Item recommendation method based on importance of item in conversation session and system thereof

Patent metadata

Field	Value
Publication number	US-12333589-B2
Application number	US-202217693761-A
Country	US
Kind code	B2
Filing date	Mar 14, 2022
Priority date	May 25, 2020
Publication date	Jun 17, 2025
Grant date	Jun 17, 2025

Abstract

The present disclosure provides an item recommendation method and system based on importance of item in a conversation session and a system thereof. In the present disclosure, an importance extracting module extracts an importance of each item in the conversation session, and then a long-term preference of a user is obtained in combination with the importance and the corresponding item, and then a preference of the user is obtained accurately in combination with a current interest and the long-term preference of the user, and finally item recommendation is performed according to the preference of the user. In this way, the accuracy of the item recommendation is improved.

First claim

What is claimed is:

1. A system for performing an item recommendation method based on importance of item in a conversation session, configured to predict an item that a user is likely to interact at a next moment from an item set as a target item to be recommended to the user, comprising:
a processor;
a non-transitory computer readable memory coupled to the processor containing program instructions for performing the item recommendation method; and
an input device;
wherein the input device is configured to record an input from the user; and
execution of the program instructions by the processor causes the processor to perform steps of an item recommendation model comprising:
obtaining an item embedding vector by embedding each item in a current conversation session to one d-dimension vector representation, and taking an item embedding vector corresponding to the last item in the current conversation session as a current interest representation of the user, wherein the current conversation session is a voice recording of the user via a microphone of a conversation conducted by the user;
obtaining an importance representation of each item according to the item embedding vector, and obtaining a long-term preference representation of the user by combining the importance representation with the item embedding vector, wherein the obtaining the importance representation of each item further comprises:
converting an item embedding vector set formed by each item embedding vector corresponding to each item in the current conversation session to a first vector space and a second vector space respectively so as to obtain a first conversion vector Q and a second conversion vector K respectively and the first conversion vector Q and the second conversion vector K are calculated according to:

Q
=

sigmoid
⁡
(

W
q

⁢
E

)

(
1
)

K
=

sigmoid
⁡
(

W
k

⁢
E

)

(
2
)

where W q ∈R d×l and W k ∈R d×l are trainable parameters corresponding to a query and a key, respectively; l is a dimension of an attention mechanism adopted while performing formulas (1) and (2); and sigmoid is a conversion function learning information from the item embedding vector in a nonlinear manner;
adopting a cross entropy function as an optimization target to learn the trainable parameters;
implementing a back propagation algorithm to train the item recommendation model;
obtaining an association matrix C between the first conversion vector Q and the second conversion vector K, wherein the association matrix C is calculated according to:

C
=

si
⁢

gmoid
⁢

(

Q
⁢

K
T

)

d

(
3
)

where √{square root over (d)} is used to reduce an attention pro rata;
blocking a diagonal line of the association matrix by one blocking operation during a process of obtaining the importance representation according to the association matrix, thereby removing irrelevant items in the current conversation session;
obtaining an importance score α i using the association matrix C, wherein the importance score α i is calculated according to:

α
i

=

1
t

⁢

∑

j
=
1

,

j
≠
i

t

C

i
⁢
j

;

(
4
)

and
obtaining the importance representation β i using a softmax layer and the importance score α i , wherein the importance representation β i is calculated according to:

β
i

=

exp
⁡

Assignees

• National Univ Of Defense Technology

Inventors

• Cai Fei
• Chen Wanyu
• Pan Zhiqiang
• Song Chengyu
• Wang Yitong
• Ling Yanxiang
• Zhang Xin
• Chen Honghui

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N5/04

  Inference or reasoning models · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/048

  Activation functions · CPC title