Keyphrase Generation With Relevance-Diversity Balancing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing keyphrase generation approaches lack diversity and fail to model similarities among generated phrases, resulting in duplicative and non-diverse output.

Innovation Solution

A method that uses a trained keyphrase generation model to generate a set of keyphrases, calculates relevance scores, and adjusts them using a diversity balancing function, incorporating techniques like Maximal Marginal Relevance (MMR) and Discounted Cumulative Gain (DCG) to enhance diversity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If existing keyphrase generation approaches are used, then keyphrases can be generated from documents, but the output lacks diversity and contains duplicative phrases

Engineering Contradiction:

Improvekeyphrase diversityVSAvoidoutput quality

Core Design Contradiction:

Quantity of substanceVSEase of operation

Solution Approach 1:

The system calculates similarity scores between generated keyphrases and feeds this information back into the selection process. The diversity balancing function uses this feedback to adjust selection probabilities, reducing the likelihood of selecting duplicative keyphrases while maintaining relevance to the document.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The invention introduces a diversity balancing function that dynamically adjusts the selection parameters based on the similarity between keyphrases. By modifying the selection criteria to account for diversity, the system generates a more varied set of keyphrases without sacrificing document relevance.

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If present keyphrases are generated only from explicit text, then extraction is simple, but the summarizer cannot link text to concepts not explicitly mentioned

Engineering Contradiction:

Improveconcept inference capabilityVSAvoidsummarizer complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary processing by generating candidate keyphrases from the document text before applying the diversity balancing function. This preliminary extraction of present keyphrases provides a foundation that the subsequent diversity-adjusted selection process builds upon, enabling both explicit and implicit concept representation.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If keyphrase generation models generate phrases one at a time without referencing previous phrases, then generation is simple, but similarities among generated phrases are not modeled

Engineering Contradiction:

Improvesimilarity modeling accuracyVSAvoidgeneration efficiency

Core Design Contradiction:

ReliabilityVSProductivity

Solution Approach 1:

The diversity balancing function incorporates feedback from previously generated keyphrases by calculating similarity scores. This feedback mechanism allows the system to model similarities among generated phrases while maintaining an efficient generation process through probabilistic sampling rather than exhaustive evaluation.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12436988B2Keyphrase generation

Publication Date: 2025.10.07 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Using a trained keyphrase generation model, a set of keyphrases corresponding to an input document is generated, a keyphrase in the set of keyphrases comprising a word summarizing a portion of a document. A relevance score measuring a similarity between the keyphrase and the document is calculated for a keyphrase in the set of keyphrases. The relevance score is adjusted according to a diversity balancing function.