Language Storage Vector Dimension Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language storage methods for interactive robots face challenges in efficiently determining whether to store a speaker's spoken language, as simple vectorization leads to high processing time or reduced accuracy due to the large number of vectors.
Innovation Solution
A language storage method that involves vectorizing spoken language using word vectors, applying dimension reduction techniques like t-distributed Stochastic Neighbor Embedding (t-SNE) or principal component analysis, and using classification methods such as Gaussian mixture models or Support Vector Machines to determine whether to store the language, thereby reducing the number of dimensions and improving processing speed and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If simple vectorization of language is performed to determine whether to store spoken language, then the processing can be done, but the number of vectors becomes enormous and processing time becomes very long
Solution Approach 1:
The patent extracts only the essential semantic information from the full language vector by identifying and retaining key vectors that represent the core meaning, while discarding redundant dimensional information. This extraction process reduces the vector dimension from thousands to a manageable number while preserving the essential meaning needed for determination.
Solution Approach 2:
The patent introduces an intermediate processing step between full vectorization and final determination. This intermediate representation uses a reduced set of key vectors that serve as mediators, capturing the essential semantic information without the computational burden of the complete high-dimensional vector.
2Loss of time
If the number of vectors of the language is reduced to shorten processing time, then the processing time becomes short, but the accuracy of the determination whether to learn this language is reduced
Solution Approach 1:
The patent applies local quality by treating different vectors differently - some vectors are identified as key vectors that capture essential semantic information and are retained, while other non-key vectors are discarded. This selective retention ensures that the reduced vector set maintains high determination accuracy by preserving locally important information.
Solution Approach 2:
The patent performs preliminary identification of key vectors before the actual determination process. By pre-analyzing which vectors contain essential semantic information and marking them as key vectors in advance, the system ensures that the reduced vector set will maintain high accuracy during the determination process without needing to process all original vectors.
3Measurement precision
If high-dimensional vectors are used to maintain determination accuracy, then the accuracy is high, but the processing time becomes excessively long
Solution Approach 1:
The patent segments the high-dimensional vector into multiple components, identifying which segments (key vectors) contain essential semantic information. By dividing the vector processing into key and non-key segments, the system can process only the essential portions, maintaining accuracy while dramatically reducing processing time.
Solution Approach 2:
The patent applies partial action by processing only a subset of the full vector - specifically, only the key vectors that contain essential semantic information. This partial processing approach achieves sufficient determination accuracy without the excessive computational cost of processing the complete high-dimensional vector.
Data Source
AI summary
A language storage method according to one aspect of the present invention includes vectorization steps for vectorizing language accumulated in a dialog log database using a word vector created in advance for each word; determination step for determining whether to employ the language that has been vectorized as response data; and a storage step for accumulating the language that has been determined to be employed as the response data in a response database, in which a dimension reduction step for reducing the number of dimensions of the vector of the language vectorized in the vectorization step is included between the vectorization step and the determination step, and in the determination step, it is determined which one of an employed group and an unemployed group the vector of the language whose number of dimensions has been reduced in the dimension reduction step belongs to.