mpdx 0.1.0a1

Semantic, graph-based document data model for text and tables

MPDX — Multi-Parent Document Exchange

MPDX is a semantic, graph-based document data model for representing textual and tabular documents independently of visual layout.

Rather than encoding documents as trees or grid coordinates, MPDX models documents as meaning-centered semantic graphs, enabling the same document semantics to be rendered, queried, analyzed, and reused across multiple forms.

This repository provides the reference specification, examples, and minimal tooling for MPDX.

---

Motivation

Most existing document formats—such as DOCX, HTML, and spreadsheets—encode documents as:

• Tree-based structures (DOM, XML)
• Row–column–oriented tables
• Layout-driven representations

While effective for rendering, these approaches make it difficult to:

• Explicitly represent semantic relationships between document elements
• Reuse document meaning across different layouts
• Apply documents directly to automation, analysis, or LLM-based processing

In particular, tabular data is usually defined by coordinates, not by meaning.
A cell is typically identified as “row × column,” even when its true meaning is a semantic combination (e.g., planned amount × student labor).

MPDX addresses this limitation by modeling documents as semantic graphs with multi-parent relationships, where values are defined by semantic intersections, not fixed positions.

---

Core Concepts

1. Semantic Graph Model

• All document elements are represented as nodes
• Nodes may have multiple parents, enabling semantic intersections
• The model is not constrained to tree structures

This allows document meaning to be expressed directly, rather than inferred from layout.

---

2. Minimal Node Types

MPDX intentionally defines a small, extensible set of node types:

Type	Description
table	Root node representing a document or table
title	Title of a document or table
t	Semantic text node (headers, labels, categories)
v	Value node representing a semantic intersection

---

3. Values as Semantic Intersections

In MPDX, a value is not defined as “row × column,” but as the intersection of multiple semantic dimensions.

Example:

• Parent A: Student Labor
• Parent B: Planned Amount

→ The resulting value represents:
“Planned amount of student labor”

This interpretation is independent of how the table is visually arranged.

---

MPDX Serialization

MPDX can be serialized using simple, human-readable, text-based formats.

This repository includes a TSV-based reference serialization:

• Each row represents a node
• Parent relationships are explicitly listed
• Designed as an intermediate representation, not a final rendering format

Example:

id parents type text
1 0 table
2 1 title Simple Budget Example
3 2 t Category
4 2 t Item
5 2 t Planned Amount
10 3 t Direct Cost
11 [10,4] t Student Labor
12 [11,5] v 19260

The serialization format is only one possible representation. MPDX as a model is not tied to any specific file format.

---

Rendering Independence

A key property of MPDX is the separation of meaning and rendering.

The same MPDX data can be rendered as:

• Hierarchical tables with merged cells
• Fully flattened analytical tables
• Pivoted or transposed views
• Database-friendly representations

All of these are different projections of the same underlying semantic model.

---

Use Cases

• Document automation and transformation
• Complex table modeling (budgets, reports, forms)
• LLM-friendly document structuring
• Semantic document analysis
• Intermediate representation between authoring and rendering
• Database-backed document systems

---

Project Status

MPDX is currently in an early, research-oriented stage.

• The core data model is stable enough for experimentation
• Reference serialization and examples are provided
• Tooling is intentionally minimal

This repository is published to:

• Establish the conceptual model
• Enable discussion and early adoption
• Support academic reference and extension

---

Related Publication

This repository accompanies the research paper:

A Semantic Graph-Based Document Model for Tabular and Textual Data MPDX: A Multi-Parent Document Exchange Format

(Preprint / publication details will be added.)

---

Contributing

MPDX is an open research and standardization-oriented project. Contributions, discussions, and experimental implementations are welcome.

Suggested contribution areas include:

• Alternative serializations
• Rendering rules
• Query models
• Tooling and converters
• Case studies and real-world applications

---

License

This project is released under the MIT License.

---

Contact

For questions, discussion, or collaboration:

• GitHub Issues
• Pull Requests

---

MPDX is not a document format. It is a semantic document model.

사용 방법

import mpdx

mp = mpdx.from_html("a.html")

inspect

for n in mp.find(type="t"): print(n.text)

modify

mp.findtype="t", text="19260".text = "20000"

render back

html = mp.to_html()