Use of Text Mining Techniques for Recommender Systems

Yanelys Betancourt

and Sergio Ilarri

2 a

Universidad de Zaragoza, Ediﬁcio Ada Byron, Mar

ıa de Luna, 1, 50018 Zaragoza, Spain

I3A, Universidad de Zaragoza, Ediﬁcio Ada Byron, Mar

ıa de Luna, 1, 50018 Zaragoza, Spain

Keywords:

Recommender Systems, Text Mining, Opinion Mining.

Abstract:

Recommender systems help users to reduce the information overload they may suffer in the current era of

Big Data, by offering them recommendations of relevant items according to their tastes/preferences and/or

context (location, weather, time of the day, etc.). We argue that text mining techniques can be exploited for the

development of recommender systems. Thus, they can be applied to detect user preferences (user proﬁling)

and also to extract context data. For this purpose, text mining can be applied on user reviews, text descriptions

associated to the items, and other texts written by the user (e.g., posts in social networks). In this paper,

we provide an overview of works exploiting text mining techniques in the ﬁeld of recommender systems,

characterizing them according to their purpose and the type of textual data analyzed.

1 INTRODUCTION

Nowadays, recommender systems (RS) (Ricci et al.,

2011; Ricci et al., 2015) have become very popu-

lar, as they can help users to reduce the information

overload they may suffer in today’s Big Data era,

where advanced data management techniques are re-

quired. These systems provide recommendations to

the users according to their tastes and preferences, al-

lowing them to ﬁlter, from a great amount of differ-

ent types of items (e.g., music, movies, news, books,

places, services, shopping areas, restaurants, points

of interest, etc.), those that can be of special rele-

vance to them. Given their current specialization,

recommender systems keep being a hot area of re-

search. In mobile environments in particular, rec-

ommender systems that take into account not only

the user’s preferences but also their context (location,

time, weather, trafﬁc conditions, etc.) are of particular

interest, which has given rise to the so-called context-

aware recommender systems (CARS) (Adomavicius

and Tuzhilin, 2011).

We argue that the use of text mining tech-

niques (Berry, 2004; Gupta et al., 2009) can be ex-

ploited for the development of recommender systems

from various perspectives. Popular text mining tasks

include text clustering, text classiﬁcation, informa-

tion extraction, text summarization, sentiment analy-

sis/opinion mining, etc. For example, sentiment anal-

https://orcid.org/0000-0002-7073-219X

ysis techniques could be applied to quantify user pref-

erences based on their comments expressed in natu-

ral language or even to detect possible inconsisten-

cies (opinion spam) if a user provides both a textual

comment and a numerical rating (mining of opinions

about items). In addition, they can be used to ex-

ploit textual descriptions of the different elements to

recommend (mining of items’ textual data). Finally,

the possible analysis of other texts written by the

user himself/herself (e.g., in blogs, social networks,

his/her agenda, etc.) may reveal information about

his/her preferences, interests, and even his/her con-

text.

A relevant complement of text mining techniques

is the use of semantic techniques, usually based on

the use of ontologies. As an explicit formal speci-

ﬁcation of a shared conceptualization (Gruber et al.,

1993), an ontology contains a shared vocabulary to

denote the types, properties and interrelations of those

concepts (Subramaniyaswamy et al., 2019). Besides

the possibility to use ontologies as a support for data

mining tasks such as information extraction, ontolo-

gies can also be exploited directly by a recommender

system (Subramaniyaswamy et al., 2019; Tarus et al.,

2018), for example to establish the relationships be-

tween users and their preferences on the recommen-

dation topic (Tarus et al., 2018). Ontology-based

recommender systems using text mining often use a

topic ontology with associated controlled vocabulary

to identify topics from text messages written in nat-

780

Betancourt, Y. and Ilarri, S.

Use of Text Mining Techniques for Recommender Systems.

DOI: 10.5220/0009576507800787

In Proceedings of the 22nd International Conference on Enterprise Information Systems (ICEIS 2020) - Volume 1, pages 780-787

ISBN: 978-989-758-423-7

ural language (comments, descriptions, opinions, re-

views, etc.).

In this paper, we provide an overview of works

that exploit text mining techniques in the ﬁeld of rec-

ommender systems (see Figure 1). The structure of

the rest of this paper is as follows. In Section 2, we fo-

cus on works in the ﬁeld of recommender systems that

apply text mining techniques on user reviews. In Sec-

tion 3, we consider recommender systems that per-

form text mining on other texts written by the user. In

Section 4, we consider works using textual data about

the items themselves. Finally, in Section 5, we sum-

marize our conclusions and future work.

2 RECOMMENDER SYSTEMS

USING TEXT MINING ON

USER REVIEWS

In this section, we focus on text mining of user re-

views (see Table 1). Sometimes users provide opin-

ions about the items they consume, through text de-

scriptions posted in blogs, microblogs (like Twitter),

product review websites, chat rooms, online feed-

back systems of companies, and different user com-

munities or social networks (such as Facebook, Twit-

ter, LinkedIn, or Instagram). These opinions contain

very valuable information to detect the interests of

the users, which can be exploited for recommenda-

tion purposes (Roul and Arora, 2019; Kim and Chun,

2019).

As these opinions are provided in natural lan-

guage, they cannot be directly exploited to learn the

user preferences and recommend the user other items

that he/she is expected to like. Therefore, a typical

use of text mining for recommender systems is to try

to transform textual user reviews into scores (e.g., in

a range of 0 to 5) or predeﬁned categories such as

positive opinion, neutral opinion, or negative opin-

ion, that can be used to build the user-item matrices

that recommender systems use in the recommenda-

tion process; speciﬁcally, sentiment analysis or opin-

ion mining techniques (Liu and Zhang, 2012; Hegde

and Padma, 2017) can be applied.

In (Aciar et al., 2007), text mining techniques are

used to transform opinions into structured data, us-

ing a translation ontology that encompasses both the

user’s skill and the user’s experience with the product.

Scores for each feature are computed in order to rank

the products according to the comments available

about that product. The system requires users to ex-

plicitly request a recommendation/assessment about a

speciﬁc product they are interested in, and they have

to select the features that interest them most. Beyond

that, it should be noted that this type of recommenda-

tion is not personalized, as the individual preferences

of the user are not considered when making the rec-

ommendation. The proposal takes into account the

level of experience of the reviewers to ensure a reli-

able recommendation.

The reliability of the reviews is also a key as-

pect considered in other proposals. Thus, (Abuein

et al., 2017) presents the design and implementa-

tion of a recommender system based on informa-

tion provided by trusted people using the concept of

trust level (LoT) (Abuein et al., 2016). In (Roul and

Arora, 2019), the reviewer’s credibility is also consid-

ered, but in this case the focus is on generating sum-

maries of multiple reviews for a given product and the

credibility (computed by analyzing how much the re-

viewer’s rating deviates from most people’s ratings)

is used to decide on the importance of a sentence. For

the automatic summarization of user reviews, fuzzy

c-means clustering is used to group similar sentences

and then one representative sentence from each group

is selected in order to compose the ﬁnal summary.

This work’s goal is to generate a representative sum-

mary of reviews to provide along with the recommen-

dation of a product, rather than proposing a novel rec-

ommender system.

In (Musto et al., 2019), the authors present an ap-

proach to generate natural language justiﬁcations to

support the suggestions returned by a generic recom-

mendation algorithm, based on natural language pro-

cessing and sentiment analysis techniques to identify

the relevant and distinctive aspects that characterize

an item. The experimental results included in the pa-

per show that the justiﬁcations generated can be rich

and satisfying and that the users prefer review-based

justiﬁcations as opposed to other explanation strate-

gies (feature-based explanations); in particular, the

authors’ proposal is compared experimentally with

ExpLOD (Musto et al., 2016). As a ﬁnal example, a

recommender system applying text mining on opin-

ions written in Chinese is presented in (Miao and

Lang, 2017). An item-feature matrix is built to calcu-

late the similarity of item characteristics, with the goal

to improve the accuracy of item similarity and so the

quality of the recommendations. A technique called

ItemCF is used to generate the recommendation (Shi

et al., 2014), which supports the pre-calculation of

similar items.

There are other works that exploit both numerical

ratings provided by the users and textual user reviews,

based on the hypothesis that this joint use can lead

to more accurate recommendations. Thus, in (Jakob

et al., 2009), that focuses on the domain of movies,

Use of Text Mining Techniques for Recommender Systems

781

Figure 1: Using text mining in recommender systems: general framework.

a phrase-level opinion extraction from free-text ﬁlm

reviews is performed to identify statements with posi-

tive and negative opinions, and this information is in-

corporated into a recommender system called HYRES

(HYbrid REcommendation System). Several movie

aspect clusters are deﬁned and global numerical val-

ues are computed for each cluster. The experimen-

tal results show that the Latent Dirichlet Allocation

(LDA) clustering approach leads to the best results.

User reviews could also be analyzed to extract

contextual information. For example, (Li et al., 2010),

which focuses on reviews about restaurants, uses text

mining techniques to extract information about 4 dif-

ferent context attributes: companion (whether the

user is alone or with other people), occasion for the

event (e.g., whether it is a holiday or a birthday), time

of the day, and location (in this case extracted from

the user proﬁle, as the authors argue that it is usu-

ally not available in the review texts). Then, the au-

thors propose a Probabilistic Latent Relational Model

(PLRM) to integrate the context data and provide the

recommendations.

3 RECOMMENDER SYSTEMS

USING TEXT MINING ON

OTHER USER’S TEXTS

In this section, we review works that mine textual data

written by the user (other than reviews about items,

which were already analyzed in Section 2) in order

to retrieve useful information for the recommenda-

tion process (see Table 2). This could mean analyz-

ing texts to obtain information about the user’s con-

text, such as his/her location (Drymonas and Pfoser,

2010; Feldman et al., 2015), and/or information about

his/her preferences (e.g., the topics he/she is inter-

ested in). The input textual data could be, for ex-

ample, tweets written by the user. Besides, many re-

cent studies have begun to explore other sources of in-

formation, such as social networks, where larger text

messages can be posted and shared. The diary of the

user could be another relevant text source. Finally,

several recent studies have also demonstrated the ef-

fectiveness of leveraging collective knowledge to en-

rich the users’ interest proﬁles (Faralli et al., 2015).

Twitter can be an interesting data source to try to

infer the interests of users (Piao and Breslin, 2018).

As an example (Xu et al., 2011), starts from the idea

that users often post noisy messages about their lives

or create conversations with friends that are not re-

ally related to their topics of interest. Therefore, they

propose a framework to address this problem by in-

troducing a modiﬁed author-subject model called the

Twitter user model, which is a generative model that

extends LDA to include information about authorship.

A latent variable is used to indicate if a tweet is related

to the author’s interests. According to the authors, this

model can be considered as an initial work for many

tasks in Twitter, such as recommending friends, rat-

ing users, and analyzing social networks. As another

example, in the proposal presented in (Zarrinkalam

et al., 2015) each topic of interest is a conjunction of

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

782

Table 1: Examples of recommender systems using text mining on user reviews.

Items Data used

Recommendation

Approach

Supporting

tools

Text mining

techniques

Reference

Products

Reviews

Collaborative Filtering

Ontology, measures

such as the OFQ

(Overall Feature Quality)

Rule-based

classiﬁcation techniques,

TMSK (Text-Miner

Software Kit),

Riktext software,

ontology (opinion quality

and product quality).

(Aciar et al., 2007)

Items

Reviews

(written in

Chinese)

Collaborative Filtering

Euclidean distance,

ItemCF

(Shi et al., 2014)

Word segmentation

(jieba), stop words

removal, POS Templates,

HowNet (emotional

dictionary), clustering,

chi-square statistics,

logistic regression

(Miao and Lang, 2017)

Tourist

attractions

Reviews Knowledge-based

Supervised algorithms

Probit algorithm, binomial

logistic regression,

decision trees

(CHAID, C&RT and

Random Forest)

IBM SPSS Modeler

Text Analytics, removing

non-linguistic entities

(Guerreiro and Rita, 2019)

Shoppings Reviews

TRS LoT (approach based

on the trust level)

LoT (Level of Trust)

Stop-word removal,

word indexing,

word comparison

(Abuein et al., 2017)

Movies

Reviews,

ratings

Collaborative Filtering HYRES recommender

Clustering, stop

word removal,

Explicit Semantic Analysis

(ESA), LDA-based clustering,

subjectivity clue lexicon

(Jakob et al., 2009)

Restaurants

Restaurant’s

review

CARS (recommendation

based on contextual

information)

Boolean model,

Probabilistic Latent

Relational Model (PLRM)

Bag of words,

rule-based classiﬁers,

GATE tools (ANNIE,

which is a time named

entity recognizer)

(Li et al., 2010)

several concepts, which are temporarily correlated on

Twitter. Based on this, active themes within a given

time interval are extracted and the user’s inclination

towards those themes is determined. Based on this

idea, a concept graph is built and community detec-

tion methods are applied to detect active topics of in-

terest in a given time interval, considering that the re-

lationships between two topics in a social network can

change over time. Besides, a technique to determine

the standpoint of a given user with respect to the ac-

tive topics is proposed. The proposal presented was

applied to design a personalized news recommender

system, which was evaluated experimentally.

Due to the existing information overload in Twit-

ter, recommender systems have also been created to

recommend useful tweets in which users could be re-

ally interested. For example, in (Chen et al., 2012),

the following elements are considered for the recom-

mendation: factors related to the thematic/topic level

of the tweet (used to capture users’ common interests

in the content of the tweets), factors of the social rela-

tionships of the user, and explicit characteristics such

as the authority of the editor and the quality of the

tweet. The experimental results show that the combi-

nation of all these elements can help to improve the

performance of tweet recommendations.

Besides tweets, another popular type of user-

written data can be text messages posted on social

networks. KBRS (Rosa et al., 2019) is a knowledge-

based recommender system that, based on ontologies

and the analysis of feelings, is able to analyze sen-

tences published on online social networks with the

purpose of detecting users with potential psychologi-

cal disorders (depression and stress). Then, if needed,

a recommender system is used to send messages of

happiness, calm, relaxation, or motivation. Text sen-

tences are analyzed using machine learning algo-

rithms, including the CNN model, BLSTM-RNN, and

considering the feeling metric (eSM2). eSM2 was

modeled to improve the performance of KBRS, tak-

ing into account parameters of the user proﬁle, his/her

geographic location, and the subject of the sentence,

Use of Text Mining Techniques for Recommender Systems

783

Table 2: Examples of recommender systems using text mining on other user’s texts.

Items Data used

Recommendation

Approach

Supporting

tools

Text mining

techniques

Reference

Tweets

Tweets, user

social relations

and other features

Collaborative Filtering

Factorization model,

Jaccard similarity

Latent factors (Chen et al., 2012)

Messages

Messages,

user proﬁle

parameters,

geographical

location

Knowledge-based

Deep Learning,

recurrent neural

networks (RNN),

BLSTM-RNN model,

the Nuadu ontology

Machine learning algorithms,

sentiment metric (eSM2)

(Rosa et al., 2019)

Tourism

POIs

Textual

messages in

a web chat

Content-Based

Filtering

Probabilistic techniques

(fuzzy reasoning)

ontology

Rocchio’s and Naive Bayes algorithms,

tourism ontology

(Loh et al., 2003)

Tags Text in tags Collaborative Filtering

Association

rules, LDA

TF-IDF (Krestel et al., 2009)

to identify the intensity of the feeling identiﬁed in a

message. The paper shows the effectiveness of using

an ontology and a personalized feeling analysis.

Text mining techniques can also be used to dis-

cover user interests from text exchanged in a chat con-

versation. For example, (Loh et al., 2003) exploits

texts in chats between a customer and a travel agent to

discover and recommend travel options for customers,

especially for those who do not really know where to

go and what to do there. The proposed system queries

a tourism ontology to identify key topics in the text

messages and then queries a database to retrieve ap-

propriate touristic options (such as cities and attrac-

tions). In this case the recommender system is hidden

from the customer, as it is the travel agent the one who

receives the recommendations, that support him/her

to better guide the client. Rather than using extensive

forms with options, attributes, and requirements, the

needs and desires of the customer arise during a natu-

ral conversation, and the data retrieved by the system

can complement the potential lack of experience or

knowledge of the travel agent regarding the speciﬁc

interests of the customer.

Some work has also been done to extract infor-

mation from texts written by users in informal com-

munications (instant messaging, recordings, meeting

minutes, emails) as a potential source of knowledge

within an organization. This could be very useful,

for example, to create user proﬁles for use in rec-

ommender systems. In (Gentile et al., 2011), the au-

thors claim that it is possible to model the experience

of people automatically by tracking informal commu-

nication exchanges (e-mails) and through the seman-

tic annotation of their content to derive dynamic user

proﬁles. The proﬁles are then used to estimate the

similarity between people (using the Jaccard index).

A tool called SimNET (Similarity and Network Ex-

ploration Tool) interactively displays content and user

networks as part of the search and navigation capa-

bilities provided by the knowledge management sys-

tem. Three techniques are used to build the proﬁles:

Proﬁle based on keywords, using the Java Automatic

Term Recognition Toolkit (JATR v1.02) for keyword

extraction; Proﬁle based on the Entity, using the Open

Calais web service for the extraction of named enti-

ties; and Proﬁle based on the Concept, using the Wik-

ify web service for concept extraction.

Another relevant type of written text that can be

considered is text present in tags. Tagging systems

have become important for the Web, as they allow

users to create tags that annotate and categorize dif-

ferent types of contents and share them with oth-

ers. However, an important difﬁculty is that tags are

not limited by a controlled vocabulary and annotation

guidelines, and they tend to be noisy and sparse. The

aim of the approach presented in (Krestel et al., 2009)

is to overcome the problem of cold start for tagging

new resources. Speciﬁcally, LDA is used to obtain

latent topics from texts and, based on this, other tags

belonging to those themes can also be recommended.

According to the experimental evaluation presented,

this approach achieves a better performance (in terms

of accuracy and recall) than an alternative approach

using association rules.

4 RECOMMENDER SYSTEMS

USING TEXT MINING ON

ITEMS’S DATA

In this section, we review some works that apply text

mining on data associated to the items that can be rec-

ommended (see Table 3).

With geo-positioning and geo-tagging of objects

that have both a geographic location and a text de-

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

784

Table 3: Examples of recommender systems using text mining on item’s data.

Items Data used

Recommendation

Approach

Supporting

tools

Text mining

techniques

Reference

Tourism

POIs

Keywords,

GPS location,

POIs’ descriptions

ReCoSKQ CoSKQ

Any text mining

technique

(Hermoso et al., 2019)

Apps

Version metadata,

textual descriptions

Hybrid

Semi-supervised

topic model, PMF,

LDA

Topic-word

distribution,

latent topics

(Lin et al., 2014)

Apps

Context’s description,

app’s title and

description,

app’s metadata

IR-based (context-aware) approach

Lucene search engine,

Context-Aware Browser

(Coppola et al., 2010)

TF-IDF,

clustering

(Mizzaro et al., 2014)

Events

GPS location,

news

Location-aware

Learning model

disambiguation

and matching

sLDA, SVM

(polynomial,

Gaussian, lineal)

(Ho et al., 2012)

scription, the so-called spatial keyword queries that

exploit both the locations and the text descriptions are

attracting growing attention (Cao et al., 2011). More

speciﬁcally, the aim of a Collective Spatial Keyword

Query (CoSKQ) is to retrieve, from a spatial database,

a group of spatial elements in such a way that the de-

scription of the elements included in that set, based on

keywords, is completely covered by the keywords of

the query and also that the elements retrieved are as

close as possible to the location of the query and have

the smallest possible distances between the elements

themselves. In (Hermoso et al., 2019), the authors ar-

gue that the use of this concept could be useful for the

development of recommender systems for tourism, in

order to provide the user with a set of points of in-

terest (POIs) that satisfy his/her query (e.g., a query

expressing his/her current interest or need) both geo-

graphically and semantically; the idea of Re-CoSKQ

is presented, which is an adaptation of CoSKQ for the

ﬁeld of recommender systems. Although text mining

techniques are not explicitly applied in Re-CoSKQ,

they are expected to be needed as a pre-processing

step, in order to obtain the keywords that describe the

different items and/or the user proﬁle.

A recommender system of mobile applications

that exploits the description of the application and as-

sociated metadata is presented in (Lin et al., 2014).

As an updated version of an app may bring signif-

icant changes, it ﬁrst generates latent themes from

the version’s features using a semi-supervised theme

model to characterize each version. Themes are dis-

criminated based on metadata and a recommenda-

tion technique called VSR (version-sensitive recom-

mendation) is proposed. For experimental evaluation,

two alternative recommendation approaches are con-

sidered as baselines: probability matrix factorization

(PMF), which is a collaborative ﬁltering (CF) tech-

nique, and LDA, which is a content-based ﬁltering

(CBF) technique. The authors show the accuracy

of the recommendations obtained when these tech-

niques are used independently and also when they

are combined (CF+CBF, CF+VSR, CBF+VSR, and

CF+CBF+VSR). This work reports that a hybrid rec-

ommender system incorporating the version-sensitive

model proposed achieves better results.

Other mobile app recommender systems also take

into account the context of the user. Thus, App-

CAB (Mizzaro et al., 2014) is a recommender system

of mobile apps that provides a proactive and fully au-

tomated procedure for querying the mobile app mar-

ket, capable of retrieving a set of applications and

classifying them according to the user’s current situa-

tion. The description of the user’s context is generated

by a Context-Aware Browser (Coppola et al., 2010) to

ﬁnd the suitable applications for the user’s needs in

that speciﬁc context. To categorize contexts, all the

words in the title and description of the applications

that have the same category are grouped, creating sev-

eral sets of words that represent the categories of the

market. The word indexing process is performed us-

ing Lucene to build a category index. When the sys-

tem analyzes the current context, it checks each word

that describes the context in the index in order to com-

pute relevance scores and obtain a list of potentially-

relevant categories. In this way it is possible to weigh

the pertinence of different types of applications for

the given context. Besides, if a word that appears in

the context is part of the title of the application, then

+10% is added to the original score, to emphasize the

importance of a match at the level of the application

name. Finally, a ﬁltering step is applied to eliminate

duplicate results before presenting them to the users.

Finally, in (Ho et al., 2012), an approach to extract

future space-time events from the Web, to be used

as candidate items in a location-aware recommender

system of events, is presented. This work proposes

Use of Text Mining Techniques for Recommender Systems

785

a procedure for the extraction of events from news

articles, which consists of two main steps: recogni-

tion and matching. In the recognition stage, place

names and future time patterns are identiﬁed and ex-

tracted. In the matching stage, operations for spatial-

temporal disambiguation, de-duplication and match-

ing, are performed. A sentiment variable (positive,

negative or neutral) is attached to each event, as an

aid for the recommendation application. So, the fu-

ture event identiﬁed consists of its geographic loca-

tion, time pattern, sentiment variable, news title, key

phrase, and URL of the news article.

5 CONCLUSIONS

In this paper, we have provided an overview of works

exploiting text mining techniques in the ﬁeld of rec-

ommender systems, characterizing them according to

the type of textual data analyzed (user reviews, other

texts written by the user, or textual data associated

to the items) and considering their purpose (extract

information about the context of the user or perform

user proﬁling).

Beyond user reviews, exploiting other types of

texts written by the user seems to be an area that re-

mains quite unexplored, particularly concerning more

personal textual data such as emails received or sent,

the user’s personal agenda, or the user’s diary; this

could be partly due to privacy concerns. Besides,

most works focus only on a speciﬁc type of text. As

future work, we intend to propose an integrated ap-

proach that combines and adapts several text mining

techniques as a support tool to build a context-aware

recommender and evaluate its performance.

ACKNOWLEDGEMENTS

This work has been supported by the project

TIN2016-78011-C4-3-R (AEI/FEDER, UE) and

DGA-FSE (COSMOS research group).

REFERENCES

Abuein, Q., Shatnawi, A., and Al-Sheyab, H. (2017).

Trusted recomendation system based on level of trust

(TRS LoT). In 2017 International Conference on En-

gineering and Technology (ICET), pages 1–5.

Abuein, Q., Shatnawi, M., Bani Yassein, M., and Batiha,

R. (2016). A framework for social media and text-

based content analysis for event management pur-

poses. International Review on Computers and Soft-

ware, 11:388–394.

Aciar, S., Zhang, D., Simoff, S., and Debenham, J. (2007).

Informed recommender: Basing recommendations on

consumer product reviews. IEEE Intelligent Systems,

22(3):39–47.

Adomavicius, G. and Tuzhilin, A. (2011). Recommender

Systems Handbook, chapter Context-Aware Recom-

mender Systems, pages 217–253. Springer US,

Boston, MA.

Berry, M. W., editor (2004). Survey of Text Mining. Springer

New York.

Cao, X., Cong, G., Jensen, C. S., and Ooi, B. C. (2011).

Collective spatial keyword querying. In ACM SIG-

MOD International Conference on Management of

Data (SIGMOD 2011), pages 373–384, New York,

NY, USA. ACM.

Chen, K., Chen, T., Zheng, G., Jin, O., Yao, E., and Yu,

Y. (2012). Collaborative personalized tweet recom-

mendation. In 35th International ACM SIGIR Con-

ference on Research and Development in Information

Retrieval (SIGIR 2012), pages 661–670. ACM.

Coppola, P., Della Mea, V., Di Gaspero, L., Menegon, D.,

Mischis, D., Mizzaro, S., Scagnetto, I., and Vassena,

L. (2010). The context-aware browser. IEEE Intelli-

gent Systems, 25(1):38–47.

Drymonas, E. and Pfoser, D. (2010). Geospatial route ex-

traction from texts. In First ACM SIGSPATIAL Inter-

national Workshop on Data Mining for Geoinformat-

ics (DMG 2010), pages 29–37. ACM.

Faralli, S., Stilo, G., and Velardi, P. (2015). Recommenda-

tion of microblog users based on hierarchical interest

proﬁles. Social Network Analysis and Mining, 5(1).

Feldman, D., Sung, C., Sugaya, A., and Rus, D. (2015).

IDiary: From GPS signals to a text- searchable diary.

ACM Transactions on Sensor Networks, 11(4):60:1–

60:41.

Gentile, A. L., Lanfranchi, V., Mazumdar, S., and

Ciravegna, F. (2011). Extracting semantic user net-

works from informal communication exchanges. In

Aroyo, L., Welty, C., Alani, H., Taylor, J., Bernstein,

A., Kagal, L., Noy, N., and Blomqvist, E., editors, The

Semantic Web – ISWC 2011, pages 209–224. Springer.

Gruber, T. R. et al. (1993). A translation approach to

portable ontology speciﬁcations. Knowledge Acqui-

sition, 5(2):199–221.

Guerreiro, J. and Rita, P. (2019). How to predict ex-

plicit recommendations in online reviews using

text mining and sentiment analysis. Journal of

Hospitality and Tourism Management. DOI:

10.1016/j.jhtm.2019.07.001. In Press, Corrected

Proof.

Gupta, V., Lehal, G. S., et al. (2009). A survey of text min-

ing techniques and applications. Journal of Emerging

Technologies in Web Intelligence, 1(1):60–76.

Hegde, Y. and Padma, S. K. (2017). Sentiment analysis

using random forest ensemble for mobile product re-

views in kannada. In 2017 IEEE 7th International

Advance Computing Conference (IACC), pages 777–

782.

ICEIS 2020 - 22nd International Conference on Enterprise Information Systems

786

Hermoso, R., Ilarri, S., and Trillo-Lado, R. (2019). Re-

CoSKQ: Towards POIs recommendation using collec-

tive spatial keyword queries. In ACM RecSys Work-

shop on Recommenders in Tourism (RecTour 2019),

in conjunction with the 13th ACM Conference on Rec-

ommender Systems (RecSys 2019), Copenhagen (Den-

mark), volume 2435, pages 42–45. CEUR Workshop

Proceedings, ISSN 1613-0073.

Ho, S.-S., Lieberman, M., Wang, P., and Samet, H. (2012).

Mining future spatiotemporal events and their senti-

ment from online news articles for location-aware rec-

ommendation system. In First ACM SIGSPATIAL In-

ternational Workshop on Mobile Geographic Infor-

mation Systems (MobiGIS 2012), pages 25–32, New

York, NY, USA. ACM.

Jakob, N., Weber, S. H., M

uller, M. C., and Gurevych, I.

(2009). Beyond the stars: Exploiting free-text user

reviews to improve the accuracy of movie recom-

mendations. In First International CIKM Workshop

on Topic-Sentiment Analysis for Mass Opinion (TSA

2009), pages 57–64, New York, NY, USA. ACM.

Kim, E.-G. and Chun, S.-H. (2019). Analyzing online car

reviews using text mining. Sustainability, 11(6):1611.

Krestel, R., Fankhauser, P., and Nejdl, W. (2009). Latent

Dirichlet Allocation for tag recommendation. In Third

ACM Conference on Recommender Systems (RecSys

2009), pages 61–68. ACM.

Li, Y., Nie, J., Zhang, Y., Wang, B., Yan, B., and Weng,

F. (2010). Contextual recommendation based on text

mining. In 23rd International Conference on Compu-

tational Linguistics (COLING 2010): Posters, pages

692–700. Association for Computational Linguistics.

Lin, J., Sugiyama, K., Kan, M.-Y., and Chua, T.-S. (2014).

New and improved: Modeling versions to improve

app recommendation. In 37th International ACM SI-

GIR Conference on Research & Development in Infor-

mation Retrieval (SIGIR 2014), pages 647–656, New

York, NY, USA. ACM.

Liu, B. and Zhang, L. (2012). A Survey of Opinion Mining

and Sentiment Analysis, pages 415–463. Springer US,

Boston, MA.

Loh, S., Lorenzi, F., Salda

na, R., and Licthnow, D. (2003).

A tourism recommender system based on collabora-

tion and text analysis. Information Technology &

Tourism, 6(3):157–165.

Miao, D. and Lang, F. (2017). A recommendation system

based on text mining. In 2017 International Confer-

ence on Cyber-Enabled Distributed Computing and

Knowledge Discovery (CyberC), pages 318–321.

Mizzaro, S., Pavan, M., Scagnetto, I., and Zanello, I.

(2014). A context-aware retrieval system for mo-

bile applications. In 4th Workshop on Context-

Awareness in Retrieval and Recommendation (CARR

2014), pages 18–25. ACM.

Musto, C., Lops, P., de Gemmis, M., and Semeraro, G.

(2019). Justifying recommendations through aspect-

based sentiment analysis of users reviews. In 27th

ACM Conference on User Modeling, Adaptation and

Personalization, UMAP ’19, pages 4–12, New York,

NY, USA. ACM.

Musto, C., Narducci, F., Lops, P., De Gemmis, M., and Se-

meraro, G. (2016). ExpLOD: A framework for ex-

plaining recommendations based on the Linked Open

Data Cloud. In 10th ACM Conference on Rec-

ommender Systems (RecSys 2016), pages 151–154.

ACM.

Piao, G. and Breslin, J. G. (2018). Inferring user interests in

microblogging social networks: a survey. User Mod-

eling and User-Adapted Interaction, 28(3):277–329.

Ricci, F., Rokach, L., and Shapira, B. (2011). Recommender

Systems Handbook, chapter Introduction to Recom-

mender Systems Handbook, pages 1–35. Springer US,

Boston, MA.

Ricci, F., Rokach, L., and Shapira, B. (2015). Recommender

Systems Handbook, chapter Recommender Systems:

Introduction and Challenges, pages 1–34. Springer

US, Boston, MA.

Rosa, R. L., Schwartz, G. M., Ruggiero, W. V., and

Rodr

ıguez, D. Z. (2019). A knowledge-based rec-

ommendation system that includes sentiment analysis

and deep learning. IEEE Transactions on Industrial

Informatics, 15(4):2124–2135.

Roul, R. K. and Arora, K. (2019). A nifty review to

text summarization-based recommendation system for

electronic products. Soft Computing, 23(24):13183–

13204.

Shi, Y., Larson, M., and Hanjalic, A. (2014). Collaborative

ﬁltering beyond the user-item matrix: A survey of the

state of the art and future challenges. ACM Computing

Surveys, 47(1).

Subramaniyaswamy, V., Manogaran, G., Logesh, R., Vi-

jayakumar, V., Chilamkurti, N., Malathi, D., and

Senthilselvan, N. (2019). An ontology-driven person-

alized food recommendation in IoT-based healthcare

system. The Journal of Supercomputing, 75(6):3184–

3216.

Tarus, J. K., Niu, Z., and Mustafa, G. (2018). Knowledge-

based recommendation: a review of ontology-based

recommender systems for e-learning. Artiﬁcial intel-

ligence review, 50(1):21–48.

Xu, Z., Ru, L., Xiang, L., and Yang, Q. (2011). Discov-

ering user interest on Twitter with a modiﬁed author-

topic model. In 2011 IEEE/WIC/ACM International

Conferences on Web Intelligence and Intelligent Agent

Technology, volume 1, pages 422–429.

Zarrinkalam, F., Fani, H., Bagheri, E., Kahani, M., and Du,

W. (2015). Semantics-enabled user interest detection

from twitter. In 2015 IEEE/WIC/ACM International

Conference on Web Intelligence and Intelligent Agent

Technology (WI-IAT), volume 1, pages 469–476.

Use of Text Mining Techniques for Recommender Systems

787