IRJET- Cross-Domain Sentiment Encoding through Stochastic Word Embedding

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 09 | Sep 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 1393
CROSS-DOMAIN SENTIMENT ENCODING THROUGH STOCHASTIC WORD
EMBEDDING
Swapna Mohan. M1, J. Umarani2
1Swapna Mohan. M, Research Scholar, Thanthai Hans Reover College (Autonomous), Elambalur, Perambalur
2J.Umarani, M.Sc, M. Phil, B.Ed, (Ph.D)., Assistant Professor, Dept. of Computer Applications, Thanthai Hans Reover
College (Autonomous), Elambalur, Perambalur
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Sentiment analysis is an importanttopicregarding
identification of feelings, attitudes, emotions and opinions
from text. A critical challenge for computerizing such analysis
is the high manual annotation cost when conducting large-
scale learning. However, the cross-domain methodology is a
key solution for this. It operates commented reviews across
domains and its success principally depend on the learning of
a robust common characterization space across domains to
support the sentiment classifier transfer. Over recent years,
consequential effort has been pervaded to improve the cross-
domain representation learning by designing increasingly
more complex and elaborate model inputs and architectures.
We support that it is not necessary to focus on design twist as
this inevitably consumes more time for model training.
Instead, we propose to traverse throughasimpledepicting the
word contradiction and occurrence information and encode
such information more accurately whilst aiming at lower
machanical costs. The proposed methodology is unique and
takes advantage of the stochastic embedding technique to
tackle cross-domain sentiment alignment. Its effectiveness is
benchmarked with over ten data tasks constructed from two
review corpora, and is compared against ten classical and
state-of-the-art methods.
Key Words: Security, Information, Outsourcing, Encryption,
Protection etc…
1. INTRODUCTION
Sentiment classification plays a significant role in many
applications related to opinion mining and sentiment
analysis, such as opinion extraction and summarization,
review spam identification, user feeling analysis, contextual
advertising, etc. The goal of sentiment classification is to
automatically identify the sentiment polarity of a given text
object, for instance, in terms of being positive, negative or
neutral. Typical examples of such text objects include
product reviews, which are generated by movie viewers,
hotel customers, merchandise buyers, etc. The emotional
tendency modeled through identifying sentiment polarityof
the reviews can serve as a succinct yet informative indicator
of the consumer attitude and opinion. This can potentially
result in not only improved efficiency in the information
sharing between the users, but also improved business
solutions and services.
Focusing on reviews of a target product, a standard
sentiment classifier can be built by training with a set of
annotated example reviews of thisproduct. Here,annotation
refers to the process of assigning each review example a
ground-truth sentiment polarity label. The sentiment
polarities of new reviews for the same product can then be
predicted by this trained classifier, [8]. Performance of such
a system heavily relies on the availability and quality of the
labelled example reviews. However, the processofmanually
annotating explosively growing online product reviews is
very expensive and can be impractical. Therefore, there has
been increasing interest on studying effective ways of
reusing labeled reviews across different products. This is
known as cross-domain sentiment classification, where a
domain is referred to as a collection of reviews for a
particular product.
2. EXISTING SYSTEM
Comparing the domain adaption strategies used by various
state-of-the-art techniques, we can see that, in addition to
the main task of sentiment classification, they usually
enhance their learning through preparing extra tasks like
detecting whether a pivot co-occurs with a domain-specific
word, whether the different versions of the samepivot word
in different domains possess similar enough representation
vectors, whether a review contains a pivot, or whether the
reviews from the source and target domains can be
distinguished in the representation space, and so on; we
refer to these as the auxiliary learning tasks. The learning
algorithms are mostly built on spectral approaches which
explore and preserve inherentdata structurethroughmatrix
decompositions or neural networks which directly learnthe
review representations throughstructuredprocessingofthe
content words based on different network architecturesand
exhaustive training.
We argue that instead of creating many auxiliary
learning tasks and constructing complex models with
elaborate design and input configurations, satisfactory
domain adaptation can be achieved by preserving simple
polarity and occurrence information of words in reviews.
These are actually parts of the classical information utilized
in early cross-domain sentiment classification works, that
however failed to achieve good performance. We support
that the unsatisfactory performance of past endeavors were
potentially caused by the employedspectral approachesthat

were incapable of preserving accurately the desired
information in their embedding spaces.Similarobservations
on poor neighbor preservation ability of spectral
embeddings are also reported in the data visualization field.
To tackle this issue, we propose a novel cross-domain
sentiment representation learning model with its design
inspired by the stochastic neighbor embedding method. It is
built upon a simple mapping architecture to ease the
computational cost, but we propose a more sophisticated
approach for optimizing the mapping variables to achieve
more accurate similarity structure preservation.
2.1 Survey
A. EMOTION RECOGNITION AND BRAIN MAPPING FOR
SENTIMENT ANALYSIS: A REVIEW
The meteoric growth of the Internet has caused the increase
in the amount of textual information available, such as in
blogs, discussion forums and review sites on the web, where
the texts surely have the emotion content. Emotion is one
appearence of people behaviour and it is an important
performance in human computer interaction (HCI). Human
convey the emotion in the form of facial expression, speech
and writing text. Recently, researchers in computational
linguistic (CL) areas are fascinated in the attention of
emotion for Sentiment Analysis (SA). SA naturally perceives
the emotion conveyed by a text, and at the same time,
distinguishing positive and negative valence.Thewideareas
of CL research, literally appreciable for scrutinizing the
emotion dimension detection and searching the approaches
and techniques in the term of emotion recognition (ER).
There are two significant drifts of research in the area, the
emotion recognition based on state affective computing and
the real time using brain signal machines. The two areas
have the same goal for getting the upgraded result in
sentiment analysis with the mapping of emotionrecognition
provided. The revelaive work on emotion detection is
comparatively rare and lacks empirical evaluation research.
This paper demonstrates the overview of past and recent
research on emotion recognitionaswell assomeapproaches
and techniques used and shows the linked between both SA
and ER.
B. FEATURE BASED OPINION SUMMARIZATION OF
ONLINE PRODUCT REVIEWS
With the strikable development of Web 2.0, an amazing
growth of the social-media and e-commerce is being
witnessed. This expansion of e-commerce has been
characterized by the availability of vast number of products
online. The websites selling these products allow its
customers to express their views freely about a purchased
product and these reviewsmayreachuptohundreds.Hence,
it becomes very difficult for the customers to read each and
every review and keep track of all the pros and cons of a
particular product with respect to all the features that it
possesses. The present study targets this problem.
Subsequently, a feature based opinion summary of the
product reviews is proposed as a solution. The proposed
algorithm based on frequent item-set mining gives a set of
candidate features for a product. Numerous feature filtering
methods like subset filtering, superset filtering and distance
based filtering are utilized to get rid of the redundant and
very basic features. Finally, sentiment lexicon is used to
determine the popularity of the opinion word associated
with the extracted feature(s) and a final summary is
developed.
Achieve our design goals of both system security and
usability, we divide each CG into three index vectors based
on the structure of the CG, the type of concept and the value
of concept. We applyorderpreservingsymmetricencryption
(OPSE) to our scheme to enhance security. Experimental
results exhibit the efficiency of our proposed scheme.
C.TASC: TOPIC-ADAPTIVE SENTIMENT CLASSIFICATION
ON DYNAMIC TWEETS
Sentiment classification is a topic-delicative task, i.e., a
classifier trained from one topic will perform worse on
another. This is particularly a problem for the tweets
sentiment analysis. Since the topics in Twitter are very
manifold, it is impossible to train a universal classifier for all
topics. Moreover, compared to product scrutiny, Twitter
lacks data labeling and a rating mechanism to acquire
sentiment labels. The extremely sparse text of tweets also
attain the performance of a sentiment classifier. In this
paper, we propose a semi-supervised topic-adaptive
sentiment classification (TASC) model, which starts with a
classifier built on common features and mixed labeled data
from various topics. It depreciate the hinge loss to adapt to
unlabeled data and features including topic-related
sentiment words, authors' sentiments and sentiment
connections derived from“@” mentions of tweets, named as
topic-adaptive features. Text and non-text features are
extracted and naturally split into two views for co-training.
The TASClearningalgorithm updatestopic-adaptivefeatures
based on the collaborative selectionofunlabeleddata,which
in turn helps to select more reliable tweets to boost the
performance. We also design the adapting model along a
timeline (TASC-t) for dynamic tweets. A demonstration on 6
topics from published tweet corpuses demonstrates that
TASC outperforms other well-known supervised and
ensemble classifiers. It also beats those semi-supervised
learning methods without feature adaption. Meanwhile,
TASC-t can also achieve impressive accuracy and F-score.
Finally, with timeline envision of “river” graph, people can
intuitively grasp the ups and downs of sentiments'
evolvement, and the intensity by color gradation.
D. AN EFFECTIVE HYBRID CUCKOO SEARCH WITH
HARMONY SEARCH FOR REVIEW SPAM DETECTION
In the recent years, online comments are one of most
important source of customer opinion. Nowadaysconsumer
can attain knowledge about the products and service from
online review resources, using which they can make

decisions. This may predominate Opinion Spam, where
spammers may manipulate and fake reviews to promote
artificially or devalue the products and other services.
Opinion spam detection is done by extricating meaningful
features from the text, and identifying the spam reviews
using machine learning techniques. This demonstration
results in a very high dimensional feature space. These
features are inapplicable, redundant, and noisy which may
affect the performance of the classifier. Therefore, a good
accent selection method is required in order to speed up the
processing rate, predictive accuracy. Evolutionary
algorithms for accent selection can be used to handle these
high-dimensional feature spaces which eliminate the noisy
and irrelevant features. In this work, an constructive hybrid
feature selectionmethodusingCuckooSearchwithHarmony
search is proposed and Naive Bayes is used for classifying
the review into spam and ham.
E. POLARITY SHIFT DETECTION APPROACHES IN
SENTIMENT ANALYSIS: A SURVEY
Sentiment analysis is a process of recognizing and
categorizing opinions expressed in a piece of text. It
categorizes the text into positive, negative or neutral.
Lexicon-based and Supervised Machine Learning-based are
the two main techniquesinsentimentanalysis.Bag-of-words
model is used to represent the text as a course of
independent words and machine learning algorithms are
used for classification.
3. PROPOSED SCHEME
In this paper, we solve the problem of how to enable a
searchable encryption system with the support of semantic
extension. Our work is one of only a few to study ranked
search over encrypted data represented by CGs in Cloud
Computing. We choose CG among various modes of
knowledge representation as our semantic representation
tool. In our scheme, we apply a state-of-the-arttechniquei.e.,
text summarization and Tregex, a tool for simplifying
sentences to summarize the document. We introduce the
existing scheme of constructing CGs to helpgenerateCGs. To
attain our design goals for both system security and
usability, we divide each CG into three index vectors in
response to the structure, the type of concept and the value
of the concept of CG. Ranked search greatlyimprovessystem
usability by returning the matching files in ranked order
with regard to certain relevance criteria. Thus, to stipulate
how the document convince the query and ranks the
returned file, we use the – ”text summarization score”(TSS),
which can gauge the extent to which the documents match
their summarizations according to the relevance score.
Moreover, if the score is higher, the document is more
coincident with the original document. To protect the
privacy of the TSS, we then integrate order preserving
symmetric encryption. At the same time by using OPM (One
to Many Preserving Mapping) technique more number of
search results are stored in an graphical representation
which reduces the storage space. The top results are ranked
which is a benefit for the user i.e.: the search results are
ranked according to the accuracy level and basedonthattop
k ranked results, the search outputs are provided to the
users in an efficient manner.
Advantages
•Increases more efficient search results.
•The search results increases the efficiency of the results.
•As the search results uses top k rank search results it is
more prior.
•Reduces the cloud storage as it stores the results in a
graphical representation and provides the searchable
content to the user in a more specific manner.
•Exact search results are maintained in a distributed
database which enhances the user preferences in an
enormous way to attain the search contents.
Fig -2: Architecture Diagram
Module Description
Cloud Owner Login
At the initial the cloud owner has to provide their login
details to login to their preferred cloud login account. After
login process the admin can concentrate on the user
activities such as searching process and so on. So that the
entire process has been under the control of the admin such
as user accounts, search results and so on.
User Registration
To access the account details the user have to register their
preferred details to gain the account details from the cloud
owner. After registration each user has been provided with
unique ID and password to access their account info.So with

the help of the registered queries the admin gathers the
basic details of the user.
User Login
After registration the user can login to their preferred
account to access or search the preferred details. Thedetails
of each user and their data has been stored with privacy and
security as they have separate username and login to access
the data and as well as for search.
Encrypted Data
The data and the info stored by the user has been in an
encrypted format for privacy preference. The encrypted has
been stored and can be accessed by the separate user by
their desired access info. The cloud database stores the user
data in an encrypted format with the help of the AES
algorithm.
Search Request
The user used to search their desired search request by
using the proposed techniques and the results for the
searched data has been provided to the user with the higher
efficiency. The results for the searched data are topk ranked
results that has been provided by the cloud server. The
search process of the USSCG scheme is an algorithm named
CGM (conceptual graphs match).
Access Control
The scheme is designed to prevent the cloud server from
learning additional information regarding the document
collection and query. The scheme is designed to prevent the
cloud server from learning additional informationregarding
the document collection and query. There are still multiple
index vectors in the query that cannot be distinguished well,
we will rank the result according to the encrypted relevance
scores by the modified OPSE We denote OPM as our one-to-
many order-preserving mapping function with parameter:
Search Control
Text Summarization (TS) always tries to determine the
”meaning” of documents. Essentially, TS techniques are
categorized as Extractive and Abstractive. In this paper, we
focus on extractive summarization. Extractive summaries
(ES) produce some of the most significant sentences
extracted from the document instead of rebuilding. To solve
the problem of semantic search based on conceptual graphs
over encrypted outsourced data effectively in the above
system model, our system has the following design goals.
Ranked search: The proposed scheme is designed to
provide not only semantic query based on CG but also
accurate result ranking.
Search Efficiency: The scheme aims to achieve low
communication and computation overhead.
Algorithm Used
Algorithm 1 CGM
Input: F, D1, D2, D3, Q1, Q2, Q3, TSS
Output: F(Q1)
for each document Fi in F containing D1, D2, D3 do
if RScore(D1;Q1) > Score then
Insert D2, D3 into F(Q1);
RScore(D2;Q2);
RScore(D3;Q3);
Insert a new element (RScore(D2;Q2),
RScore(D3;Q3), FID) into RList;
else
return;
end if
end for
for each Fj , Fk in F(Q1) do
if RScore(Dj2;Qj2) == RScore(Dj2;Qk2) then
if RScore(Dj3;Qj3) == RScore(Dj3;Qk3) then
Compare the TSS of Fj , Fk to sort the elements of
F(Q1);
else
According to the RScore(D3;Q3) to sort the elements
of F(Q1);
end ifelse
According to the RScore(D2;Q2) to sorttheelements
of F(Q1);
end if
end for
Algorithm 2 One-to-many Order-preservingMapping-OPM
Input: D, R, m, id(F)
Output: c
while jDj! = 1 do
fD;Rg Binary Search(K;D;R;m);
end while
coin R TapeGen(K; (D;R; 1 jj m; id(F)));
c coin R;
Return c;.
4. CONCLUSION
In this paper, we define the problem of semantic search
based on conceptual graphs over encryptedoutsourceddata
for the first time. We choose CGs among various methods of
knowledge representation to represent the documents. To
generate the CGs, we apply a state-of-the-art technique,
ie.,text summarization and Tregex a tool for simplifying
sentences in our method. And to achieve our design goals of
both system security and usability, we divide each CG into
three index vectors based on the structureoftheCG,thetype
of concept and the value of concept. We apply order
preserving symmetric encryption (OPSE) to our scheme to
enhance security. Experimental results exhibitthe efficiency
of our proposed scheme.

REFERENCES
[1] E. Cambria,“Affectivecomputingandsentimentanalysis,”
IEEE Intelligent Systems, vol. 31, no. 2, pp. 102–107, 2016.
[2] L.-W. Ku, Y.-T. Liang, H.-H. Chen et al., “Opinion
extraction, summarization and tracking in news and blog
corpora.” in AAAI spring symposium: Computational
approaches to analyzing weblogs, vol. 100107, 2006.
[3] S. Liu, X. Cheng, F. Li, and F. Li, “Tasc: topic-adaptive
sentiment classification on dynamic tweets,” IEEE
Transactions on Knowledge and Data Engineering, vol. 27,
no. 6, pp. 1696–1709, 2015.
[4] F. Li, M. Huang, Y. Yang, and X. Zhu, “Learning to identify
review spam,” in IJCAI Proceedings-International Joint
Conference on Artificial Intelligence, vol. 22, no. 3, 2011, p.
2488.
[5] P. H. Calais Guerra, A. Veloso, W. Meira Jr, and V. Almeida,
“From bias to opinion: a transfer-learning approach to real-
time sentiment analysis,” in Proceedings of the 17th ACM
SIGKDD international conference on Knowledge discovery
and data mining, 2011, pp.150–158.

IRJET- Cross-Domain Sentiment Encoding through Stochastic Word Embedding

More Related Content

What's hot (20)

Similar to IRJET- Cross-Domain Sentiment Encoding through Stochastic Word Embedding (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Cross-Domain Sentiment Encoding through Stochastic Word Embedding