ĈһWYҩQÿӅ
+LӋQQD\YӟLVӵSKiWWULӇQPҥQKPӁFӫDLQWHUQHWYjӭQJGөQJFӫDPi\KӑF FQJYӟL OѭӧQJWK{QJWLQNKәQJOӗFӫDFiFQJѭӡLGQJ YuYұ\VӕOѭӧQJQKӳQJEӝ SKLPWăQJOrQ FjQJQJj\ÿDGҥQJYjSKRQJSK~FҧQӝLGXQJYjWKӇORҥL9ҩQÿӅÿһWUDOjÿӇWuPNLӃP PӝWEӝSKLPKRһFVҧQSKҭPÿӅXÿҫXWLrQOjQJѭӡLGQJVӁJ}WUrQWKDQKF{QJFөtìm NLӃPFӫDWUDQJZHEFөWKӇ&iFF{QJFөWuPNLӃPFyWKӇÿѭDUDGDQKViFKVҧQSKҭP NK{QJFKtQK[iFQӃXEӓTXDQJӳFҧQKFөWKӇ Hình 1-1 YtGөQJѭӡLGQJFҫQWuPiRVѫ
Amazon is no longer the only game in town for online shopping Competitors like Walmart, Target, and eBay are all vying for a piece of the e-commerce pie To stay ahead of the competition, Amazon is investing in new technologies and expanding its product offerings The company is also placing a greater emphasis on customer service By offering a wider selection of products, better prices, and improved customer service, Amazon is well-positioned to maintain its dominance in the e-commerce market.
+ӋWKӕQJNKX\ӃQQJKӏ5HFRPPHQGHU6\VWHPYLӃWWҳWOj56OjKӋWKӕQJVӱGөQJFiFNӻ WKXұWYjF{QJFөSKҫQPӅP[ӱOêGӳOLӋXQJѭӡLGQJYjGӳOLӋXYӅVҧQSKҭPQKҵPÿѭD UDJӧLêYӅPӝWVҧQSKҭPKRһF GӏFKYөFKRQJѭӡLGQJ 0ӝWKӋWKӕQJNKX\ӃQ QJKӏ NK{QJWKӇÿѭӧFWULӇQNKDLQӃXFKѭDTXDÿiQKJLi9LӋFÿiQKJLiPӝWKӋWKӕQJNKX\ӃQ
Hình 1-1 M{SKӓQJKӋNKX\ӃQQJKӏ
2 QJKӏ OjPӝWJLDLÿRҥQFҫQWKLӃWYuKLӋXTXҧFӫDPӝWKӋWKӕQJNKX\ӃQ QJKӏ NK{QJFKӍSKө WKXӝFYjRÿһFÿLӇPGӳOLӋXPjFzQSKөWKXӝFYjRPөFÿtFKJӧLý
&ҩX WU~FFӫDPӝWKӋNKX\ӃQQJKӏJӗPFiFWKjQKSKҫQFKtQKWұSKӧSQJѭӡLdùng, WұS KӧSFiFPөFGӳOLӋX YjWұSKӧSFiFPӕLTXDQKӋJLӳD QJѭӡLGQJYjPөFGӳOLӋX x Mөc dӳ liӋu (Item) là thuұt ngӳ FKXQJÿӇ chӍ nhӳQJJuQJѭӡLGQJFyWѭѫQJWiF trong hӋ thӕng khuyӃn nghӏ Item có thӇ là sách, phim, truyӋn, tin tӭc x 1Jѭӡi dùng (User): là tұp hӧp nhӳQJQJѭӡLGQJWѭѫQJWiFWURQJKӋ thӕng x Mӕi quan hӋ giӳ QJѭӡi dùng và mөc dӳ liӋu có thӇ là quan hӋ ÿiQK giá, yêu thích, PXDÿӑc, x Sessions: là nhӳng lӏch sӱ WѭѫQJWiFFӫDQJѭӡi dùng trên nhӳng mһt hàng, sҧn phҭm (vớ dө nhӳQJÿiQKJLiSKLPOѭӧWPXDKjQJô
7K{QJWKѭӡQJKӋWKӕQJNKX\ӃQQJKӏVӁÿѭӧF[k\GӵQJSKKӧSYӟLPӝWORҥLKuQKVҧQ SKҭPQKҩWÿӏQKÿӇÿҧPEҧRKLӋXTXҧNKX\ӃQQJKӏFKRVҧQSKҭPÿy +ӋWKӕQJÿҫXWLrQ (collaborative filtering - CF) ÿѭӧFUDÿӡLQăP [1] &iFWUDQJWKѭѫQJPҥLÿӅXKѭӟQJ WӟLVӱGөQJKӋNKX\ӃQQJKӏÿӇWăQJWtQKWLӋQtFKYjWKXK~WOѭӧQJOӟQNKiFKKjQJFKR GӏFKYө1ӃXFyVӵWUӧJL~SFӫDFiFKӋWKӕQJNKX\ӃQQJKӏWKuQJѭӡLGQJVӁWLӃWNLӋP QKLӅXWKӡLJLDQÿӇWuPUDVҧQSKҭPSKKӧSYӟLQKXFҫX FӫDKӑJLӳDKjQJWULӋXPһWKjQJ WUrQFiFWUDQJWKѭѫQJPҥLÿLӋQWӱ
+ӋWKӕQJVӁGӵDWUrQVӵWѭѫQJWiFFӫDQJѭӡLGQJÿiQKJLiWURQJKӋWKӕQJWKHRQJӳFҧQKWKӡLJLDQ, tӯÿyKӋWKӕQJVӁÿѭDUDGDQKViFKQKӳQJVҧQSKҭPSKKӧSQKҩW +LӋQQD\FyUҩWQKLӅXQJKLrQFӭXOLrQTXDQYӅVӱGөQJF{QJQJKӋKӑFVkXFKRKӋNKX\ӃQQJKӏQKҵPWuPUDJLҧLSKiSÿHPOҥLNӃWTXҧFDR[2]&iFKӋWKӕQJNKX\ӃQQJKӏWKѭӡQJGQJFKӍFyhai WKӵFWKӇOjQJѭӡLGQJXVHUVYjVҧQSKҭPLWHPV Hình 1-2 WKӇKLӋQPӕLOLrQKӋWKtFKJLӳDQJѭӡLGQJYjVҧQSKҭP
3 Hình 1-20{SKӓQJWѭѫQJWiFFӫDQJѭӡLGQJYjVҧQSKҭP
0DWUұQWѭѫQJWiF1JѭӡLGQJ± 6ҧQSKҭP8WLOLW\0DWUL[8VHU± ,WHPPDWUL[OjPӝW FѫVӣGӳOLӋXP{WҧVӣWKtFKFӫDPӛL8VHUYӟLWӯQJ,WHPWURQJKӋWKӕQJ
Hình 1-30DWUұQWѭѫQJWiFXVHU± item
1KӳQJ JLiWUӏWURQJEҧQJGӳOLӋXOjQKӳQJÿiQKJLiFӫDQJѭӡLGQJWUrQWӯQJPһWKjQJ QKӳQJGҩXKӓLOjQKӳQJÿiQKJLiFKѭDELӃWFӫDQJѭӡLGQJ7UrQWKӵFWӃPDWUұQWѭѫQJ tác 1J˱ͥLGQJ± 6̫QSḴP QyLWUrQWKѭӡQJEӏNKX\ӃWJLiWUӏӣUҩWQKLӅXô, YҩQÿӅQj\ ÿѭӧF JӑLOjY̭Qÿ͉GͷOL XWK˱D (data sparsity) [3] %rQFҥQKÿyWUѭӡQJKӧSWURQJKӋ WKӕQJ[XҩWKLӋQQJѭӡLGQJKRһFLWHPPӟLFKѭDFyWѭѫQJWiFQjRWKuPDWUұQ1J˱ͥL dùng ± 6̫QSḴP VӁ[XҩWKLӋQFiFKjQJFӝWEӏWUӕQJKRjQWRjQ9ҩQÿӅQj\ÿѭӧFJӑLOj
YҩQÿӅNKӣLÿӝQJQJXӝLcold-start problem) [3] 0ӝWSKLrQFӫDQJѭӡLGQJVHVVLRQ EDRJӗPPӝWKRһFQKLӅXPөFSKLPÿѭӧFQJѭӡLQj\ÿiQK giá OjPӝWFKXӛLGӳOLӋXFy WUuQKWӵ &iFKWLӃSFұQFӫDÿӅWjLQj\KӑFviên VӱGөQJNӻWKXұWKӑFVkXGӵDWUrQ WѭѫQJ WiFQJѭӡLGQJJӑLOj session-based ĈӇFyÿѭӧFKӋNKX\ӃQQJKӏFyÿӝFKtQK[iFFDROjPӝWYҩQÿӅWKiFKWKӭFFiFQKjQJKLrQFӭXNK{QJQJӯQJWLӃSWөFFҧLWKLӋQÿӇWuPUDP{KuQKFKRUDNӃWTXҧFjQJQJj\FjQJFKtQK[iFKѫQ /XұQYăQ này KӑFYLrQ ÿӅ[XҩWÿӅWjL³HӋNKX\ӃQQJKӏ SKLPҧQK cho
QJѭӡLGQJVӱGөQJF{QJQJKӋKӑFVkX´FKRKӋWKӕQJ[HPSKLPFyWKӇKӛWUӧQJѭӡL
GQJGӉGjQJWuPNLӃPQKӳQJEӝSKLPSKKӧSYӟLÿLӅX NLӋQ QJӳFҧQK[XQJquanh, dӵD trên QKӳQJWLӃSQӕLWULWKӭFFӫDQKӳQJF{QJWUuQKÿmFyQKҵPFҧLWKLӋQWӕWKѫQYӅP{ KuQKKӋNKX\ӃQQJKӏ
0өFWLrXÿӅWjL³HӋNKX\ӃQQJKӏ SKLPҧQKFKRQJѭӡLGQJVӱGөQJF{QJQJKӋKӑF sâu´ x Nghiên cӭu lý thuyӃW Yj FiF SKѭѫQJ pháp xây dӵng hӋ khuyӃn nghӏ dӵa trên chuӛi thӡi gian x Xây dӵng hӋ khuyӃn nghӏ FKRQJѭӡi dùng xem phim bҵng sӱ dөng công nghӋ hӑc sâu x KiӇm thӱ mô hình vӟi dӳ liӋXYjÿiQKJLikӃt quҧ
1.3 1KLӋP YөÿӅWjL ĈӇÿҥWÿѭӧFPөFWLrXÿӅUDFҫQ SKҧLWLӃQKjQKYӟLFiFQKLӋPYө x 3KkQWtFKFiFSKѭѫQJSKiSP{KuQKFyVҹQÿmWӗn tҥi x ThiӃt kӃ và xây dӵng hӋ thӕng khuyӃn nghӏ vӟLP{KuQKÿӅ xuҩt LSTM dӵa trên lӏch sӱ WѭѫQJWiFQJѭӡi dùng có yӃu tӕ thӡi gian x ĈiQKJLiNӃt quҧ và so sánh
Vì thӡi gian nghiên cӭXÿӅ tài có hҥn nên phҥm vi nghiên cӭu FNJQJÿѭӧc giӟi hҥQQKѭ sau: x Dӳ liӋu là nhӳng bӝ phim thӵF Pj QJѭӡL GQJ ÿm WѭѫQJ WiF (tұp dӳ liӋu MovieLens) x 6ӱGөQJmô hình KӑFVkX/670 QKҵPKXҩQOX\ӋQGӳOLӋX
ĈӕL WѭӧQJQJKLrQFӭX
ĈӅWjLQj\KѭӟQJWӟLYLӋFQJKLrQFӭX FiFÿӕLWѭӧQJ:
5 x Tұp dӳ liӋu MovieLens cӫa GroupLens [4] x Word2vec x Mô hình LSTM
1JKLrQFӭXWәQJTXDQYӅOêWKX\ӃWYjFiFSKѭѫQJSKiSNKX\ӃQ QJKӏ YӟLGӳOLӋXWKӡL gian và FѫVӣWKӵFQJKLӋPÿӇ[k\GӵQJKӋWKӕQJNKX\ӃQQJKӏ
%jLEiRFiRÿѭӧFWәFKӭFWKjQKFKѭѫQJYӟLFiFQӝLGXQJFөWKӇQKѭVDX
&KѭѫQJ*LӟLWKLӋXÿӅWjL - 7UuQKEj\VѫOѭӧFYӅÿӅWjLPөFtiêuJLӟLKҥQÿӕLWѭӧQJ
YjSKҥPYLQJKLrQFӭXFӫDÿӅWjL
&KѭѫQJ&ѫVӣOêWKX\ӃW - 7UuQKEj\PӝWVӕNLӃQ WKӭF lirQTXDQÿӃQFiFNӻWKXұWKӑF
VkXFyOLrQTXDQÿѭӧFÿӅFұSÿӃQWURQJÿӅWjLnày QKѭJLӟLWKLӋXYӅPҥQJQѫ-ron, RNN, /670YjNӻWKXұWWord2vec
&KѭѫQJ&{QJWUuQKQJKLrQFӭXOLrQTXDQ - 6ѫ OѭӧFTXDFiFF{QJWUuQKQJKLrQFӭX
OLrQTXDQÿӃQÿӅ tài TӯÿyÿӅ[XҩWUDKѭӟQJQJKLrQFӭXVӁWKӵFKLӋQÿӅWjLOXұQYăQ
&KѭѫQJ0{KuQKÿӅ[XҩW - Trình bày mô hình ÿӅ [XҩWKӋNKX\ӃQQJKӏSKLPҧQKFy
\ӃXWӕWKӡLJLDQ /XұQ YăQVӱGөQJP{KuQK/670ÿӅ[XҩW
&KѭѫQJ7KӱQJKLӋPYjNӃWTXҧ - 7UuQKEj\FiFWұSGӳOLӋXVӁÿѭӧFWKӵFQJKLӋP
FiFKFjLÿһWYjKXҩQOX\ӋQP{KuQK7ӯÿyÿѭDNӃWTXҧWURQJWKӵFQJKLӋPĈӗQJ WKӡL FNJQJVRViQKNӃWTXҧWKӵFQJKLӋPFӫDP{KuQKÿӅ[XҩWYӟLP{KuQKFѫVӣ
&KѭѫQJ.ӃWOXұQ - ĈѭDUDNӃWOXұQYӅP{KuQKÿӅ[XҩWÿiQKJLiѭXYjQKѭӧFÿLӇP
FӫDQyÿӗQJWKӡLÿӅUDKѭӟQJQJKLrQFӭXWURQJWѭѫQJODL
9jRQăP 5XPHOKDUWYjFiFÿӗQJQJKLӋSFӫD{QJÿm[XҩWEҧQPӝWEjLEiRQJKLrQ FӭXFyWӵDÿӅ³Learning internal representations by error propagation´[5] Trong nghiên FӭXQj\PӝWNLӃQWU~FPҥQJFәÿLӇQÿӇ[ӱOêGӳOLӋXWXҫQWӵÿѭӧFJӑLOj0ҥQJQѫ-ron WUX\KӗL 0ҥQJQѫ-URQWUX\KӗLOjPӝWNLӃQWU~FPҥQJQѫ-URQPjӣÿyWUҥQJWKiLFӫDP{ KuQKWҥLEѭӟFOLӅQWUѭӟFVӁÿѭӧFGQJOjPÿҫXYjRFӫDEѭӟFKLӋQWҥL1yLFiFKNKiF PӝWP{KuQKPҥQJQѫ-URQWUX\KӗLVӁKRҥWÿӝQJWKHRGҥQJiQK[ҥGӳOLӋXÿҫXYjRYj WUҥQJWKiLWUѭӟFÿyFӫDP{KuQKWKjQKGӳOLӋXÿҫXUD [6] Hình 2-1 là YtGөYӅNLӃQWU~F FӫDPӝWOӟSPҥQJQѫ-ron truy KӗL
. Hình 2-1 LӃQWU~FFӫDPӝWOӟSPҥQJQѫ-URQWUX\KӗLÿѭӧFELӇXGLӉQYӟLÿѭӡQJQӕL
9ӟLPҥQJQѫ-URQWUX\ӅQWKӕQJGҥQJ IHHG-IRUZDUGGӳOLӋXFKӍÿѭӧFWUX\ӅQWKHRPӝW FKLӅXWtQKWRiQYjÿѭD UDNӃWTXҧJyLJӑQWURQJPӝWSKDGӵDKRjQWRjQYjRGӳOLӋXLQSXW NK{QJWӗQWҥLÿѭӡQJQӕLYzQJWӭFOjFyÿѭӡQJOLrQNӃWWӯPӝWWҫQJÿӃQPӝWWҫQJWUѭӟF ÿy9ӟLPҥQJQѫ-URQWUX\KӗLYLӋFWK{QJWLQWӯOҫQWUX\ӅQWUѭӟFÿѭӧFOѭXOҥLWҥLPӝW
WU̩QJWKiḺQ (hiGGHQVWDWHFKROҫQWtQKWRiQKLӋQWҥLWҥRUDPӝWÿ˱ͥQJQ͙LYzQJ WUrQÿӗ
WKӏWtQKWRiQ1yLFiFKNKiFKLGGHQVWDWHÿyQJYDLWUzQKѭPӝWEӝQKӟFӫDOӟSQѫ-ron WUX\KӗLĈLӅXQj\ÿһFELӋWKӳXGөQJNKL[ӱOêGӳOLӋXGҥQJFKXӛLNKLWDFyWKӇiSGөQJ
Shape 2-2 Image OjYtGөOXӗQJWtQKWRiQWUrQVHTXHQFH511
1ăP+RSILHOGÿmÿѭDUDNLӃQWU~F +RSILHOGQHWZRUN+RSILHOGEOjPӝW GҥQJPҥQJQѫ-URQYӟLNKҧQăQJOѭXJLӳOҥLWUҥQJWKiLFNJWUѭӟFÿyÿӇKӛWUӧFKROҫQWtQK WRiQKLӋQWҥL+RSILHOGQHWZRUNOjPӝWGҥQJÿһFELӋWFӫDPҥQJQѫ-URQWUX\KӗL7UӑQJ VӕWUrQPҥQJ+RSILHOGÿѭӧFFұSQKұWEҵQJSKѭѫQJSKiSWӕLѭXOӗLNKiFYӟLJUDGLHQW GHVFHQWӣPҥQJQѫ-URQWUX\KӗLKLӋQÿҥL0ҥQJQѫ-URQWUX\KӗLFKtQKWKӭFÿѭӧFÿѭDUD GӵDWUrQF{QJWUuQKFӫD'DYLG5XPHOKDUWQăP
511ÿѭӧFWKLӃWNӃÿһFWKÿӇJLҧLTX\ӃWGӳOLӋXGҥQJVHTXHQFH0ӝWVӕѭXÿLӇP, QKѭӧF ÿLӇP FӫD511VRYӟLPҥQJIHHG-IRUZDUGNLӇXWUX\ӅQWKӕQJFyWKӇNӇÿӃQQKѭ ˰X ÿL͋P x KҧQăQJ[ӱOêGӳOLӋXYӟLÿӝGjLEҩWÿӏQK x tFKWKѭӟFP{KuQKOjFӕÿӏQKYjÿӝFOұSYӟLÿӝGjLÿҫXYjR x 9LӋFWtQKWRiQFyWұQGөQJÿѭӧFWK{QJWLQWӯTXiNKӭ x 6ӱGөQJFKXQJPӝWEӝWUӑQJVӕFKRWRjQEӝVHTXHQFHWӭF là iSGөQJFQJPӝWORJLFWtQKWRiQWUrQWӯQJÿLӇPGӳOLӋXWURQJVHTXHQFH
8 x &KLDVҿWUӑQJVӕWURQJTXiWUuQKWtQKWRiQFzQJL~SOjPJLҧPVӕOѭӧQJWKDPVӕ FҧLWKLӋQWtQKWәQJTXiWKyDUHJXODUL]DWLRQYjWUiQKVӵTXiNKӟS
1K˱ͫFÿL͋P x Tính toán tuҫn tӵ: KӃt quҧ cӫa lҫn tính toán hiӋn tҥi dӵa vào kӃt quҧ cӫa lҫn tính WRiQWUѭӟc Vì vұy, dӳ liӋu tӯ các lҫn truyӅn không thӇ ÿѭӧc thӵc hiӋn song song, dүQÿӃn viӋc không tұn dөQJÿѭӧc GPU x Có khҧ QăQJEӏ "quên" dӳ liӋu
'511FyÿѭӧFWKLӃWNӃÿӇOѭXWUӳWK{QJWLQWӯTXiNKӭWX\QKLrQYӟLFiFWKLӃWNӃFѫ EҧQFK~QJYүQFKӍOѭXÿѭӧFPӝWOѭӧQJWK{QJWLQWӯPӝWVӕKӳXKҥQFiFOҫQOһSWUѭӟF ÿy/êGRFKRVӵTXrQQj\OjGRYLӋFQKkQOһSÿLOһSOҥLFӫDPӝWJLiWUӏWURQJFKXӛi chain-UXOHӣTXiWUuQKODQWUX\ӅQQJѭӧFGүQÿӃQYLӋFvanishing và exploding gradient [7] ĈӇJLҧLTX\ӃWYҩQÿӅQj\PӝWSKLrQEҧQSKiWWULӇQFӫD511VÿmÿѭӧF+RFKUHLWHU Yj6FKPLGKXEHUÿѭDUDYjRQăP ÿѭӧFJӑLOj%ӝQKӟQJҳQ dài KҥQ/670 LSTM NKҳFSKөFÿѭӧFQKӳQJKҥQFKӃFӫD511 [6] [8] (Long Short-Term Memory) YӟLNKҧ QăQJ FKӑQ OӑF WK{QJ WLQ ÿӇ JLӳEӓ WURQJ TXi WUuQK ODQ WUX\ӅQ WKXұQ Yj WUiQK YLӋF vanishing/exploding gradient WURQJTXiWUuQKODQWUX\ӅQQJѭӧF [6]
LSTM PDQJOҥLKLӋXVXҩWFKtQK[iFKѫQEҵQJFiFKVӱGөQJ PӝWOӟSҭQ (hidden layer) OjP{QKӟ(memory cell) thay vì {OһSOҥLrecurrent cell) Hình 2-3
Các mô hình ngôn ngữ lớn (LLM) là các hệ thống xử lý ngôn ngữ tự nhiên được đào tạo trên một lượng lớn dữ liệu văn bản Chúng có khả năng thực hiện nhiều tác vụ liên quan đến ngôn ngữ, bao gồm tạo văn bản, dịch ngôn ngữ và trả lời câu hỏi LLM được sử dụng trong nhiều ứng dụng, chẳng hạn như chatbot, công cụ tìm kiếm và hệ thống đề xuất.
/670JӗPFyFәQJ: x Cәng vào (Input) x Cәng ra (Output) x Cәng quên (Forget)
.LӇPVRiWOXӗQJWK{QJWLQErQWURQJP{KuQK/670ÿѭӧFELӇXGLӉQEҵQJWRiQKӑFQKѭ sau (1a) - (1e) ݅ ൌ ߪሺܹ ݔ ௧ ܹ ݄ ௧ିଵ ܾ ሻ (1a) ݂ ൌ ߪሺܹ ݔ ௧ ܹ ݄ ௧ିଵ ܾ ሻ (1b) ൌ ߪሺܹ ݔ ௧ ܹ ݄ ௧ିଵ ܾ ሻ (1c) ሺܥ ௧ ሻ ൌ ሺܥሻ ௧ିଵ ٔ ሺܹ ݔ ௧ ܹ ݄ ௧ିଵ ܾ ሻ (1d) ݄ ௧ ൌ ٔ ሺሺܥሻ ௧ିଵ ሻ (1e)
7URQJF{QJWKӭF1a) - (HLIR&KELӇXGLӉQWKHRWKӭWӵOjFәQJYjR FәQJTXrQ FәQJUDinternal state và OӟS ҭQhidden layer) Và ݔ WѭѫQJӭQJYӟLÿҫXYjRӣEѭӟFWKӭ i ܹ ǡ ܹ ǡ ܹ ǡ ܹ và ܾ ǡ ܾ ǡ ܾ ǡ ܾ ÿҥLGLӋQFKRZHLJKWYjELDVFӫDFәQJYjPHPRU\ FHOOWѭѫQJӭQJ HjPNtFKKRҥWsigmoid ı JL~SP{KuQK/670NLӇPVRiWOXӗQJWK{QJ tinKjPNtFKKRҥWVLJPRLGFySKҥPYL>@1ӃXJLiWUӏOjWKuWҩWFҧWK{QJWLQVӁEӏEӓ ÿLQJѭӧFOҥLWK{QJWLQVӁÿѭӧF ÿLTXD 7ѭѫQJWӵӣFәQJUDFNJQJVӱGөQJKjPNtFKKRҥW sigmoid ı JL~SNLӇPVRiWWK{QJWLQYjWUӑQJVӕVӁÿѭӧFFұSQKұWOjSKpSQKkQelement- wise FӫDFәQJUDYjinternal state ÿѭӧFNtFKKRҥWEҵQJKjPSKLWX\ӃQtanh Vì LSTM
Fy{QKӟPHPRU\FHOOOjFKӫ FKӕWFKӭD WK{QJWLQFәQJYjRFәQJUDFәQJTXrQÿӅX ÿyJL~S/670NKҳF SKөFÿѭӧFKҥQFKӃFӫD511.KҧQăQJJKLQKӟFiFJLiWUӏWURQJNKRҧQJWKӡLJLDQW\êEҵQJFiFKÿLӅXFKӍQKOXӗQJWK{QJWLQErQWURQJmemory cell
10 L670FyNKҧQăQJKRҥWÿӝQJUҩWWӕWNKLKӑFFiFWtQKQăQJWӯGӳOLӋXWXҫQWӵQKѭWjL OLӋXGRFXPHQW[ӱOờJLӑQJQyLSKiWKLӋQEҩWWKѭӡQJ ô
Word2vHFOjPӝWWURQJQKӳQJNӻWKXұWÿѭӧFVӱGөQJSKәELӃQQKҩWWURQJOƭQKYӵF;ӱ lý QJ{QQJӳWӵQKLrQWord2vec ÿѭӧFWҥRUDYjF{QJEӕYjRQăPEӣLPӝWQKyP FiFQKjQJKLrQFӭXGүQÿҫXEӣL7RPDV0LNRORYӣ*RRJOHYjÿmÿѭӧFÿăQJNêEҧRKӝ TX\ӅQSKiWPLQKViQJFKӃ0LNRORYHWDO
Word2vec is a natural language processing (NLP) technique that generates word embeddings, representing words as numerical vectors that capture their semantic and syntactic relationships Encoders, such as autoencoders [9], process these embeddings to reduce their dimensionality while preserving important features By incorporating context words into the encoder input, the resulting embeddings not only encode the word's meaning but also its context-dependent representation This enhances the encoder's ability to handle ambiguities and variations in word usage.
Ví GөGLҧVӱNKRQJӳOLӋXFӫDWDJӗPWjLOLӋX'''
D3: 1Jj\PDLW{LNK{QJK͕FO̵SWUuQK
6DXEѭӟFWLӅQ[ӱOêNKRQJӳOLӋXWUrQVӁÿѭӧFELӃQÿәLWKjQKWұSFiFWӯSKkQELӋWÿѭӧF VҳS [ӃS WKHR WKӭ Wӵ EҧQJ FKӳ FiL QKѭ VDX ^FNJQJ K͕F K{PBQD\ K{PBTXD NK{QJ
7ӯPөFWLrX 7ӯQJӳFҧQK hôm_qua tôi tôi K{PBTXDKӑF
%ҧQJ2-1 7ӯPөFWLrXYjWӯQJӳFҧQKWѭѫQJӭQJFӫD'FyNtFKWKѭӟFFӱDVәOj
11 Word2vec có 2 mô hình: x Skip-gram x CBOW
7URQJÿӅWjLQj\KӑFYLrQVӱGөQJP{KuQK&%2:
CBOW (Continuous Bag of Words) FyÿҫXYjROjWӯPөFWLrXÿҫXUDOjFiFWӯWURQJQJӳ FҧQKéWѭӣQJ FKtQKFӫDP{KuQK&%2:OjGӵÿRiQWӯPөFWLrXGӵDYjRFiFWӯQJӳ FҧQK[XQJTXDQKQyWURQJPӝWSKҥPYLQKҩWÿӏQK &KRWӯPөFWLrXݓ ௧ WҥLYӏWUtWWURQJ FkXYăQEҧQNKLÿyÿҫXYjROjFiFWӯQJӳFҧQKܹ ௧ି ǡ ǥ ǡ ܹ ௧ିଵ ǡ ܹ ௧ାଵ ǡ ǥ ǡ ܹ ௧ା ) xung TXDQKWӯܹ ௧ WURQJSKҥPYLP
3 CễNG 75ẻ1+1*+,ầ1&Ӭ8/,ầ148$1 ĈҫXWLrQOXұQYăQVӁJLӟLWKLӋXYӅPӝWVӕP{KuQKJӧLêWUX\ӅQWKӕQJWUѭӟF NKLÿLYjR FiFQJKLrQFӭXFiFKӋWKӕQJGӵDWUrQSKѭѫQJSKiSKӑFVkX
&iFKӋWKӕQJNKX\ӃQQJKӏWUX\ӅQWKӕQJ56EDRJӗP
12 Hình 3-1 6ѫ ÿӗKӋNKX\ӃQQJKӏWUX\ӅQWKӕQJ
/ӑFFӝQJWiF (Collaborative filtering - CF) [10]
Các SKѭѫQJSKiSOӑFFӝQJWiFGӵDWUrQYLӋFWKXWKұSYjSKkQWtFKPӝWOѭӧQJOӟQWK{QJ WLQYӅKjQKYLKRҥWÿӝQJKRһFVӣWKtFKFӫDQJѭӡLGQJYjGӵÿRiQQKӳQJJuQJѭӡLGQJ VӁWKtFKGӵDWUrQVӵWѭѫQJÿӗQJFӫDKӑYӟLQJѭӡLGQJNKiF0ӝWOӧLWKӃTXDQWUӑQJFӫD SKѭѫQJSKiSOӑFFӝQJWiFOjQyNK{QJGӵDYjRQӝLGXQJSKkQWtFKPi\YjGRÿyQyFy NKҧQăQJÿӅ[XҩWFKtQK[iFFiFPөFSKӭFWҥSQKѭSKLPPjNK{QJ\rXFҫX³KLӇXELӃW´ YӅPөFÿy 1KLӅXWKXұWWRiQÿmÿѭӧFVӱGөQJÿӇÿROѭӡQJVӵJLӕQJQKDXFӫDQJѭӡLGQJ KRһFVӵWѭѫQJÿӗQJYӅPһWKjQJWURQJFiFKӋWKӕQJJLӟLWKLӋX/ӑFFӝQJWiFGӵDWUrQ JLҧÿӏQKUҵQJQKӳQJQJѭӡLÿmÿӗQJêWURQJTXiNKӭVӁÿӗQJêWURQJWѭѫQJODLYjUҵQJ KӑVӁWKtFKFiFORҥLPһWKjQJWѭѫQJWӵQKѭKӑWKtFKWURQJTXiNKӭ
3KѭѫQJSKiSOӑFGӵDWUrQQӝLGXQJGӵDWUrQP{WҧYӅKҥQJPөFYjKӗVѫYӅFiFW\FKӑQFӫDQJѭӡLGQJ7URQJKӋWKӕQJNKX\ӃQQJKӏGӵDWUrQQӝLGXQJWӯNKyDÿѭӧFVӱGөQJ ÿӇP{WҧFiFKҥQJPөFYjKӗVѫQJѭӡLGQJÿѭӧF[k\GӵQJÿӇFKӍUDORҥLKҥQJPөF mà QJѭӡi dùng này thích CiFWKXұWWRiQQj\FӕJҳQJÿӅ[XҩWFiFPһWKjQJWѭѫQJWӵYӟLFiFKҥQJPөFPjQJѭӡLGQJÿmWKtFKWURQJTXiNKӭ &өWKӇFiFPөFÿӅ[XҩW khác nhau ÿѭӧFVRViQKYӟLFiFPөFÿѭӧFÿiQKJLiWUѭӟFÿk\EӣLQJѭӡLGQJYjFiFPөF nào phù KӧSQKҩW YӟL[iFVXҩWFDRQKҩWVӁ ÿѭӧFÿӅ[XҩW &iFKWLӃSFұQQj\FyQJXӗQJӕFWӯ YLӋFWKXWKұS thông tin và QJKLrQFӭX OӑFWK{QJWLQ
ĈӝÿR
+ӑF viên VӱGөQJÿӝÿR055Mean Reciprocal Rank), Pricision@k và Recall@k ÿӇ ÿiQKJLiP{KuQK
6ӱGөQJÿӝÿRÿӇÿiQKJLiFKҩWOѭӧQJFӫDP{KuQKWUrQWұSGӳOLӋXWHVW7URQJEjLOXұQ YăQQj\m{KuQKWKDPNKҧRYjP{KuQKÿӅ[XҩWKӑFYLrQ VӱGөQJÿӝÿR0550HDQ 5HFLSURFDO5DQNĈӝÿRQj\UҩWWKѭӡQJGQJWURQJFiFKӋNKX\ӃQQJKӏ
3KѭѫQJSKiSÿiQKJLiGӵDWUrQGDQKViFK[ӃSKҥQJFӫDQKӳQJEӝSKLPÿѭӧFWUҧYӅ (Hình 4-7)
7URQJÿy ranki WKDPFKLӃXWӟLYӏWUtFӫDQKӳQJKҥQJUDQNSKKӧSÿҫXWLrQFKRFkX WUX\YҩQWKӭL (Hình 4-8)
26 Ngoài ra KӑFYLrQ FzQVӱGөQJÿӝÿRNKiFQKѭ x Precision là tӹ lӋ giӳa sӕ Oѭӧng các gӧi ý phù hӧp và tәng sӕ các gӧi ý dӵ ÿRiQ Precision bҵQJFyQJKƭDOjWҩt cҧ các kiӃn nghӏ ÿӅu phù hӧp ܲݎ݁ܿ݅ݏ݅݊ ൌ ݏዎ݈ዛዘ݊݃݃ዘ݅ý݄ợ݄ዘ ݏዎ݈ዛዘ݊݃݃ዘ݅ý݀ዠ¯݊ x 5HFDOOÿѭӧFÿӏQKQJKƭDEӣi tӍ lӋ giӳa sӕ Oѭӧng các gӧi ý phù hӧp và sӕ Oѭӧng các mөc dӳ liӋXPjQJѭӡLGQJÿmÿiQK giá 5HFDOOÿѭӧc sӱ dөQJÿӇ ÿRNKҧ QăQJKӋ thӕQJWuPÿѭӧc nhӳng mөc dӳ liӋu phù hӧp so vӟi nhӳQJJuPjQJѭӡi dùng cҫn ܴ݈݈݁ܿܽ ൌ ݏዎ݈ዛዘ݊݃݃ዘ݅ý݄ợ݄ዘ ݏዎ݈ዛዘ݄݊݃݅݉¯ዛዘܿ¯݄ܾ݊݃݅ዖ݅݊݃ዛዕ݅݀ợ݊݃ x F-VFRUHÿѭӧc sӱ dөQJÿӇ ÿiQKJLiKLӋu quҧ tәng thӇ cӫa hӋ thӕng bҵng cách kӃt hӧp hai chӍ sӕ Recall và Precision ܨ െ ݏܿݎ݁ ൌʹ כ ܲݎ݁ܿ݅ݏ݅݊ כ ܴ݈݈݁ܿܽ ܲݎ݁ܿ݅ݏ݅݊ ܴ݈݈݁ܿܽ
+ӑFYLrQ VӱGөQJWұSGӳOLӋX0MovieLens [4]'ӳOLӋXÿѭӧFP{WҧWҥL4.1
7ұSGӳOLӋXEDRJӗP x ~ 1000000 WѭѫQJWiFrating x aQJѭӡi dùng x ~ 4000 bӝ phim ĈӇP{KuQKFKRUDNӃWTXҧNKX\ӃQQJKӏWӕWQKҩW KӑFYLrQ FKӍ FKӑQQKӳQJÿiQKJLi KRһFVDRFӫDQJѭӡLdùng
7ұSGӳOLӋXVDXNKLFKӑQOӑF QKѭVDX x ~ 300000 WѭѫQJWiFÿiQKJLi x ~ 3220 QJѭӡi dùng x ~ 3410 bӝ phim
BiӇXÿӗ Hình 5-1 thӇ hiӋn sӕ OѭӧQJÿiQKJLiFӫDQJѭӡi dùng
'ӵD WUrQNӃWTXҧWKӵFQJKLӋPFiF WKDPVӕÿmWKӱQJKLӋPWUrQP{KuQKFKRUDNӃWTXҧ WӕWQKҩW
7KDPVӕ *LiWUӏ batch size 64 embedding dimension 64 learning rate 0.001 l2 0.0001
%ҧQJ5-1 &jLÿһWWKDPVӕP{KuQK 'ӵDWUrQTXiWUuQKWKӵFQJKLӋPYӟLWұS GӳOLӋXPүXWKuKӑFYLrQQKұQWKҩ\YӟLJLiWUӏ LQWHUYDO P{KuQKVӁFKRUDNӃWTXҧWӕWQKҩW
7RjQEӝTXiWUuQKKXҩQOX\ӋQYjNLӇPWKӱÿѭӧFWKӵFKLӋQWUrQPi\ Intel® Core (TM) i7-10510U CPU @ 1.8GBz (8 CPUs)
5.3 Cách hXҩQOX\ӋQP{KuQK
0{KuQKÿѭӧFKXҩQOX\ӋQYjNLӇPWKӱWUrQWұSGӳOLӋXMovieLens.
7tQKPҩWPiWORVVIXQFWLRQWURQJTXiWUuQKWUDLQGӵDWUrQ hàm Cross Entropy
Hình 5-3 Cross Entropy Loss ӃWTXҧKXҩQOX\ӋQP{KuQKWUrQWұS0RYLHLens VRYӟLP{KuQKWKDPFKLӃX Ĉӝ ÿR 0{KuQKWKDPFKLӃX 0{KuQKÿӅ[XҩW
%ҧQJ5-2 ĈӝÿRFӫDP{KuQKWKDPFKLӃXYj mô hình ÿӅ[XҩW
%ҧQJ5-2 FKRWKҩ\P{KuQKÿӅ[XҩWFyKLӋXTXҧKѫQP{KuQKWKDPFKLӃXĈLӅXQj\FNJQJFKӭQJPLQKYLӋFNKX\ӃQQJKӏGӵDWUrQ\ӃXWӕWKӡLJLDQVӁFKRUDNӃWTXҧWӕWKѫQ
7әQJ NӃW WKjQKTXҧFӫDFK~QJWDWURQJTXiWUuQKQJKLrQFӭXOêWKX\ӃWYӅPҥQJQѫ-ron WUX\KӗL/670Word2vHFFiFP{KuQKFӫDF{QJWUuQKQJKLrQFӭXOLrQTXDQWӯÿyÿѭD UDP{KuQKÿӅ[XҩW/6700{KuQKÿӅ[XҩWFyWKӇNKX\ӃQ QJKӏQKӳQJEӝSKLPWKHR\ӃX
This article explores the significance of keyword research in search engine optimization (SEO) and its role in driving organic traffic to a website It emphasizes the importance of identifying relevant keywords that align with the target audience's search intent Successful keyword research enables businesses to optimize their website content, improve search rankings, and connect with potential customers.
.ӃWTXҧQJKLrQFӭXFyWKӇOjPWjLOLӋXWKDPNKҧRFKRQKӳQJDLPXӕQOjPTXHQYӟLKӋ WKӕQJNKX\ӃQQJKӏGӵDWUrQVHVVLRQ-based YӟL\ӃXWӕWKӡLJLDQQKҵPJL~SWLӃWNLӋPÿѭӧF WKӡLJLDQWLӃSFұQ
0{KuQKÿӅ[XҩWFzQQKLӅXKҥQFKӃ x Mô hình cҫn phҧi có nhiӅu cҧi tiӃn KѫQÿӕi vӟLQJѭӡi dùng mӟLFKѭDFyGӳ liӋu thì mô hình không thӇ ÿӅ xuҩt x Mô hình chӍ sӱ dөng lӏch sӱ WѭѫQJWiFFӫDQJѭӡLGQJPjNK{QJGQJWtQKWѭѫQJ quan giӳa các user
9LӋFVӱGөQJP{KuQKVHVVLRQ-based NӃWKӧSYӟLGӳOLӋXWKӡLJLDQOjPFKRTXiWUuQK KXҩQOX\ӋQFKRUDNӃWTXҧKѫQYӟLP{KuQKNK{QJNӃWKӧS\ӃXWӕWKӡLJLDQ 9uYұ\P{ KuQKNKX\ӃQQJKӏWKHRFiFKWLӃSFұQQj\VӁJL~SFKRKӋNKX\ӃQQJKӏFKRUDNӃWTXҧchính xác KѫQ
0{KuQKÿӅ[XҩWFKѭDFK~WUӑQJYjRPӕLWѭѫQJTXDQJLӳDFiFQJѭӡLGQJÿӇWtQKÿѭӧFQKӳQJVӣWKtFKWѭѫQJWӵQKDX+ѭӟQJSKiWWULӇQWLӃSWKHRFyWKӇOjNӃWKӧSYӟLP{KuQK
31 WUX\ӅQWKӕQJODLJLӳDOӑFFӝQJWiFYjFӝQJQӝLGXQJFөWKӇOj P{KuQKTXDQWkPÿӃQQӝi GXQJWLrXÿӅSKLPWiFJLҧGLӉQYLrQÿӝWXәLQJѭӡLGQJYjFNJQJTXDQWkPÿӃQQKӳQJQJѭӡLGQJWѭѫQJÿӗQJYӟLQKDXYj NӃWKӧSWKrP\ӃXWӕWKӡLJLDQYtGөTXDQWkPVӣWKtFKQJҳQKҥQFӫDQJѭӡLdùng) WӯÿyQkQJFDRFKҩWOѭӧQJFӫDP{KuQKKѫQQӳD
[1] L Yu, L Liu and X Li, "A hybrid collaborative filtering method for multiple-interests and multiple-content recommendation in E-Commerce," in Expert Systems with
[2] S Zhang, L Yao, A Sun and Y Tay, "Deep Learning based Recommender System: A Survey and New Perspectives," ACM Computing Surveys, vol 52, pp 1-38, 2019
[3] G Guibing, "Resolving Data Sparsity and Cold Start in Recommender Systems," in
User Modeling, Adaptation, and Personalization, Berlin, 2012
[4] "MovieLens 10M Dataset." Internet: https://grouplens.org/datasets/movielens/10m/ [Accessed 22 10 2021]
[5] D E Rumelhart, G E Hinton and R J Williams, "Learning representations by back- propagating errors," Nature, vol 323, pp 533-536, 1986
[6] Q T Tho, "Modern Approaches in Natural Language Processing," in VNU Journal of
Science: Comp Science & Com Eng., vol 37, no 1, pp 10-12, 2021
[7] Y Bengio, P Simard and P Frasconi, "Learning long term dependencies with gradient descent is difficul," in IEEE Transactions on Neural Networks, vol 5, iss 2, 1994 [8] H Sepp and S Jurgen, "Long Short-Term Memory," Neural Computation, vol 9, iss 8, pp 1735-1780, 1997
[9] I Goodfellow, Y Bengio and A Courville, Deep Learning Cambridge, MA: MIT Press, 2016
[10] G Linden, J York and J York, "Amazon.Com Recommendations:Item-to-Item
Collaborative Filtering," IEEE Internet Computing, vol 7, iss 1, pp 76-80, 1 7 2003 [11] M J Raymond, and R Loriene, "Content-Based Book Recommending Using Learning for Text Categorization," in Proceedings of the SIGIR-99 Workshop on Recommender
Systems: Algorithms and Evaluation, Berkeley, CA, August 1999
[12] L Pasquale, S Giovanni and G Marco de, "Content-based Recommender Systems: State of the Art and Trends," in Recommender Systems Handbook, pp 73-105, 2010 DOI 10.1007/978-0-387-85820-3_3
[13] L Yifang, Z Xu, Q An, Y Yi and Yanzhi, "Simultaneous Relevance and Diversity: A New Recommendation Inference Approach" 3HUVRnDOHPDLO6HS 2020
[14] D B Tran, and T T S Nguyen, "Long Short-Term Memory Based Movie
Recommendation," in School of Computer Science and Engineering, International
[15] R Steffen, F Christoph , G Zeno and S.-T Lars , "BPR: Bayesian Personalized
Ranking from Implicit Feedback," in Proceedings of the 25th Conference on
Uncertainty in Artificial Intelligence, Montreal, Quebec, Canada, 2009
[16] J Daniel and H M James, "Neural Networks and Neural Language Models," in Speech and Language Processing, 2 nd Edition Pearson Education,Inc., 2021
&jLÿһWP{LWUѭӡQJ x python: 3.8 trӣ lên x matplotlib: 3.3.2 trӣ lên x numpy: 1.19.4 trӣ lên x torch: 1.7.1 trӣ lên x tqdm: 4.51.0 trӣ lên
ĈiQK JLiNӃWTXҧ
7әQJ NӃW WKjQKTXҧFӫDFK~QJWDWURQJTXiWUuQKQJKLrQFӭXOêWKX\ӃWYӅPҥQJQѫ-ron WUX\KӗL/670Word2vHFFiFP{KuQKFӫDF{QJWUuQKQJKLrQFӭXOLrQTXDQWӯÿyÿѭD UDP{KuQKÿӅ[XҩW/6700{KuQKÿӅ[XҩWFyWKӇNKX\ӃQ QJKӏQKӳQJEӝSKLPWKHR\ӃX
SEO is vital for creating valuable content, as it allows search engines to understand and rank your content Search engine optimization is the practice of improving the visibility and ranking of a website or web page in search engine results pages (SERPs) By incorporating relevant keywords, structuring content effectively, and building backlinks, you can enhance your content's visibility and attract more organic traffic.
.ӃWTXҧQJKLrQFӭXFyWKӇOjPWjLOLӋXWKDPNKҧRFKRQKӳQJDLPXӕQOjPTXHQYӟLKӋ WKӕQJNKX\ӃQQJKӏGӵDWUrQVHVVLRQ-based YӟL\ӃXWӕWKӡLJLDQQKҵPJL~SWLӃWNLӋPÿѭӧF WKӡLJLDQWLӃSFұQ
0{KuQKÿӅ[XҩWFzQQKLӅXKҥQFKӃ x Mô hình cҫn phҧi có nhiӅu cҧi tiӃn KѫQÿӕi vӟLQJѭӡi dùng mӟLFKѭDFyGӳ liӋu thì mô hình không thӇ ÿӅ xuҩt x Mô hình chӍ sӱ dөng lӏch sӱ WѭѫQJWiFFӫDQJѭӡLGQJPjNK{QJGQJWtQKWѭѫQJ quan giӳa các user.
ĈyQJJySFӫDÿӅWjL
9LӋFVӱGөQJP{KuQKVHVVLRQ-based NӃWKӧSYӟLGӳOLӋXWKӡLJLDQOjPFKRTXiWUuQK KXҩQOX\ӋQFKRUDNӃWTXҧKѫQYӟLP{KuQKNK{QJNӃWKӧS\ӃXWӕWKӡLJLDQ 9uYұ\P{ KuQKNKX\ӃQQJKӏWKHRFiFKWLӃSFұQQj\VӁJL~SFKRKӋNKX\ӃQQJKӏFKRUDNӃWTXҧchính xác KѫQ
0{KuQKÿӅ[XҩWFKѭDFK~WUӑQJYjRPӕLWѭѫQJTXDQJLӳDFiFQJѭӡLGQJÿӇWtQKÿѭӧFQKӳQJVӣWKtFKWѭѫQJWӵQKDX+ѭӟQJSKiWWULӇQWLӃSWKHRFyWKӇOjNӃWKӧSYӟLP{KuQK
31 WUX\ӅQWKӕQJODLJLӳDOӑFFӝQJWiFYjFӝQJQӝLGXQJFөWKӇOj P{KuQKTXDQWkPÿӃQQӝi GXQJWLrXÿӅSKLPWiFJLҧGLӉQYLrQÿӝWXәLQJѭӡLGQJYjFNJQJTXDQWkPÿӃQQKӳQJQJѭӡLGQJWѭѫQJÿӗQJYӟLQKDXYj NӃWKӧSWKrP\ӃXWӕWKӡLJLDQYtGөTXDQWkPVӣWKtFKQJҳQKҥQFӫDQJѭӡLdùng) WӯÿyQkQJFDRFKҩWOѭӧQJFӫDP{KuQKKѫQQӳD
[1] L Yu, L Liu and X Li, "A hybrid collaborative filtering method for multiple-interests and multiple-content recommendation in E-Commerce," in Expert Systems with
[2] S Zhang, L Yao, A Sun and Y Tay, "Deep Learning based Recommender System: A Survey and New Perspectives," ACM Computing Surveys, vol 52, pp 1-38, 2019
[3] G Guibing, "Resolving Data Sparsity and Cold Start in Recommender Systems," in
User Modeling, Adaptation, and Personalization, Berlin, 2012
[4] "MovieLens 10M Dataset." Internet: https://grouplens.org/datasets/movielens/10m/ [Accessed 22 10 2021]
[5] D E Rumelhart, G E Hinton and R J Williams, "Learning representations by back- propagating errors," Nature, vol 323, pp 533-536, 1986
[6] Q T Tho, "Modern Approaches in Natural Language Processing," in VNU Journal of
Science: Comp Science & Com Eng., vol 37, no 1, pp 10-12, 2021
[7] Y Bengio, P Simard and P Frasconi, "Learning long term dependencies with gradient descent is difficul," in IEEE Transactions on Neural Networks, vol 5, iss 2, 1994 [8] H Sepp and S Jurgen, "Long Short-Term Memory," Neural Computation, vol 9, iss 8, pp 1735-1780, 1997
[9] I Goodfellow, Y Bengio and A Courville, Deep Learning Cambridge, MA: MIT Press, 2016
[10] G Linden, J York and J York, "Amazon.Com Recommendations:Item-to-Item
Collaborative Filtering," IEEE Internet Computing, vol 7, iss 1, pp 76-80, 1 7 2003 [11] M J Raymond, and R Loriene, "Content-Based Book Recommending Using Learning for Text Categorization," in Proceedings of the SIGIR-99 Workshop on Recommender
Systems: Algorithms and Evaluation, Berkeley, CA, August 1999
[12] L Pasquale, S Giovanni and G Marco de, "Content-based Recommender Systems: State of the Art and Trends," in Recommender Systems Handbook, pp 73-105, 2010 DOI 10.1007/978-0-387-85820-3_3
[13] L Yifang, Z Xu, Q An, Y Yi and Yanzhi, "Simultaneous Relevance and Diversity: A New Recommendation Inference Approach" 3HUVRnDOHPDLO6HS 2020
[14] D B Tran, and T T S Nguyen, "Long Short-Term Memory Based Movie
Recommendation," in School of Computer Science and Engineering, International
[15] R Steffen, F Christoph , G Zeno and S.-T Lars , "BPR: Bayesian Personalized
Ranking from Implicit Feedback," in Proceedings of the 25th Conference on
Uncertainty in Artificial Intelligence, Montreal, Quebec, Canada, 2009
[16] J Daniel and H M James, "Neural Networks and Neural Language Models," in Speech and Language Processing, 2 nd Edition Pearson Education,Inc., 2021
&jLÿһWP{LWUѭӡQJ x python: 3.8 trӣ lên x matplotlib: 3.3.2 trӣ lên x numpy: 1.19.4 trӣ lên x torch: 1.7.1 trӣ lên x tqdm: 4.51.0 trӣ lên