An Efficient Trust Evaluation using Fact-Finder Technique

AN EFFICIENT TRUST EVALUATION USING FACT-FINDER
TECHNIQUE
K.T.Senthil Kumar1
and Dr.R.Ponnusamy2
1
Part-Time Research Scholar in Computer Science, R&D Department, Bharathiar University, Coimbatore, India.
2
Professor, Department of CSE, Sri Lakshmi Ammal Engineering College, Chennai, India
ABSTRACT
Provide individualized suggestions
of data or products related to users’ needs
by Recommender systems (RSs). Even
if RSs have created substantial progresses
in theory and formula development and
have achieved many business successes, a
way to operate the wide accessible info in
online social Networks (OSNs) has been
mainly overlooked. Noticing such a gap in
the existing research in RSs and taking
into account a user’s choice being greatly
influenced by his/her trustworthy friends
and their opinions; this paper proposes a,
Fact Finder technique that improves the
prevailing recommendation approaches by
exploring a new source of data from
friends’ short posts in microbloggings as
micro-reviews.Degree of friends’
sentiment and level being sure to a user’s
choice are known by victimisation
machine learning strategies as well as
Naive Bayes, Logistic Regression and
Decision Trees. As the verification of the
proposed Fact finder, experiments
victimisation real social data from Twitter
microblogger area unit given and results
show the effectiveness and promising of
the planned approach.
Index Terms — Recommender systems,
machine learning, trust, sentiment analysis,
microblogging, Fact-finder
I. INTRODUCTION
The eventual characteristic of
internet 2.0 is the free generated content by
users. Users have the power to reveal their
Opinions in Online Social Networks OSNs
such as Facebook, Wikis and Twitter. In
fact, an exponential growth of information
has become available to users.
Accordingly, two cases are imposed: one
is the difficulty for users to find contents
that are relevant to their own interest
among vast amount of alternatives. The
other is the demand to a modern technique
which can provide personalised
recommendations by exploiting
information in the current environment of
World Wide Web.
In fact, traditional Recommender
Systems (RSs) play a very important role
in providing recommendations and those
they are deployed within
the business, like Amazon, Netflex and
EBay Typically, there are two main
techniques of RSs: Collaborative Filtering
recommenders (CF) and Content-Based
recommenders (CB) [1], [2]. In CF
techniques, recommenders can predict
relevant items for an active user by
utilising his previous history of ratings
from similar users as neighbours [3] or
similar things [4]. On the opposite hand,
CB recommenders enrich recommendation
by building user-item rofiles supported the
foremost vital options of item contents [5],
[6]. Usually, this technique depends on
things containing matter data, for instance,
content of websites recently, within
the try of rising ancient RSs; some studies
International Journal of Computer Science and Information Security (IJCSIS),
Vol. 15, No. 9, September 2017
10 https://sites.google.com/site/ijcsis/
ISSN 1947-5500

enhance the recommendation by including
trust relations. Trust increases the ability
of RSs to approach more trusted users and
as a consequence more reliable products to
be suggested [7]–[9]. Another
methodology proposed to empower
recommendation by extracting the
sentiment data from long reviews that
users have written concerning merchandise
[10]–[12].
However, there are several
limitations which still appear in the
aforementioned approaches. First of all,
the user-item rating matrix in CF
techniques suffers from the well-known
problem of sparsity as people tend to rate
few items, and this produces less dense of
the available ratings in RSs as pointed out
in [4]. Further, the challenge of the cold-
start problem when new users or items do
not have any rating history. Traditional CF
and CB recommenders assume the
existence of sufficient amount of ratings or
content information in order to generate
powerful recommendation but this is not
true in many cases [1]. Second, most of
trust-enhanced RSs are not practical as
they do not replicate the influence of real
social connections in providing
recommendations. In globe, we have a
tendency to tend to trust our friends’
opinions about books, movies and
restaurants. Third, recommendation
extracted from users’ product reviews
don’t use any benefits which might be
controlled to change recommendations
from friends in OSNs such as Twitter,
since vast amount of information and
opinions are available in such networks.
In this paper, we aim to propose a
solution to the above problems and model
a recommender which can involve a user’s
OSNs to draw the user’s preferences even
in the case that he/she does not have any
rating history. We propose Fact Finder
technique. It is based on the assumption
that users tend to be influenced by their
friends’ opinions even if they have
different interests. We have a tendency to
argue that OSNs, microbloggings in
particular, will be an upscale supply of
data to change recommendation. Our
analysis interest is, for an energetic user, to
harness his/her friends’ sentiments about
products by exploitation the trust degree
between the user and his/her friends in
social network. We explore whether, for a
user, his/her trusted friends’ posts could
also be thought of as short reviews to
empower recommendation.
To the Simplest of our information, this
can be the primary work using sentiment
and trust from microbloggings to generate
personalized recommendations. The
contributions of this paper are:
Employing OSNs as a replacement supply
of knowledge so as to use the short posts
messages as micro-reviews in the
projected Fact-Finder recommendation
framework.
1) Inferring multiple score ratings from
friends’ posts in microbloggings by
victimization sentiment analysis technique,
as these posts are unit short and embrace
informal use of language.
2) Using intercommunication between
friends as the trust indicator to the
importance of friends’ opinion to a user.
3) Improving the prediction performance
using different machine learning
classification algorithms, in particular, for
new users.
The rest of this paper is organised as
follows. In Section II, we provide an
overview of some major studies and
approaches for recommender systems. In
Section III, we explain the environment of
social network data that we utilise. Section
ISSN 1947-5500

IV details the problem we study in this
paper. Our approaches in predicting ratings
are presented in Section V. The results of
the experimental analysis are presented in
Section VI, followed by the conclusion
and future work in Section VII.
II RELATED WORK
In this section we tend to highlight
related important approaches: 1) ancient
collaborative recommender, 2) trust-
increased recommender, and 3) reviews-
based recommender.
First, the standard collaborative
filtering approaches are either memory-
based or model-based. These strategies are
based on the rating history from users
within the memory-based strategies,
similarity computation is a primary part.
They use a heuristic utility of similarity
between users’ vectors like Pearson
Correlation Coefficient (PCC) or
trigonometric function similarity measure
(VCC) [1], [5], [13], [14]. On the opposite
hand, the model-based strategies used
machine learning models to predict
product ratings [15], [16]. For example,
Sarwar et al. [4], [17] implemented
clustering algorithms to identify groups of
customers who rated similar products and
these clusters can be seen as likeminded
neighbours. Since k clusters are created,
recommendation prediction can be
computed by averaging the ratings in that
cluster. Miyahara and Pazzani [5]
proposed a RS based on Naive Bayes
classifier and they only considered items
which co-rated between users. They
manipulated two classes: like and don’t
like and features are selected in a pre-
processing step. Recent proposals focused
in the accuracy of predictions such as
matrix factorization for collaborative
filtering. The approach proposed in [8]
involved social connections data in
providing recommendation by assigning
social regularization terms in order to
constraint matrix factorization objective
function. They assumed that friends rate
products and hence they used PCC and
VCC to measure similarity as intermediate
step.
Second, additional studies
have targeted on trust-enhanced
recommenders. Some studies applied trust
by building trust network supported the
belief that users will get additional
accurate recommendation from folks they
trust [18], [19]. These styles
of strategies used direct evaluations of
trust from users. Golbeck et al. [18]
propagated trust from trust network so-
referred to as net Of Trust
WOT. Solely friends whose
trust analysis exceeds a threshold are going
to be concerned in recommender encounter
Recommendations square
measure obtained by weighted average of
ratings alongside the
trust price victimisation
Film Trust dataset. In another context,
Massa et al. [19] used trust to filter the set
of neighbours and solely their things
would be thought of in predicting ratings
to an vigorous user. once filtering
neighbours, they applied the standard
recommendation algorithm. The
experiments were supported Opinion
dataset that contains each user’s ratings
and also the trust values from users
towards one another.
Third, recent researches are done to
exploit the sentiment in the textual reviews
to augment ratings in collaborative
recommenders [11], [12], [20]. Authors in
[11] tried to enhance the RSs
by leverage topic and sentiment data at
ISSN 1947-5500

sentences level. They inferred ratings from
textreviews written by
users regarding restaurants in multi-point
rating scale instead of solely positive or
negative polarities. They applied text
regression model to estimate scaled
sentiment ratings. They’re the
primary WHO integrated
the helpful data in reviews into RSs.
Lenug et al. [12] planned a probabilistic
sentiment logical thinking framework.
They applied
Natural language techniques to compute
sentiment orientation in reviews. They
designed their rating inference model
supported on the Naive mathematician
classifier. Then, they integrated between
the logical thinking ratings from reviews
and a CF algorithmic program to
extend users’ preferences and achieved
encouraging results. Esparza et al. in [21]
investigated how to obtain
recommendation from online
microblogging services. They proposed a
solution to exploit short posts written by
users as product reviews. These posts are
used to build user-item profile. Then a
query search algorithm is applied to
retrieve relevant item profiles based on a
twitter-like review service called
blipper.com. This study is similar to our
work in using microblogging as a source
of recommendation.
Some inherent drawbacks still have not
been solved in the above mentioned
methods. Most of these approaches require
users to produce some structured data first
such as trust evaluations and ratings to
allow the corresponding systems to work
properly. In fact, this is not practical and
usually not available. Nevertheless, the
weaknesses of sparsity and cold start
problems appear in the case of trust
network as it is in the user-item rating
matrix. On the other hand, review based
recommenders require a user to write
reviews and rate products to generate the
suggestions. Unlike existing studies, our
novel approach fact-finder overcomes the
need of ratings or written reviews by users
and reflects the real hidden social trust
relations. In our work we personalise
recommendations from microbloggings
using sentiment analysis and trust between
friends.
III MICROBLOGGING SERVICES
In this section we have a tendency to
introduce our target social network
Twitter. Users with in the microblogger
Twitter will publish short posts in 140
character limit questionable tweets. Today,
Twitter users will generate over 300
Million tweets each day [22] about
different topic and interest. For example,
people can generate brief posts about their
personal experience in reading books,
watching movies, breaking news or even
the release of new electronic gadgets.
Additionally, users have the selection to
determine relationships among every other
for social links, seeking info or
distinguishing following/followers friends.
Fig.1Intercommunication with User’s
and friends
ISSN 1947-5500

Measuring the various levels of hidden and
subjective trust relationships between
friends in Twitter is crucial in our analysis.
Therefore, we developed a tool to
mechanically collect social network
information by using Twitter API. This
tool extracts the specified interactions
between friends, and it is known as Twitter
Interaction Extractor (TIE), additional
details in Section VI. It is important to
high light that we access only accounts
that available to public and not any
protected accounts for private security.
however, some regulations that Twitter
service applies increase the challenge of
getting such social network information,
for instance, the rate limit of accessing and
requesting information from Twitter
service and the dynamic change in the
relations and contents. Moreover, some
users choose to apply more privacy
constraints on their accounts in order to
avoid their info to be revealed to general
public.
IV METHODOLOGY
Fact-Finder Cluster Technique
We believe that the social relations
showing among users and friends in OSNs
influence users’ buying behaviour. As the
quantity of data and opinions regarding
products and services increase and diverse
additional and additional, harnessing
friends’ opinions and incorporating trust
relations between them to enhance
recommendation are getting a vital need.
Fig. 1 is an example for an online active
user who has not experienced lots of items
or a completely new user in a retailer
website. During this state of affairs,
drawing the user’s taste and preferences
isn’t offered and most of the existing RSs
algorithms cannot give the individualized
suggestions.
What usually folks do once they measure
lack of information is to raise their friends
as they trust them. People also tend to
point out interest and curiosity in things
(movies, books, restaurants ...etc) that their
friends like. Being influenced by friends’
style could be a common feature within the
real world. For the example in Fig. 1, the
active user has five friends who
broadcasted different messages, for
example, Let us assume the domain of
movie recommenders, then the
investigation of his friends’ circles shows
that Friend1, Friend2 and Friend3 have
some knowledge about movies. The
challenge is that the active user definitely
has different relationships with his peers.
We need to analyse these social ties and
understand that friend that the active user
could trust his opinion the foremost. To do
so, it’s vital to require in to consideration
the communication behaviours between
the active user and his friends as a trust
indicator. Samples of these interactions are
the action of resending messages from his
friends (RE) and saving his friends’ posts
in favourite list (FV). It is also highly
needed to search out the extent of
sentiment orientation in friends’ posts
about regarding movies and contemplate
the informal use of languages and icons.
Next section explains how to compute the
implicit social trust, and then sentimental
analysis of the Post (or) comments is
introduced in Section IV-B.
ISSN 1947-5500

A. Implicit Social Trust
In this section, we will observe some
necessary communication activities
between peers to indicate how people trust
one another in OSNs. We tend to believe
that the interactions between friends in
OSNs will indicate what proportion trust
they’ll hold towards completely different
friends. Some actions show however
friends perceive each other such as re-
tweeting, mentioning others, favouring
others’ posts and number of followers
[23]. Communication activities in Twitter
that we consider in this paper are defined
as: the action of re-tweeting which implies
that a user re-sends a tweet to all his/her
friends to show the interest, and will be
denoted as RT. We would like to compute
the trust relation between user u and one
friend f among the cluster of friends F
since f ∈ F.
Intuitionally, trust is identified as a
normalised average
Trustu, f =
,

where we denote trust between user u and
friend f as u, f trust . And u, f RT is the
number of re-tweeted messages done by u
to friend f in a given period of time which
is the total re-tweeted messages done by u
to all friends in Cluster F denoted as u, f
RT in that period of time.
Fig2. Proposed System
Due to the fact that people interactions
vary over the time and relations are not
static, we define the periods of times as
T = {t1, t2, ...tW } and then the same
computation of trust in equation is applied
for each time period tj ∈T ,
Trustu, f (t j) =
,

Based on the above equation, we can
detect the trust between u and f over all
periods of time T as,
TRUSTu, f (T) = ∑ (trust) ( tj)
B. Sentimental Analysis using Natural
language processing
How will we have a tendency to
learn the probabilities P(c) and P (fi |c)?
Let’s first take into account the utmost
change estimate. We’ll simply use the
ISSN 1947-5500

frequencies within the knowledge. For the
document previous P(c) we ask what
percentage of the documents in our
training set are in each class c. Let Nc be
the number of documents in our training
data with class c and Ndoc be the total
number of documents.
(en.wikipedia.org/wiki/Emoticon) such as
the emoticon” :)” and
abbreviation”OMG”. Positive and negative
intensifiers are also counted such as
“cooool”. Binary vector of options is
enforced manually for each micro-review
and any existence of the sw will be given
worth one even positive or negative,
otherwise it will take the zero value.
In contrast to the binary sentiment ways
that offer solely the polarity of reviews as
negative or positive, we want to get
additional precise sentiment analysis that
describes over easy like or dislike ways..
For a given set of micro-reviews every is
represented by a set of sentiment words
We want to infer sentiment rating sˆr to
hold a class of ratings for example, since
our goal is to allocate a rating to describe
the strength of an opinion in micro-
reviews. Now we will infer sentiment
ratings by aggregating all the existence of
positive sw’s normalised by the full
variety of existence mentioned options
previously either positive or negative,
similar methodology is used in [11] but
they worked solely at sentence level while
not together with the special language
option. The subsequent equation illustrates
however we have a tendency
to cipher the reasoning sentiment rating sr
from mr:
Sr f,I =
( , )
( , ) ( , )
Number of the class categories used in
recommendation, for example, some
systems based on five or ten score rating
scale. This can be further explained by,
Positive and negative words are tokenism
and future improving their levels.
V. ITEMS INFERENCE RATINGS
The crucial step in RSs is to generate
prediction of ratings. We develop two
different techniques one based on heuristic
foundation to predict ratings in Section V
(A) and the second is to apply machine
learning models in Section V (B).
A. Heuristic Prediction
The group of friends who have
opinion of the items then among the all
friends
1) It is important to say that trust between
one user u and his
friend f may be not bidirectional, hence:
if u → f and f → u then trust u→f ≠ trust
f→u
2) If (trustu,f1 ) > (trustu,f2 ) then, srf1
contributes more to Ru, i than srf2 .
3) MI Nf ∈ TF (srf,i ) ≤ Ru,i ≤ M AXf ∈
TF (srf,i )
B. Models-Based Prediction
Used to validate all of the information’s
of all users and friends can contribute to
products ratings estimation by using
classification learning algorithms.
Naive Bayes classification, Logistic
regression and Decision tree. The
importance of these algorithms comes
from yielding good result in different
domain and the availability of the related
software tools [24]. Applying these
machine learning methods on our social
data SD = {trust u, f , srf,i }
needsto identifythe relations between
features terms by IR which indicates the
friend’s opinion (sr)on user-rating about
IR( u , f , i ) = g (trustu , f , srf , i )
ISSN 1947-5500

Where g is an unknown function we need
to define. And then we aggregate the
impact function I R to obtain the final user
ratings as:
The three well-known machine learning
models are used to represent or
approximate the unknown function g as the
output will be one of the class ratings and
feature vector we examined the
classification models on the social data
dataset SD to predict one of five nominal
classes of ratings Extreme Like, Like,
Neutral, Dislike, Extreme Dislike. In fact,
the core engine behind these three
classification algorithms is quite different.
Next paragraphs describe brief details
about each.
P c X P c X for i, j ∈ {1, .. m}
There are some premises to apply NB. It is
simple technique to use and tends to be
optimal for particular domain classes with
highly independent and irrelevant features.
Moreover, the probabilistic nature of NB
allows it to handle missing values [25].
We describe two more algorithms might
achieve better results.
A decision tree model is also applied.
It is a nested set of rules that used to split
the data. This recursive algorithm
constructs a tree structure automatically
starting from root features and ending with
leaf nodes. When splitting the data a
decision rule is applied for every feature
then the feature that minimises the cost
function is chosen to build tree branches.
The leaf node at the end of each branch is
a class. There are many decision tree
algorithms in the literature, in this work
we adopt C4.5 algorithms. The metric that
used to measure the best splitting of data
by C4.5 algorithm is called the information
gain IG measurement derived from the
dataset itself to split the tree branches. Let
pi is the probability that a subset of SD
labeled by ci , then:
I ( p ) = p i log2 pi
VI. EXPERIMENTAL ANALYSIS
A. Dataset
The dataset that we like should contain
friends revealed posts (micro-reviews) from
OSNs and trust relations besides the classic
users and items ratings info. In fact, there is
no adequate dataset with the predefined
requirements since the available datasets
contain only ratings, only reviews or ratings
with reviews. Given the absence of friends
information, therefore, we tend to designed
a package tool TIE to scrawl Twitter and
prepare the friends information using the
Twitter API for JAVA (contact the authors
for the collected dataset).
We could collect social data about 111
users’ information as follows: firstly, we
randomly choose movies form the popular
movie lens dataset after that we used the
search tool TIE to gather information about
a person who posted a tweet about the
chosen movie such as name and twitter ID.
Secondly, since we were able to allocate the
publisher information then we start
to detect the re-tweeting messages activities
rate between this person and his friends. On
the other hand, we used three
B. Metrics
To compare the performance between
the three applied machine learning
classification algorithms we applied
different metrics. We used Accuracy metric
to indicate the percentage of the correctly
classified instances in the test set. However,
this metric is not enough because it is not
sensitive to class distribution or the chance
of being correct. Hence, we also used the
standards evaluation measurements that are
widely used in information retrieval and
classification such as Precision and Recall
defined in equations (17), (18)
ISSN 1947-5500

respectively. These two metrics test the
accuracy of classification algorithm in
predicting ratings [25].
Precision =

Recall =

F-measure is considered as the harmonic
mean between the two metrics Precision and
recall to overcome any conflict between
them. It is given as follow:
F- Measure = 2.
.
For testing, we randomly split the dataset
into ten non-overlapped folds to apply 10-
fold cross-validation. The experiments are
repeated on the ten folds. Every fold is
used as test set and the rest nine folds used
as training set. Our results are computed
based on the average of all the ten folds
Models.
We also applied the statistical accuracy
metrics such as Mean Absolute Error
(MAE) to evaluate the recommendation
algorithm. It is the most widely used and
acceptable in the recommendation
community because it is easy to apply and
we can interpret comparisons directly.
MAE is defined,
where N is the size of the test set, and ru,i
is the rating assigned by user u to movie i,
and rˆu,i indicates the rating estimated by
the proposed recommendation algorithm.
Obtaining small results of MAE shows
more accurate performance of the system.
Neutral or Extreme Like which are
semantically relevant rather than going
further distance classes such as Dislike or
ExtermeDislike.
Fig. 3. Precision, recall and F-measure
for each ratings class produced by
decision trees.
c. Evaluation Result
Experiment is applied to evaluate the
recommendation classification algorithms
accuracy given the computed trust and
sentiment values derived from the
collected data described in Section VI-A.
F-measure of NB, logistic regression and
decision tree were 0.57, 0.67 and 0.72
respectively. This means that decision tree
has power of prediction higher than the
two others algorithms. It is clear that NB
classifier has the worst accuracy
percentage while the best performance is
given by decision tree it can correctly
predict 72% of test ratings. More
specifically, Fig. 3 sheds the light on these
metrics results using decision trees
according to each class category. There is
more determination about negative classes
Dislike and ExtermeDislike than positive
classes. Meanwhile the Neutral class gains
the lowest accuracy and this may due to
the ambiguous nature with this class since
it holds uncertainty about opinion. It is
difficult to know whether Neutral class is
closer to which polarities (negative or
positive) unless further contextual
information is included such as
0
1
2
3
4
5
6
7
8
Accuracy
Rating Classes
Precision
Recall
F-measure
ISSN 1947-5500

demographic information or more trust
indicators are used.
Algorith
ms
Accuracy Precision Recall F-measure
Naïve
Bayes
57.65% 0.58 0.58 0.57
Logistic-
Regressi
on
67.6% 0.66 0.68 0.67
Fact-
finder
72.1% 0.73 0.72 0.72
d)Recommendation Results
We will evaluate the performance of our
approach having group of friends. In this
experiment, we assume the existence of
totally 12 new users. Each user is assigned
a group of friends of size 10 sampled from
the dataset SD. Moreover, we applied a
threshold of the lowest acceptable trust
value. Only friends who exceeding this
threshold will be involved in
recommendation process. We chose the
value of the threshold to be 0.17 as this
value of trust is the information gain in
decision tree model. This threshold will be
applied for all baselines. After selecting
the trusted friends, this local community
will be treated as user’s neighbourhood.
Table III reports different values based on
the average of MAE from friends groups
of the 12 users. B4 achieved the highest
error among all the methods. This is
because it is based on the traditional
collaborative filtering method which
requires both common ratings between
friends or items, and user average ratings.
Thus, traditional collaborative filtering
lead to poor recommendation in new user
situations. Due to the large error obtained
by B4 we excluded it in the next
experiment. It can be observed that both of
our methods I ST S1 and I ST S2 gained
the smallest error results 1.2 and 0.836
respectively.
Algorithms MAE
B1 1.292
B2 1.735
B3 1.278
B4 3.236
ISTS₁ 1.20
ISTS₂ 0.836
VII CONCLUSION
In this paper, we have tendency to
explore the potential of social info derived
from microbloggings as a supply of user
relevant recommendations. In distinction
to ancient RSs which are based mainly on
structured data, we investigate the data
comes from the current web environment.
We propose the approach Fact finder that
can exploit two factors from OSNs: the
sentiment orientation in friends posts
regarding sure things and therefore the
trust relations between friends. Our
evaluations are applied on real social
knowledge from Twitter. The results show
that the these short and inconsistent posts
can empower the users preferences data in
particular when no preferences of history
were available. Several machine learning
classification algorithms were accustomed
classify a score rating, and the tree
decision model performs the best accuracy
metrics results. In the future challenge, we
have tendency to believe that user’s taste is
absolutely very important in
personalisation. Therefore we have a
tendency to attempt to fuse user’s own
preferences - once exists - with social info
comes from OSNs to counterpoint and
augment collaborative filtering
recommenders.
ISSN 1947-5500

REFERENCES
[1] G. Adomavicius and A. Tuzhilin,
“Toward the next generation of
Recommender systems: A survey of the
state-of-the-art and possible extensions,”
IEEE Transactions on Knowledge and
Data Engineering, vol. 17, no. 6, pp. 734–
749, 2005.
[2] F. Ricci, L. Rokach, and B. Shapira,
“Introduction to recommender systems
handbook,” Recommender Systems
Handbook, pp. 1–35, 2011.
[3] P. Resnick, N. Iacovou, M. Suchak, P.
Bergstrom, and J. Riedl, “Grouplens: An
open architecture for collaborative filtering
of netnews,” in Proc. the 1994 ACM
Conference on Computer Supported
Cooperative Work, ACM, 1994, pp. 175–
186.
[4] B. Sarwar et al., “Item-based
collaborative filtering recommendation
algorithms,” in Proc. the 10th
International Conference on World
Wide Web, ACM, 2001, pp. 285–295.
[5] M. Pazzani and D. Billsus, “Learning
and revising user profiles: The
identification of interesting web sites,”
Machine Learning, vol. 27, no. 3, pp. 313–
331, 1997.
[6] L. Si and R. Jin, “Flexible mixture
model for collaborative filtering,” in
Proc. Machine Learning-International
Workshop then Conference, vol. 20, no. 2,
2003, p. 704.
[7] H. Ma, I. King, and M. R. Lyu,
“Learning to recommend with social trust
ensemble,” in Proc. the 32nd International
ACM SIGIR Conference on Research and
Development in Information Retrieval,
ACM, 2009, pp. 203–210.
[8] H. Ma, D. Zhou, C. Liu, M. R. Lyu,
and I. King, “Recommender systems with
social regularization,” in Proc. the Fourth
ACM International Conference on Web
Search and Data Mining, ACM,
2011, pp. 287–296.
[9] Y. Kim and R. Phalak, “A trust
prediction framework in rating-based
experience sharing social networks without
a web of trust,” Information Sciences,
2012.
[10] B. Pang, L. Lee, and S.
Vaithyanathan, “Thumbs up?: Sentiment
classification using machine learning
techniques,” in Proc. the
ACL-02 Conference on Empirical Methods
in Natural Language
Processing, vol. 10, 2002, pp. 79–86.
[11] G. Ganu, N. Elhadad, and A. Marian,
“Beyond the stars: Improving
rating predictions using review text
content,” presented at 12th
International Workshop on the Web and
Databases, 2009.
[12] C. W.-K. Leung, S. C.-F. Chan, F.-L.
Chung, and G. Ngai, “A
probabilistic rating inference framework
for mining user preferences
from reviews,” World Wide Web, vol. 14,
no. 2, pp. 187–215, 2011
[13] J. A. Konstan, B. N. Miller, D. Maltz,
J. L. Herlocker, L. R. Gordon,
and J. Riedl, “Grouplens: applying
collaborative filtering to usenet
news,” Communications of the ACM, vol.
40, no. 3, pp. 77–87, 1997.
[14] X. Su and T. M. Khoshgoftaar, “A
survey of collaborative filtering
techniques,” Advances in Artificial
Intelligence, p. 4, 2009.
[15] K. Miyahara and M. J. Pazzani,
“Improvement of collaborative
filtering with the simple bayesian
classifier,” IPSJ Journal, vol. 43, no.
11, pp. 3429–3437, 2002.
[16] X. Su and T. M. Khoshgoftaar,
“Collaborative filtering for multi-class
ISSN 1947-5500

data using belief nets algorithms,” in Proc.
International Conference on Tools with
Artificial Intelligence, IEEE, 2006, pp.
497–504.
[17] B. M. Sarwar, G. Karypis, J. Konstan,
and J. Riedl, “Recommender systems for
large-scale e-commerce: Scalable
neighborhood formation using clustering,”
in Proc. the Fifth International Conference
on Computer and Information Technology,
vol. 1, 2002.
[18] J. A. Golbeck, “Computing and
applying trust in web-based social
networks,” 2005.
[19] P. Massa and P. Avesani, “Trust-
aware recommender systems,” in Proc. the
2007 ACM Conference on Recommender
Systems, ACM, 2007, pp. 17–24.
[20] F. Peleja, P. Dias, and J. Magalha˜es,
“A regularized recommendation algorithm
with probabilistic sentiment-ratings,” in
Proc. 12th
International Conference on
Data Mining Workshops (ICDMW),
IEEE, 2012, pp. 701–708.
[21] S. G. Esparza, M. P. O’Mahony, and
B. Smyth, “Mining the real time web: A
novel approach to product
recommendation,” Knowledge-Based
Systems, vol. 29, pp. 3–11, 2012.
[22] S. M. Kywe, E.-P. Lim, and F. Zhu,
“A survey of recommender systems
in twitter,” Social Informatics, Springer,
2012, pp. 420–433. [23] B. Hajian and T.
White, “Modelling influence in a social
network: Metrics and evaluation,” in Proc.
Third International Conference on
Privacy, Security, Risk and Trust (Passat),
IEEE, 2011, pp. 497–500.
[24] I. H. Witten and E. Frank, Data
Mining: Practical Machine Learning
Tools and Techniques, Morgan Kaufmann,
2005.
[25] X. Amatriain, A. Jaimes, N. Oliver,
and J. M. Pujol, “Data mining
methods for recommender systems,”
Recommender Systems Handbook,
Springer, 2011, pp. 39–71.
[26] S. L. Cessie and J. V. Houwelingen,
“Ridge estimators in logistic regression,”
Applied Statistics, pp. 191–201, 1992.
[27] L. Rokach and O. Maimon, “Top-
down induction of decision trees
classifiers-a survey,” IEEE Transactions
on Systems, Man, and Cybernetics, Part C:
Applications and Reviews, vol. 35, no. 4,
pp. 476–487, 2005.
[28] P. Victor, M. D. Cock, and C.
Cornelis, “Trust and recommendations,”
Recommender Systems Handbook,
Springer, 2011, pp. 645–675.
ISSN 1947-5500

An Efficient Trust Evaluation using Fact-Finder Technique

More Related Content

What's hot (18)

Similar to An Efficient Trust Evaluation using Fact-Finder Technique (20)

Recently uploaded (20)

An Efficient Trust Evaluation using Fact-Finder Technique