A LOCATION-BASED MOVIE RECOMMENDER SYSTEM USING COLLABORATIVE FILTERING

International Journal in Foundations of Computer Science & Technology (IJFCST), Vol.5, No.4, July 2015
DOI:10.5121/ijfcst.2015.5402 13
A LOCATION-BASED MOVIE RECOMMENDER SYSTEM
USING COLLABORATIVE FILTERING
Kasra Madadipouya
Department of Computing and Science, Asia Pacific University of Technology &
Innovation
ABSTRACT
Available recommender systems mostly provide recommendations based on the users’ preferences by
utilizing traditional methods such as collaborative filtering which only relies on the similarities between
users and items. However, collaborative filtering might lead to provide poor recommendation because it
does not rely on other useful available data such as users’ locations and hence the accuracy of the
recommendations could be very low and inefficient. This could be very obvious in the systems that locations
would affect users’ preferences highly such as movie recommender systems. In this paper a new location-
based movie recommender system based on the collaborative filtering is introduced for enhancing the
accuracy and the quality of recommendations. In this approach, users’ locations have been utilized and
take in consideration in the entire processing of the recommendations and peer selections. The potential of
the proposed approach in providing novel and better quality recommendations have been discussed
through experiments in real datasets.
KEYWORDS
Recommender system, Movie recommender system, Location-based recommender, Collaborative filtering,
Pearson correlation, Context-aware
1. INTRODUCTION
In traditional recommender systems the recommendation procedure is done with considering
about users’ past rating history and items features (tagging) and basically no contextual
information is taken into account for generating recommendations [1]. However, there are usually
various factors influencing users’ decision on in reality. Besides, users’ demand might change
with context as well (e.g. time, location and weather etc) which cannot be accomplished through
tradition recommender systems.
As a result just having information about users, items and/or few contextual parameters are not
sufficient enough to give accurate real-time emotional-based recommendations, however, more
information is necessary in order to provide items to users under certain conditions [2]. For
instance, adding the temporal, day/night context, in recommender systems would provide more
accurate recommendations which also might reflect user's emotions as well. Additionally, in the
personalized content delivery for a recommender system, this fact is crucial to determine whether
particular content should be delivered to users or not and if it has to be recommended to users
when it should be recommended [2].
With the creation of the new generation recommender system, contextual information has become
one of the most valuable knowledge sources to improve recommendations and provide more user
specific recommendations under that similar circumstance (i.e., contexts) are related with similar
user preferences [1][3] [4].

14
A better Recommendation System (RS) is the one which delivers recommendations that best
match with users' preferences, needs and hopes at the right moment, in the right place and on the
right media. This can't be achieved without designing a RS that takes into account all information
and parameters that influence user's ratings. This information may concern demographic data [5]
preferences about user's domain of interest, quality and delivery requirements as well as the time
of the interaction, the location, the media, the cognitive status of the user and his availability.
Hence, giving precise recommendations to user is deepened on the relevant gathered information
about the user's preferences such as user's location and other parameters.
The rest of the paper is organized as follows; in the next section a review regarding different
movie recommender systems are provided. In section three the proposed algorithm is explained in
detail which is based on the collaborative filtering and Pearson Correlation Coefficient. Then in
the next segment, the results of the experiment is analyzed and presented. Finally, in the last
section, conclusions and future enhancement are discussed.
2. RELATED WORK
In recommender systems variety of research had been conducted to deliver the most relevant
content to user [6][7] [8] in different recommendation scopes for users, such as music
recommendation [9][10][11], location-based services [12][13] or even recommendation system
for tourism [14][15][16].
In movie recommender system some of the works focus on inputting rich and detailed meta-data
(genre and other features) to the system in order to enhance the quality of recommendations. In a
movie recommender system by [17], this facility was provided to allow user to even set new
keywords for movies which resulted in providing more accurate recommendations based on
features which set users. The big advantage of this solution is that it adapts to changes regarding
what the users find important, something which can change over time. However, manually
categorization of content can be expensive, time-consuming, error-prone and highly subjective.
Due to this, many systems aim to automate the procedure.
Another approach of providing high quality recommendation can be achieved by utilizing
contextual-information [18] such as time of the day, users’ mood, etc. According to [19] most of
the users who are located in the same geographical location have the similar taste to each other.
For instance, users who are located in Florida mostly like to watch “Fantasy”, “Animation” types
of the movie however, the user from Minnesota State are mostly interested to “War”, “Drama”
type of the movie. Therefore, in recommending items the location of the user is not considered in
any steps, as a result it is possible for a user who is looking for a resultant in Chicago get
recommendations about some restaurants that are located in other areas such as Seattle [19].
Various algorithms examine by researchers in order to enhance the traditional recommender
systems. For instance, in [20], authors presented a novel hybrid approach for movie recommender
systems whereby items are recommended based on the collaborative between users as well as
analysing contents. In another work by Bogers [21], an algorithm based on Markov random walk
proposed which applied on a movie dataset by considering regarding some contextual
information. In addition, Rendle et al. [22] applied matrix factorization in a movie dataset with
the aim of enhancing the quality and accuracy of the recommendation as well as reducing the
complexity of building their model. Simon Funk proposed Regularized singular value
decomposition (RSVD) to enhance the accuracy of the recommendation which reduces the
dimensionality of the original rating matrix [23].

15
3. THE PROPOSED ALGORITHM
The new method was made to reach to numbers of goals and objectives. Firstly, by contrast of the
baseline collaborative filtering that does not take into account users’ locations in providing
recommendations, in this framework current user location is considered in the system in which
user gets different recommendations based on the location. This reflects the fact that if the user
changes his/her location the received recommended items will be affected because of changing in
the location. Secondly, user preferences that lead him/her to select an item, and correspondingly
potentially generates better recommendations based on the user location which increases the
locality of the recommended items. Lastly, the main goal is to improve the locality of the
recommendations in both peers selection stage and item recommendations with gathering some
information about user location and considering about user location in all recommendation
processes.
The only recommendation method that is used in the framework contains the following steps as
demonstrated in below,
1- Keeping a copy of the user profile in the memory (memory-based approach).
2- Applying similarity index function for the user to find the peers (Pearson Correlation
Coefficient).
3- Adding α parameter for those peers who are in the same place with the active user.
4- Applying some filtrations on selected peers to reduce and remove non-related peers.
5- Selecting items for providing recommendations to the active user.
6- Showing the recommended items to the active user.
Figure_1 (Overview of collaborative filtering process)
Figure_2 (Pearson Correlation)
Since, the main algorithm (Figure_2) just relies on one parameter (similarity index) to select peers
to the active user and seems all the users with no difference, in this framework we introduced (α)
parameters refers to the priority index which directly affects the outcome of the similarity index
function that returns a value between (-1 to 1). In prioritize the users those users who are located
in the same geographical place with the active user get higher priority with add the constant value
(α) to the result of similarity index.

16
As it depicts in Figure_1 the framework is based on the collaborative filtering in which first the
users necessary information about the active user will be stored in different database from the
user profile in memory since some temporary analysis and changes might be applied in the user
profile which will not be overwritten on the user profile or change it. After copying it, the user
profile will pass through from peers selection stage, in that stage, firstly, similarity function is
applied for the active user against other users of the system to find which user has the similar
taste to the current user. Then peers are selected for the active user. In other word, users will be
classified in different groups which each group has the users with the same taste. After that, peers
location will be compared with the active user and those who are located in the same
geographical location with the active user, their similarity indexes will be increased by adding
constant value which is known as α. Then some filtrations will be applied to the peers who are
called threshold-based selection, according to which users whose similarity exceeds a certain
threshold value are considered as neighbors of the target user [24] and be selected as final peers to
give recommendations to the active user based on them. Hence, in this stage all peers who have
less than 0.5 similarity index (threshold) will be aborted and will not considered in further steps.
After applying filtration to the users in different peers groups, top peers of the active user will be
nominated for providing recommendations. Finally, top rated items of the peers who gained the
highest similarity indexes will be selected for recommendations. In the final stage, those
recommended and selected items are shown to the active users as recommendations.
4. EXPERIMENTAL RESULTS
For experimental test, LDOS-CoMoDa [25] dataset which is provided by Laboratorij za digitalno
in French is utilized for testing the framework against the baseline collaborative (user-based)
filtering approach and therefore, necessary elements from the dataset are extracted for
experiment. In the dataset, information about the users is available as part of the user profile and
in another section of the dataset list of the rated items, available items and many other parameters
are saved. All information in the dataset was gathered by crawling some agencies and was from
real user involvement in the system. Hence, there was no direct user involvement in the testing
system. However, for implementing the system in the real world all above elements are necessary
and should be available for completeness of the system.
Overall the dataset contains 2296 records (rows) which are recorded 120 users’ behaviors. The
total fields of dataset after modification is six fields, which are userID, itemID, Country (User
location), Rating, item description, Related to (Country of item). The rating range is from 1 to 5
which one expresses that the user is highly dislike the item and five represents that the user is
highly interested in the item. Number of ratings for each user is various and is not dependent to
any parameters such as user location or anything else. The reason for that, it to make the dataset
to what happens in reality which the number of ratings items for each user could be different and
it is not dependent to any obvious parameter. Therefore, this dataset is obtained by gathering data
from different agencies and hence, it is exactly recorded users’ behaviors. Table_1 demonstrates
all mentioned information about the dataset in summery.
Table_1 (Dataset information)

17
The users of the dataset are from five different countries (Denmark, France, Germany, UK, and
US). Each country contains different number of the users and this parameter is independent
parameter, but in average base on the number of total user each country has 24 users. Table_2
demonstrates, the exact number of each user for each mentioned country.
Table_2 (User distribution in countries-table)
The experiment has been applied for both baseline Pearson Correlation algorithm and the
proposed framework in the form of memory-based user-based filtering. In this experiment, the
framework is tested against various value of α (0.1, 0.2, 0.3) separately the results are
demonstrated in the following section.
By applying the baseline Pearson Correlation in the dataset, the average gained peers locality was
14.10% which means that only 14.10% of peers are located in the same location with the active
users whereas the rest of the peers locations are not the same with the active user. Applying the
proposed framework with α = 0.1, gives different result which is demonstrate shows 1.20 per cent
(15.33% peers locality rate) improvement in comparison with the baseline approach.
By increasing α to 0.2, the improvement in locality rate reached near to 1.73 per cent (15.33%
peers locality rate). Finally for the last experiment α value is set to 0.3 and the framework has
been tested against the dataset which shows 1.87 percent (15.97% peers locality rate)
enhancement in peers selection stage.
Changing α value on the other hand affects the recommendation results as well in terms of the
locations that movie produced. In baseline algorithm only 1.46% of items related to the same
location as active user, however, by applying α = 0.1, this amount reached to 1.7% and the
number increased to 1.88% and 1.99 by applying α = 0.2 and α = 0.3 respectively.
5. CONCLUSION AND FUTURE WORK
In this paper a novel framework to improve locality in user-based collaborative filtering for movie
recommender system is proposed and experimental results of the evolution of is presented which
gathered results from different tests on dataset analyzed and presented. In addition to that,
compare and comparison between the baseline algorithm and the proposed approach is provided
in details based on the experimental results.
As already mentioned earlier, the dataset which has been used for this work is part of LDOS-
CoMoDa dataset that gathered from various movies agencies. Hence, for future work in order to
reach to more accurate result and better analysis, the experiment should be performed on other
and bigger datasets which contains hundred thousand of records and have different level of
sparseness. In addition, the experimental performed under the lab condition and there was no
direct user involvement in experiment the framework and method for getting results. Hence, the
proposed method could be implemented in the real environment against the real user experiment.

18
Moreover, the proposed method in this work applied on one class of filtering method which is
Pearson correlation coefficient. Thus, various algorithms from both collaborative and content
based approaches could be tested against the dataset and the results compared with the proposed
framework in order to provide more accurate analysis.
REFERENCES
[1] Adomavicius, G., Tuzhilin, A. and Er. 2011. Context-aware recommender systems. Springer, pp. 217-
-253.
[2] Junmanee, S. and Hinze, A. (2006). Advanced recommendations in a mobile tourist information
system.
[3] Baltrunas, Linas et al. 'Context Relevance Assessment And Exploitation In Mobile Recommender
Systems'. Pers Ubiquit Comput 16.5 (2011): 507-526. Web. 7 May 2015.
[4] Gorgoglione, Michele, Umberto Panniello, and Alexander Tuzhilin. 'The Effect Of Context-Aware
Recommendations On Customer Purchasing Behavior And Trust'. Proceedings of the fifth ACM
conference on Recommender systems - RecSys ' 11 (2011): n. pag. Web. 7 May 2015.
[5] Krulwich, Bruce.: ‘Lifestyle finder: intelligent user profiling using large-scale demographic data’. AI
Magazine 18(2), 37–45 (1997).
[6] Adomavicius, G., Sankaranarayanan, R., Sen, S. and Tuzhilin, A. (2006). Incorporating contextual
information in recommender systems using a multidimensional approach. ACM Transactions on
Information Systems (TOIS), 23 (1), pp. 103-145.
[7] Wang, Yiwen et al. 'Recommendations Based On Semantically Enriched Museum Collections'. Web
Semantics: Science, Services and Agents on the World Wide Web 6.4 (2008): 283-290. Web.
[8] Bobillo, Fernando, Miguel Delgado, and Juan Gómez-Romero. 'Representation Of Context-
Dependant Knowledge In Ontologies: A Model And An Application'. Expert Systems with
Applications 35.4 (2008): 1899-1908. Web. 7 May 2015.
[9] Wang, Xinxi, Ye Wang, and David Rosenblum. "A daily, activity-aware, mobile music recommender
system." Proceedings of the 20th ACM international conference on Multimedia. ACM, 2012.
[10] Lampropoulou, P. S., A. S. Lampropoulos, and G. A. Tsihrintzis. 'A Mobile Music Recommender
System Based On A Two-Level Genre-Rating SVM Classifier Enhanced By Collaborative Filtering'.
New Directions in Intelligent Interactive Multimedia Systems and Services - 2 (2009): 361-368. Web.
7 May 2015.
[11] Lampropoulos, Aristomenis S., Paraskevi S. Lampropoulou, and George A. Tsihrintzis. 'A Cascade-
Hybrid Music Recommender System For Mobile Services Based On Musical Genre Classification
And Personality Diagnosis'. Multimedia Tools and Applications 59.1 (2011): 241-258. Web. 7 May
2015.
[12] Daly, Elizabeth M., and Werner Geyer. 'Effective Event Discovery'. Proceedings of the fifth ACM
conference on Recommender systems - RecSys '11 (2011): n. pag. Web. 7 May 2015.
[13] Barranco, Manuel J. et al. 'A Context-Aware Mobile Recommender System Based On Location And
Trajectory'. Advances in Intelligent Systems and Computing (2012): 153-162. Web. 7 May 2015.
[14] Noguera, José M. et al. 'A Mobile 3D-GIS Hybrid Recommender System For Tourism'. Information
Sciences 215 (2012): 37-52. Web. 7 May 2015.
[15] Kuo, Mu-Hsing, Liang-Chu Chen, and Chien-Wen Liang. 'Building And Evaluating A Location-
Based Service Recommendation System With A Preference Adjustment Mechanism'. Expert Systems
with Applications 36.2 (2009): 3543-3554. Web. 7 May 2015.
[16] Savage, Norma Saiph, et al. I’m feeling loco: A location based context aware recommendation
system. Springer Berlin Heidelberg, 2012.
[17] Garden, M. and Dudek, G. (2006) "Mixed collaborative and content-based filtering with user-
contributed semantic features", paper presented at AAAI'06 proceedings of the 21st national
conference on Artificial intelligence , AAAI Press, p.1307-1312.
[18] Wang, L. (2011). Understanding and using contextual information in recommender systems. pp.
1329--1330.
[19] Lev, Oski, J., Sarwat, M., Eldawy, A. and Mokbel, M. (2012). Lars: A location-aware recommender
system. pp. 450--461.
[20] Lekakos, G., & Caravelas, P. (2008). A hybrid approach for movie recommendation. Multimedia
tools and applications, 36(1-2), 55-70.

19
[21] Bogers, T. (2010, October). Movie recommendation using random walks over the contextual graph. In
Proc. of the 2nd Intl. Workshop on Context-Aware Recommender Systems.
[22] Rendle, S., Gantner, Z., Freudenthaler, C., & Schmidt-Thieme, L. (2011, July). Fast context-aware
recommendations with factorization machines. In Proceedings of the 34th international ACM SIGIR
conference on Research and development in Information Retrieval (pp. 635-644). ACM.
[23] Funk, S. (2006). Netflix update: Try this at home.
[24] Gong, S. (2010). A collaborative filtering recommendation algorithm based on user clustering and
item clustering. Journal of Software, 5 (7), pp. 745--752.
[25] http://recsyswiki.com/wiki/LDOS_Comoda

A LOCATION-BASED MOVIE RECOMMENDER SYSTEM USING COLLABORATIVE FILTERING

Recommended

More Related Content

Similar to A LOCATION-BASED MOVIE RECOMMENDER SYSTEM USING COLLABORATIVE FILTERING (20)

More from ijfcstjournal (20)

Recently uploaded (20)

A LOCATION-BASED MOVIE RECOMMENDER SYSTEM USING COLLABORATIVE FILTERING