Neural Network Action Extraction for Natural Language Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems using AI for natural language processing often fail to accurately extract the 'action of interest' from user queries, leading to inaccurate responses as they primarily focus on user intent without considering the action, resulting in misinterpretation of user intent.
Innovation Solution
A method and device that create an input vector for each word in a sentence using Part of Speech (POS) vectors, word embeddings, and dependency labels, processed through a trained neural network with Rectified Linear Units (RELU) activation to identify and tag the 'action of interest', enabling the extraction of relevant action text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If conventional systems focus only on user intent identification, then the system simplicity is maintained, but the accuracy of understanding user queries deteriorates
Solution Approach 1:
The patent segments user intent into two distinct components: action of interest (AOI) and center of interest (COI). By dividing the intent identification task into separate extraction processes for actions (verbs) and centers (nouns), the system achieves more precise understanding without excessive complexity increase. The neural network is trained separately to identify AOI tags, and then COI is extracted based on the identified actions.
2Productivity
If only center of interest is mined without action of interest, then the processing speed is maintained, but the accuracy of response generation deteriorates
Solution Approach 1:
The patent applies preliminary action by first identifying the action of interest (AOI) through neural network processing before extracting the center of interest (COI). The AOI tags are assigned to words in advance, which then guides the COI extraction process. This preliminary identification of actions ensures that the subsequent COI extraction is more accurate, leading to better response generation while maintaining efficient processing through the streamlined two-stage approach.
3Adaptability or versatility
If multiple verbs are present in user utterance, then the versatility of language understanding is improved, but the difficulty of identifying the correct action of interest increases
Solution Approach 1:
The patent employs feedback mechanisms through the neural network that processes each word in the sentence and assigns AOI tags based on contextual information from surrounding words. The network uses feedback from the input vector (including POS vectors, word embeddings, and dependency labels) to accurately identify which verb among multiple options is the action of interest. This contextual feedback allows the system to handle multiple verbs effectively by evaluating each in its linguistic context.
Data Source
AI summary
A method and device for extracting Action of Interest (AOI) from natural language sentences is disclosed. The method includes creating an input vector comprising a plurality of parameters for each target word in a sentence inputted by a user. The method further includes processing for each target word, the input vector through a trained neural network with RELU activation, which is trained to identify AOI from a plurality of sentences. The method includes assigning AOI tags to each target word in the sentence based on processing of associated input vector through the trained neural network with RELU activation. The method further includes extracting AOI text from the sentence based on the AOI tags assigned to each target word in the sentence. The method further includes providing a response to the sentence inputted by the user based on the AOI text extracted from the sentence.