Predictive Model based on Sentiment Analysis for Peruvian SMEs in

the Sustainable Tourist Sector

Gianpierre Zapata

, Javier Murga

, Carlos Raymundo

Jose Alvarez

and Francisco Dominguez

Escuela de Ingeniería de Sistemas y Computación, Universidad Peruana

de Ciencias Aplicadas (UPC), Lima, Lima, Perú

Departamento de Informática, Universidad Carlos III, Getafe, Madrid, Spain

Facultad de Informática, Universidad Rey Juan Carlos, Mostoles, Madrid, Spain

Keywords: Sentiment Analysis, Big Data, Cloud Computing, Travel Management Process, Tourism Sector.

Abstract: In the sustainable tourist sector today, there is a wide margin of loss in small and medium-sized enterprise

(SMEs) because of a poor control in logistical expenses. In other words, acquired goods are note being sold,

a scenario which is very common in tourism SMEs. These SMEs buy a number of travel packages to big

companies and because of the lack of demand of said packages, they expire and they become an expense, not

the investment it was meant to be. To solve this problem, we propose a Predictive model based on sentiment

analysis of a social networks that will help the sales decision making. Once the data of the social network is

analyzed, we also propose a prediction model of tourist destinations, using this information as data source it

will be able to predict the tourist interest. In addition, a case study was applied to a real Peruvian tourist

enterprise showing their data before and after using the proposed model in order to validate the feasibility of

proposed model.

1 INTRODUCTION

In recent years tourism has become a powerful

transformative force that has had a decisive influence

on the lives of thousands of people. This is because it

is one of the main employment generation sectors in

the world (Scowsill, D., 2017) being the sector that

presented a growth of 3.1% in 2016, it contributes a

9.8% to the Gross Domestic Product (GDP)

worldwide. That is why the United Nations (UN;

2015) declares the year 2017 as the International Year

of Sustainable Tourism for Development, seeking to

encourage a change in policies, practices of tourism-

related businesses and to evaluate consumer behavior

in order to promote a more sustainable tourism sector.

One of the limitations of the current tourism

model is associated with the progressive growth of

productivity in the international market, facing

important challenges that emerge from the need to re-

establish its comparative advantages over other

competing destinations / countries. Faced with this

need, the search for new innovative solutions,

understood as new products, products, processes, new

marketing techniques or organizational improvement

to minimize costs and to differentiate the product of

the offered service and, ultimately, target all these

strategies to increased productivity of the system is

underway. In several tourism subsectors there is a

rapid process induction to new innovation

technologies that are changing the bases of

production and market structure. Without deviating,

the tourism sector seeks to evaluate consumer

behavior in order to be able to make decisions, so it

has taken an increased interest to exploit the countless

amount of data generated by the social networks,

where users shed their unbiased opinions of any

subject (Thomas H. Davenport, 2013).

However there are websites that try to cover the

chain value by providing services based on an

information catalog and supported by previous

opinions and experiences as seen on TripAdvisor,

Booking, HotelsCombined, Agoba, Kayak, among

others. But these platforms cannot measure the

interests of consumers and in a market where inter-

national competition is growing, a tourism forecasting

model must be able to deal with the development of

competitive advantages which will allow a better

performance to select a tourist destination.

Zapata G., Murga J., Raymundo C., Alvarez J. and Dominguez F.

Predictive Model based on Sentiment Analysis for Peruvian SMEs in the Sustainable Tourist Sector.

DOI: 10.5220/0006583302320240

In Proceedings of the 9th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (KMIS 2017), pages 232-240

ISBN: 978-989-758-273-8

In this sense, we propose a tourist interest

prediction model based on the sentiment analysis of

social networks and their results. This research is

divided into the section that describes the research

that are the basis for the proposed model, modeling,

implementation results, and conclusions.

2 BACKGROUND

2.1 Sentiment Analysis

Sentiment analysis refers to the use of word

processing and analysis tools to quantify the

sentiment expressed in words. This analysis helps

discover trends and reflect the real world in social

networks (Kiran Garimella et al., 2016). It also helps

predict different topics, in this case the tourist traffic

and time series of social networks.

In order to carry out this analysis, one can proceed

in different ways, within which one is making the

analysis based on a lexicon another being the

ontological analysis (Ali Marstawi et al., 2017). In the

first type of analysis, the classification is based on the

present words and the number of occurrences, taking

into consideration the semantic orientation of the

words, which must be added to a dictionary

classifying and rating them in negative and positive,

according to their intensity level (Anna Jurek et al.,

2015). In the ontological analysis a model is

generated by classifies the feelings based on concepts

(Pratik Thakor et al., 2015).

It is clear that the sentiment analysis is a discipline

that is generating interest in the scientific community,

resources have been produced such as Word2vec

(Dongwen Zhang et al., 2015), WordNet Affect,

SentiWord Net, among others (R. Linares et al.,

2015), which allows the generation of new research

cases and then a predictive analysis to examine in this

case the tweets and be able to analyze the origin

location, if it is classified in a positive or negative

way, among others established parameters (Eric

Baucom et al., 2013).

Thus, observing case studies is revealed that this

sentiment analysis allows us to make a prediction on

a specific topic as the case of the stock market

analysis, where values can be predicted according to

consumer trends (John Kordonis et al., 2016). This

allows us to affirm that this analysis is scalable to

diverse scenarios like the tourism sector.

2.2 Prediction Model

A predictive model is a way of observing the data that

one has, classifying it and then after an analysis, being

able to predict, based on the chosen parameters, a

future result. To do this, the data has to be

parameterized according to certain features chosen

according to the characteristics to be predicted. In

order to succeed, we must have the precise number of

variables, if insufficient variables are specified, the

model produces partial estimates, if there is an excess

of variables the model produces low accuracy

estimates. These variables will be defined according

to the chosen field, in this case study will be related

to tourism.

Figure 1: Technological Model.

The consulted research, shows a relation between

social networks and the information they store, which

gives us the ability to predict events, such as

presidential elections, tourism expectation or sales of

products. In addition, in a case study it was possible

to demonstrate how the power of social interactions

in the form of an advertising campaign and the tweets

generated by it can be analyzed to predict to a certain

extent the results of a future presidential campaign

(Andranik Tumasjan, et al. 2010).

Therefore, we can say that the expectation has

been seen as a point of reference that consumers use

to determine the satisfaction or assessment for the

performance of products or services (Chunyang

Wang et al., 2016), but it is contrasted with how the

expectation is related to the result of a number of

factors generated in the particular case of a film

shown in cinemas (Yang Liu et al., 2007). These

statements generate a motivation about tourism,

which can be classified into two forces that describes

how individuals are motivated to make a decision, in

this case a trip and how they are attracted to a

particular destination by its qualities. On this subject,

the image that each individual has of the destination

is closely related to the motivation he has to travel to

that particular place (Chunyang Wang et al., 2016).

3 PROPOSED MODEL

3.1 Analysis of the Model

According to the research that was carried out for the

development of the model and based on previous

researches that presents scenarios where the use of the

internal processes of the tourism sector can be used as

sources of information that allows the analysis of the

characteristics and behaviors of consumers (Carlos

Raymundo et al., 2017), a need was found in the

tourism sector. Thus the reason a user interest pre-

diction model is proposed. Under this approach, we

asked the following question: What are users looking

for? This can be answered by enhancing its decision-

making management and in the future to predict

tendencies according to the interests of the users.

Consequently, a model for predicting the tourist

interest is proposed as it helps to significantly

increase the decision making capacity. The main idea

of the proposal is to develop a model that has

technological aspects as well as to improve the

process of travel management by adding a new

channel to existing activities. It should be mentioned

that the prediction model is part of the technological

mode. Taking in consideration that in order to

develop it first one must know the business and after

making the process improvements, propose a

technological solution that meets the need. The model

presents 5 components: information input, functional

model, transformation and solution, technological

support and lastly benefits and indicators (See Figure

1). With each one being able to retrieve the necessary

information from the sector and recognize the needs

of the business, to later define the company's pro-

cesses and identify opportunities for improvement.

Based on this analysis, it is possible to define an

implementation model, one that can identify benefits

and business indicators. Finally, to improve processes

and automate them through a technological solution

that allows analyzing the information of social

networks´ history and it’s in real time and processes

to generate new business opportunities.

3.2 Components

3.2.1 Input Information

Although companies in the tourism sector have limited

capacity in their resources, as well as its processes,

each has its own complexity and information

extraction mechanisms (be it quantitative or

qualitative). For this reason, some of the types of

information needed to have a general framework with

respect to tourism companies are shown, including

information on the current situation of companies in

the tourism sector that allow the identification and

processes As-Is, in addition to looking for a way to

solve the problems that arise in the sector. Therefore,

an external investigation was carried out in order to

understand the following question: What is the most

commonly used information search channel? A survey

was carried out with a sample of 300 users distributed

in 4 well-defined groups and organized by age ranges,

from 17 to 24 years, from 25 to 40 years, from 41 to 60

years and 60 from more. The group with the highest

percentage as shown in Table 1 is the one that includes

ages from 25 to 40 years with 40%. Taking into

account that 55% are male and 45% female.

The search channel type resulting of the surveys

is demonstrated in a ranking Table 1.

Table 1: Search channels ranking.

Search cannel type %

Social Networks (Facebook, twitter, youtube,

among others).

46%

Travel and Tourism Companies 29%

Friend recommendation 16%

Newspapers and magazine 7%

Other 2%

Being the main interest of this study to investigate

the trends of travel and interest of users, the 46% of

the interviewed assess their travelling options with

social networks as a search channel. As a result,

companies in the sector can use social networks as a

new way for collecting information and making

decisions based on the needs and trends of the user.

3.2.2 Functional Model

Each company in the tourism sector is different and

has a complexity of its own according to its structure,

but all of them have shared aspects. That is why the

best practices can be gathered to model their

processes, data, applications and networks. The

functional model presented is segmented to have the

ideal processes in terms of travel management, taking

into account the addition of a new information search

channel. Thus, in the event that at the moment of

releasing information from the tourism sector the

company does not have the defined processes, the

functional model can be supported to fill that gap with

a structured and defined process for them, in addition

to contemplating all the data and documents to

determine the improvement of the processes of the

company. With regard to networks and applications,

it is the reflection of how to model these aspects that

will be modeled in the next stage.

3.2.3 Technological Support

The technological model is determined by several

levels as is grouped by applications and deployment

or networks. In turn, these are classified by

Application Servers, Database, Operating Systems

and Cloud Services (See Figure 2).

Figure 2: Deployment model.

The levels shown in this Technological Support

component are detailed from the bottom to the top,

since part of the need for a solution to be available in

the required time and at the same time that the costs

associated with the solution are adequate to the

situation in which the company in the sector is due to

its limited purchasing power. The levels of the

component are as follows:

Google Cloud Platform: Due to the situation that

companies in the tourism sector are found, it is

proposed the use of Cloud Computing services as a

lower level of infrastructure. Within the cloud-

oriented services there are types, which are: Software

as Services (SaaS), Platform as Services (PaaS) and

Infrastructure as Service (IaaS). For this reason, the

PaaS was chosen because of the need for a platform

to work on the development of the solution.

Likewise, we performed an analysis among the

other cloud service providers where we found well-

known companies as Azure, AWS, IBM SoftLayer

and Google. Among the aforementioned, Google

Cloud Plataform (GCP) it’s starting to have a greater

impact on the market as it has cheaper tariffs and the

ability to compete with the other suppliers. In this

sense, for this contribution, this provider is taken and

services called Cloud DataStore API and Machine

Learning Engine.

Framework, Flexibility and PaaS: For the

development of the system we use the ASP.Net MVC

4 framework, which provides a suitable development

environment for the application, since we can clearly

separate the data loads and use of them to be

displayed in the application. The flexibility is found

in the cloud platform, as it is scalable in nature, it

adapts the ingestion of data loads, either little or if

need be, massive amounts of data. The ability to

perform this task was of vital importance to be able to

build the system, since it adapts to variable data loads.

Within the cloud services we choose PaaS, since the

model covers the entire development cycle from

planning to implementation and testing. It was also

used because it reduces the costs of maintenance,

having a constant monitoring tool and the greater

availability of service use.

The application was developed based on these

three concepts, the framework gives us the ease to

develop and maintain; PaaS is in the integration of

database services, sentiment analysis and continuous

deployment and the flexibility is found in the

platform capacity to deal with variable amounts of

data.

Big Data - Sentiment Analysis: To know the interest

of the user in the places expressed in the Tweets it

was decided to carry out a sentiment analysis, which

was carried out through the Google Prediction API.

For the training of the model a package of 379

examples was used for the classification in four

categories, good, regular, tedious and bad. With this

information we can know in detail the user's feeling.

Data extraction and analysis was performed

following the steps of the following illustration (See

Figure 3).

Figure 3: Load and analysis pseudo-code.

The terms used in the pseudocode for extracting

data from Twitter and the subsequent sentiment

analysis are as follows. A Tweeter query object is

initialized, to which the search parameters are

assigned. In the case of the project, a list of HashTags

related to travel and tourism was used, as the program

has a period of periodic execution, the extraction of

tweets was limited from the previous day to the

present day, and finally, to determine the place from

where the tweet was launched, a flag was used to

collect only tweets containing geo-location. Once

collected, the list is sent to the Google Prediction API

where the sentiment analysis is performed and returns

the tweet's rating as a response.

Prediction Model: To create the prediction model, a

neural network of the Deep Feed-Forward type was

used to predict the tourist interest. The neural

network, takes the input data in the first layer and

after being classified by the hidden layers of the

network, returns the result that was defined according

to the model. This model was trained with data from

the sentiment analysis to have a more accurate

prediction. Taking into consideration that for the

development the wide and deep models were joined,

this gives the neural network a high capacity of

abstraction (See Figure 4). In the proposed model,

horizontal characteristics are taken as day or country,

in addition to using cross features to give the model

more certainty. Within the vertical characteristics

were taken the entry columns. These models are

linked to the data output of the linear regression of the

DNN * (Deep Neural Network). This union gives us

speed and accuracy of calculation, with which we

could predict the tourist interest.

Figure 4: Neural network pseudocode.

3.2.4 Transformation and Solution

This component will take as input the processes As-

Is improved and adapted to what is collected from the

functional model. At this point, To-Be processes are

defined which are recognized as the final processes.

It should also be taken into account that these

processes have process improvements by adding the

proposed new channel, an implementation model to

be developed in the company and the web tool that

supports it.

3.2.5 Benefits and Indicators

Then the section of Transformation and Solution,

should have results that make the tourism company

more efficient, so we detail the following benefits and

indicators of success:

Financial Stability: This indicator will allow the

company to control its expenses, besides allowing to

negotiate discounts with supplier companies by

having a clear idea of the consumers need.

Time Saving: This indicator will reduce the sales

decision making and creation of tourism packages

time, because it will allow to visualize the need of the

consumer in the system.

Sales Decisions Improvement: This benefit allows

the company to visualize forecasting analysis of the

user interest so that decisions can be made in a shorter

time without depending on the companies that control

the market.

Market Opportunity: This benefit is the most

important, because having a new information search

channel will give the company means to compete

against large companies by having real-time

information on the consumer’s needs.

4 VALIDATION

To validate the presented prediction model, we will

use a case study to show that the proposal

successfully solves the needs of tourism companies.

Var Query = new TweetQuery();

Query.HashTags = listHashTags();

Query.Since = Yesterday();

Query.Until = Today();

Query.HasGeotag = True;

ListTweets = SearchTwitter(Query);

Foreach Tweet in ListTweets{

TweetSent = Predict(Tweet);

}

InputColumn = CSV_Column (activity, day, place,

country);

WideColumn = [crossed (activity, place), day, country]

DeepColumn = [activity, day, place, country]

DNNLinearCombinedClassifier(WideColumn,

DeepColumn )

It can be said that it is validated as a part of the model

in order to ensure its correct functioning in the case

study. The components involved will be

transformation and solution for the practical

validation through the case study. In addition, it is

mentioned that for confidentiality purposes the real

company of the sector will be represented as OT

S.A.C.

4.1 OT S.A.C.

OT S.A.C is a small company dedicated to the sell

and distribution of tourist packages, a business that

has 10 employees and a monthly turnover of

approximately $ 29,447.85. It is located in the district

of Santiago de Surco, in the city of Lima, Peru. Its

main suppliers are the wholesale companies in the

sector, which give a list of packages for sale and

distribution. In turn, this company sells custom

packages according to customer's requirement.

This company, since its beginnings, according to

the national regulations regarding the package

distributors, was favored by the client portfolio that

already had because of related businesses. But as new

competing companies appeared in the sector with

same products, it led to a price contest that gave a

sudden growth of companies in this area.

Under this circumstance, the company owners and

managers focused on the vision of using new

technologies and trends to keep the company's sales

afloat. That is why it is proposed to this company to

implement a prediction model in collaboration with

it’s the workers, as well as the direct managers of the

company.

4.2 Implementation

The schematic to implement the proposed model can

be visualized in Figure 5, which details the necessary

steps that the company OT S.A.C has to implement

for this proposal, the graph is read from top left to the

bottom right of the graph.

Figure 5: Technological model implementation.

Face Meetings: The implementation was made

through weekly meetings with the owners and

workers involved, during this period were explained

the company processes. OT S.A.C has the following

processes:

- Travel Management

- Sales

- Logistics

- Accounting

- IT Management

- Legal

Processes As-Is: When carrying out an information

survey, it was noted that the company OT SAC did

not have well-defined processes, so it was not

possible to design an As-Is, but it was based on all the

functions performed by each worker to recognize

their activities and then the processes they performed.

In the process of developing processes As-Is, we

detect a dependency in wholesalers companies that

control the market, so that a new information

collection channel is presented where companies can

make decisions based on the needs of users and trends

shown on social networks.

To-Be Processes: With the identified processes, a

detailed work was done to be able to define which key

processes the company should carry out to have an

optimal performance, without any voids that can

affect it. Reason why it was based on the functional

model to reference the components and thus be able

to contrast their information gaps.

In this particular case, we identify the following

wrongdoings inside the Travel Management process.

- The process does not have proper

documentation or any additional information

other than the mayor market controlling

companies.

- The input information they have is outdated,

as is based on the historical record of other

companies.

- The user requirements response time is

limited to the response that can be provided

by another company that has the information.

These are the main findings compared to the

Functional Model. To solve these problems we

introduced a new information search channel within

the process supported by an application developed for

the company. In this way the company can make real-

time decisions according to the user's tendencies and

the predictions that tool generates.

Implementation Model: It is the graphic representa-

tion of the phases that the company must carry out in

order to implement the proposed model. Taking into

account that training is planned in the company to

perform a correct management of the organizational

change and prepare the workers to use the new tools

and technologies in this sector.

Predictive Model: Taking into account the benefits of

implementing a sentiment analysis along with a

predictive model, the solution was implemented in

the company OT S.A.C. To implement the model, the

training was carried out with test data from the

sentiment analysis. Figure 6, shows the learning

process of the neural network through each iteration.

Once a job is sent to train the neural network with the

model, it makes use of the neurons to be able to infer

the results. The progressive increase of accuracy in

each step of the training phase, demonstrates the

learning capacity of the network.

Figure 6: Accuracy graph.

Web Tool: To develop the prediction model for the

company OT SAC, a web tool was devised in order

to research and collect information from a social

network, in this case twitter. This information will

analyze the users' comments and identify their

preferences, tastes and mood, etc. This will provide

rating information of a city, country and travel

package, as well as to generate input data to

subsequently predict travel destinations with a greater

acceptance range.

For this part, we used the concepts of Sentiment

Analysis presented in section 3.2.3; which allowed us

to analyze the tweet of users worldwide. This helped

us to create a database with the following fields ID,

Activity (Museum, Beach, Trekking, Adventure

Tourism), Country, City (Place), Year, Day, Month,

Range (Good, Bad , Regular, Tedious) and finally the

tweet itself to analyze. (See Figure 7a and 7b).

Figure 7a: Database of analyzed tweets. Part 1.

Figure 7b: Database of analyzed tweets. Part 2.

With these results and using the prediction model

integrated into the web tool, it is possible to predict

the trend and activities the users will have in different

parts of the world (see Figure 8). In addition, we can

identify which days are highest rated among users to

plan a trip (see Figure 9). Taking it into account, a

metric to analyze this information was constructed,

dividing it into three categories as described in the

following lines, good, average and bad, as seen in

Table 2. Furthermore it grants the user a view of the

raked packages from other users. (See Figure 10).

Figure 8: User tendency map.

Table 2: Ranking metrics by city, country or package.

Metric Category

[0- 0.25] Bad

[0.251- 0.75] Regular

[0.751- 1] Good

Figure 9: Rating reports by city.

Figure 10: User trends map showing activities by country.

4.3 Results

Once implementation of the technological model in

OT SAC was completed, it can be observed that the

benefits and indicators in the processes using the new

trends and technologies assure a greater quantity of

sales. It is also considered that using tools of massive

data analysis, in accordance to International Data

Corporation (IDC) (Carlos Raymundo et al., 2017),

progressively increase sales by a 12% margin. In

order to test this assertion, an analysis of the loss

indicator due to logistical expenditures was made, in

Fig. 15. It presents all the logistical movements that

are considered expenses. The 57% of the total annual

expense and costs are due to two mayor causes. The

first cause is the difference in purchase prices,

amounting to 17.6% of the total. The other reason is

the unnecessary purchases taking a 39.9% of this

total. After the implementation of the technological

model and thanks to the support provided by the

prediction model, the company OT S.A.C was able to

reduce their logistical expenses as explained in Table

3. Considering a 40% threshold.

Table 3: Loss caused by logistical expenses.

Loss caused by logistical expenses ($)

Before After

Annual

expenses

cause by price

difference.

4824.24

Annual

expenses

cause by price

difference.

3600

Annual loss

cause by

purchases.

11,104.3

Annual loss

cause by

purchases.

7773.01

Loss %

57%

Loss %

27%

Indicator

Critic

Indicator

Positive

Upon a closer inspection of the results, the

following can be said:

- Before the implementation of the technologi-

cal model, the company had a large margin of

loss in the purchase and separation of

packages. The reason for this behavior can be

explained because the company did not have a

clear reference on the need of its customers.

Now with the model, not only can company

recognize the customers’ need, it can also

negotiate with its suppliers new rates that will

generate a higher gross sales income.

- The feasibility study (see Fig. 11), shows the

investment and recovery periods projected

after the model implementation. It shows a $

24,212.92 investment over 12 month. It

includes both the workforce cost for the

implementation and cloud services costs.

Culminating that period the only cost would be

the cloud services. At this time frame the

company will have an estimated profit of $

25,953.93 due to insured sales. These last

values are made possible by having an

investment return of approximately 11

months, with a ROI of 114%, being totally

profitable for a company in the tourism sector.

Figure 11: Feasibility study.

- We arrive at the conclusion that the joint

implementation of a technological and

predictive model will bring benefits to the

tourism sector as it generates new market

opportunities.

5 CONTINUITY

In future proposals new instances of the business

model could be applied and be escalated across the

company, not only generating new sales business

channels, allowing us to make more accurate

decisions, but it could also be used in other

administrative process, reducing their cost.

In addition, it opens a myriad of possibilities for

related and unrelated fields where the analysis of

massive social network data could reduce the

operating costs and increase the company's revenue.

6 CONCLUSION

To answer the problem of controlling the logistical

expenses based on a technological model through a

reengineering, process improvement, the use of

Cloud Computing and Big Data is needed to support

a company in the tourism sector and to obtain both

financial and operational stability.

It was demonstrated that the use of technologies

such as Cloud Computing and Big Data oriented in a

free software guideline is reliable for companies in

the tourism sector, since the cost of these services are

adequate to the purchasing capacity of small and

medium enterprises.

Finally, it was demonstrated that a control vision

provides utilities for an industry in an indirect way,

reducing the losses in logistical expenses, saving time

creating packages and money as a result.

REFERENCES

Scowsill, D., 2017. Global Economic Impact & Issues

2017. Economic impact research. World Travel &

Tourism Council. Retrieved June 9, 2017 from

https://www.wttc.org/-/media/files/reports/economic-

impact-research/2017-documents/global-economic-

impact-and-issues-2017.pdf

UN, 70/193, General Assembly, 70th Sess, Supp No 81, UN

Doc A/RES/70/193 (2015).

Thomas H. Davenport. 2013. At the Big Data Crossroads:

turning towards a smarter travel experience.

Kiran Garimella, Gianmarco De Francisci Morales,

Aristides Gionis, and Michael Mathioudakis. 2016.

Quantifying Controversy in Social Media. In

Proceedings of the Ninth ACM International

Conference on Web Search and Data Mining (WSDM

'16). ACM, New York, NY, USA, 33-42. DOI:

https://doi.org/10.1145/2835776.2835792

Ali Marstawi, Nurfadhlina Mohd Sharef, Teh Noranis

Mohd Aris, and Aida Mustapha. 2017. Ontology-based

Aspect Extraction for an Improved Sentiment Analysis

in Summarization of Product Reviews. In Proceedings

of the 8th International Conference on Computer

Modeling and Simulation (ICCMS '17). ACM, New

York, NY, USA, 100-104. DOI: https://doi.org/

10.1145/3036331.3036362

Anna Jurek, Maurice Mulvenna, Yaxin Bi, Improved

lexicon-based sentiment analysis for social media

analytics, Security Informatics, Volume 4, 2015, ISSN

2190-8532, http://dx.doi.org/10.1186/s13388-015-

0024-x

Pratik Thakor, Sreela Sasi, Ontology-based Sentiment

Analysis Process for Social Media Content, Procedia

Computer Science, Volume 53, 2015, Pages 199-207,

ISSN 1877-0509, http://dx.doi.org/10.1016/j.procs.

2015.07.295

Dongwen Zhang, Hua Xu, Zengcai Su, Yunfeng Xu,

Chinese comments sentiment classification based on

word2vec and SVM, Expert Systems with Applications,

Volume 42, Issue 4, 2015, Pages 1857-1863, ISSN

0957-4174, http://dx.doi.org/10.1016/j.eswa.2014.09.

011

R. Linares, J. Herrera, A. Cuadros and L. Alfaro,

"Prediction of tourist traffic to Peru by using sentiment

analysis in Twitter social network," 2015 Latin

American Computing Conference (CLEI), Arequipa,

2015, pp. 1-7. DOI=https://doi.org/10.1109/CLEI.

2015.7360051

Eric Baucom, Azade Sanjari, Xiaozhong Liu, and Miao

Chen. 2013. Mirroring the real world in social media:

twitter, geolocation, and sentiment analysis. In

Proceedings of the 2013 international workshop on

Mining unstructured big data using natural language

processing (UnstructureNLP '13). ACM, New York, NY,

USA, 61-68. DOI=http://dx.doi.org/10.1145/2513549.

2513559

John Kordonis, Symeon Symeonidis, and Avi Arampatzis.

2016. Stock Price Forecasting via Sentiment Analysis

on Twitter. In Proceedings of the 20th Pan-Hellenic

Conference on Informatics (PCI '16). ACM, New York,

NY, USA, Article 36, 6 pages. DOI=https://doi.org/

10.1145/3003733.3003787

Andranik Tumasjan, et al. 2010. Predicting Elections with

Twitter: What 140 Characters Reveal about Political

Sentiment. In Proceedings of the Fourth International

AAAI Conference on Weblogs and Social Media.

(Washington, DC, USA, May 23-26, 2010).

Chunyang Wang, Hailin Qu, Maxwell K. Hsu, Toward an

integrated model of tourist expectation formation and

gender difference, Tourism Management, Volume 54,

2016, Pages 58-71, ISSN 0261-5177, http://dx.doi.org/

10.1016/j.tourman.2015.10.009

Yang Liu, Xiangji Huang, Aijun An, and Xiaohui Yu. 2007.

ARSA: a sentiment-aware model for predicting sales

performance using blogs. In Proceedings of the 30th

annual international ACM SIGIR conference on

Research and development in information retrieval

(SIGIR '07). ACM, New York, NY, USA, 607-614. DOI=

https://doi.org/10.1145/1277741.1277845

Carlos Raymundo, Cesar Mérida, Richer Rios and Alfred

Kobayashi, 2017, Modelo arquitectónico de

información para una plataforma de Big Data para el

Sector Turístico. In Proceedings of the Décima Sexta

Conferencia Iberoamericana en Sistemas, Cibernética

e Informática (CISCI 2017), Orlando, Florida, EE.UU.

Anon. 2017. Big Data and Business Analytics Revenues

Forecast to Reach $150.8 Billion This Year, Led by

Banking and Manufacturing Investments, According to

IDC. (March 2017). Retrieved June 10, 2017 from

http://www.idc.com/getdoc.jsp?containerId=prUS4237

1417.