Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Question-answer matching method fusing deep representation and interaction model

A matching method and model technology, applied in biological neural network models, special data processing applications, instruments, etc., can solve problems such as easy omission of important semantic features and weak interactive models, and achieve the effect of overcoming semantic drift and strong resistance.

Active Publication Date: 2019-10-11
SHANDONG SYNTHESIS ELECTRONICS TECH
View PDF7 Cites 31 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Because the interactive model is too weak in feature extraction, it is easy to miss important semantic features

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Question-answer matching method fusing deep representation and interaction model
  • Question-answer matching method fusing deep representation and interaction model
  • Question-answer matching method fusing deep representation and interaction model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0035] This embodiment discloses a question-and-answer matching method for the fusion depth representation of the loose combination mode and the interaction model, such as figure 1 As shown, it is a model architecture that uses loose combination to deeply integrate the representational semantic matching model and the interactive matching model. The specific steps for question-answer matching based on the model architecture are:

[0036] S01), firstly carry out the complete deployment of the system, including the configuration of hardware resources, the construction of the development environment, the coding of the model, the training and evaluation of the model, and the pre-entry of dependent data. The training of the model is to use models such as ELMo and BERT to pre-train word vectors and character vectors, solidify the pre-trained model parameters locally, obtain a vocabulary list from a large-scale corpus, and traverse the vocabulary list to find the corresponding embeddin...

Embodiment 2

[0047] This embodiment discloses a question-and-answer matching method for a fusion depth representation of a tight combination mode and an interactive model, such as figure 2 As shown, it is the model architecture diagram of the tight combination of the depth representation and the interaction model. The external interface parameters of the tight combination model are consistent with those of the loose combination model. Both of them perform similarity calculations on the standard question traversal that is pre-entered in the knowledge base. , and then sort the results to recommend topN similar answers on the semantic level for users. Therefore, the two models can be easily replaced with each other without any modification to the external call function of the interface.

[0048] The specific process of using the tight combination mode for question-answer matching is as follows:

[0049] S01), first perform word segmentation and padding operations on the request text input b...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a question-answer matching method fusing deep representation and an interaction model. The method comprises the following steps: firstly, pre-training word vectors and character vectors of a text pair consisting of spoken questions input by a user and standard questions in a knowledge base; and then fusing the fusion depth representation and the interaction model in a loosecombination or tight combination mode, and performing question and answer matching on spoken questions input by the user by using the fused model. The universality of the service can be enhanced, therecognizable spoken language category can be expanded, the answer matching accuracy can be improved, and accurate semantic matching of wide semantic recognition can be realized.

Description

technical field [0001] The invention relates to a question-answer matching method that integrates depth representation and an interactive model, and belongs to the field of artificial intelligence, especially the field of artificial intelligence question-answer systems. Background technique [0002] Question-answer matching refers to the query proposed by the user. In the pre-recorded answer library, the most similar answer to the question is selected and fed back to the user. It is the core technology of the retrieval question answering system. The question-answer matching model is generally divided into two stages: recall and sorting. In the recall phase, traditional question-answer matching models such as keywords, BOW, TFIDF, etc. can only recall candidate answers literally. For example, a query related to "potato" cannot recall text containing "potato". In order to achieve semantic recall, a deep learning model based on large-scale text statistical training must be int...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06F16/332G06F17/27G06N3/04
CPCG06F16/3329G06F40/205G06F40/284G06N3/045
Inventor 王太浩朱锦雷张传锋申冲
Owner SHANDONG SYNTHESIS ELECTRONICS TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com