k-nearest neighbor (knn) based product ranking model with ... · malhotra, dhairya et. al. has used...
TRANSCRIPT
INTERNATIONAL JOURNAL OF TECHNOLOGY AND COMPUTING (IJTC)
ISSN-2455-099X,
Volume 2, Issue 8 August 2016
IJTC201608006 www. ijtc.org 402
K-Nearest Neighbor (kNN) based Product Ranking
Model with Minimum Cost Evaluation
Manpreet Kaur1, Dr. Raman Chadha2 1Research Scholar, 2Professor and Head,
CGCTC, Jhanjeri, Mohali [email protected], Mohali, [email protected]
Abstract--The product ranking model is the model to create the ranking lists over the product listing data obtained from the e-
commerce portals. The product ranking models are utilized to produce the variety of product ranking lists in order to fulfill the
requirement to show the multiple lists over the e-commerce ranking lists. The e-commerce portals use the variety of lists to
increase the usability and the accessibility by creating the maximum visibility of the output results, which are shown for the
similar product ranking, products suggested by other users, most selling products, newly added products, etc. A perfect product
ranking model must be capable of producing the content-based and collaborative indexing over the input data. The content-based
product ranking models are utilized to produce the product ranking by evaluating the input query by the users. The collaborative
filtering is the algorithm used to create the product ranking lists according to the user’s profile and other similar users, friends,
etc. In this thesis, the proposed model has been designed by using the effective combination of the content-based and collaborative
filtering methods for the realization of the robust product ranking model.
Indexed Terms—Product ranking, e-commerce ranking, k-nearest neighbor, non-probabilistic classification
I. INTRODUCTION
E-commerce stands for Electronic commerce. It is that the
shopping for and commercialism of products and services
employing a network resembling web. E-commerce is
employed in on-line searching, bill payments, ebooks, e-
mail , etc. The businesses offer these services square
measure Amazon, flipkart, quikr, snapdeal, olacabs and
paytm. E-commerce is principally used as a result of it
reduces dealings value and improve level of client services.
It conjointly has worldwide market accessibility. Numerous
classes of e-commerce are: business-to-business, business-
to-customer, customer-to-customer or customer-to-business.
The ranking system is extremely helpful particularly in e-
commerce. A number of the advantages of ranking are:
It helps the shoppers to choose that choice to be chosen
among the multiple choices and in less quantity of your
time.
The personalised e-commerce ranking system has
redoubled the trust and loyalty relationship between
customers and web site homeowners.
Provides comparison between totally different
corporations and websites that increase competition
between the web site homeowners which ends up in
higher and improved services.
Ranking product may be a method to rank numerous
product. First off one product is graded then grouping of
product are performed and also the teams of product is
graded. Let's say let’s take a product Mobile phones, they
there are sorted consistent with their corporations like
Samsung, Nokia, Apple, HTC etc. And that they are sub
sorted consistent with their models and at every step
ranking is performed.
Semantic internet is employed to store knowledge within
the system while not the steering of human and might be
simply clear to humans in type of websites. In linguistics
internet knowledge stores are created on internet. It
develops the common framework that allows to share the
information among corporations, communities or
applications. The linguistics internet is amalgamated with
the ranking system and during this dynamic consolidation is
performed. In dynamic consolidation, if a brand new
product is entered in an exceedingly graded list then, it'll
mechanically generate the rank of the merchandise by
examination it with different product within the list and then
it updates the ranking.
II. LITERATURE REVIEW
Hepp, Martin et. Al. Has worked for the e-commerce
notably schema.org and good relations for researchers and
practitioners on the online of the info. Within the paper, the
author has given the introduction and first steerage on the
abstract structure of schema.org. They need created the
patterns for demand and possession that embody the range
of things like piece of furniture, apparels, natural
philosophy devices, cosmetics, books, etc. And have created
a full tool chain for manufacturing and overwhelming the
actual information. The author have conjointly mentioned
the subject like authentication (e.g. With webid), identity,
access control; information management problems from the
publisher and client perspective and micropayment services.
The drawback of this application is restricted to micro-data
that isn't applicable for e-commerce product ranking
system.
Sessoms, Matthew, and Kemafor Anyanwu has worked
model and algorithms for sanctionative a Package question
criterion on the linguistics net. The package question is that
the combination of multiple queries that helps to induce
resource combination on linguistics net. The taxonomic
category of such queries is “skyline package queries”. In
distinction to package queries on one relative models, the
RDF model have injected the challenge of deciding the
skyline package of ternary relations over multiple joins. The
various combination of latest operators for skyline package
queries relative question operators and RDF information
storage models have developed the four ways for analysis.
The author lacks within the use of further techniques for
optimization love prefetching additionally because the
integration of top-k techniques.
IJTC.O
RG
INTERNATIONAL JOURNAL OF TECHNOLOGY AND COMPUTING (IJTC)
ISSN-2455-099X,
Volume 2, Issue 8 August 2016
IJTC201608006 www. ijtc.org 403
Malhotra, Dhairya et. Al has used the rear propagation
neural network on Intelligent net mining to upgrade website
rank. Due to the rise in data resources, the online is
developing at quicker rate. However its large size will
increase the difficulties throughout the analysis method in
extracting the desired data from net. This drawback are
often overcome by exploitation the personalised net search
however the user should provide his personal data to keep
up privacy. Within the paper, the author addresses all the on
top of mentioned problems by exploitation the rear
propagation neural network for implementing page ranking
method.
Mital, Monika et. Al. Have planned The integrative
framework within the context of e-procurement and ERP to
spot determinant of selection for saas. During this paper, the
author has tried to classify, determine and rank the size that
are influencing saas sourcing call. Exploitation extended
AHP (analytic hierarchy process) technique, the framework
is analyzed that helped in distinguishing quality and prices
so weights criteria are known exploitation the info that was
collected by eight users and nine service suppliers of saas
supported ERP and e-procurement.
III. EXPERIMENTAL DESIGN
The proposed model has been designed for the product
ranking and listings over the product data obtained from the
e-commerce portals. The multivariate ranking of the
products plays the vital role in the listing of the products.
The products are required to be listed in the multiple types
according to the requirement. The proposed model has been
primarily defined in the two major stages for the product
ranking according to the content based ranking and user’s
preference based ranking. The first phase of the ranking is
computed on the basis of the product properties and
popularity given in the form of product rating, manufacturer
rating, popularity and other similar factorization. The
proposed model evaluates the product entities on the basis
of the individual product properties as per given in the
database extracted from the online source. The proposed
model evaluates the product properties in the factorization
method called popularity, accessibility and trust (PAT)
assessment or evaluation. The proposed model design for
the content based ranking based upon the PAT features has
been enlisted in the following algorithm design:
Algorithm 1: The content based product ranking
1. Acquire the input data
2. Extract the popularity, access and trust features
for all products
3. Apply the averaging mean factor (AMF) over the
popularity, access and trust features for each
product entry
4. Add all of the AMF factors to create the
cumulative mean
5. Update the product list with the cumulative means
6. Apply the content ranking based upon the
cumulative mean value
7. Sort the product entries in the database according
to the cumulative mean value
8. Return the ranking list
Algorithm 2: Collaborative user-similarity based ranking
1. Acquire the product data with PAT features
2. Acquire the current user’s browsing history
3. Compute the weighted mean based upon the
browsing history
4. Construct the weighted mean vector from the
product ranking data
5. Acquire the all user mean vector database
(AUMVD)
6. Initiate the k-nearest neighbor model factors
7. Run the iteration for all entries in the AUMVD
a. Obtain the current user’s mean vector
b. Obtain the mean vector from AUMVD on
current index
c. Compute the squared distance between
the two vectors in (a) and (b)
d. Update the similarity data matrix (SDM)
e. Update the kNN classification result data
f. If the last entry on AUMVD
i. Return the iteration
8. Sort SDM in the descending order
9. Extract the N user list from the top
10. Acquire the overall weighted mean vector
(OWMV) from the top N users
11. Run the iteration for all entries in the product list
a. Obtain the current product vector
b. Obtain the OWMV
c. Compute the squared distance between
the two vectors in (a) and (b)
d. Update the Product Similarity Matrix
(ProSiM)
e. Update the kNN classification result data
f. If the last entry on product list
i. Return the iteration
12. Sort SDM in the descending order
13. Return the ProSim Matrix as the result
IV. RESULT ANALYSIS
The proposed model named PAT model, which has been
compared against the existing model of product ranking
based upon web parameters over the given e-commerce
product list in this simulation scenario with similar structure
and environment. The detailed analysis of the simulation
results has been conducted by analyzing the results obtained
from the existing and proposed model simulations. The
proposed model has been described efficient and effective
while evaluated on the basis of the projected resources and
entropy for the coalition of the network load and uniqueness
respectively. Minimum of the 10 percent improvement has
been observed in the favor of the proposed model when
compared on the basis of various performance parameters in
the variety of simulation scenarios.
The adverse situations have been analyzed theoretically and
the effective in the results has been observed in the
proposed model to mitigate the errors and repetitions with
the new product ranking model than the existing model. The
newly proposed scheme has been primarily evaluated for
the utilization of the resources while using the ranking
model the given product data in the simulation. The
utilization of the resources, measured in the form of
projected resources has been recorded significantly lower
than existing model, which clearly indicates the robustness
of the proposed model. The following figure 1 indicates the
robust performance of the proposed model.
IJTC.O
RG
INTERNATIONAL JOURNAL OF TECHNOLOGY AND COMPUTING (IJTC)
ISSN-2455-099X,
Volume 2, Issue 8 August 2016
IJTC201608006 www. ijtc.org 404
Figure 1: Projected Resources comparison between existing and proposed model
Additionally, the entropy parameter signifies the higher
uniqueness level in the ranking data using the proposed
model than the existing model. The proposed model
evaluation over the entropy has been described in the
following figure 2. The high uniqueness of the product table
data increases the additional robustness to the ranking
models associated with the arrangement of the entries in the
perfect form over the given set of data. The result of the
proposed model has been observed nearly double than the
existing model, which can be clearly indicated by the
following figure:
Figure 2: Entropy comparison between existing and proposed models
V. CONCLUSION
The individual entity relationship based similarity
evaluation has been utilized for the content based filtering,
which utilizes the popularity, accessibility and trust (PAT)
factors based content filtering algorithm. The PAT features
have added the flexible and robust content-filtering model.
The k-nearest neighbor classification model has been
utilized for the collaborative filtering model for the
evaluation of the similarity of the other users, when the new
users are enlisted over the e-commerce engines. The
historical data of the existing users is evaluated using the
averaging factors to compute the first level test vector,
ExistingPropose
dExisting
Proposed
ExistingPropose
dExisting
Proposed
ExistingPropose
d
S1:E100 S2:E500 S3:E1000 S4:E2000 S5:E5000
Average 2.6499 1.5469 13.3108 7.7344 26.4594 15.4688 52.7602 30.9375 110.9954 77.3438
Minimum 1.8221 1.0977 7.8656 5.4883 15.0027 11.25 35.7773 21.9531 80.7805 54.8828
Maximum 3.7667 1.9648 19.7488 9.8242 38.6813 19.6484 77.2214 39.2969 160.2389 98.2422
0
20
40
60
80
100
120
140
160
180
Average Minimum Maximum
Existing Proposed Existing Proposed Existing Proposed Existing Proposed Existing Proposed
S1:E100 S2:E500 S3:E1000 S4:E2000 S5:E5000
Average 2.0207 2.4678 2.0131 2.6344 2.0712 2.7662 1.9829 2.5944 2.1466 2.5038
Minimum 1.7632 2.0544 1.609 2.155 1.7707 1.991 1.5884 1.9055 1.926 1.9444
Maximum 2.2706 2.9963 2.4393 3.1638 2.3884 3.2897 2.3225 3.0931 2.2644 3.1026
0
0.5
1
1.5
2
2.5
3
3.5
Average Minimum Maximum
IJTC.O
RG
INTERNATIONAL JOURNAL OF TECHNOLOGY AND COMPUTING (IJTC)
ISSN-2455-099X,
Volume 2, Issue 8 August 2016
IJTC201608006 www. ijtc.org 405
which is further classified with the k-nearest neighbor
model for the production of the collaborative filtering. The
proposed model results have been evaluated in the form of
various performance parameters associated with the time
complexity, duplicate entry evaluation and resource usage.
The proposed model has outperformed the existing model
on the basis of all of the above listed parameters during the
performance evaluation phase
REFERENCES
[1]. Verma, Neha, Dheeraj Malhotra, Monica Malhotra,
and Jatinder Singh. "E-commerce Website Ranking
Using Semantic Web Mining and Neural
Computing." Procedia Computer Science 45 (2015):
pp. 42-51, ELSEVIER.
[2]. Hepp, Martin. "The Web of Data for E-Commerce:
Schema. org and GoodRelations for Researchers and
Practitioners." In Engineering the Web in the Big
Data Era, pp. 723-727. Springer International
Publishing, 2015.
[3]. Sessoms, Matthew, and Kemafor Anyanwu.
"Enabling a Package Query Paradigm on the
Semantic Web: Model and Algorithms."
In Transactions on Large-Scale Data-and
Knowledge-Centered Systems XIII, pp. 1-32.
Springer Berlin Heidelberg, 2014.
[4]. Malhotra, Dhairya. "Intelligent web mining to
ameliorate Web Page Rank using Back-Propagation
neural network." In Confluence The Next Generation
Information Technology Summit (Confluence), 2014
5th International Conference-, pp. 77-81. IEEE,
2014.
[5]. Scioscia, Floriano, Michele Ruta, Giuseppe Loseto,
Filippo Gramegna, Saverio Ieva, Agnese Pinto, and
Eugenio Di Sciascio. "A Mobile Matchmaker for the
Ubiquitous Semantic Web." International Journal on
Semantic Web and Information Systems (IJSWIS) 10,
no. 4 (2014): 77-100.
[6]. Furukawa, Takao, Kaoru Mori, Kazuma Arino,
Kazuhiro Hayashi, and Nobuyuki Shirakawa.
"Identifying the evolutionary process of emerging
technologies: A chronological network analysis of
World Wide Web conference sessions."
Technological Forecasting and Social Change 91
(2015): 280-294.
[7]. Mital, Monika, Ashis Pani, and Ram Ramesh.
"Determinants of choice of semantic web based
Software as a Service: An integrative framework in
the context of e-procurement and ERP." Computers
in Industry 65, no. 5 (2014): 821-827.
[8]. D.T. Green and J. M. Pearson, “The examination of
two web site usability instruments for use in B2C e-
commerce organizations,” Journal of Computer
Information Systems, Vol. 49, No. 4, 2009, pp. 19-32
[9]. T. Wang and Y. Lin, “Accurately predicting the
success of B2B ecommerce in small and medium
enterprises,” Expert Systems with Applications, Vol.
36, No. 2, published by Elsevier, 2009, pp. 2750–
2758.
[10]. M. Lazarica and I. Lungu, Aspecte privind
proiectarea sistemelor de comert electronic, ASE
Publishing House, 2007, pp. 147.
IJTC.O
RG