Optimization and Practical Research on English Reading Machine Intelligence Algorithm Based on Transformer Architecture

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216208%3A11320%2F26%3AHHJNCKU9" target="_blank" >RIV/00216208:11320/26:HHJNCKU9 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1016/j.procs.2025.04.405" target="_blank" >http://dx.doi.org/10.1016/j.procs.2025.04.405</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.procs.2025.04.405" target="_blank" >10.1016/j.procs.2025.04.405</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Optimization and Practical Research on English Reading Machine Intelligence Algorithm Based on Transformer Architecture

Popis výsledku v původním jazyce

Some early English reading comprehension systems relied on rules and templates. Although these systems performed well in some specific tasks, they lacked flexibility and scalability and had difficulty in handling complex and changing language phenomena. This study deeply explores the core mechanisms of the Transformer model and its application in natural language processing tasks, focusing on the analysis of self-attention mechanism, multi-head attention mechanism, position encoding, and encoder-decoder structure. The self-attention mechanism adjusts the word vector representation by calculating the similarity between each input word and other words, thereby capturing the dependencies between words in the sequence. Position encoding solves the problem that the Transformer model lacks a time order processing mechanism, provides position information for each Token, and ensures the understanding of the sequence order. In addition, this paper also analyzes the encoder and decoder structure of Transformer, where the encoder is responsible for converting the input sequence into a context representation, and the decoder combines the output of the encoder to generate the target sequence. Experimental results show that the Transformer model has significant advantages in understanding complex and long texts and performing deep semantic reasoning, especially in multi-level reasoning tasks and cross-paragraph information integration. The model's personalized recommendation capabilities and contextual adaptability are also excellent, and it can provide accurate understanding and recommendations based on the needs of different users. © 2025 The Authors. Published by Elsevier B.V.

Název v anglickém jazyce

Optimization and Practical Research on English Reading Machine Intelligence Algorithm Based on Transformer Architecture

Popis výsledku anglicky

Some early English reading comprehension systems relied on rules and templates. Although these systems performed well in some specific tasks, they lacked flexibility and scalability and had difficulty in handling complex and changing language phenomena. This study deeply explores the core mechanisms of the Transformer model and its application in natural language processing tasks, focusing on the analysis of self-attention mechanism, multi-head attention mechanism, position encoding, and encoder-decoder structure. The self-attention mechanism adjusts the word vector representation by calculating the similarity between each input word and other words, thereby capturing the dependencies between words in the sequence. Position encoding solves the problem that the Transformer model lacks a time order processing mechanism, provides position information for each Token, and ensures the understanding of the sequence order. In addition, this paper also analyzes the encoder and decoder structure of Transformer, where the encoder is responsible for converting the input sequence into a context representation, and the decoder combines the output of the encoder to generate the target sequence. Experimental results show that the Transformer model has significant advantages in understanding complex and long texts and performing deep semantic reasoning, especially in multi-level reasoning tasks and cross-paragraph information integration. The model's personalized recommendation capabilities and contextual adaptability are also excellent, and it can provide accurate understanding and recommendations based on the needs of different users. © 2025 The Authors. Published by Elsevier B.V.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

—

Návaznosti

—

Rok uplatnění

2025

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Procedia Comput. Sci.

ISBN

—

ISSN

18770509

e-ISSN

—

Počet stran výsledku

10

Strana od-do

780-789

Název nakladatele

Elsevier B.V.

Místo vydání

—

Místo konání akce

Haikou

Datum konání akce

1. 1. 2026

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—