Effectiveness of Profile Injecting Attack on Recommender System

Verified

Added on 2024/07/12

AI Summary

This research proposal investigates the effectiveness of profile injecting attacks against recommender systems, particularly focusing on the Netflix platform. It explores the motivation behind such attacks, including push and nuke attacks, and their potential impact on user experience and system integrity. The research questions delve into the role and consequences of profile injecting attacks, the current and future status of these threats, and suitable approaches to mitigate them. The proposal references several research articles discussing shilling attacks, hybrid attacks, and methods for detecting anomalous ratings. The methodology outlines the research approach, including literature review, data collection, and analysis, with a detailed research schedule presented in a Gantt chart. Ultimately, the research aims to contribute to a better understanding of the security challenges faced by recommender systems and to identify effective strategies for protecting user profiles and system performance.

Research Proposal
Research Topic “Investigating the
effectiveness of profile injecting attack
against recommender system”
Student Name:
Student ID:
1

Paraphrase This Document

Need a fresh take? Get an instant paraphrase of this document with our AI Paraphraser

Table of Contents
Research Project Motivation......................................................................................................... 3
Research Question.........................................................................................................................4
Thesis Description..........................................................................................................................5
Article: Shilling attacks against recommender systems: a comprehensive survey.........................6
Article: Hybrid attacks on model-based social recommender systems..........................................7
Article: Estimating user behaviour toward detecting anomalous ratings in rating systems..........8
Article: Recommender Systems— Beyond Matrix Completion......................................................9
Article: Android fine-grained permission control system with real-time expert recommendations
..................................................................................................................................................... 10
Methodology...............................................................................................................................11
Research schedule....................................................................................................................... 13
Gantt Chart:.................................................................................................................................14
References:.................................................................................................................................. 15
2

Research Project Motivation
Recommender system has been used in the number of the field such as digital media, news, e-
commerce, social networks, and tourism sector. There is various recommender system such as
Amazon, Netflix and many more. Although they handle the information and gives personalized
services, they have great privacy concerns. In this research proposal, we have research the
various attacks occurred on the recommender system “Netflix” and investigate the efficiency of
profile that has injected against the Netflix recommender system. There are two attacks, one is
Push attack and the other is Nuke attack. Push attack tries to protect the recommendations that
are positive and Nuke attack tries to protect the recommendations that are negative. The main
motivation of this research proposal is to investigate the effectiveness of profile injecting the
Netflix system. Netflix is a media streaming system that provides tv shows, online videos,
movies and many things. On researching and investigating this proposal, it is found that various
attacks happened on these recommender systems like Hybrid attack, and Shilling attack. With
the increasing demand and popularity of the digital social platform, there are more chances of
happening of such types of attacks. Because of these rating systems and comment area section
under the Netflix recommender system, they are vulnerable to such malicious attacks or
malware. In this research paper, we are presenting several attacks and their impacts on the
Netflix recommender system in many aspects. The malicious users inset the forged profiles in
the item box or the details of the user’s profiles so that the predicted ratings can be affected by
them in place of their benefits, Rashid (2007).
3

Research Question
The main aim of this research proposal is to deeply study and analyze the topic” “Investigating
the effectiveness of profile injecting attack against recommender system”. This research
proposal covers the following research questions:
 Study the role and working of the efficiency of outline injecting attack in contradiction of
Netflix recommender system.
 What are the consequences of this profile injecting attacks?
 Study the current and future status of profile injecting attacks.
 The main objective of researching this proposal is to study the security concerns of the
users and the Netflix system.
 Study the suitable and appropriate approaches to overcome these attacks and provide
security to the Netflix recommender system from these attacks.
 What are the challenge and security issues faced by these recommender systems and how
user secure their profile on the Netflix recommender system from this profile injecting
attacks?
4

Paraphrase This Document

Need a fresh take? Get an instant paraphrase of this document with our AI Paraphraser

Thesis Description
A recommender system is an application from which users can extract personalized and high
valued information from the huge sources. According to Wang, Tang (2013), recommender
system helps the users to find the stuff that they want and care from the large unmanageably
number of things by matching up with the list. The Netflix recommender system is the platform
that provides the online video streaming services to the millions of users. In general, this
recommender system has helped the people to find the things according to their field of
interest among the large pool area. Both users and operators have the mutual benefits from
the Netflix recommender system. Users have benefits as they do not go through thousands of
available media items. As the internet is becoming more popular, users watch Tv shows, movies
and music online and Netflix is the most popular used recommender system. But there are
many profiles injecting attacks that affect the Netflix recommender system. In profile injection
attack users make the fake profile that is often mostly correlated and they have the power to
influence the more people and the things available on that recommender sites in order to affect
the predicted ratings or diminished the system’s performance in place of the system goals. The
purpose of some attacks is to increase the popularity of some targeted data stuff and such
types of attacks are patch attack, while some kinds of attack have the intention to decrease the
popularity of the items that are mainly targeted and such attacks are known as nuke attack.
Now it becomes very imperative to handle this type of profile injecting attacks, and that’s why
researchers have been paying more attention to such attacks. The thesis studies major research
areas about the profile injecting attacks and the solutions to get secure from them and also to
avoid them in the future. We have researched the five peer research papers that cover the
different malicious attacks, attack types of profile injection, detection systems, an algorithm
that are used to overcome such profile injection attacks and analyze them to provide security
and performance. This research proposal investigates the harmfulness of the fusion attack and
likened all the attacks, the loss that the Netflix recommender system suffers from such attacks,
and also analyzed that the feedback links given by the users to the spammers will be more
affected or not.
5

Article: Shilling attacks against recommender systems: a comprehensive survey
According to Gunes, Kaleli, Bilge, Polat (2012), the main objective of all the shilling attacks is to
either nuke or push the popularity of particular product of the Netflix recommender system in
order to increase the economic profits than other competitors. There are two shilling attacks,
one is pushing attack and the other is a nuke attack, they are designed to increase and decrease
the product’s popularity respectively. The attacker must know about some details of the
recommender system to attacks. They generate fake profiles of popular and well-known items
and in this way, the injected profiles easily push the prophecies to the target item and
associated with other users in terms of similarity. There are some methods that are designed to
obfuscate the standard attacks, like noise injection, is used to mask the monogram of common
attacks and random numbers are created from Gaussian distribution to add in each rating list
inside the fake profile injections. To detect the shilling attack in Netflix recommender system,
some techniques are designed like statistical, data-reduction techniques, and clustering. In the
statistical technique, anomalies in the database are identified that can arise due to suspicious
ratings. X-bar control limit and confidence interlude control edge are there that determine the
measurement of how far is the item from overall mean and assume the rating list respectively.
The clustering technique periodically clusters the database and checks if there is any unwanted
change or not. It is concluded from this research paper that more the algorithm to detect them
strategies should be created to secure from these fake profile injections.
6

Article: Hybrid attacks on model-based social recommender systems
According to Yu et al, (2017), social recommender system like Netflix is the combination of
social network services and recommender systems and both depend on the profiles of the user.
In CF techniques, attackers inject the fake profiles to manipulate the recommendations, and
they also established the bogus relationships with the users to expand their link farm or social
spammers. It is analyzed from this research paper that social spammers and hybrid attackers
are mostly injecting the fake profiles in the Netflix recommender system. Spammers give farm
links to the users to increase the follower, due to this Netflix recommender system may create
the incorrect recommendations to the users and this creates benefits for the spammers. The
profile made by the user to access the Netflix system contains rating and relation profile. The
score profile is used to differentiate them and to encourage the objects and the relative profile
is used to seizure more followers to increase the profit. In this research paper, we discover the
influence of cross attack on Netflix system and recommendation algorithms and measures for
evaluating and solving these attacks. These recommendations include co-factorization methods,
collaborative methods, and regularization methods. By taking benefits from these
recommendation methods, Netflix recommender system can improve the quality and
protection from such fake links and injected profile.
7

Paraphrase This Document

Need a fresh take? Get an instant paraphrase of this document with our AI Paraphraser

Article: Estimating user behaviour toward detecting anomalous ratings in rating
systems
In this research paper, we research and identify the recognition method to determine such
attacks and the main objective is to keep tracks on the all attacks and explore the similarity
between the patterns of the fake infected profile. According to Yang (2016), the discovery rate
and untrue alarm rate is used to analyse the effectiveness of profile injecting attacks against
recommender systems. Detection rates mean the division of the number of suspected attackers
by the number of attackers and false alarm rate is a division of number sincere attackers that
assumes an attacker by the number of actual genuine users. These detection methods can be
used to measure the fake injected profiles identified in the Netflix recommender system by
evaluating the performance of proposed methodologies compared to the baselines. Three
methods are used to verify the detection performance which is DeR-TIA, Bp-based method, and
PCA method. It has been seen that the detection presentation of the suggested method gives
the nearly outdoes of these systems. The main objective of performing the detection
performance is to stand-in all attackers and sieve out the honest users. These investigational
results validated the effectiveness of the projected method. Emerging a rapid and operative
detection technique for those significant datasets is a vulnerable matter. In addition, inspecting
a detection method to protecting bouts for model-based concerted clarifying is an engrossed
task in the future work.
8

Article: Recommender Systems— Beyond Matrix Completion
Jannah (2016), stated that around 75% of the download and rentals in Netflix has come from
their recommender systems. In some methods, the methods that produce these
commendations are remarkable. The Netflix Prize with its attention on correctness has certainly
strapped recommender structures research onward. Though, it has also partly outshined many
other significant tasks when constructing a recommender system and today even Netflix states
“there are much superior customs to help society find videos to watch than concentrating only
on those with a high forecast star rating”. From the research viewpoint, context is a
multilayered concept that has been considered in various investigation disciplines. Over the last
10 years’ substantial advancement has been made also in the meadow of context-aware
recommenders and the first proportional assessments and benchmark methods such as DeR-
TIA and HHT-SVM were issued. To relate with the benchmarked methods (DeR-TIA and HHT-
SVM), the planned method achieves the best detection performance particularly for identifying
the fake profile injection in the recommender system of Netflix. In addition, these benchmark
method demonstrates the advanced detection enactment than HHT-SVM in the trend (both
arbitrary and usual) and converse such type of injected fake profiles. Generally, researchers
tend to contain as many structures as possible in an endeavour to increase the
recommendation performances. However, comprising more structures sustains more privacy
concerns because the service provider’s requirement to track the behaviour of the fake profile
to find these structures in reality. The majority of present solutions only target to protecting the
rating paths for the users and do nothing more. Users suggest that present privacy-protection
solutions such as anti-tracking systems can be assimilated to deliver extra privacy security.
9

Article: Android fine-grained permission control system with real-time expert
recommendations
The main objective of studying this research work is to build some processes to suspect the fake
profile injections in the recommender systems and approaches to protect from them.
Cooperative filtering, as one of the most effective approaches to form recommender systems,
also has numerous challenges. This approach may suffer from the problem of detecting such
attacks which means it cannot deliver a precise recommendation facility to new users or items.
This problem can be alleviated by hybrid method based schemes and additional information,
like demography and societal network. Scalability is another problem. An ideal privacy-
preserving solution should assurance privacy without mislaying any efficacy. There are two
approaches used to protect the data such as Data perturbation approach in which the Netflix
system bother their peculiar data and relief them to each other. In this approach, the main
determination is to secure the privacy of individuals logged in the datasets. Other is secure
combined reckoning approach in which commendations are produced through a secure
combined reckoning protocol Rashidi (2016).
10

Paraphrase This Document

Need a fresh take? Get an instant paraphrase of this document with our AI Paraphraser

Methodology
The main aim of this report is to enlarge the methods that are used in establishing the
scheming of this research work. This method quantity will vibrant the approaches that are using
in the research on the effectiveness of fake profile injected in the recommender systems and
also help to fix the tasks or an issue that arises and delivers the devices to solve these
difficulties. The quantitative method is used to comprehensive this study work. Researches
have operated separately to ample the research on the base of the research topic. The
reckonable technique is used because it helped inconsiderate the topic: “Investigating the
effectiveness of profile injecting attack against the recommender system”.
In this research work, I have intensely studied the Netflix’s recommender system and its impact
and the adoptions among user’s behaviour businesses. This research proposal also represents
the approaches of protecting from such attacks and fake profiles that have injected into the
system to increase or decrease the rating list of the recommender system of Netflix. Nearly all
literature have used the measurable procedure which I have used in my research work. I have
considered the research paper that was published in the year 2003-2014 and to know the
present status of the topic, I have considered the papers which are issued between the years
2015-2018.
There are numerous research approaches that are used to resolve the problem arises
throughout the research. These issues are resolved by using and examining the approaches.
Following stages are used in smearing the research technique:
 In the chief step, difficulties are recognized. I have recognized the difficulties in my
research work are to discover the suitable method on expending the recommender
system.
 In the second step, scheming of literature work has been done. I have studied five
literature papers that help in examining my topic.
 The power and weakness have been examined in the third step.
 In this step, assessment has been completed among diverse methodologies of the
literature paper.
11