Semantic Parsing for Short Text Using Syntactic Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing natural language processing (NLP) methods struggle to effectively parse short texts due to their sparsity and low number of features, which limits the ability to capture word co-occurrence information.
Innovation Solution
A method is proposed that involves obtaining a set of sentences associated with a short text, determining syntactic features such as constituency and dependency relations, and using these features to determine a semantic structure of the short text, thereby improving parsing accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing NLP methods are used to parse short text, then the processing can be performed, but the parsing accuracy is poor due to text sparsity and low feature count
Solution Approach 1:
The patent transforms the two-dimensional word co-occurrence matrix into a three-dimensional structure by introducing sentence-level semantic representations as an additional dimension. This allows the model to capture semantic relationships between words that co-occur in the same sentence, effectively increasing the feature dimensionality without adding more words to the input text.
Solution Approach 2:
The patent performs preliminary sentence-level semantic parsing before the main word-level parsing task. By first obtaining semantic representations of complete sentences using existing NLP models, the system prepares enriched feature vectors that capture semantic relationships in advance, which then guide the subsequent word-level parsing more accurately.
2Measurement precision
If more sophisticated NLP methods are used to improve parsing accuracy, then parsing performance improves, but computational cost and complexity increase
Solution Approach 1:
The patent introduces sentence-level semantic representations as an intermediary layer between the input short text and the final word-level parsing output. This intermediary captures semantic relationships at the sentence level and then feeds this information back to guide word-level parsing, effectively mediating between simple word processing and complex semantic understanding without requiring entirely new complex methods.
Solution Approach 2:
The patent makes existing NLP models serve multiple functions: they first perform sentence-level semantic representation generation, then their outputs are reused as features for word-level parsing. This multi-functional use of existing models improves parsing accuracy while avoiding the need to develop and deploy entirely new complex parsing systems.
Data Source
AI summary
Embodiments of the present disclosure relate to semantic parsing for short text. According to embodiments of the present disclosure, a method is proposed. The method comprises: obtaining a set of sentences associated with a short text, each of the set of sentences containing all of words in the short text; determining a set of syntactic features associated with the set of sentences, each of the set of syntactic features indicating at least one of a constituency relation and a dependency relation of the corresponding sentence; and determining a semantic structure of the short text based on the set of syntactic features.