Topic-Based Paper Classification for Structured Proposal Reviews
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing literature review generation methods for research proposals lack structured subsections and fail to adequately map scientific articles to high-level thematic categories, leading to monolithic and biased reviews, especially in the early stages of proposal writing where citation texts are unavailable.
Innovation Solution
A processor-implemented method involving a two-stage approach using a topic-based retriever (TR) and classifier (TC) model to classify scientific papers based on topic, utilizing pseudo-positive and negative reference text spans, and enhanced regular expressions to extract relevant text chunks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If existing literature review generation methods are used, then automated review generation is achieved, but the reviews lack structured subsections and fail to map articles to thematic categories, resulting in monolithic and biased reviews
Solution Approach 1:
The patent segments the literature review into multiple thematic subsections by classifying retrieved scientific articles into different topic categories. The system divides the monolithic review into structured sections, each dedicated to a specific theme or topic, thereby preserving thematic information while maintaining automation.
Solution Approach 2:
The patent introduces an intermediary classification system that acts as a mediator between the automated retrieval process and the final literature review. This intermediary layer maps articles to thematic categories, ensuring that structured organization is maintained throughout the automated generation process.
2Measurement precision
If citation text based retrieval is used, then precise article retrieval is achieved, but citation texts are unavailable in early stages of research proposal writing
Solution Approach 1:
The patent performs preliminary retrieval of scientific articles using available proposal information (title, abstract, keywords) before citation texts are generated. By retrieving articles in advance based on preliminary criteria, the system eliminates waiting time while maintaining retrieval quality through subsequent refinement.
Solution Approach 2:
Instead of waiting for citation texts to become available before retrieving articles, the patent inverts the approach by first retrieving articles using available proposal metadata, then later refining the retrieval using generated citation texts. This reversal enables early-stage automation without sacrificing precision.
3Adaptability or versatility
If existing classification approaches are used, then article categorization is achieved, but the classification lacks alignment with proposal-specific topics and research contexts
Solution Approach 1:
The patent applies local quality by customizing the classification process for each specific research proposal. Instead of using a generic classification system, the model is trained and adapted to recognize topic-specific terminology, research contexts, and thematic structures relevant to each proposal, thereby ensuring reliable alignment with proposal-specific topics.
Solution Approach 2:
The patent dynamically adjusts classification parameters based on the specific research proposal being processed. The system modifies topic definitions, classification thresholds, and model weights to match the unique characteristics of each proposal, ensuring that article categorization reliably reflects the specific research context and topics.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
For categorizing literature for a specific target proposal, authors face challenges while organizing their papers in diverse ways. Embodiments of the present disclosure provide method and system for classification of scientific papers to research proposal based on topic. A constructed dataset with a positive and negative reference text spans are augmented to obtain an extended dataset. Top-k chunks from a reference paper relevant to the citation text are considered as the reference text spans. A topic-based retriever model is trained on subset of the extended dataset by the positive reference text span, and the negative reference text span. A topic classifier model is trained using research proposal title, the proposal topic, and the reference text span from the reference paper to classy if the reference paper is aligned to the topic. A reference paper to be relevant to the proposal topic is classified with corresponding topics in the research proposal.