SIWAM: Using Social Data to Semantically Assess the Difﬁculties in

Mountain Activities

Javier Rinc´on Borobia, Carlos Bobed, Angel Luis Garrido and Eduardo Mena

IIS Deparment, University of Zaragoza, Zaragoza, Spain

Keywords:

Semantic Web, Information Extraction, Ontologies, Social Network.

Abstract:

In the last few years, the amount of people moving to the mountains to do several activities such as hiking,

climbing or mountaineering, is steadily increasing. Not surprisingly, this has come along with a raise in the

amount of accidents, which are mainly due to the inexperience of the people, and the lack of information and

proper planning. Although one could expect to ﬁnd appropriate updated information about this issue on the

Internet, most of the information related to mountain activities is stored in personal blogs, or in Web sites

that are not exploiting the possibilities that the Semantic Web and the Social Web offer regarding content

generation and information processing.

In this paper, we present SIWAM, a semantic framework oriented to share and evaluate the difﬁculties of

mountain activities. It provides a thematic social network front-end to enable users to share their descriptions

about their own experiences. Using text mining techniques on these descriptions, it extracts relevant facts

about these experiences, which are used to evaluate the difﬁculty of the particular activity. The evaluation

is done according to a well-established standard for evaluating the difﬁculty of mountain activities (MIDE),

which is modeled in the system using ontologies.

1 INTRODUCTION

Mountain activities comprise a set of sports that are

amongst the most practiced in the world. The amount

of people practising them is increasing year by year

all around the world (Global Industry Analyst, Inc.,

2012; Jenkins, 2013). This steady increment of prac-

titioners along with the fact that mountain activities

are considered extreme sports make the security be an

important and recurring matter of study. There are re-

search areas (specially medical ones) (Chamarro and

Fern´andez-Castro, 2009), and organizations, such as

the International Mountaineering and Climbing Fed-

eration, which are specially concerned about moun-

tain accidents (UIAA Mountaineering Commission,

2004). However, in spite of all the efforts, the amount

of accidents is, not surprisingly, increasing as more

and more people move to the mountains. From the

yearly reports of the rescue groups in Spain (Min-

isterio del Interior, 2012), we can point out that the

main problems regarding accidents are the inexperi-

ence of the people, and the lack of information and

proper planning.

While mountain guide books can be a source of in-

formation about the tracks and the environmentwhere

the activity is held, it seems pretty safe to assume that,

as the main cause of accidents is lack of information,

people are not using them to plan their activity (when

even they plan it). Instead, in these times, people go to

the Internet to look for information as fast as possible,

regardless the possible security implications. In the

case of mountain activities, this information is mainly

available as descriptions of the different experiences

in text format (e.g., blog entries).

However,trusting these on-line descriptions might

be a double-edged sword: On the one hand, they

might be really useful as they hold information about

the activity; but, on the other hand, they are describing

unique experiences whose setup and conditions might

not be applicable to other situations. A climbing route

might have many variations, shortcuts, forks, etc.,

within the same path. In a mountain environment, a

slight detour might turn an easy track into a challeng-

ing one. Besides, the weather conditions also affect

strongly to the mismatch between the descriptions and

the track difﬁculties. As an example, in high moun-

tain, the orientation is mainly guided by milestones

that may be covered by snow during winter season.

Moreover, the level of expertise of the person de-

scribing the activity might lead to dangerous biases,

Rincón Borobia J., Bobed C., Garrido A. and Mena E..

SIWAM: Using Social Data to Semantically Assess the Difﬁculties in Mountain Activities.

DOI: 10.5220/0004812600410048

In Proceedings of the 10th International Conference on Web Information Systems and Technologies (WEBIST-2014), pages 41-48

ISBN: 978-989-758-024-6

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

as, for an experienced mountaineer, several actions

might be considered too easy to be worthy of being

described, although they may suppose a danger for an

amateur practitioner. Therefore, there is a need for

methods to support the correct evaluation of the dif-

ferent activities taking into account the expertise and

physical conditions of the practitioner.

Currently, in the mountain Web sites, the main

method to introduce information about an activity

is form-oriented, with several predeﬁned ﬁelds fre-

quently including a textual one, where users can pro-

vide a more detailed description of their experiences.

In fact, the information that can make a difference

may be hidden in these textual descriptions (e.g., “the

track is quite easy, but there is a fork that might lead

to a . ..”). An improvement could be to include more

detailed forms; however, the amount of details to be

included might be overwhelming for the users, result-

ing in a decrease of user collaboration.

In this paper, we present SIWAM

, a system that

provides a semantic framework to improve users’ in-

formation in the domain of mountain activities with

the main goal of improving their security. It con-

sists of three main modules: 1) a social Web front-

end (RSAM), where users can share their experiences

and provide information about their proﬁles; 2) a text-

extraction module (MECMIDE) that is in charge of

detecting and extracting the relevant facts from the

activity descriptions provided by the users; and, ﬁ-

nally, 3) an activity semantic evaluator that, given

the relevant facts, classiﬁes the different activities ac-

cording to a well-deﬁned mountain activities standard

(MIDE), which is modeled in the system by an ontol-

ogy network. The complete set of individual evalua-

tions can be taken into account to provide a more ac-

curate evaluation of the different activities depending

on different aspects such as the weather conditions,

the season, the user’s proﬁle, and so on.

The rest of the paper is as follows. In Section 2,

we present the architecture of our system, detailing

the main modules. In Section 3, we introduce the

system’s ontology and the MIDE standard (Roche,

2002), and how we have modeled the latter to be ex-

ploited by SIWAM. Section 4 and Section 5 detail the

information extraction module (MECMIDE) and the

evaluation module (VALMIDE),respectively. A com-

plete example of how SIWAM works is presented in

Section 6. We discuss some related work in Section 7.

Finally, the conclusions and future work are drawn in

Section 8.

SIWAM stands for Sistema de Informaci´on Web para

Actividades de Monta˜na (in Spanish, which means Moun-

tain Activities Web Information System).

2 ARCHITECTURE OF THE

SYSTEM

In this section, we present a general overview of SI-

WAM, describing its aim and generalarchitecture. SI-

WAM is conceived to be a social site where users of

all expertise share their experiences performing dif-

ferent mountain activities. Using text-mining and Se-

mantic Web techniques, SIWAM is capable of pro-

cessing this information to assess the difﬁculty of the

different tracks and activities according to a well-

established mountain activities standard. To do so,

SIWAM exploits the knowledge stored in the System

Ontology (see Figure 1). In this way, SIWAM offers

much more precise and updated information about the

different activities, having as a ﬁnal objective to im-

prove the mountaineers security (no matter their ex-

perience levels).

As it can be seen in Figure 1, SIWAM consists of

three main modules:

• RSAM (Social Network for Mountain Activities:)

This is the Web front-end of SIWAM, and sup-

ports the common features of a social site (e.g.,

user proﬁles, instant messaging, groups, contacts,

etc.). Its functionality is speciﬁcally extended

with features oriented to the management and

sharing of information about mountain activities

(e.g., including new ones, adding descriptions and

experiences, sharing maps, GPS routes, etc.). SI-

WAM processes these shared descriptions and ex-

periences in background to obtain the information

about the safeness of a particular activity.

• MECMIDE (Concept Extraction Module:) It is

the module in charge of processing each of the

descriptions and extracting the relevant facts out

of them. To do so, it uses Freeling (Carreras

et al., 2004) to analyze the texts, looking for

relevant patterns which might contain informa-

tion. These patterns are mapped to different con-

cepts and properties of the System Ontology, so

MECMIDE outputs a set of axioms in the form of

RDF

triples which are correctly aligned with this

ontology.

• VALMIDE (Evaluator Module:) This module

takes as input the facts that MECMIDE has ex-

tracted from the text, and, with the help of a De-

scription Logics reasoner (Baader et al., 2003)

(DL reasoner from now on), evaluates the difﬁ-

culty of the activity according to the System On-

tology. This ontology models the MIDE (Roche,

2002) standard to evaluate the difﬁculty of moun-

tain activities regarding several criteria. The

http://www.w3.org/TR/rdf-primer/

WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies

SIWAM

RSAM

VALMIDE

Activities

MECMIDE

Reasoner

Facts

Social Network

Activities

Management

Applications

Additional

User

Search,

others, ...

Inserting

Activity

Social

Activity Description

Difficulty Assessment

Activities DB

System

Ontology

Figure 1: Architecture of SIWAM.

global evaluation of a particular activity is ob-

tained combining the single evaluations of all the

people that have done it. Considering all the

descriptions makes the evaluation more robust

against omitted information.

In the following, we will focus on the semantic part

of SIWAM. First, we present how we have modeled

the MIDE evaluation standard in the System Ontol-

ogy. Then, we overview how MECMIDE extracts the

information from the texts using the vocabulary de-

ﬁned in the used ontology. Finally, we explain how

this knowledge is used by VALMIDE to assess the

difﬁculty of a particular activity.

3 SYSTEM ONTOLOGY

The System Ontology stores all the information

needed to perform the evaluation of the different ac-

tivities. Following the directives given in NeON

Methodology (Su´arez-Figueroa, 2012), we have built

an ontology network to make it possible to follow a

modular development. In Figure 2, we can see the

inner structure of this ontology:

Criteria Module

MIDE Path

Criteria Module

GCDE Module

....

System Ontology

Other Evaluation

Modules

MIDE Environment

Figure 2: Inner structure of the ontology used by SIWAM.

• The Global Criteria Excursions Deﬁnition

(GCDE) ontology module integrates the different

modules that model different evaluation criteria,

and stores information about how they must be

used to evaluate an activity. In particular, for each

of the activity types that SIWAM handles, GCDE

stores the evaluation modules that are applicable

to it and the methods to be used to do so. This

enables the system to react automatically to the

addition of new evaluation modules and methods.

Currently, it only contains the description of one

method of evaluation, which is the one applied to

the MIDE evaluations (we will see it later in Sec-

tion 5); however, it is interesting to specify its role

in the system as it provides our system with a ﬂex-

ible method to add/remove evaluation modules.

• For each evaluation criteria to be used, there is an

ontology module that models it in the system. In

particular, we have selected the MIDE criteria for

SIWAM, as it is a well-established standard for

mountain activities evaluation.

The motivation for using ontologies to model the

different evaluation criteria is two-fold: On the one

hand, so far, there has not been not any initiative to

model the domain of mountain activities and provide

a shared vocabulary; on the other hand, using ontolo-

gies provides us with a logical framework to perform

the activity evaluation in a more tractable and man-

ageable way than using pure rule systems, as we will

see in Section 5. Moreover, the use of an ontologi-

cal model as a base for the text information extraction

has been successfully applied in many works such

as (Garrido et al., 2012; Vogrincic and Bosnic, 2011;

Garrido et al., 2013; Kara et al., 2012).

Thus, in the rest of the section, we focus on how

we have modeled the evaluation criteria knowledge in

this ontology. To do so, we ﬁrstly overview the MIDE

SIWAM:UsingSocialDatatoSemanticallyAssesstheDifficultiesinMountainActivities

standard, and then we detail how we have captured its

information in the appropriate ontology module.

3.1 MIDE Evaluation Standard

The MIDE standard (Roche, 2002) is focused on the

evaluation of mountain tracks. To do so, MIDE pro-

poses to tag a particular excursion (experiences using

the SIWAM terminology) with two different kinds of

information: 1) information about the track (name,

type, accumulated height difference, etc.) and the

conditions of the actual activity performing (hour,

season, weather conditions, etc.); and 2) evaluation

information, which provides difﬁculty values ranged

in 1 – 5 for different aspects of the track, namely En-

vironment, Path, Journey, and Physical Effort.

Most of the ﬁrst kind of information can be cap-

tured using simple forms (e.g., times, height differ-

ences, distances, etc.). For the second type of infor-

mation, MIDE provides guidelines to evaluate each

of those values. While the Physical Effort can be

obtained using different formulae, the rest of values

(Environment, Path and Journey) are assessed using

a set of criteria that might or might not be present in

the track. Then, depending on the number of crite-

ria the activity fulﬁls, it is assigned a difﬁculty value

from 1 to 5 for that particular aspect.

Capturing these criteria using formularies is more

complicated. They would require complex forms,

which would be long and time-consuming to ﬁll, a

task that not all the users are prone to do. Fortu-

nately, this is the kind of information that is usually

comprised in the textual descriptions of the user’s ac-

tivities. Thus, the ﬁrst step to perform an automatic

evaluation of the track is to model these criteria to

be used by SIWAM. In the following subsection, we

present our approach to do this task.

3.2 Modeling MIDE with Ontologies

The development of the ontology modules that model

the MIDE evaluation criteria was carried out in two

well differentiated stages:

1. Modeling the mountain domain: The objective of

this ﬁrst stage was to obtain and organize the el-

ements (concepts and properties) within the do-

main of the mountain activities. We did not aim

at modeling the whole domain at once, but in-

crementally, taking each of the MIDE evaluation

criteria as the competence questions (Gr¨uninger

and Fox, 1994) for each iteration. This way, we

could assure that we had modeled all the elements

needed to model the MIDE knowledge.

2. Modeling the criteria: In the case of MIDE, the

evaluation is performed by checking whether a

particular activity fulﬁls or not an speciﬁc (and

complex) condition. From an ontological point of

view, this is equivalent to say that the activity be-

longs to a particular type of activities deﬁned by

this condition. Thus, we used deﬁned concepts to

model them. In Description Logics (Baader et al.,

2003), the underlying formalism of OWL

, a de-

ﬁned concept provides a complete deﬁnition of

its members, that is, it establishes necessary and

sufﬁcient conditions for an instance to belong to

it. This kind of deﬁnitions enables DL reason-

ers to classify the instances according to them. In

the following section, we will see how VALMIDE

module takes advantage of this issue.

Moreover, we added also the information about

how many criteria the activities must fulﬁl to be as-

signed each of the 1 – 5 values. We modeled this in-

formation also as deﬁnitions, but they do not affect

to the reasoning process. They are just consulted by

VALMIDE to obtain the mapping between the num-

ber of criteria and the ﬁnal value.

We now present two different examples to illus-

trate how these deﬁnitions comprise the knowledge

about the criteria. We focus on the subdomain of the

Environment module, this is, criteria about the harsh-

ness of the environment:

• Criteria “crossing a place farther than 1 hour

(walking time) from a inhabited place” is modeled

as a new concept whose deﬁnition is

(actionPerformed

some

CrossRemoteArea3H)

along with the following deﬁnitions

CrossRemoteArea3H

equivalentTo

(Cross

and

(actionPerformedIn

some

RemoteArea3H))

RemoteArea3H

equivalentTo

(SingularElement

and

(distantFrom

some

InhabitedPlace)

and

(isWalkingDistance

some

integer[>=3]))

• Criteria “high probability of temperatures un-

der 0

◦

C ” and “high probability of temperatures

under 10

◦

C” are modeled respectively as

(hasMinTemperature

some

integer[<0])

and

(hasMinTemperature

some

integer[<-10])

http://www.w3.org/TR/owl-primer/

WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies

These deﬁnitionsenable VALMIDE to classify the

texts with the help of a DL reasoner. In the following

section, we present the MECMIDE module, that en-

ables SIWAM to obtain the facts to be classiﬁed out

from the text.

4 MECMIDE MODULE

The MECMIDE module is in charge of extracting the

relevant information out from the descriptions pro-

vided by the users. It works as follows: For each fact

or concept in the System Ontology that we want to be

able to detect, MECMIDE has a list of search patterns

that are to be looked up in the texts. For instance, the

concept “Use hands on step” has a list of patterns such

as “utilize hand”, “use hand”, “require hand use”, etc.

That list contains the lemmatized form of each word

used. The patterns do not include articles, conjunc-

tions, prepositions, or other words that lack of intrin-

sic semantic value.

To enrich the pattern-search analysis, MECMIDE

includes two additional advanced semantic features:

• The processing window: MECMIDE searches

each of the words composing the patterns in

an area deﬁned according to a given number of

words. This feature enables MECMIDE to de-

tect a pattern in the text regardless the word order.

This is useful because sentences can hold different

structures with the same words

• The use of synonyms: Instead of looking just

for the exact words of each pattern, MECMIDE

considers also their synonyms. For example the

word “way” has the same meaning as “traverse”,

“track”, “trail”, “path”, “route”, or “itinerary”.

The use of synonyms is solved using a lexical

database, like WordNet (Miller, 1995), which al-

lows MECMIDE to transform the words that in-

tegrate each pattern in synsets, i.e., the canonical

form of its meaning. Then, the search is carried

out using these synset, which broadens the vocab-

ulary coverage of the patterns.

In many cases, it is not sufﬁcient that a single pat-

tern is recognized to deduce that the text is related

to a particular concept. So, on top of these text pat-

terns, MECMIDE uses a set of rules to decide if we

can actually associate a text with a fact or concept in

the VALMIDE concept (e.g., the need of a minimum

number of patterns to deduce whether a particular de-

scription is related to an ontology concept). Thus, the

This structure richness is very typical when processing

Spanish texts.

process followed by MECMIDE to process each text

is composed by three steps:

1. Lemmatization of the texts: Freeling (Carreras

et al., 2004) is used as a tagger and lemmatizer, to

ﬁlter stop words and to obtain the lemma of each

word of the text, respectively.

2. Looking for patterns: To do this, the words that

form the patterns are converted into synsets to im-

prove the quality of searches. For example, “dan-

gerous way” is equivalent to “slippery path”. In

our current implementation, we have chosen Eu-

roWordNet (Vossen, 1998) as our lexical database

because our prototype is in Spanish.

3. Applying the rules: The set of rules is evaluated to

see whether the text must be linked to a concept,

i.e., whether a fact can be derived from the text.

The result of this analysis is a set of concepts and

facts that are associated to the text. In the following

section, we present how this set of facts is used by

VALMIDE to evaluate the activity according to the

knowledge stored in the System Ontology.

5 VALMIDE MODULE

As we have seen in Section 3, the ontology used

by SIWAM is an ontology network composed by

two levels: One with information about the evalua-

tion methods used for each evaluation criteria (GCDE

Module), and the other with the actual knowledge

needed to perform the actual evaluation (criteria mod-

ules). VALMIDE uses this information to perform

the evaluation of each single activity. This process

is composed by the following steps:

1. Deciding the evaluation modules: When it re-

ceives the description text along with the extracted

facts, VALMIDE consults the GCDE module to

see which evaluation modules are applicable to

the type of activity that is being described.

2. Obtaining the evaluation methods: GCDE con-

tains information about the speciﬁc method to be

used for each evaluation module. VALMIDE con-

sults it to know how to handle an speciﬁc module.

This way, we can attach evaluation modules and

methods in a ﬂexible and decoupled manner.

3. Evaluating the activity description: For each mod-

ule, VALMIDE asserts the facts in a copy of the

evaluation module and classiﬁes it. The classiﬁ-

cation and inferring capabilities of DL reasoners

enable VALMIDE to make it explicit knowledge

that otherwise would remain implicit. This infor-

mation is used to calculate the actual evaluation.

SIWAM:UsingSocialDatatoSemanticallyAssesstheDifficultiesinMountainActivities

4. Combining evaluations: Depending on the eval-

uation module, VALMIDE can consider whether

to just evaluate the activity as a single one, or

to combine previous evaluations to achieve an

agreed evaluation. The information of the method

to use is stored in the GCDE module.

Regarding the MIDE Modules, we havedeveloped

an evaluation method based on the activity descrip-

tion classiﬁcations. As we have explained in Sec-

tion 3, each MIDE criteria is modeled as a deﬁned

concept. The MIDE evaluation method obtains the

agreed number of criteria fulﬁlled using the follow-

ing formulae:

MIDE value(A

instance

) =

|Crit|

∑

i=1

ful fil(crit

)

with

ful fil(crit

) =











1 if

|Desc|

∑

k=1

crit

(desc

)

|Desc|

> CT

0 otherwise

where A

instance

is the activity we are evaluating, Crit

is the set of criteria to be evaluated (the set of con-

cept deﬁnitions), Desc is the set of individual descrip-

tions for A

instance

,CT is a conﬁdence threshold in 0..1,

and crit

(desc

) is a function that evaluates whether a

description is an instance of a particular criteria (this

function is evaluated with the help of the DL reasoner,

and returns 1 if the belonging relationship is entailed).

The above method counts how many of the criteria

the activity fulﬁls by asking the DL reasoner whether

the activity description belongs to each of the criteria

deﬁnition concepts. Then, the MIDE information for

each particular activity (the evaluation of each of its

descriptions) is combined to establish whether a par-

ticular criterion is met. This is done by calculating

whether it is present in the descriptions in a percent-

age above a particular conﬁdence threshold. Finally,

SIWAM translates the agreed number of criteria into

the MIDE ﬁnal value (ranged in 1 – 5 values) by con-

sulting the information also stored in the module.

In the following section, we present two excerpts

of an actual description to illustrate each of the steps

that our system takes to process the information from

text to the actual evaluation.

6 COMPLETE EXAMPLE

Although our system is in its early stages of imple-

mentation, we have already carried out some concept

proofs that support the approach. In particular, we

have used descriptions that correspond to real experi-

ences of four different mountain ascensions located in

the Pyrenees, a range of mountains that forms a nat-

ural border between France and Spain. The original

texts are in Spanish, but we have translated the inter-

esting excerpts used in this section to illustrate how

SIWAM works.

In the ascension to the Petit Vignemale (3032 m),

one of the mountaineers wrote:

“... It took us a little more than 3 hours to

reach the Refuge of Oulettes de Gaube, which

was closed (we already knew it). This refuge

is located at a height of 2151 m., in an isolated

high mountain place, ...

... The temperature at such height is -18

◦

with a 20 km/h wind, ...”

From these excerpts (ex1), MECMIDE extracts

the following facts:

• From the subject of the description (is given by

RSAM):

ex1 isA Hike

• From the location excerpt:

refugeOulettesGaube isA InhabitedPlace

zone2 isA SingularElement

zone2 distantFrom refugeOulettesGaube

zone2 isWalkingTime 3

crossZone2 isA Cross

crossZone2 actionPerformedIn zone2

ex1 actionPerformed crossZone2

where zone2 is the area which the mountaineers

were traversing, and crossZone2 is the action of

traversing it.

• From the temperature excerpt:

ex1 hasMinTemperature -18

The instances created in the extraction are related

to the activity instance, as MECMIDE assumes that a

text contains the description of a single activity. Thus,

VALMIDE asserts these axioms in the Environment

module, and, with the help of a DL reasoner,infers the

following (recall the deﬁnitions for the criteria pre-

sented in Section 3):

• From the location axioms:

1. zone2 is a RemoteZone3H as it fulﬁls that is

a SingularElement and is distant an Inhabited-

Place (refugeOulettesGaube), and is at a walk-

ing distance of more than 3 hours.

2. crossZone2 is a CrossRemoteZone3H as it is a

Cross action performed in a RemoteZone3H.

WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies

3. ex1 is an instance of the concept deﬁnition as-

sociated to the MIDE criterion of crossing a re-

mote area at 3 hours (walking distance) as it has

an action that is a CrossRemoteZone3H.

• From the temperature axioms:

1. ex1 is an instance of the concept deﬁnition as-

sociated to the criterion about temperatures be-

low -10

◦

2. ex1 is also an instance of the concept deﬁnition

associated to the criterion about temperatures

below 0

◦

VALMIDE thus calculates the ﬁnal evaluation

counting the criteria that ex1 is instance of

. Note

how ex1 is added two points in the MIDE criteria due

to its temperature as it fulﬁls two differentcriteria (be-

low 0

◦

C and below 10

◦

C). This is coherent with the

way the MIDE standard evaluates the activity.

7 RELATED WORK

The spread and presence of mountain activities the-

matic sites on the Web have been quite low compared

to other ﬁelds. To the best of our knowledge, there is

no such an approach as ours in any of the current Web

sites about mountain activities.

Regarding the System Ontology, we have not

found any other ontology modeling the domain of

mountain activities. The closest works are related

to the tourism domain (Fodor and Werthner, 2005;

Prantner et al., 2007; Barta et al., 2009; Mouhim

et al., 2011), although they have different aims as ours

as they are mainly oriented to model the tourism do-

main within the context of the e-commerce. Anyway,

we have considered all of them to capture the vocab-

ulary of our ontology.

The most related works to SIWAM can be found

in the Information Extraction (IE) ﬁeld. In particular,

according to (Wimalasuriya and D., 2010), SIWAM

can be classiﬁed as an Ontology-Based Information

Extraction (OBIE) system, which is extended with the

reasoning capabilities of VALMIDE module. In this

context, there are several approaches oriented to auto-

matic content annotation (Cimiano et al., 2004; Buite-

laar et al., 2008). PANKOW (Cimiano et al., 2004)

processes Web pages looking for instances of a given

ontology, thus automatically annotating the Web page

with metadata about its content. SOBA (Buitelaar

et al., 2008) is oriented to obtain structured informa-

tion out from semi-structured resources (populate a

In this example, we are considering just one description

so the formulae in the previous section is reduced just to

count the criteria concepts which ex1 belongs to.

knowledge base). However, these systems aim only

at annotation and fact extraction, while SIWAM uses

this extracted information to perform the evaluation

of the different activities exploiting the model in the

ontology and the DL reasoner classifying capabilities.

Without leaving the Information Extraction ﬁeld,

it is worthy mentioning several approaches (Wu et al.,

2008; Cimiano and V¨olker, 2005) that have as its

main goal to construct the ontology that is behind the

processed information. For example, Kylin (Wu et al.,

2008) uses extraction techniques against Wikipedia’s

articles to obtain a structured schema out from them.

It uses also WordNet along with machine learning

techniquesto obtain the ﬁnal ontology. However,con-

structing the ontology is not the objective of SIWAM.

SIWAM exploits the domain model to detect impor-

tant facts in the extraction stage, and then, the model

is used to evaluate different aspects of each of the in-

put activities (via their descriptions).

8 CONCLUSIONS AND FUTURE

WORK

In this paper, we have presented SIWAM, a complete

framework to share information about mountain ac-

tivities. Apart from its social network features, SI-

WAM extracts and infers new information from the

descriptions provided by the users to help assessing

the difﬁculties of the different activities. This is done

to improve the information for the practitioners in or-

der to reduce the risks in the mountains. Moreover,

to the best of our knowledge, we have developed the

ﬁrst ontology aimed at modeling mountain activities

and at modeling a standard evaluation method such as

MIDE. Our system has the following features:

• It uses the descriptions provided by the users to

extract relevant facts aligned to an ontology. This

is a source of information which was almost un-

exploited in the mountain domain.

• It uses the extracted information to evaluate the

agreed difﬁculty of each activity with the help of

a DL reasoner. To do so, it exploits the inferring

capabilities of the reasoner, along with the evalu-

ation criteria deﬁnitions.

• It is completely ontology guided: the adoption of

a modular evaluation method makes it possible

to extend SIWAM with different evaluation stan-

dards in a ﬂexible and efﬁcient way.

We cannot make it available by the time we are writing

the paper due to several project restrictions.

SIWAM:UsingSocialDatatoSemanticallyAssesstheDifficultiesinMountainActivities

Currently, the prototype is under development, al-

though the preliminary results are very promising. As

future work, apart from testing the user behaviour, we

want to extend the modeled activities and include fur-

ther evaluation methods taken from the ﬁeld of rec-

ommender systems.

ACKNOWLEDGEMENTS

This research work has been supported by the CICYT

project TIN2010-21387-C02-02 and DGA-FSE.

REFERENCES

Baader, F., Calvanese, D., McGuinness, D., Nardi, D., and

Pastel-Scheneider, P. (2003). The Description Logic

Handbook. Theory, Implementation and Applications.

Cambridge University Press.

Barta, R., Feilmayr, C., Pr¨oll, B., Gr¨un, C., and Werthner,

H. (2009). Covering the semantic space of tourism:

An approach based on modularized ontologies. In

Proc. of the 1st Workshop on Context, Information and

Ontologies (CIAO’09), Heraklion (Greece), pages 1–

8. ACM.

Buitelaar, P., Cimiano, P., Frank, A., Hartung, M., and

Racioppa, S. (2008). Ontology-based information

extraction and integration from heterogeneous data

sources. International Journal of Human-Computer

Studies, 66(11):759–788.

Carreras, X., Chao, I., Padr´o, L., and Padr´o, M. (2004).

FreeLing: An open-source suite of language analyz-

ers. In Proc. of the 4th Intl. Conf. on Language Re-

sources and Evaluation (LREC’04), pages 239–242.

European Language Resources Association.

Chamarro, A. and Fern´andez-Castro, J. (2009). The per-

ception of causes of accidents in mountain sports: A

study based on the experiences of victims. Accident

Analysis & Prevention, 41(1):197–201.

Cimiano, P., Handschuh, S., and Staab, S. (2004). Towards

the self-annotating web. In Proc. of the 13th Intl. Conf.

on World Wide Web (WWW’04), New York (NY, USA),

pages 462–471. ACM.

Cimiano, P. and V¨olker, J. (2005). Text2Onto: A frame-

work for ontology learning and data-driven change

discovery. In Proc. of the 10th Intl. Conf. on

Natural Language Processing and Information Sys-

tems (NLDB’05), Alicante (Spain), pages 227–238.

Springer Verlag.

Fodor, O. and Werthner, H. (2005). Harmonise: A step to-

ward an interoperable e-tourism marketplace. Interna-

tional Journal of Electronic Commerce, 9(2):11–39.

Garrido, A. L., Buey, M. G., Ilarri, S., and Mena, E.

(2013). GEO-NASS: A semantic tagging experience

from geographical data on the media. In Proc. of the

17th East-European Conf. on Advances in Databases

and Information Systems (ADBIS’13), Genoa (Italy),

pages 56–69. Springer Verlag.

Garrido, A. L., G´omez, O., Ilarri, S., and Mena, E. (2012).

An experience developing a semantic annotation sys-

tem in a media group. In Proc. of the 17th Intl. Conf.

on Natural Language Processing to Information Sys-

tems (NLDB’12), Groningen (The Netherlands), pages

333–338. Springer Verlag.

Global Industry Analyst, Inc. (2012). Extreme sports: A

global industry outlook.

Gr¨uninger, M. and Fox, M. S. (1994). The role of com-

petency questions in enterprise engineering. In Proc.

of the IFIP WG5.7 Workshop on Benchmarking - The-

ory and Practice, Tronheim (Norway), pages 22–31.

Springer.

Jenkins, M. (2013). Maxed out on Everest. National Geo-

graphic, 32(6):94–113.

Kara, S., Alan,

O., Sabuncu, O., Akpınar, S., Cicekli, N. K.,

and Alpaslan, F. N. (2012). An ontology-based re-

trieval system using semantic indexing. Information

Systems, 37(4):294–305.

Miller, G. A. (1995). WordNet: A lexical database for En-

glish. Communications of the ACM, 38(11):39–41.

Ministerio del Interior (2012). Anuario es-

tad´ıstico del Ministerio del Interior.

http://www.interior.gob.es/ﬁle/62/62261/62261.pdf,

accessed September 13, 2013.

Mouhim, S., Aouﬁ, A. E., Cherkaoui, C. E., Douzi, H., and

Mammas, D. (2011). A knowledge management ap-

proach based on ontologies: The case of tourism. In-

ternational Journal of Computer Science & Emerging

Technologies, 2(6):362–369.

Prantner, K., Ding, Y., Luger, M., Yan, Z., and Herzog, C.

(2007). Tourism ontology and semantic management

system: State-of-the-arts analysis. In Proc. of IADIS

Intl. Conf. WWW/Internet 2007, Vila Real (Portugal),

pages 111–115. IADIS Press.

Roche, A. P. (2002). MIDE: M´etodo de Informaci´on de

Excursiones. Federaci´on Aragonesa de Monta˜nismo.

http://www.montanasegura.com/MIDE/manualMIDE

.pdf, accessed September 13, 2013.

Su´arez-Figueroa, M. C. (2012). NeOn Methodology for

Building Ontology Networks: Speciﬁcation, Schedul-

ing and Reuse. IOS Press.

UIAA Mountaineering Commission (2004). To Bolt or not

to Be.

Vogrincic, S. and Bosnic, Z. (2011). Ontology-based multi-

label classiﬁcation of economic articles. Computer

Science and Information Systems, 8(1):101–119.

Vossen, P. (1998). EuroWordNet: a multilingual database

with lexical semantic networks. Kluwer Academic

Boston.

Wimalasuriya, D. C. and D., D. (2010). Ontology-based in-

formation extraction: An introduction and a survey of

current approaches. Journal of Information Science,

36(3):306–323.

Wu, F., Hoffmann, R., and Weld, D. S. (2008). Information

extraction from Wikipedia: Moving down the long

tail. In Proc. of the 14th ACM Intl. Conf. on Knowl-

edge Discovery and Data Mining (SIGKDD’08), Las

Vegas (NV, USA), pages 731–739. ACM.

WEBIST2014-InternationalConferenceonWebInformationSystemsandTechnologies