A Knowledge-based Approach for Personalised Clothing

Recommendation for Women

Hemilis Joyse Barbosa Rocha

, Evandro de Barros Costa

, Emanuele Tuane Silva

Natalia Caroline Lima

and Juliana Cavalcanti

¹Federal University of Alagoas-UFAL Institute of Computing,

Av Lourival Melo Mota, S/N - Tabuleiro dos Martins, Maceió - AL, Brazil

²Federal Institute of Alagoas-IFAL, Informatics in Campus Viçosa, AL, Brazil

Keywords: Knowledge-based Recommender System, Clothing Personalization, User Modeling, Intelligent and Decision

Support Systems.

Abstract: Currently, recommendation system technology has been assumed as a promising approach to contribute to

fashion domain in terms of electronic commerce. In this paper, we propose an approach for a clothing

personalized recommendation system that is able to help the women to identify appropriate clothing categories

together with models linked to clothing images, mainly based on their fashion styles and body types. To

achieve this, besides an intelligent user interface, our recommendation approach deals with two main

components: the user modeling and the clothing recommendation, which is responsible for recommending

fashion clothing items to women. The user modeling is responsible for creating and updating the user model,

including two main knowledge-based mechanisms: the first is responsible for automatically identifying the

fashion style, and the second is responsible for detecting body type. We evaluated our recommendation

approach and preliminary results indicate that it significantly supports the women with choices.

1 INTRODUCTION

There has been an increasing amounts of digital data

and online resources in several domains readily

available to users, forming a diversity of options to

make choices. This is not a trivial issue, mainly when

we are considering fashion domain. In recent years,

particularly, there has been a rapid growing of fashion

industry leading to very positive social and economic

impacts worldwide in the global digital marketplace.

Hence, this market segment has demanded academic

research mainly in the fields of artificial intelligence

(Russell, 2010) and visual computing, including

computer vision and image processing in fashion

domain (He and McAuley, 2016) and (Jagadeesh et.

al 2014). In particular, recommendation system

technology has been assumed as a promising

approach to contribute to fashion context in terms of

electronic commerce, suggesting clothing and other

fashion items to users (Sha, 2016), therefore,

facilitating and helping them on making appropriate

choices.

Building personalized systems for clothing

recommendation is a difficult task, mainly by

considering the diversity of choice possibilities and

styles. In this context, one relevant and challenging

research problem is on how to make effective

clothing recommendations to the users based on their

personal characteristics. Unfortunately, in a broad

sense, little research has been done on this issue, just

some specific parts of this issue has been addressed,

for instance fast item detection (Jagadeesh et. al

2014) and recognition (Sha, 2016) of apparel and

accessories in real-world images, followed by search

for similar items (Fukuda et. al, 2011) in online

shops.

The present paper addresses the mentioned

clothing recommendation problem by proposing an

approach for clothing personalized recommendation

system that is able to help the women to identify

appropriate clothing categories together with models

(here model means a kind of subcategory) linked to

clothing images, mainly based on their fashion styles

and body types. To achieve this, besides an

intelligent user interface that unable the system to

interact with the user, our recommendation approach

610

Rocha, H., Costa, E., Silva, E., Lima, N. and Cavalcanti, J.

A Knowledge-based Approach for Personalised Clothing Recommendation for Women.

DOI: 10.5220/0006337306100617

In Proceedings of the 19th International Conference on Enterprise Information Systems (ICEIS 2017) - Volume 1, pages 610-617

ISBN: 978-989-758-247-9

deals with two main components: the user modeling

and the clothing recommendation, which is

responsible for recommending fashion clothing items

to women. The user modeling is responsible for

creating and updating the user model, including two

main knowledge-based mechanisms to generate

relevant information to the user model: the first is

responsible for automatically identifying the fashion

style, and the second is responsible for detecting body

type. Particularly, we have considered the five

clothing categories(dresses, coats, tops, pants, skirt),

five body types, and nine clothing fashion styles.

Associated to each category, we define a clothing

database containing a collection of models linked to

women´s clothing images in order to be filtered by the

system and then help the user to select the appropriate

clothing option. Here, it is important to state that this

study belongs to a Project where we investigate the

feasibility and the reachability of the use of such

technology for clothing recommendation.

The rest of this paper is organized as follows. In

Section 2, we provide some background knowledge

and related work. In Section 3, we describe our

approach for clothing recommendation. In Section 4,

we discuss how our approach has been evaluated and

the current obtained results. Some conclusions and

proposal for some future work are presented in

Section 5.

2 BACKGROUND KNOWLEDGE

AND RELATED WORK

This section presents some background knowledge

relevant to understand the proposed approach for

clothing recommendation. Moreover, we discuss

some related work.

2.1 Some Clothing Fashion Concepts

2.1.1 Overview of Body Types

The female body types are differentiated by the basic

combinations of the measures of the shoulder, the

waist and the hips. We emphasize here that we are not

focusing their size, only the symmetry of the apparent

combination between them. Basically, we highlight

the following models and their proportions, and in

this case, most women fall into one of five body

types:

i) Hourglass: The measurements of the shoulders and

hips are practically the same, presenting the different

measures of the waist, being able to be only

demarcated or much narrower than the other

measures, that is, curvy but evenly proportioned;

ii) Inverted Triangle or Apple: In this female body

type the measurement of the shoulders is greater than

the waist measurement and the hips;

iii) Triangle or Pear: This female body type has

narrower shoulders and waist thin than hips, that is

larger at the bottom;

iv) Rectangle: This format shows the measurements

of the shoulders of the waist and of the hips being

practically the same, without many variations of the

measures, that is straight up and down; and

v) Oval: This type of female body is highlighted by

the width of the waist is much larger than the

measured shoulders or hips.

2.1.2 Overview of Clothing Fashion Styles

The concept of style originates in the Latin term stilus

which, in turn, derives from the Greek language. On

the other hand, habitual use refers to the taste,

elegance, or distinction of a person or thing. Thus, in

this work, we call "fashion style" the way each person

dresses and their preferences for pieces of clothing,

accessories and the image they want to convey

(Aguiar, 2003).

We describe below, in a very simplified way, the

9 main fashion styles that served as the basis for the

present recommendation system:

i) LadyLike: This style transmits romanticism and

sweetness. It is characterized by light colors, delicate

prints, marked waist and details of laces, lace and

ruffles;

ii) Classic: It refers to the name given to the style

that does not change due to trends, being a more

conservative and formal style;

iii) Rocker: It transmits boldness and it is

generally highlighted by dark colored pieces, plaid,

leather, studs and torn;

iv) Casual: It is characterized by the union of

classic and informal parts. It is versatile and very

bare;

v) Glam: It has characteristics like clothes with

soft lines, delicate jewelry, stiletto heels, sequins and

luxury items;

vi) Cool: It is characterized by the use of parts

that are always in fashion and the overlap of parts;

vii) Boho: It has as main element the comfort. It

is characterized by wide and fluid parts with light

fabrics and earth elements;

viii) Sexy: Transmits sensuality by showing some

parts of the body with just and short parts;

A Knowledge-based Approach for Personalised Clothing Recommendation for Women

611

ix) Activewear: Style of clothing intended for

sporting activities.

2.2 Overview of Recommender

Systems

Recommender systems (RS) are software

applications that aim to find the information or

product or even people that matches the user needs

and preferences, often attempting to reduce

information overload (Resnick, 1997). These systems

have proved to be helpful tools for various

information seeking and filtering tasks on several

domains on the Web, for instance in films, music,

books (He and McAuley, 2016). Their main goal is

to recommend items of interest to the end users based

on their preferences and other data collected. To

achieve that, most Recommender Systems exploit the

collaborative and content-based filtering approaches,

but there are other approaches, for instance

Demographic Filtering, and utility and knowledge-

based approaches (Fukuda et. al, 2011).

In the present research work, we are focusing on

knowledge-based recommender systems, mainly

involving knowledge representation and reasoning

techniques from the field of artificial intelligence.

Such systems make decisions and provide items or

services suggestions to users, according to their

preferences and needs and the knowledge

representation of the application domain. Therefore,

this kind of system generates recommendations to a

user by consulting its knowledge base of the product

domain, and then reasoning what items will best

satisfy the user’s requirements.

2.3 Related Work

Some previous work in the literature related to

clothing recommender system has focused on

collaborative filtering approach (He and McAuley,

2016). Following the line of fashion coordination

support systems, the work in (Fukuda et. al, 2011)

proposes a recommender system called Talking

Closet, which considers the user preferences to

suggest clothing combinations according to the user

closet, automating the daily cloths choice process. In

(Qingqing et. al, 2010), the authors describe a

recommender system for helping customers to find

their most suitable fashion choices in mass fashion

information in the virtual space based on multimedia

data mining from the Web. This system considers

some characteristics of the users, such as: color

preferences and clothing styles, as well body

parameters like skin tone. With a different proposal,

by directly considering fashion domain knowledge,

the work in (Vogiatzis, 2012) proposes a knowledge-

based recommender system for style advice in the

fashion domain, taking into account the knowledge of

domain expertise and user interaction data with

fashion sites. The present work is very similar to the

work in (Resnick, 1997) in the sense that it also has

invested in a knowledge-based approach for

recommendation. However, this work differs from

the other mentioned works, mainly in terms of the

used recommendation approach.

3 THE PROPOSED

RECOMMENDATION

APPROACH

In this section we discuss our approach and describe

the system architecture with its main components, as

well as, we describe the interactions between these

components. Moreover, we provide an overview

about implementations aspects of this

recommendation system.

3.1 Preliminary

Our approach for personalized clothing

recommendation to women is mainly based on their

fashion styles and body characteristics. It firstly

consists of creating and initializing with personal

data of a user model and subsequently executing

mechanisms to be performed in four main steps: the

first is responsible for automatically identifying the

fashion style, and the second is responsible for

identifying the user's body type. The third is

responsible for mapping pairs formed by fashion style

and body types into clothing models associated to

categories and finally, the fourth is responsible for

recommending clothing models associated with

clothing images for the women, taking into account

the clothing categories linked to a selected pair

<body type, fashion style> from a relationship

between a body type set and a fashion style set. Thus,

associated to each category, we have clothing models

with several linked to women´s clothing images,

forming a clothing database.

3.2 The Architecture

The recommendation system architecture, as shown

in Figure 1, consists of the following main

components: Intelligent User interface, user

modeling with the modules fashion style engine and

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

612

body type selector, clothing recommender with

models and clothing images, as well as three

knowledge bases and a clothing images database.

The proposed recommendation system was mainly

designed following the conceptual architecture of a

classical knowledge-based system.

Before going to describe each one of these

components, let us first to present a formalization of

some sets used in the recommendation system. Let S,

C, BT, and BP be four nonempty sets, where: S =

{s1, s2, …, sn}, n is the cardinality of S, that is, the

number of elements that belong to S, and S denotes a

set of clothing fashion Styles ; C = {c1, c2, …, ck},

k is the cardinality of C, and C denotes a set of

clothing Categories; BT = { bt1, bt2,...,bt5}, denoting

a set of Body Types; and BP = {bp1, bp2, …,bpm},

m is the cardinality of BP, and denotes a set of Body

Parameters; M = {m1, m2, …, mt}, t is the cardinality

of M, and M denotes a set of clothing Models

associated with the categories; IM = {im1, im2, …,

ims}, s is the cardinality of IM, and IM denotes a set

of clothing images associated with the models.

Figure 1: Recommender System Architecture.

3.2.1 The Intelligent User Interface (IUI)

This component has the responsibility of interacting

with the user and with the rest of the components of

the recommendation system. Thus, the user interacts

with the recommendation system via IUI, which

provides overall control over the whole process. This

component includes question-answer facilities that

are generated by the engines in the architecture,

enabling the system to interact with the user for the

purpose of asking questions and providing answers.

It is a kind of intelligent interface that includes a

controller mechanism playing the role of the control

of interactions in the recommendation system,

working as an intermediary between the user and the

two components: User modeling and Clothing

Recommendation Engine involving clothing

categories, models with clothing images. It takes the

responsibility for controlling the interactions among

the components and presenting information to the

user or to getting information from the user. It

contains rule interactions with knowledge about the

system components, that is, a kind of meta-

knowledge on the other components of the

recommendation system, having the meta-rules for

interacting with the user. Then, the main functions of

the controller are to interpret the user action and to

select the appropriate component for acting, to

coordinate the operations of the five components, and

to interact with users. Hence, it is able to recognize

all actions that an user or any component might

produce, and deciding what action should be taken.

Thus, IUI uses, for instance, rules in the form IF the

current user action is about the task T

THEN call

component

3.2.2 The User Modeling - UM

This component is responsible for creating and

updating the user model. To achieve this, it requests

and stores in the KBUM knowledge base from the

user model: (i) information directly elicited from the

user by the interface and (ii) information inferred by

the two modules in UM: SE, BT, and also (iii)

information inferred by the CCRE component.

 The Clothing Fashion Style Engine - SE

This module is responsible for interacting with the

user, via questionnaire generated from the rules in the

style knowledge base, aiming to identify her clothing

fashion style. Hence, to select the clothing fashion

style, the engine works on a knowledge-base in the

form of production rules and then, sometimes, by

asking questions to the user, keeping user's responses

in a working memory. Therefore, from these rules

and user's inputs, the SE makes a decision about user's

fashion style. Its inference engine uses backward

chaining to explore the rule base.

Formally, SE can be defined as follows:

Input: a rule base and some pairs <attribute,

value> obtained from the questionnaire answered by

the user taking into account questions from the rules

asked her.

Process: Rule-based inference Engine.

Output: si⋲S, representing a single element from

 The Body Type Selector - BTS

This module is responsible for selecting the user's

body type.

Formally, BTS can be defined as follows:

A Knowledge-based Approach for Personalised Clothing Recommendation for Women

613

Input: rule base and set of values for the used body

parameters.

Process: Inference on a set of IF THEN Rules.

Output: bt

, representing a specific body type.

3.2.3 The Clothing Category Recommender

Engine - CCRE

This component is responsible for providing users

with personalized clothing recommendation

containing to each category a collection of models

with linked clothing images, based on the

relationships that exist between her body type

information and her fashion style. The representation

of the clothing domain knowledge is in the form of

production rules with their conclusions parts

containing actions to query a table and to generate a

clothing recommendation output, according to the use

of an inference engine, which explores the rules.

Mathematically, using the relation notation, we can

express the recommendation system (rs) as follows:

rs: (S x BT) → ((C x M) x IM).

Formally, CCRE can be defined as follows:

Input: rule base and a table with pairs <btj , si >

mapping into < C, set of models to each category, set

of clothing images to each model>, meaning a pair

containing a particular body type and a particular

fashion style.

Process: rule-based inference engine with output

actions to query to the decision table containing styles

versus body types linking with each clothing category

and associated clothing model with a collection of

images.

Output: clothing category, from the cartesian style

and body type, associated with clothing models

linked to a collection of clothing images.

3.2.4 The Knowledge Bases and the

Clothing Database

User Model as Knowledge Base: A user model is a

representation of the user characteristics that may be

relevant for her interaction with the recommendation

system. Thus, this component stores information

about the user, as a kind of internal representation of

her. We have considered body parameters like skin

color, hair color, waist, bust, hip, height and weight,

where such information is obtained when the user is

asked by the system by means of a form filled.

Addionally, there is other information about the

user’s characteristics and that is also updated in each

recommendation session. It is captured, by a dynamic

user modeling component, according to the inference

processes and to the function computation. The

knowledge is represented by facts with pairs

<attribute, value> expressed in a relational database.

Fashion Clothing Style Knowledge Base via

Rules(KBS): This knowledge base contains a

collection of 41 IF-THEN production rules and a

connected working memory, to be used together by

the inference engine, allowing to answer question on

what is the fashion style of a given user. These rules

were directly obtained from a constructed decision

tree.

Clothing Domain Knowledge via Rules(CDKD):

This knowledge base contains a rule base and an

associated decision table involving a relationship

between styles and body types, mapping into clothing

categories linked to models and clothing images

collections. In this case, an example of rule from

Knowledge Body Type is IF BodyType = hourglass

and Style = LadyLike THEN skirt1=Eve and

skirt2=Flared, where a particular body type and style

yields two models for the category skirt. Each model

is linked to a collection of clothing images.

Clothing Database: It contains collections of models

associated with each clothing category, and a set of

women´s clothing images connected to each one of

those models.

3.3 Overview of User-System

Interactions

The interaction process begins when the user accesses

the system and then starts a basic dialogue with the

interface component. The interface first checks if the

user is entering in the system for the first time. If so,

it requests the user modeling to create the user model

with her profile. Next, after this basic registration

with some personal data, the system continues the

interaction process as follows.

In general, from now on, the interaction between

the user and the system in the recommendation

process of clothing is achieved taking into account

interaction components that deal with physical

characteristics and with fashion styles. Then, the full

recommendation process is performed by following

three main steps: first, the system, via user modeling,

identifies the user’s style based on the execution of

the SE module working on the Fashion Style

Knowledge Base and on the user’s answers to the

questions generated by the processed rules. The

second, the system, via user modeling, uses body

parameters of the user to infer the user's body type.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

614

The third, the recommendation system uses the output

of the first and second steps to select clothing

categories with models linked to clothing image

collection, associated with each category, which

matches with the identified fashion style and a

particular body type.

3.4 Design and Implementation

Aspects

This section explains the implementation of the

recommendation system, in line with the architecture

and description presented. We first discuss the used

knowledge bases considering acquisition and

knowledge representation aspects, as well as, the

database solution. Then, we discuss the used rule-

based inference engines. In general the

recommendation system is implemented in Java

technology, by considering the multiplatform

characteristics of this technology, and as a database

implementation we used relational database

technology adopting MySQL as a database

management system.

3.4.1 Knowledge Bases and Inference

Engines

Our approach involves the design of a

recommendation system using knowledge expressed

in three knowledge bases represented by production

rules. Specifically, to model the knowledge about

fashion styles we first used a decision tree and then

mapped this tree into a set of If-Then statements for

the development of the rule-base. To model the

knowledge about body types, as well as the

knowledge for clothing recommendation, we used a

set of If-Then rules. Thus, as discussed before, three

rule-based systems have been designed, one for the

inference of fashion style, other for the inference of

body type, and finally, another to clothing

recommendation. We have focused on utilizing

expert knowledge bases to detect both fashion style,

to identify body type, and to clothing

recommendation. The knowledge base for the style

detection was stored as related tables in a relational

database using MySQL. The rule has the following

structure: IF preconditions THEN action , where the

rule's preconditions and action are expressed in terms

of triples <attribute, operator, value>, denoting: (1)

the preconditions (IF) that involves a set of conditions

to be satisfied; (2) the consequent (THEN) that

contains actions to be executed or new knowledge to

be produced if the consequent is true. The knowledge

base for representing body types knowledge is also

expressed in If-Then rules. Moreover, the clothing

knowledge base is also expressed in If-Then rules, as

illustrated in one example in 3.2.7.

The inference engine uses rule chaining with

backward chaining algorithm to obtain its

conclusions. In general the recommendation system

is implemented in Java technology, by considering

the multiplatform characteristics of this technology.

We implemented the three used inference engines

using the rules Engine Drools shell that is

implemented in Java, allowing easy integration with

some module written in Java. It is possible to invoke

Java methods from Drools and vice versa. This

enables us to develop more powerful and flexible

rules.

3.4.2 Clothing Database Model

It contains a set of models and a set of women´s

clothing images connected to each one of those

models, which then are associated with each clothing

category. In this database model we represent the

main relations in the system, as well as, the

relationship among them. Thus, in Figure 2 is

presented part of the relational database model

containing the following relations: style,

style_has_category, category, bodytype,

bodytype_has_category, models and images. The

Style relation stores data related clothing styles and

each style has one or more categories, as represented

by the Category relation, as well as, the same category

can be in more than one style. The BodyType relation

represents the body types and can be associated to

more than one category and vice-versa. Moreover,

there is a transitive relationship among the relations

Category, Models and Images, and in this sense, a

category has one or more models and a model has one

or more clothing images.

Figure 2: Clothing Database Model.

A Knowledge-based Approach for Personalised Clothing Recommendation for Women

615

4 PRELIMINARY EVALUATION

AND RESULTS

In this section we are going to describe a preliminary

experiment, involving twelve women, we have

conducted to evaluate our approach. The main

purpose of this experiment is twofold: (i) to evaluate

the quality of the knowledge bases for fashion styles

and for body types of the system in terms of their

accuracies and (ii) to measure if the recommendation

system significantly helps the users to find suitable

clothing choices.

The participants of this experiment consisted of

12 women, where all of them are between 22 and 30

years of age, as well as, 8 are undergraduate students

from a public university and 4 have a graduate degree.

Additionally, in this study we have a support of a

Personal Stylist. We explained to each participant the

procedure of the experiment and what kind of data

would be collected, as well as, the aim of the

experiment. In Table 1, there are two examples of

body measures of two women W1 e W2 that

participated in the experiment. The data in the table

were used as input to the body type engine in order to

it infer the body type of the women.

Table 1: An example of inputs to detect body type: Body

measures for each woman.

Women

Bust

Waist

Hip

Height

Weight

92.5

69.4

83.5

1.65

109.4

112.00

98.4

1.69

Additionally, the system was tested by one

personal stylist aiming to check its functionality in

different scenarios to cover a variety of situations.

Table 2: Example of agreement degree between the system

inferences and Stylist opinions, regarding body type and

clothing style.

Women

System Output

Personal Stylist Opinion

Body Type

Style

Body Type

Style

Apple

Ladylike

Apple

Classic

Hourglass

Glam

Hourglass

Glam

In Table 2 is an example from the experiment to

illustrate the input and output of the system and of the

stylist opinion, concerning data of two women with

respect to the identification of body type and of

fashion style. There was a total agreement with regard

to the detection of body type. However, concerning

the style, there was a little divergence between the

two evaluations, once that to W1 the system assigned

to “Ladylike”, whereas the stylist assigned to

"Classic". Such divergence, however, was explained

by the expert, considering that the two styles are very

close in terms of some similar characteristics.

All participants answered all the proposed

questions. Overall, this brief evaluation indicated that

the system is feasible (meaning that it works: In all

the input-output tests accomplished, the results were

satisfactory) and that all participants mentioned that

they found the clothing recommendations useful and

that the information contained in the

recommendations helped them to improve their

understanding about fashion clothing domain.

Women also agreed that the system is useful. But,

concerning the comparison between the system

output and the stylist opinion, there was some

disagreement between them with respect to style, in

just two situations among twelve. This situation,

however, was not considered a relevant problem to

the system quality in terms of its accuracy.

5 CONCLUSION AND FUTURE

WORK

In this paper, we presented a personalized approach

for clothing recommendation, explaining how the

system generates personalized recommendations for

women. We reported the first evaluation results,

showing that the quality aspects of the constructed

knowledge bases and then the significant help

provided by the recommendations to the users. But,

of course, even with preliminary positive results until

now, there is an urgent need for more and more

evaluations with the same experiment to better test

the system, involving other users with more different

characteristics.

As immediate future work, we will focus on

provide more experiments toward better

understanding and limitations of the proposed

approach. In parallel, we will extend the

recommendation system by including one more filter

in the last phase in order to promote more specificity

in the clothing recommendation, as well as we will

improve the user model in its dynamic information

part to capture other user's preferences and

characteristics.

ICEIS 2017 - 19th International Conference on Enterprise Information Systems

616

REFERENCES

Russell, S. and Norvig, P. Artificial Intelligence: A

Modern Approach. Pearson, third edition, 2010.

R. He and J. McAuley. Ups and downs: Modeling the visual

evolution of fashion trends with one-class collaborative

filtering. In WWW 2016: 25th International World

Wide Web Conference. 2016.

V. Jagadeesh, R. Piramuthu, A. Bhardwaj, W. Di, and N.

Sundaresan. Large scale visual recommendations from

street fashion images. August 2014, KDD'14:

Proceedings of the 20th ACM SIGKDD international

conference on Knowledge discovery and data mining.

Dandan Sha et. al, 2016. An Approach for Clothing

Recommendation Based on Multiple Image Attributes.

Springer Book chapter: Web-Age Information

Management, Volume 9658 of the series Lecture Notes

in Computer Science, pp 272-285.

Fukuda, Mio; Nakatani, Yoshio. “Clothes Recommend

Themselves: A New Approach to a Fashion Coordinate

Support System”. Proceedings of the World Congress

on Engineering and Computer Science (WCECS),

2011.

Aguiar, Titta. Personal Stylist: guia para consultores de

imagem. São Paulo: Editora Senac São Paulo, 2003.

Resnick, Paul; Varian, Hal R. “Recommender Systems”.

Communications of the ACM, ACM, New York, Ny,

USA, Vol. 40, nº 3, p. 56-58, March, 1997.

Qingqing, Tu; Le, Dong. “An Intelligent Personalized

Fashion Recommendation System”. In Proceedings of

the International Conference on Communications,

Circuits and Systems (ICCCAS), 2010, pp. 479–485.”

D. Vogiatzis, et. al., 2012. “Expert and community based

style advice”. Expert Systems with Applications: An

International Journal, vol. 39, no. 12, pp. 10647–10655,

September 2012

A Knowledge-based Approach for Personalised Clothing Recommendation for Women

617