Compound Word Splitting With LLM Voting for Search Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Search engines struggle to accurately split compound words in languages like German, leading to incorrect search results due to improper word splitting, which affects vector-based and keyword-based systems, especially in domain-specific contexts.
Innovation Solution
Utilize a large language model (LLM) with varied prompts and a voting technique to generate and select the most appropriate word split for compound words, incorporating domain-specific data to enhance contextual relevance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional linguistic rules or statistical methods are used to split compound words, then the process is simple and fast, but the accuracy of word splitting is poor leading to incorrect search results
Solution Approach 1:
The system segments the compound word splitting task into multiple independent LLM prompts, each generating candidate splits. Different prompts with varying temperatures explore different splitting possibilities, and the results are aggregated through voting to determine the final split. This segmentation allows the system to achieve high accuracy without requiring a single complex model.
Solution Approach 2:
The system changes the temperature parameter of the LLM across different prompts to generate diverse candidate word splits. By varying this parameter, the system explores different levels of creativity and randomness in splitting, allowing it to find the most accurate split through voting rather than relying on a single fixed-parameter model.
2Measurement precision
If multiple domain-specific models are trained to handle different contexts, then the accuracy for domain-specific contexts improves, but the computational cost and training time increase significantly
Solution Approach 1:
The system uses a single universal LLM that can handle multiple domains and contexts through prompt engineering rather than training separate domain-specific models. The same model adapts to different domains (e.g., medical, legal, technical) by receiving domain-appropriate prompts and example compound words, eliminating the need for multiple specialized models and reducing computational costs.
Solution Approach 2:
The system performs preliminary preparation by collecting example compound words from domain-specific data sources and incorporating them into prompts before the actual splitting task. This preliminary action allows the LLM to understand domain-specific compound word patterns without requiring domain-specific training, achieving domain adaptability through data preparation rather than model training.
3Reliability
If a single LLM prompt is used to generate word splits, then the process is fast and simple, but the reliability of the selected word split is low
Solution Approach 1:
The system merges multiple LLM prompt results through a voting mechanism to determine the final word split. By combining the outputs of several prompts with different temperatures and configurations, the system achieves higher reliability through consensus, selecting the split that appears most frequently across different prompts rather than relying on a single prompt result.
Solution Approach 2:
The voting mechanism provides feedback by evaluating the frequency and consistency of different candidate splits across multiple prompts. The system uses this feedback to identify the most reliable split, effectively using the collective output of multiple prompts to guide the selection process and improve reliability beyond what any single prompt could achieve.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The technology relates to determining word splits for compound words using a large language model (LLM). It can be used to enhance search engine performance in languages where words are often combined as compound words, such as German and Dutch. An example method involves prompting the LLM with different prompts to generate multiple candidate word splits for a compound word. A voting technique is applied to select the most appropriate word split. The method may include using different LLM temperatures and compound word-word split pairs from a domain-specific dataset as examples within the prompts. The voting technique may identify the word split that appears most frequently. If no majority, the method selects a candidate word split based on the number of splits, either the highest or lowest, and in some cases, selects a random word split from candidate word splits with the highest or lowest number of splits.