Granted Patent B2

US 12,417,344 · App. 17/296,821 · Granted Sep 16, 2025

Inventors: James Russell Geraci (Suwon-si, KR); Francisco Pena (Dublin, IE); Aonghus Lawlor (Dublin, IE); Barry Smyth (Dublin, IE); Ilias Tragos (Dublin, IE); Neil Hurley (Dublin, IE)

Assignee: Samsung Electronics Co., Ltd.

G06F40/216G06F40/284

Abstract

Claims (49)

1. An electronic device comprising:

a memory; and

a processor configured to:

obtain importance data representing an importance of each of a plurality of words included in text data using text data related to a plurality of items written by a plurality of users,

obtain a topic model representing a relationship between a topic and a word by applying a topic modelling algorithm to the obtained importance data,

obtain preference data representing preference of each of the plurality of users for the topic and relationship data representing a relationship between the plurality of items and the topic, based on the obtained topic model, and

train a recommendation model to output result data including an estimated preference for the plurality of items of the plurality of users, based on the preference data and the relationship data,

wherein the recommendation model comprises a first latent factor and a second latent factor, and

wherein the processor is further configured to:

set a numeral value included in the preference data as an initial value of the first latent factor,

set a numeral value included in the relationship data as an initial value of the second latent factor, and

train the recommendation model so that a difference value between a computation value of the initial value of the first latent factor and the initial value of the second latent factor and a pre-defined learning value is minimized.

2. The electronic device of claim 1 , wherein the processor is further configured to:

identify a noun word among the plurality of words included in the text data,

generate a bag of words including the identified noun word, and

obtain the importance data based on the generated bag of words.

3. The electronic device of claim 2 , wherein the processor is further configured to obtain the importance data by applying a term frequency-inverse document frequency (TF-IDF) algorithm to the generated bag of words.

4. The electronic device of claim 1 , wherein the processor is further configured to:

group the text data by the plurality of users,

obtain first frequency data representing a frequency of words used by each of the plurality of users using the grouped text data, and

obtain the preference data by performing a computation between the first frequency data and the topic model.

5. The electronic device of claim 1 , wherein the processor is further configured to:

group the text data by the plurality of items,

obtain second frequency data corresponding to the item using a frequency of a word included in the grouped text data, and

obtain the relationship data by performing computation between the second frequency data and the topic model.

6. The electronic device of claim 1 , wherein the recommendation model is further configured to, based on the difference value being minimized, perform computation between the first latent factor and the second latent factor and output the result data.

7. A controlling method of an electronic device, the controlling method comprising:

obtaining importance data representing an importance of each of a plurality of words included in text data using text data related to a plurality of items written by a plurality of users;

obtaining a topic model representing a relationship between a topic and a word by applying a topic modelling algorithm to the obtained importance data;

obtaining preference data representing preference of each of the plurality of users for the topic and relationship data representing a relationship between the plurality of items and the topic, based on the obtained topic model; and

training a recommendation model to output result data including an estimated preference for the plurality of items of the plurality of users, based on the preference data and the relationship data,

wherein the recommendation model comprises a first latent factor and a second latent factor, and,

wherein the training comprises:

setting a numeral value included in the preference data as an initial value of the first latent factor;

setting a numeral value included in the relationship data as an initial value of the second latent factor; and

training the recommendation model so that a difference value between a computation value of the initial value of the first latent factor and the initial value of the second latent factor and a pre-defined learning value is minimized.

8. The method of claim 7 , wherein the obtaining the importance data comprises:

identifying a noun word among the plurality of words included in the text data;

generating a bag of words including the identified noun word; and

obtaining the importance data based on the generated bag of words.

9. The method of claim 8 , wherein the obtaining the importance data comprises obtaining the importance data by applying a term frequency-inverse document frequency (TF-IDF) algorithm to the generated bag of words.

10. The method of claim 7 , wherein the obtaining the preference data comprises:

grouping the text data by the plurality of users;

obtaining first frequency data representing a frequency of words used by each of the plurality of users using the grouped text data; and

obtaining the preference data by performing a computation between the first frequency data and the topic model.

11. The method of claim 7 , wherein the obtaining the relationship data comprises:

grouping the text data by the plurality of items;

obtaining second frequency data corresponding to the item using frequency of a word included in the grouped text data; and

obtaining the relationship data by performing computation between the second frequency data and the topic model.

Assignments (1)

Priority Claims (2)

Continuity (1)

References Cited (33)