Context-Enriched Key Phrase Extraction With Embedding Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing key phrase extraction methods, such as using parts-of-speech tagging and generating all potential noun phrase candidates or n-grams, often fail to produce optimal key phrases, leading to inefficiencies in natural language processing tasks.
Innovation Solution
A two-stage key phrase extraction method is employed, involving candidate phrase extraction followed by a contextual feature-based method using an enriched knowledge base. This approach vectorizes candidate phrases and context data, calculates similarity scores, and selects the top candidate phrases based on their relevance to the context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If parts-of-speech tagging and n-gram generation are used for key phrase extraction, then the process is simple and fast, but the accuracy and relevance of extracted key phrases deteriorates
Solution Approach 1:
The extraction process is divided into two distinct stages: (1) candidate phrase generation using simple n-gram methods for speed, and (2) contextual filtering using LLMs for accuracy. This segmentation allows each stage to optimize for its specific function without compromising the other.
Solution Approach 2:
An enriched knowledge base acts as an intermediary between the simple extraction process and the final key phrase selection. It provides contextual information that mediates between the speed of n-gram generation and the accuracy requirements of relevant phrase selection.
2Adaptability or versatility
If all potential noun phrase candidates and n-grams are generated, then comprehensive coverage is achieved, but computational resources and time are wasted
Solution Approach 1:
The system generates more candidate phrases than strictly necessary (excessive action) to ensure comprehensive coverage, then uses efficient filtering to eliminate unnecessary candidates. This approach ensures no relevant phrases are missed while minimizing processing time through selective retention.
Solution Approach 2:
Candidate phrase generation is performed as a preliminary step to create a comprehensive pool of potential key phrases. The computationally intensive contextual filtering is then applied only to this pre-generated list, optimizing the overall process efficiency.
3Measurement precision
If contextual information is incorporated into key phrase extraction, then key phrase relevance is improved, but system complexity increases
Solution Approach 1:
The LLM serves multiple functions: it generates candidate phrases, retrieves contextual information from the knowledge base, and filters relevant phrases. This multi-functionality reduces overall system complexity by consolidating multiple components into a single versatile model.
Solution Approach 2:
Instead of building a complex custom extraction system, the patent leverages pre-trained LLM capabilities as a copy of existing sophisticated NLP functionality. This approach achieves high relevance without the complexity of building such systems from scratch.
Data Source
AI summary
A plurality of candidate phrases are extracted from a first field of a received textual input. Also, first context data is extracted from a second field of the received textual input. Additionally, second context data is retrieved from one or more data sources related to the received textual input. The first context data and the second context data are combined to form combined context data. Then, the plurality of candidate phrases and the combined context data are vectorized. Next, for each candidate phrase of the plurality of candidate phrases, a similarity score is calculated between a vectorized version of the candidate phrase and a vectorized version of the combined context data. Then, a subset of candidate phrases are selected having highest calculated similarity scores. Next, the subset of candidate phrases are provided to one or more applications which are generating responses to the textual input.