Retrieval Reply Dialogue Method and System Combining Keywords and Semantic Understanding Representation

A technology for semantic understanding and keywords, applied in the field of retrieval-based reply dialogue methods and systems, can solve problems such as failure to capture key information, failure to achieve response effects, failure to filter out information, etc., to improve fluency and naturalness, The effect of eliminating errors and improving the quality of dialogue

Active Publication Date: 2022-03-25
ZHEJIANG LAB
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In the current technology, the RNN-based model is often used to obtain the semantic representation of the dialogue text. When the text is too long, the key information is often not captured, and the redundant information cannot be filtered out. The quality of the relevant replies retrieved not tall
However, simply using keyword representations to retrieve relevant matching replies does not achieve a smooth and natural reply effect in terms of semantics.

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
  • Retrieval Reply Dialogue Method and System Combining Keywords and Semantic Understanding Representation
  • Retrieval Reply Dialogue Method and System Combining Keywords and Semantic Understanding Representation
  • Retrieval Reply Dialogue Method and System Combining Keywords and Semantic Understanding Representation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0083] Specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the present invention, and are not intended to limit the present invention.

[0084] Such as figure 1 As shown, a search-based reply dialogue method that combines keywords and semantic understanding representations includes the following steps:

[0085] S1. According to the dialogue text corpus, obtain the single-sentence dialogue text and the word segmentation information of the single-sentence dialogue;

[0086] Collect Chinese dialogue text corpus[ ], respectively disassembled to obtain all single-sentence dialogue texts [[ ]] and word segmentation information, train the word2vec model, and save the word2vec model;

[0087] Preprocess the dialogue text, and respectively predict the dialogue [ ] is processed into a single-round...

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 search-type reply dialogue method and system combining keywords and semantic understanding representations. The system combines vector representations of two levels of granularity, which are respectively bag-of-words vector representations and semantic understanding representations. Keyword information and context-based semantic understanding are also considered, which greatly improves the performance of the retrieval reply model. In the present invention, the Chinese pre-training model Bert network model is adopted to obtain the sentence vector representation, which not only understands the sentence meaning, but also eliminates the error caused by the word vector weighting. The system uses the Bert network model to train the classification task on its own single-round dialogue—the task of whether the dialogue matches or not. Through fine-tuning, it learns the weight of the linear layer and activation function in Bert. The system uses the fine-ranking model LGMRanker, which can directly predict the relative order of replies related to the query and return a sorted list.

Description

technical field [0001] The invention relates to the field of artificial intelligence retrieval reply dialogue, in particular to a retrieval reply dialogue method and system combining keywords and semantic understanding representations. Background technique [0002] At present, the dialogue system has attracted more and more attention in various fields. It is mainly a system that allows machines to understand and process human language through dialogue. Dialogue processes that can be modeled. Dialogue modeling is not a simple task, it is a comprehensive entity involving technologies in multiple directions such as understanding, generation, and interaction. The complexity of its dialogue scenarios, such as customer service, voice assistants, chatting, etc., also creates the complexity of the dialogue system. [0003] Retrieval dialogue is a classic solution that abstracts a dialogue problem into a search problem. Early dialogue systems were implemented using this solution. U...

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 Patents(China)
IPC IPC(8): G06F16/332G06F16/33G06F16/335
CPCG06F16/3329G06F16/3334G06F16/3344G06F16/3347G06F16/3346G06F16/335
Inventor 李太豪张晓宁阮玉平郑书凯
Owner ZHEJIANG LAB
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
Try Eureka
PatSnap group products