Semantic Region Searchable Data Structure for Electronic Documents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The sheer quantity of electronic documents makes it difficult to retrieve specific information, as keyword searches fail to capture the context and layout information that humans use when searching, leading to inefficient information retrieval, especially in documents with structured or semi-structured content.
Innovation Solution
Generating searchable data structures that capture both text and context information from electronic documents, including their graphical and semantic layouts, to facilitate hierarchical arrangement and path-based searches, using machine learning models to assign category labels and generate semantic region categories.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If keyword searches are used to search electronic documents, then the search process is simple, but the search results do not capture context and layout information leading to inefficient information retrieval
Solution Approach 1:
The patent segments the document into multiple semantic regions based on layout and context (e.g., headings, paragraphs, tables, figures). Each region is independently analyzed and tagged with semantic labels, allowing the search system to retrieve information based on both keyword matching and semantic context, thereby improving retrieval efficiency while maintaining operational simplicity
Solution Approach 2:
The patent introduces an intermediary layer of semantic region annotations between the raw document text and the search query. This intermediary layer captures layout and contextual information that bridges the gap between simple keyword searching and complex information retrieval, enabling the system to understand document structure without complicating the user interface
2Measurement precision
If searchable data structures capture both text and context information including graphical layout, then information retrieval accuracy improves, but the data structure complexity increases
Solution Approach 1:
The patent extracts only the necessary layout and context information into separate semantic region annotations, rather than storing the complete original document structure. By taking out only the essential layout descriptors and semantic labels, the system achieves accurate information retrieval while reducing the complexity of the searchable data structure compared to storing full graphical representations
Solution Approach 2:
The patent adds a semantic dimension to the traditional text-based search index. Instead of only indexing text content, the system creates a multi-dimensional index that includes text, layout position, semantic labels, and contextual relationships. This dimensional expansion enables accurate retrieval of context-aware information without requiring the entire complex graphical layout to be stored verbatim
3Productivity
If the searchable data structure uses hierarchical arrangement for structured or semi-structured content, then path-based searches retrieve information more efficiently, but the processing time to generate the data structure increases
Solution Approach 1:
The patent performs preliminary analysis of document layout and semantic regions during the indexing phase, creating hierarchical structures and path-based search indices in advance. This preliminary action prepares the data structure before actual search queries are executed, enabling fast path-based retrieval during query processing while distributing the processing time across the indexing and query phases
Solution Approach 2:
The patent creates a dynamic hierarchical data structure that adapts to different document types and structures. The hierarchy is generated based on the specific layout and semantic content of each document, allowing the system to optimize the data structure for the particular search patterns needed. This dynamic generation enables efficient path-based searches while managing processing time through adaptive rather than fixed structural approaches
Data Source
AI summary
A method includes obtaining, at a device, a hierarchical structure representing a graphical layout of content items of an electronic document, the content items including at least text. The method also includes generating a word embedding representing a word of the electronic document. The method further includes determining position information of a location of the word in the electronic document. The method also includes determining a descriptor that indicates a relationship of the location to the hierarchical structure. The method further includes providing input data to a machine learning model to generate a semantic region category label of a semantic region of the electronic document. The semantic region includes the word. The input data includes the word embedding, the position information, and the descriptor.