IRJET- Finding Related Forum Posts through Intention-Based Segmentation

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2629
Finding Related Forum Posts through Intention-Based Segmentation
Bijinepalli Sai Lakshmi Mounika1, Chebiyyam Sai Pratyusha2, Davuluri Sai Meghana3
1,2,3 Student, Dept. of Computer Science & Engineering, VVIT, Nambur.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract – Here the issue of finding related discussion
presents on a post at a hand will be considered. In
conventional approach content examinationwillbeperformed
for finding related records, while in this approach we consider
each post as an arrangement of sections, each composed
because of an alternate objective. Relatedness between the
posts ought to be founded on likeness of portions that are
passing on a similar objective. This is conceivable just when
similar terms weigh diversely in relatedness score in view of
section aim in which they are found. In this division technique,
by checking number of content highlights the parts of a post
can be distinguished and critical hops happen demonstrating
purpose of division. Goal bunches will be shaped by grouping
the fragments of all posts and after that similitude’s will be
ascertained crosswise over sections with same expectation.
The effectiveness of our division strategy in finding related
gathering posts is delineated.
Key Words: Segmentation, Intention, Forums, Posts.
1. INTRODUCTION
A type of blog that lets users publishes short text updates.
Bloggers can usually use a number of services for the
updates including instant messaging, e-mail, and twitterand
so on. The posts are called micro posts, while the act ofusing
these services to update your blog is called micro blogging.
Posts:
A comment posted by the users for a particular online
discussion or forum.
Forum:
An Internet forum, or message board, isanonlinediscussion
site where people can hold conversations in the form of
posted messages.
Gatherings with regards to online client groups offer clients
the capacity to look for arrangements and settle on choices
in regards to different issues by misusing other clients'
involvement. They likewise offer organizations the capacity
to interface and bolster their client base.Existingdiscussions
go from areas like wellbeing, law and innovation. The
association of the gathering posts into classifications is a
component that encourages clients to distinguish all the
more effectively those presents related on a theme. In any
case, since perusing a substantial number of posts is baffling
and tedious, most gathering destinations offer catch phrase
seek capacities. However, catchphrase pursuitmaynotbring
about a total arrangement of related posts since the
determination of the correct watchwords isn't generally
direct. Relatedness has generally been converted into
content comparability. Content comparability registered
specifically crosswise over gathering posts is, sadly, not
exceptionally viable for this situation on the grounds that
inquiries are done under particular topical classifications,
e.g., printers, or lodgings in New York, in which the
substance of the considerable number of posts is in any case
comparative. We advocate that when we are estimating the
relatedness of two discussion posts, regarding them as
composite questions rather than solid elementscan prompt
more compelling examinations. The gathering posts
comprise of various parts each present with an alternate
objective. For example, a section may serve to portray an
issue that the writer has, another to give foundation data
keeping in mind the end goal to put the peruse into setting,a
third to expressa want, and a fourth to achieve a conclusion.
We allude to these parts of a discussion post as sections. The
relatedness of two posts would then be able to be founded
on an examination crosswise over portions that serve a
similar objective, i.e., they are proposed for a similar reason,
rather than a correlation of the two posts as wholes. The
correlation among content fragments with a similar
expectation can be performed by simple searching
techniques and SQL queries. In any case, in our approach,
given the distinctive aims of the gathering post creator, the
significance and significance of a term is assessed in view of
the section in which the term is found. Distinctive weights
for similar terms have been utilized crosswise over various
topical discussion classes or spaces. To the best of our
insight, it is the first occasion when that a weighting plan
may relegate distinctive weights to a term in posts of the
same topical classification; or even inside a similar post.
Distinguishing the sections in a discussion post is a testing
undertaking. Discussion posts arecommonlymaybeacouple
passages long, with finish sentences. They don't take after
the contracted style utilized as a part of smaller scale sites,
however in the meantime, since they are expected for
intelligent discourses, they are not verbose and they do not
have the auxiliary develops (e.g., segments) ordinarily
utilized asa part of full-content recordstodistinguishtopical
units.
We have built up a structure for finding related gathering
post that depends on the above thought. By misusing the
correspondence implies, the framework distinguishes the
distinctive portions inside every gatheringpostandpartsthe
discussion post into these sections. Portions serving a
similar aim are distinguished and assembled together.Given
a gathering post within reach, its sections are distinguished

and the coordinating score of each fragment with other
discussion posts' portions that have a similar goal is figured.
The intentions of the posts are classified as positive and
negative based on the comments provided by the users.
When the users post a comment the posts are divided based
on the set of positive and negative words provided in the
algorithm. The algorithm has the flexibility of adding any
number of words to it based on the users interest and
intention.
• We formally present a novel technique for finding
related discussion posts that regards each post as an
arrangement of portions and registers content likeness just
crosswise over fragments of a similar expectation.
• We give broad investigations genuine clients that
affirm the presence of such sections in discussion posts of
various spaces, and confirm the viability of the individual
advances and choices of our philosophy, including thefringe
determination systems, the choice of highlights, and towrap
things up the capacities and weights for catching content
element variety.
• We assess the viability of the general approach on
the suggestion of related gathering posts utilizing
evaluations and criticism by clients in 3 unique spaces.
Data Mining:
Data Mining is one of the popular techniques which are
mainly used for collecting data from different perspectives
and analyzing it into useful information. The information
which can be used to increase revenue and cost. Datamining
is software which is analyzing data from different analytical
tools. Finding correlations or patternsamongdifferentfields
in large relational databases and gathering the knowledge
from data mining techniques.
Segmentation:
Data mining is the process of extracting previouslyunknown
and actionable information from large, complex databases.
Segmentation is a key data mining technique. A segment is a
group of consumers that react in a similar waytoaparticular
marketing approach. So the key to segmentation is to decide
how to split the database up.
Clustering:
Clustering is the process of using machine learning and
algorithms to identify how different typesof data arerelated
and creating new segments based on those relationships.
Clustering finds the relationship between datapointssothey
can be segmented.
SQL:
SQL stands for Structured Query Language. It includes
simple queries to extract the informationfromlargevolumes
of data.
2. SYSTEM ARCHITECTURE
Fig 1: System Architecture
A server cluster is a collection of clusters, called nodes that
communicate with each other to make a set of services
highly available to clients. If user needs any information, he
interacts with clustering server. If related content is found
the server directly communicates with the user otherwise
server interacts with the database. If the data is found in the
database it provides information to the user (Refer Fig 1).
3. EXISTING AND PROPOSED SYSTEM
3.1 EXISTING SYSTEM:
In the Existing System there is no goal based division. There
exists significant measure of work for post coordinating in
Question Answering Communities (QAC). Individuals look
for answers to general intrigue, verifiable or educational,
questions.
Division techniques are isolated into 2 general gatherings.
The first is topical division where contiguoussets of content
pieces are thought about for general likeness in light of
terms or points.
The second gathering of division strategies comprises of
Transcribed oral-talk procedures utilized as a part of the
investigation of translated oral correspondence utilizing
semantic criteria.
DISADVANTAGES:
 Less Efficiency.
 Difficult to finding related forum posts.
 Segmentation increases cost.
 Complexity of TF/IDF algorithm is high.
 Searching is difficult.

3.2 PROPOSED SYSTEM:
The comparison among text segments with the same
intention can be performed by Information Retrieval
methods, such as one of the many TF/IDF or BM25 variants
or language model based methods, or using topicsgenerated
by topic modeling techniques like LDA, paraphrasing
techniques or even auxiliary external services,withthelatter
been used especially for documents with short and poor
content, e.g., tweets. However in our approach, given the
different intentions of the forum post author, the meaning
and importance of a term is estimated based on the segment
in which the term is found. Different weights for the same
terms have been used across different thematic forum
categories or domains. To the best of our knowledge, itisthe
first time that a weighting scheme may assign different
weights to a term in posts of the same thematic category; or
even within the same post.
The complexity of this algorithm is reduced using simple
search queries.
ADVANTAGES:
 Due to the nature of the posts, measuring the
relatedness score after having distinguished the
different segments/messagesthat the authorsintend
to communicate has been proved more effective than
the direct comparison of the whole posts.
 Easy to finding related forum posts with more
effectively and more efficiently.
 Complexity is low.
 More Efficient
4. IMPLEMENTATION
 The user cannot access his account until the Admin
authorize the user status as authorized.
 Admin can see the all the users who are present in
the database.
 The user cannot post more than 2 lines in micro
blogs.
4.1 Create Forums
Only the Admin can create the forum posts. First, the Admin
log into his account and he creates the forum post category
and then he gives the unique name to the post and
description to the forum post.
4.2 Search Posts
Here, the user first log into his account. He then searchesthe
post using keyword (Searching content based on post
description). If the post is present in the database it will be
displayed along with post name and domain.
4.3 Intention based Matching
When the user commentson the post the post is dividedinto
set of segments. Based on the segmentation algorithm the
posts are classified as positive and negative intention posts
(Refer Fig 2).
This positive and negative intention posts are displayed in
the admin home page where the admin can see how many
positive and negative posts are posted by the users for a
particular forum post. Based on the number of positive and
negative intentions for a particular post the positive post
chart (Refer Fig 5) and negative post chart (Refer Fig 6) are
created.
Fig 2: Positive and Negative Intentions
Table 1: Some of the positive and negative words for
segmentation
Positive Negative
Good Bad
Yes
No
Not
None
Do
Does
Don’t
Doesn’t
Was
Were
Wasn’t
Weren’t
Has
Have
Had
Hasn’t
Haven’
Hadn’t
Better
Cheap
Abusive
Sucks
Based on the words present in the table (Refer Table 1) the
posts will be divided as positiveor negative. We can add any

number of positive and negative words to the algorithm
based on our interest.
4.4 Recommend posts
The user can also recommend posts to his friendsif hefound
the post is useful. This is the additional feature implemented
in this paper that to this feature doesn’t exist in the micro
blogs till now.
4.5 Segmentation grouping
The same intention posts are clustered together intoasingle
cluster. The cluster consists of all the posts that are intended
for the same goal/intention.
For ex:
Positive intention posts in single cluster
Negative intention posts in single cluster
Fig 3: Positive Intention Posts
On clicking the View Positive Intention button the admin is
displayed with a page containing a cluster of all the positive
comments related to a particular post (Refer Fig 3).
If the admin clicks on the View Negative Intentionbuttonthe
admin is displayed with a page containing a cluster of all the
negative comments related to a particular post (Refer Fig4).
Fig 4: Negative Intention Posts
Delete Forums
The admin has the right to delete forums if he found more
than five bad intentions to the same post. The admin
provides the reason of why he was deleting the post. Once
the post is deleted by the admin, it will be no longeravailable
in the database.
5. ALGORITHMS
5.1 TF/IDF Algorithm
 It is often used as a weighing factor in searches of
Information retrieval.
 It is one of the simplest ranking functions and is
computed by summing the TF/IDF query term.
 This algorithm is mainly used for searchingprocess.
 Then tf–idf is calculated as:
TF-> Term Frequency
IDF->Inverse Document Frequency
Where,
t=term
d=referencing document
D=set of Documents
The equation has a higher complexity in the existing system
and also searching process takes more time. So, we reduced
the complexity of this algorithm using simple queries and
searching methods.
Example SQL query used in this paper forsearchingpurpose:
String str=”select * from allcomments where
p_name=’”+p_Name”’ and categorie=’”+p_Categorie”’;
5.2 Clustering Algorithm
Clustering is a process which partitions a given data set
into homogeneous groups based on given features such
that similar objects are kept in a group whereas
dissimilar objects are in different groups. It is
the most important unsupervised learning problem.
It deals with finding structure in a collection of unlabeled
data.

6. SEGMENTATION EVALUATION
Fig 5: Positive Segmentation chart
Fig 6: Negative Segmentation chart
Here, the segmentation chart will be formed by considering
the information of respective clusters. The positive cluster
consists of all the positive posts regarding a topic and the
negative cluster consists of all negative posts.
7. CONCLUSION
We proposed a novel approach for matchingareferencepost
to the k most related posts in a collection. Our method
identifies and exploits post segments that convey similar
author intentions. We presented several experiments
regarding the right segmentation criteria, the effectiveness
of the segmentation algorithms and the formation of
intention clusters that prove that a rather intuitive concept,
that of the author intentions to communicate a certain
message, can be effectively captured by an automated
process. Moreover, due to the nature of the posts,measuring
the relatedness score after having distinguishedthedifferent
segments/messages that the authorsintendtocommunicate
has been proved more effective than the direct comparison
of the whole posts.
REFERENCES
[1] J. Jeon, W. B. Croft, and J. H. Lee, “Finding semantically
similar questions based on their answers,” in Proceedingsof
the 28th ACM SIGIR Conference, ser. SIGIR ’05. New York,
NY, USA: ACM, 2005, pp. 617–618.
[2] T. C. Zhou, C.-Y.Lin, I. King, M. R. Lyu, Y.-I.Song, and Y.Cao,
“Learning to suggest questions in online forums.” in AAAI,
2011.
[3] L. Weng, Z. Li, R. Cai, Y. Zhang, Y. Zhou, L. T. Yang, and L.
Zhang, “Query by document via a decomposition-basedtwo-
level retrieval approach.” Association for Computing
Machinery, Inc., July 2011.
[4] V. Govindaraju and K. Ramanathan, “Similar document
search and recommendation,” Journal of Emerging
Technologies in Web Intelligence, vol. 4, no. 1, pp. 84–93,
2012.
[5] S. Robertson, S.Walker, and M. Hancock-Beaulieu,“Okapi
at TREC-7: Automatic ad hoc, filtering, VLC and interactive
track,” TREC ’98, pp. 199–210, 1998.
[6] D. M. Blei, “Probabilistic topic models,” Commun. ACM,
vol. 55, no. 4, pp. 77–84, Apr. 2012.
[7]J. Berant and P. Liang, “Semantic parsing via
paraphrasing.” in ACL (1), 2014, pp. 1415–1425.
[8] H. Wen, W. Zhongyuan, W. Haixun, Z. Kai, and Z. Xiaofang,
“Short text understanding through lexical-semantic
analysis,” in IEEE ICDE, 2015.
[9] Z.-Y. Ming, T.-S.Chua, and G. Cong, “Exploring
domainspecific term weight in archived question search,” in
Proceedingsof the 19th ACM CIKM, ser. CIKM ’10. NewYork,
NY, USA: ACM, 2010, pp. 1605–1608.
[10] D. Papadimitriou, G. Koutrika,Y.Velegrakis and J.
Mylopoulos, “Finding Related Forum Posts through Content
Similarity over Intention-Based Segmentation” in IEEE
Transactions on Knowledge and Data Engineering, vol. 29,
no. 9, pp. 1860-1873, Sep 2017.
[11] M. Chen, X. Jin, and D. Shen, “Short text classification
improved by learning multi-granularity topics,” in IJCAI,
2011, pp. 1776–1781.
[12] M. Hearst, "Texttiling: Segmenting text into multi-
paragraph subtopic passages", Comput. Ling., vol. 23, pp.33-
64, 1997.
[13] H. Misra, F. Yvon, J. M. Jose, O. Cappe, "Text
segmentation via topic modeling: an analytical study", Proc.
Conf. Inf. Knowl. Manage., pp. 1553-1556, 2009.
[14] X. Hu, N. Sun, C. Zhang, T. Chua, "Exploiting internal and
external semantics for clustering short texts using world
knowledge", Proc. Conf. Inf. Knowl. Manage., pp. 919-928,
2009.
[15] I. Hulpus, C. Hayes, M. Karnstedt, D. Greene,
"Unsupervised graph-based topic labelling using
dbpedia", Proc. ACM Int. Conf. Web Search Data Mining, pp.
465-474, 2013.

IRJET- Finding Related Forum Posts through Intention-Based Segmentation

Recommended

More Related Content

What's hot (20)

Similar to IRJET- Finding Related Forum Posts through Intention-Based Segmentation (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Finding Related Forum Posts through Intention-Based Segmentation