Sememe Prediction Using Dictionary Definitions and Negative Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sememe prediction methods using word embeddings fail to provide accurate predictions for low-frequency words and are sensitive to embedding quality, making it difficult to learn embeddings for new words.
Innovation Solution
A novel sememe prediction method that leverages online word dictionaries, employing multi-label text classification and a negative sampling strategy to improve the robustness of the prediction model by capturing semantic relevance between a word's definition and sememes, using a text matching model and two negative sampling methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If word embeddings are used for sememe prediction, then the method is simple and straightforward, but prediction accuracy deteriorates for low-frequency words and the method becomes sensitive to embedding quality
Solution Approach 1:
The patent introduces an intermediary component (dictionary-based semantic representation) between the input word and the sememe prediction. Instead of directly using word embeddings, the system first retrieves dictionary definitions and semantic information as an intermediate representation, which then feeds into the prediction model. This intermediary layer provides richer semantic context and reduces sensitivity to embedding quality, particularly for low-frequency words.
Solution Approach 2:
The patent combines multiple information sources (word embeddings, dictionary definitions, semantic relationships) into a composite representation for sememe prediction. Rather than relying solely on word embeddings, the system integrates multiple types of linguistic information, creating a more robust and accurate prediction framework that leverages the strengths of each component.
2Stability of the object's composition
If word embeddings are fixed after training, then the system is stable, but it becomes difficult to learn embeddings for new words
Solution Approach 1:
The patent performs preliminary actions by pre-loading comprehensive dictionary data and semantic information before the prediction process. This preliminary preparation creates a rich knowledge base that can immediately support new words without requiring retraining of embedding models. The system is pre-equipped with linguistic knowledge that can be applied adaptively to novel inputs.
Solution Approach 2:
The system enables new words to serve themselves by leveraging the pre-loaded dictionary and semantic resources. When a new word is encountered, the system automatically retrieves relevant semantic information from the dictionary without requiring external retraining or manual intervention. This self-service mechanism allows the system to adapt to new words while maintaining stability.
3Reliability
If all incorrect sememes are treated as negative labels, then comprehensive training is achieved, but training time increases significantly
Solution Approach 1:
The patent extracts and focuses on the most informative negative samples rather than using all incorrect sememes. By selecting a subset of critical negative examples (those most likely to be confused with positive sememes), the system achieves effective training without the computational burden of processing every possible negative label. This extraction approach maintains training reliability while improving efficiency.
Solution Approach 2:
The patent applies partial action by using a strategically selected portion of negative samples rather than the complete set. Instead of exhaustively training on all incorrect sememes, the system uses a carefully chosen subset that provides sufficient training signal. This partial approach achieves comparable or superior training effectiveness with reduced computational cost.
Data Source
AI summary
Method and apparatus for automatically predicting lexical sememes using a lexical dictionary, comprising inputting a word, retrieving the word's semantic definition and sememes corresponding to the word from an online dictionary, setting each of the retrieved sememes as a candidate sememe, inputting the word's semantic definition and candidate sememe, and estimating the probability that the candidate sememe can be inferred from the word's semantic definition.