Location-based Fast Recommendation Social Network

Iaakov Exman and Erez Nagar

Software Engineering Department, The Jerusalem College of Engineering - Azrieli

POB 3566, Jerusalem, 91035, Israel

Abstract. Location-based response to queries has the potential to improve the

quality of recommendations. But in order to actually improve quality one needs

in addition to location-based, also time-dependent evaluation of the basis for

recommendation. We propose the usage of opinion-ontologies – a short and

sharp loosely structured subjective ontology – that can be transmitted in real-

time and be integrated with previously received opinion-ontologies. This

approach has been designed and implemented into a system with wireless

smartphone clients and servers supporting information integration, evaluation

and recommendation formulation. The approach has been tested with a

prototype composed of existing software packages playing the needed roles in

the designed system.

Keywords. Location-based, Time-dependent, Opinion-ontology, Social

network, Recommendation, Conflicting opinions Integration, Wireless

Smartphone.

1 Introduction

Suppose that during a trip to a region in the north, one approaches dinner time. One

needs to decide upon a convenient restaurant, and then find a hotel to spend the night,

without forgetting about a gas station to refill the car tank. The old way to do that is to

buy in advance a printed tourist guide of the area. The modern way is to ask other

people that have travelled around, or are just on location, about a place to eat with a

good recommendation.

But are the recommendations reliable?

One should be able to compare and evaluate different recommendations.

This work describes a system for location-based queries, with capabilities to

integrate and evaluate the most fitting overall recommendation. The approach uses

opinion-ontologies [6]. These are quite short, loosely structured subjective ontologies

whose purpose is to express opinions that serve as recommendations. As these are

very short, one swiftly transmits and manipulates them, to obtain results in real-time.

In this paper we describe a recommendation system based on the opinion-

ontologies approach. It was designed and implemented with smartphone clients and a

server supporting information integration and overall personalized recommendations.

Exman I. and Nagar E..

Location-based Fast Recommendation Social Network.

DOI: 10.5220/0005167600550062

In Proceedings of the 5th International Workshop on Software Knowledge (SKY-2014), pages 55-62

ISBN: 978-989-758-051-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

1.1 Related Work

Location-based approaches have been proposed and implemented in several specific

contexts, one of them being crowdsourcing. Alt et al [1] and Bulut et al. [2] extend

crowdsourcing to location-based problems. Mashhadi and Capra [11] discuss quality

control for real-time ubiquitous crowdsourcing.

Mobility issues, especially related to social networks, called the attention of

several researchers. Chow et al. [5] investigate location-based social networking

services. Lenders et al. [8] deal with the issue of location-based trust for mobile user-

generated content. Schuster et al. [13] develop a service-based mobile real-time

collaboration application for social networks.

Opinion and ontologies have been discussed at length in the literature. Chang et al.

[4] refer to reputation ontologies. They deal with “Trustworthiness of Opinion

Ontology”. Li and Du [9] research ontology-based opinion leader identification for

marketing in online social blogs. Lu et al. [10] propose using structured ontologies to

organize scattered online opinions.

Cambria et al. [3] describe a public semantic resource for Opinion Mining, called

SenticNet. Sentic Computing enables analysis of very short documents, say one

sentence. Zobel et al. [14] use a majority rule to merge conflicting knowledge bases.

In the remaining of the paper we introduce the domain of location-based

recommendation (section 2), describe opinion-ontologies that are used in our solution

(section 3), overview the software architecture and implementation of our tool

(section 4), discuss case studies as a demonstration of the approach (section 5) and

conclude with a discussion (section 6).

2 Location-based Recommendation

Location-based recommendation means that people who are or have been in a certain

location express their opinion on a requested service or institution. The receiver of the

recommendation, without being in the referred location, has to evaluate the

recommendations, integrate them and take a decision on the action to be taken.

2.1 Locations: Acquisition, Storing and Display

Half of the problem has to do with locations. A location-based mobile

recommendation system has to be able to do the following functions:

 Acquisition – of the recommender locations which entitle him to express

his/her opinion; this certainly involves some kind of GPS functionality;

 Storing/Retrieval – the set of locations per recommender should be stored in

some standard format; retrieval should be consistent with the storing function;

the efficiency issue refers to what extent locality (the client) as opposed to a

central server is important;

 Display – use a map with zoom_in/zoom_out capabilities.

2.2 Recommendations: Acquisition, Integration and Evaluation

Recommendations should be received from several people willing to express their

opinions. Either the recommendation receiver or an automatic system or a

combination of both should be able to evaluate the quality and reliability of the

recommendation. An integration function should combine all the relevant

recommendations to extract a single decision, or a ranked small set of possible

decisions. In more details:

 Acquisition – a standard format recommendation should be acquired, stored and

eventually retrieved from memory, locally or from a central server, similarly to

the associated locations;

 Integration – the recommendation receiver or an automatic sub-system should

take the individual standard format recommendations and obtain an integrated

recommendation with the same format, thus facilitating decisions;

 Evaluation – a grading system, based upon an accepted but updatable scale,

should associate a pure number with recommendations, either the individually

received ones, or those which result from integration.

3 Opinion-Ontologies

In this work we adopted “opinion-ontologies” [6] as the standard format for

recommendations. Its main motivation is integration of any number of opinions into a

short data structure, to guarantee efficiency. Next, we shortly describe the standard

format and the integration possibilities.

3.1 Structure for Efficiency

An opinion-ontology is a quite short – with a strictly limited amount of characters –

loosely structured piece of information. An opinion-ontology has two kinds of terms:

class terms, with at most four letters and free-text words. Class terms are not reserved

words and are explained and systematically used within our application.

An example of an opinion-ontology regarding a coffee-shop is shown in Fig. 1. Its

class terms are marked by italic-bold (here stressed in red color).

Fig. 1. Sample Opinion Ontology – It refers to a coffee-shop named Maestro.

This opinion-ontology refers to a coffee-shop named Maestro, which is located

near the sea-shore. The food served includes long sandwiches, with a variety of

sauces and among its specialties one finds a perfect macchiato. The coffee-shop

decoration is just functional. But, as a negative point, the service is not so fast.

A graphical schematic representation of the same opinion-ontology is seen in Fig.

2. Similar graphs can be obtained from ontology editor tools such as Protégé [12].

“opinion-ontolo

#17: co

Maestro loc se

-shore

ood long

sandwiches, sauce variety, perfect macchiato déco functional serv

not so express.”

Fig. 2. Sample Coffee-Shop Opinion Ontology – In a schematic graph which has classes

represented by blue ellipses, and instance values represented by yellow rectangles. Maestro is

the name of an instance of a coffee-shop.

3.2 Integration

Integration should take into account two kinds of sometimes opposing concerns:

a. Semantic Concerns – such as mismatches caused by different terminologies,

term ambiguities, contradictions;

b. Size Concerns – i.e. integration of opinion-ontologies without increasing size

above the allowed number of characters, say by avoiding excess words.

Our fundamental integration principle is: whenever there is a conflict between

semantic and size concerns, the size concerns are prevalent. If size may still

accommodate expansion of opinion-ontologies, then semantic concerns may be taken

into account differently from the case in which expansion is not possible anymore.

Here are some specific rules derived from the fundamental principle:

 Reinforcement – when two or more different recommendations reinforce the

same opinion, use a numerical weight to express it, say *3 (multiplied by 3)

means that the particular preceding free-text word appeared among three of

the integrated opinion-ontologies;

 Contradictions – opposing opinions reflected in particular expressions and

their negations, may be kept, if size allows it; if size does not permit, use a

combination of positive and negative weights, say *-4 *6 (four negations and 6

affirmations) after the chosen expression;

 Different Terminologies – choose the most common term among the different

ones; if this information is lacking choose one term arbitrarily;

 Ambiguities – disambiguate terms using the several opinion-ontologies before

they are integrated;

 Excess Words – among the union of the sets of words of the individual

opinion-ontologies, discard the less surprising (less informative) words.

4 Software Architecture and Implementation

In this section we shortly describe the system software architecture and

its implementation.

4.1 Software Architecture

The system software architecture is schematically shown in Fig. 3. It basically has

three upper modules:

1. Central Server – that controls communication with any client and stores

recommendation information in a database;

2. Location dedicated server – that takes care of location related information;

3. Client – a mobile module carried by any user of the system.

Fig. 3. System Software Architecture – It has a central server with a database, another server

dedicated to deal with locations and a mobile client that may be replicated as needed.

4.2 Implementation

A prototype system was implemented by a combination of existing services and

completely new coded software, to allow actual testing of the whole approach.

The location dedicated server is based upon the API of an existing social network

service – Foursquare [7]. Users “check in” in venues they visited and the location

information is saved for later use.

The central server manipulates opinion-ontologies for storing, retrieval, integration

and evaluation. Communication with clients is made efficient by using GCM (Google

Cloud Messaging).

The client is an application on an Android smartphone operating system version

4.2. Both the server and client were written in Java. Testing was done first on an

emulator and then with real smartphones.

5 Case Studies

We have tested our system in a series of case studies. Two of them are presented here:

the “theater show” and “automobile garage” contexts.

5.1 The Theater Show Context

Typical questions regarding temporary events, besides overall recommendation

concerning quality and taste, are “did it start already?”, “are there available tickets

yet?”. In Fig. 4 one sees the Theater Show opinion-ontology used.

Fig. 4. A Theater Show opinion-ontology – It displays the theater house, location, kind of

show, cast and timing notification. Four opinion-ontologies integrated into opinion-ontology

#54 stated that the show is a comedy.

If the only question of interest is “are there available tickets yet?” one can just use the

relevant items as shown in Fig. 5.

Fig. 5. Theater Show opinion-ontology for “no-tickets” – It displays only the relevant class

terms and respective information.

Although one could think that this is a waste of space and time – why not just send

say an SMS with “no-tickets left tonight”? – an opinion-ontology has a fixed format

that may be filled-in later and reused.

5.2 The Automobile Garage Context

Here the typical situation is as follows: one is travelling by car and unexpectedly

appears a problem that allows one to continue the trip, but not for long. One needs

assistance from a garage as soon as possible to repair the car.

Fig. 6 shows a generic opinion-ontology for the automobile garage context. It

refers to professional specialization, types of vehicle and service quality and

characteristics, besides the expected location.

Fig. 6. Generic opinion-ontology for car repair garage – It shows garage name, location,

specialization, vehicle types and service considerations. In this particular opinion the service is

unreliable.

Since the “unreliable” qualification is ambiguous – does it refer only to timing or

to the overall quality of service – it could generate further inquiries from the receiver

of the recommendation.

“opinion-ontolo

#54:

ous The Arena loc Broadway show

comedy *4, bourgeois, light cast the troupe time no tickets left

tonight.”

“opinion-ontolo

#54:

ous The Arena loc Broadway show -

cast - time no tickets left tonight.”

“opinion-ontolo

#36: Gar

Peter’s Automotive loc 5 km north

spec electrical, mechanics vehc cars, trucks serv on the spot, same

day, unreliable”

6 Discussion

This work proposes the direct use of short loosely structured opinion-ontologies,

instead of completely free natural language text, for location-dependent and fast

recommendations in mobile social networks. The idea is that with structured very

short ontologies one may still make inferences in almost real-time.

The opinion-ontologies have a standard format, but it is loosely structured in the

sense that there are no reserved keywords and the number of instance values for each

class-terms is not fixed.

One of the interesting implications of making inferences upon opinion-ontologies

is that one gets implicit opinions. For instance, in the coffee-shop opinion-ontology in

Fig.1, the decoration is said to be “functional”. There is no mention of pricing. But an

inference rule could state that “functional furniture implies not-expensive”, meaning

that the referred coffee-house is not pricey. In general one should not expect

recommendations to state negative or positive opinions just by means of explicit

positive (nice) or negative (nasty) words.

Integration of various opinion-ontologies takes into account both semantic and

size conservation concerns. Our fundamental principle is that when there are conflicts

between semantic and size concerns, size concerns are prevalent. Specific integration

rules are derived from this principle.

6.1 Future Work

The current system is a prototype. Once it is comprehensively tested, we may be able

to build a second version, more mature system. System mobility and real-time

capabilities should be tested for significantly larger quantities of clients and

recommendation traffic.

Concerning opinion-ontologies, these are first tentative explorations of their

possibilities. Extensive investigation of its various aspects – communication, storage

and retrieval, inferences – is necessary to obtain a deeper and wider appraisal of their

use.

6.2 Main Contribution

The main contribution of this work is the usage of short loosely structured opinion-

ontologies for fast recommendations within a mobile social network.

References

1. Alt, F., Shirazi, A. S., Schmidt, A., Kramer, U. and Nawaz, Z. “Location-based

Crowdsourcing: Extending Crowdsourcing to the Real World”, in NordiCHI 6

Nordic

Conference on Human-Computer Interaction: Extending Boundaries, Pages 13-22, (2010).

http://www.medien.ifi.lmu.de/pubdb/publications/pub/alt2010nordichi/alt2010nordichi.pdf

2. Bulut, M. F., Yilmaz, Y.S. and Demirbas, M., “Crowdsourcing Location-based Queries”, in

IEEE Pervasive Computing and Communications Workshops (2011).

http://www.deliberslabs.com/basepapers/Dot_Net_Projects/IEEE_2011_Dot_Net_Projects/

Networking/61140.pdf

3. Cambria, E., Speer, R., Havasi,C. and Hussain, A., “SenticNet: A Publicly Available

Semantic Resource for Opinion Mining”, in AAAI Fall Symposium, pp. 14-18, (2010).

4. Chang, E., Hussain, F.K. and Dillon, T., “Reputation ontology for reputation systems”, in

SWWS Int. Workshop on Web Semantics, pp. 957-966, Springer-Verlag (2005). DOI:

10.1007/11575863_117.

5. Chow, C. Y., Bao, J. and Mokbel, F. M., “Towards Location-based Social Networking

Services”, in 2

ACM SIGSPATIAL International Workshop on Location Based Social

Networks, pp. 31-38, (2010). http://www.cs.cityu.edu.hk/~chiychow/papers/

LBSN_2010.pdf

6. Exman, I., “Opinion-Ontologies: Short and Sharp”, accepted for publication in 6

KEOD,

Int. Conf. on Knowledge Engineering and Ontology Development (2014).

7. Foursquare – http://en.wikipedia.org/wiki/Foursquare.

8. Lenders, V., Koukoumidis, E., Zhang, P. and Martonosi, M., “Location-based Trust for

Mobile User-generated Content: Applications, Challenges and Implementations”, in 9

Workshop on Mobile Computing Systems and Applications, pp. 60-64, (2008). http://kou

koumidis.com/papers/Lenders_Hotmobile_2008.pdf

9. Li, F. and Du, T.C., “Who is talking? An ontology-based opinion leader identification

framework for word-of-mouth marketing in online social blogs”, Decision Support

Systems, Vol. 51, (1) pp. 190-197, (2011). DOI: 10.1016/j.dss.2010.12.007.

10. Lu, Y., Duan, H., Wang, H. and Zhai, C.X., “Exploiting structured ontology to organize

scattered online opinions”, in Proc.COLING’10, 23

Int. Conf. on Computational

Linguistics, pp. 734-742, (2010).

11. Mashhadi, A. F., Capra, L. “Quality Control for Real-time Ubiquitous Crowdsourcing”, in

UbiCrowd 2

Int. Workshop on Ubiquitous Crowdsourcing, pp. 5-8, (2011). http://

www0.cs.ucl.ac.uk/staff/L.Capra/publications/ubicrowd11.pdf

12. Protégé - http://protege.stanford.edu/.

13. Schuster, D., Springer, T. and Schill, A., “Service-based Development of Mobile Real-time

Collaboration Applications for Social Networks”, in IEEE Pervasive Computing and

Communications Workshops, (2011). http://www.rn.inf.tu-dresden.de/uploads/Publika

tionen/percol2010.pdf

14. Zobel, C.W., Rees, L.P. and Rakes, T.R., “Automated merging of conflicting knowledge

bases, using a consistent, majority-rule approach with knowledge-form maintenance”,

Computers & Operations Research, Vol. 32, Issue 7, pp. 1809-1829 (July 2005).