays. The “second-generation multimedia presentation systems” that employ this method are distinct from language-only systems like document search or summarization. Different media deliver information from different angles.
- Integrating text creation with information visualization frameworks is the subject of recent study.
- External data sources such as medical records, numerical time-series data, online bar chart pictures, and corpora of evaluative language can be used by these systems. The intention is to supplement or replace visual presentations with text that provides a summary of the knowledge source.
Systems and Research Examples
Several examples of systems that combine information visualization and NLG are described in the give below:
- Systems (like CLEF) that give healthcare providers with a summary of a patient’s medical history.
- Systems (like the BT-45) that produce text summaries for graphical data.
- AutoBrief is a technology that uses information graphics and text to create presentations on transportation timetables. By combining text creation with an information visualization architecture, AutoBrief enables users to engage with the generated text or visuals to examine the underlying data. It links the perceptual tasks that a graphic should support to communicative objectives.
- A tool for creating interactive multimedia summaries of extensive collections of evaluative text, including product reviews. Usually, the interface has a generated written summary on one side and an interactive visualisation on the other. While the textual description serves to orient the user prior to interaction, users typically examine data through the visualisation. This system aggregates information, structures the summary, and determines the features, polarity, and strength of judgements through content selection.
- The SIGHT project aims to give visually challenged users alternative access by creating text summaries of the primary communication objective of bar charts that appear on the Web. Using criteria such as perceptual task effort, element salience, and terminology in the accompanying text, SIGHT evaluates the visual components of the bar chart, uses Bayesian plan recognition to infer its communicative goal, and creates a template-based summary using corpus-derived heuristics for data descriptions. SIGHT-generated texts were assessed by evaluators as good to very good, which is much better than baselines.
- An older example is the Ana system, which uses linguistically communicative language to make sense of raw data such as trading volumes and stock indices.
- A bar chart visualization is used in Microsoft Bing search opinion summaries of product reviews.
- Research comparing textual and graphical time series data displays to aid in medical decision-making.
In conclusion, visualization is a commonly used tool within NLP itself for data and model analysis, even though “information visualization” is a specific domain where NLG is applied to provide textual descriptions or summaries that complement visual displays.