International Journal of Emerging Technology in Computer Science & Electronics (IJETCSE)
ISSN: 0976-1353 Volume 21 Issue 3 – APRIL 2016.
BUILD RECOMMENDATION SYSTEM USING USER’S
PAST BEHAVIOR
D.CHADNRU, Mrs.N.SENTHAMARAISELVI2(SAsP)
Department of computer science engineering,IFET College Of Engineering,Villupuram-605602,India
in one of two ways – through collaborative or content-based
Abstract:
filtering.Collaborative
filtering approaches
building
a
systems (sometimes replacing
model from a user's past behavior (items previously purchased or
"system" with a synonym such as platform or engine) are a subclass
selected and/or numerical ratings given to those items) as well as
of information filtering system that seek to predict the 'rating' or
similar decisions made by other users. This model is then used to
'preference' that a user would give to an item. Recommender systems have
predict items (or ratings for items) that the user may have an interest
become extremely common in recent years, and are applied in a variety of
in.[9] Content-based filtering approaches utilize a series of discrete
Recommender system or recommendation
applications. The most popular ones are probably movies, music, news,
characteristics of an item in order to recommend additional items
books, research articles, search queries, social tags, and products in
general.
However,
there
are
also
recommender
systems
with similar properties. These approaches are often combined
for
experts, collaborators, jokes, restaurants, financial services, life insurance,
(see Hybrid Recommender Systems).
persons (online dating), and Twitter followers. Recommender systems
The differences between collaborative and content-based filtering can
typically produce a list of recommendations in one of two ways - through
be demonstrated by comparing two popular music recommender
collaborative or content-based filtering. Collaborative filtering approaches
systems – Last.fm and Pandora Radio.
building a model from a user's past behavior (items previously purchased or
selected and/or numerical ratings given to those items) as well as similar
Last.fm creates a "station" of recommended songs by observing what
decisions made by other users. This model is then used to predict items (or
ratings for items) that the user may have an interest in. Content-based
bands and individual tracks the user has listened to on a regular basis
filtering approaches utilize a series of discrete characteristics of an item in
and comparing those against the listening behavior of other users.
order to recommend additional items with similar properties.
Last.fm will play tracks that do not appear in the user's library, but
are often played by other users with similar interests. As this
I.
INTRODUCTION
Recommender
approach leverages the behavior of users, it is an example of a
systems or recommendation
collaborative filtering technique.
systems (sometimes
replacing "system" with a synonym such as platform or engine) are a
Pandora uses the properties of a song or artist (a subset of the 400
subclass of information filtering system that seek to predict the
'rating'
or
'preference'
that
a
user
would
give
to
attributes provided by the Music Genome Project) in order to seed a
an
"station" that plays music with similar properties. User feedback is
item.Recommender systems have become extremely common in
used to refine the station's results, deemphasizing certain attributes
recent years, and are applied in a variety of applications. The most
when a user "dislikes" a particular song and emphasizing other
popular ones are probably movies, music, news, books, research
attributes when a user "likes" a song. This is an example of a content-
articles, search queries, social tags, and products in general.
However,
there
are
also
experts, collaborators, jokes,
recommender
Each type of system has its own strengths and weaknesses. In the
followers.
above example, Last.fm requires a large amount of information on a
Recommender systems typically produce a list of recommendations
user in order to make accurate recommendations. This is an example
persons
(online
dating),
financial
based approach.
for
services,life
insurance,
restaurants,
systems
and
Twitter
444
International Journal of Emerging Technology in Computer Science & Electronics (IJETCSE)
ISSN: 0976-1353 Volume 21 Issue 3 – APRIL 2016.
of the cold start problem, and is common in collaborative filtering
Even so, the significant predictive relationship word analysis had
systems. While Pandora needs very little information to get started, it
with participants’ navigation has useful implications for the design of
is far more limited in scope (for example, it can only make
future tools, as we discuss in the next section. In addition, these
recommendations that are similar to the original seed).Recommender
results pro-vide independent evidence about the premises behind
systems are a useful alternative to search algorithms since they help
current systems like Team Tracks and Hipikat. The designers of these
users discover items they might not have found by themselves.
systems have conducted studies, but our study is independent, not
Interestingly enough, recommender systems are often implemented
conducted by the designers of Hi
using search engines indexing non-traditional data.Montaner provides
Specifically, the results showed that the bug reports predicted the
the first overview of recommender systems, from an intelligent
vocabulary of the queries participants used to find appropriate source
agents perspective. Adomavicius provides a new overview of
code. This supports the concept of including textual analysis, used in
recommender systems.Herlocker provides an additional overview of
systems like Hipikat. The R-squared values also showed that word
evaluation techniques for recommender systems, and Beel et al.
analysis alone was unlikely to be enough, sup-porting the concept
discuss the problems of offline evaluations. Beel et al. also provide a
behind both Hipikat and Team Tracks that multiple sources of
literature survey on research paper recommender systems.
information are needed to make good navigation predictions, and
II. SCOPE OF THE PROJECT
that a single source is unlikely to have high enough accuracy.
One approach to the design of recommender systems that has wide
Both Hipikat and Team Tracks make use of collective knowledge
use is collaborative filtering. Collaborative filtering methods are
through approaches such as navigation
based on collecting and analyzing a large amount of information on
popularity. Our results support this design choice too, showing that
users’ behaviors, activities or preferences and predicting what users
participants collectively narrowed their focus on the same files. But
will like based on their similarity to other users. A key advantage of
given that participants reached more consensus for the bug than for
the collaborative filtering approach is that it does not rely on machine
the feature request, such systems may be able to improve further by
analyzable content and therefore it is capable of accurately
factoring in which type of issue (bug or feature request) a developer
recommending complex items such as movies without requiring an
is working on. Turning to theoretical underpinnings, our results are
"understanding" of the item itself. Many algorithms have been used
consistent with a number of existing theories, and shed further light
in measuring user similarity or item similarity in recommender
upon the way developers go about debugging and maintenance.
path “wear” and class
systems.
First, our results are consistent with the well-established idea of
III. RELATED WORK
hypothesis formation as a basis of debugging [4, 10, 20]. In our
results, vocabulary in the bug reports predicted vocabulary in the
The goal of this experiment has been to contribute to prediction of
queries, suggesting the possibility of queries as a surrogate for
developers’ behavior, and the results showed significant relationships
hypotheses about “subject areas” in the source code where work will
between word-based predictions and the participants’ actual behavior.
be needed. Further, the places to which participants navigated as a
How-ever, as the R-squared values show, the linguistic pre-dictions
result of their queries were the “right” places for investigating their
did not explain all of their behavior. R-squared values can range from
hypotheses, as evidenced by the fact that they spent significant time
0 to 1. In studies of human behavior, R-squared values are common
in the files once they had gotten there. Note also that vocabulary did
in the .09 to .25 range. With one exception (.05), our R-squared
not explain all of participants’
values ranged from .26 to .31. Such R-squared values indicate, of
navigation behavior, which is
consistent with the idea of hypothesis formation as well, since it is
course, that a great deal of the humans’ behavior was not accounted
unlikely that all of the vocabulary of suitable hypotheses would be
for. That is to be expected because, first, there is inherent
present in a bug report.
measurement error in attempting to measure human behavior and,
second, it is not realistic to expect a single variable to account for all
of their behavior.
445
International Journal of Emerging Technology in Computer Science & Electronics (IJETCSE)
ISSN: 0976-1353 Volume 21 Issue 3 – APRIL 2016.
Second, our results are consistent with beacons and with Ko et al.’s
4.2 DRAWBACKS OF EXISTING SYSTEM:
model of searching, relating, and collecting. The work on beacons
emphasizes the importance of cues in comprehension, though
beacons are at a finer granularity than what we examined. Ko et al.’s
•
Low accuracy
•
Flexible
•
Scalability
•
Inefficiency
model is related to information foraging theory in its emphasis on the
importance of cues in the environment. Our results further refine
what goes on in the “search” aspect that leads developers to the right
places for relating and collecting. pikat or Team Tracks.
4.3 PROPOSED WORK
IV. SYSTEM ANALYSIS
In
a
broad
sense,
a general methodology (not
a
fixed
Recommender systems typically produce a list of recommendations
set
in one of two ways - through collaborative or content-based
of techniques) that applies a 'systems' or 'holistic' perspective by
filtering. Collaborative filtering approaches building a model from a
taking all aspects of the situation into account, and by concentrating
on
the
interactions
between
It provides a framework in
its
different
which judgments of
user's past behavior (items previously purchased or selected and/or
elements.
numerical ratings given to those items) as well as similar decisions
the experts in
made by other users. This model is then used to predict items (or
different fields can be combined to determine what must be done, and
ratings for items) that the user may have an interest in.
what is the best way to accomplish it in light of current and
future needs. Although closely associated with data or information
Content-based filtering approaches utilize a series of discrete
processing,
characteristics of an item in order to recommend additional items
the practice of
SA
has
been
in
existence
since long before computers were invented.
with similar properties. Recommender systems are a useful
alternative to search algorithms since they help users discover items
In a narrow sense, analysis of the current and future roles of
proposed computer
system in
an organization,
they might not have found by themselves. Interestingly enough,
The system
recommender systems are often implemented using search engines
analyst (usually
indexing non-traditional data. Montaner provides the first overview
software engineer or programmer) examines the flow of documents, i
nformation,
and material to design a
system
that
best
of
meets
recommender
systems,
from
an
intelligent
agents
perspective. Adomavicius provides a new overview of recommender
the cost, performance, and scheduling objectives.
systems.
4.1 EXISITING WORK
4.4 ADVANTAGES OF PROPOSED SYSTEM
These approaches make file-to-edit recommendations by mining
association rules between files frequently edited together in the
past.We developed a new powerful context formation approach and
demonstrated its efficacy
in
•
Accuracy
•
Collaborative Filtering algorithm is adopted to generate
mining programmer interaction
appropriate recommendations
histories (MI).We developed a comparative framework that helps
understand which factors have
much
recommendation performance.We demonstrated
•
influence on the
that
the
Efficiency and Scalability
rules
5. SYSTEM ARCHITECTURE
mined by our approach (MI) outperform the rules mined from edit
histories (ROSE) in recommending files to edit.We identified that the
System architecture is a conceptual model that defines the structure,
significant improvement of the recommendation performance is
behavior, and more views of a system. An architecture description is
enabled by the context further elaborated by the records of files
a formal description and representation of a system, organized in a
viewed.
446
International Journal of Emerging Technology in Computer Science & Electronics (IJETCSE)
ISSN: 0976-1353 Volume 21 Issue 3 – APRIL 2016.
way that supports reasoning about the structures and behaviors of the
[2] Baeza-Yates, R., Ribeiro-Neto, B. Modern Information Retrieval,
system.
Addison Wesley Longman, 1999.
[3] Blackwell, A. First steps in programming: A rationale for
attention investment models,IEEESymp. Human-Centric Comp.
Langs. Envs., 2002, 2-10.
[4] Brooks, R. Towards a theory of the cognitive processes in
computing programming, Int. J. Human-Computer Stud-ies51, 1999,
197-211.
[5] Chi, E., Rosien, A., Supattanasiri, G., Williams, A., Royer, C.,
Chow, C., Robles, E., Dalal, B., Chen, J., Cousins, S. The
Bloodhound project: Automating discov-ery of web usability issues
using the InfoScent simulator, ACM Conf. Human Factors Comp.
Sys., Ft. Lauderdale, Florida, 2003.
VIII. CONCLUSION
Collaborative
filtering (CF)
[6] Cubranic, D., Murphy, G., Singer, J., Booth, K. Hipikat: A project
is
a
technique
used
memory
by
for
software
development,
IEEE
Trans.
Software
Engineering31(6), 446-465, June 2005.
some recommender systems. Collaborative filtering has two senses, a
narrow one and a more general one. In general, collaborative filtering
[7] Detienne, F.,
is the process of filtering for information or patterns using techniques
Software Design -Cognitive Aspects, Springer,
2001.
involving collaboration among multiple agents, viewpoints, data
sources, etc. Applications of collaborative filtering typically involve
[8] DeLine, R., Czerwinski, M. and Robertson, G., Easing program
very large data sets.
comprehension by sharing navigation data, IEEE Symp. Visual
Langs. Human Centric Computing, 2005, 241-248.
Another common approach when designing recommender systems
is content-based filtering. Content-based filtering methods are based
[9] Glick, M. and Holyoak, K. J., Schema induction and analogical
on a description of the item and a profile of the user’s preference. In a
transfer. Cognitive Psychology1983, 15, 1-38.
content-based recommender system, keywords are used to describe
the items and a user profile is built to indicate the type of item this
[10] Ko, A., Myers, B., A framework and methodology for studying
user likes. In other words, these algorithms try to recommend items
the causes of software errors in programming systems, J. Visual
that are similar to those that a user liked in the past (or is examining
Langs. Computing 16(1-2), 2005.
in the present). In particular, various candidate items are compared
[11] Ko, A.,Myers, B., and Chau, D., A linguistic analysis of how
with items previously rated by the user and the best-matching items
people describe software problems, IEEE Symp. Vis-ual Langs.
are recommended. This approach has its roots in information
Human-Centric Computing, 2006, 127-136.
retrieval and information filtering research.
[12] Ko, A, Myers, B., Coblenz, M., and Aung, H., An ex-ploratory
REFERENCES
study of how developers seek, relate, and collect relevant information
[1] Anderson, J. R., Bothell, D., Byrne, M., Douglass, D., Lebiere, C.
during
and Qin, Y. An integrated theory of mind. Psychological
Engineering32(12), Dec. 2006.
Review111(4), 2004, 1036-1060.
software
maintenance
tasks,
IEEE
Trans.
Software
[13] Mitchell, M. and Jolley, J. Research Design Explained, 6th Ed.,
Wadsworth Publishing, 2006
447