Social Semantic Web aims at combining approaches and technologies from both Social and Semantic Web. While Social Web sites provide a rich source of unstructured information, what makes its automatic processing very limited, Semantic Web aims at giving a welldefined meaning to the Web information, facilitating its sharing and processing. Multilinguality is an emergent aspect to be considered in Social Semantic Web and its realization is highly dependent on the development of multilingual ontologies. This paper presents Hontology, a multilingual ontology for the hotel domain. Hontology has been proposed in the context of a framework for ontology-driven mining of Social Web sites content. Comments are annotated with concepts of Hontology, which are labeled in three different languages. This approach facilitates the task of comments mining, helping managers in their decision-making process.
Social Web focuses on social interaction mainly through comments in social sites.
Its rapid growth has created a huge unstructured and multilingual knowledge
base, what essentially makes its automatic processing very limited. On the other
hand, Semantic Web aims at giving a well-defined meaning to information on
the Web, better enabling cooperation between software agents and people [
Social Web and Semantic Web have been integrated into the called Social
Semantic Web [
The realization of the Multilingual Social Semantic Web is highly dependent
on the development of multilingual ontologies. Different approaches have been
proposed for dealing with ontology multilinguality [
This paper presents the main ideas behind Hontology, a multilingual
ontology for the hotel domain. Hontology has been manually created and its current
version supports English, French and Portuguese languages. Each concept and
property of Hontology are manually annotated with different labels in these three
languages. Although for dealing with the huge source of knowledge at the web
scale, automatic methods for creating and populating ontologies are required,
Hontology can be seen as a starting point to these approaches. Hontology has
been proposed in the context of a framework for annotating comments provided
by users in social web sites [
This paper is structured as follows. Section 2 gives an overview of the framework in which Hontology is being proposed. Section 3 shows the details of Hontology and describes the methodology we have followed to develop it. A comparison between Hontology and related ontologies is presented in Section 4. Section 5 discusses the main approaches we are exploiting for extending Hontology. Section 6 discusses on other related work. Finally, Section 7 concludes the paper. 2
The inspiration for a framework for annotating comments from Social Web sites
comes from the gathered needs in Customer Knowledge Management (CKM)
research [
The hotel domain is strongly affected by the comments written in social sites. These comments often help guests to decide whether making a reservation. On the other hand, Hotel managers need tools to better explore customers knowledge from Social Web to support their decision-making task. To that extend, we have proposed a framework that aims at integrating knowledge from Social Web to support CKM (Figure 1). A multilingual ontology is the central element in this framework.
The entry point of the framework is the set of comments (raw text) from Social Web sites, e.g., “booking.com” or “realtravel.com”. Some Social Web sites provide together with the comments, the user profile (e.g., family with young children or mature couples). These data (comments and profile) are then pre3 http://olp.dfki.de/ontoselect processed and stored into a knowledge base (KB). Three main processes are carried out in this pre-processing phase: Content extraction deals with the selection of multiple information sources, i.e., different Social Web sites, and format, e.g., HTML, in which these information sources are provided; Content transformation selects parts of the texts to be loaded, e.g., comment and its classification, and join data from multiple sources, e.g., join the comments of the profiles “family” with “family with young children”, since information sources have different classifications to the profiles; Loading classifies the comments, i.e., positive, negative or neutral, and store the classification into the KB. The classification task involves, basically, to identify adjectives, e.g., “good” and “satisfactory” or “not good” and “unsatisfactory” in comments in order to infer such classification.
Next, the comments are annotated with the concepts from the ontology, via the module comment annotator. This module receives as input the pre-processed comments and compares terms of these comments to labels of concepts in the ontology. Recognizing the language in which a comment is written is a precondition to annotate it. The comment is annotated with the corresponding ontology concept if the degree of similarity between them is above a specific threshold. In a first approach, we combine syntactic approaches, i.e., based on string similarity, such as string equality, sub-string and edit distance. Comments and their annotations are then stored into the KB.
Furthermore, Hontology can be augmented with new concepts, via the module ontology augmenter. This module is responsible for identifying potential new concepts in comments, which are then filtered out and validated by a user expert. This identifying process considers terms correlation, rules (lexical patterns) and synonyms, as detailed in Section 5.
Finally, the manager can navigate within the concepts of the ontology and retrieve the comments annotated with the corresponding concepts. The advantage of using a multilingual ontology-driven navigation is two fold. On the one hand, once the manager has selected a concept (in English, for instance), all comments in all languages the ontology supports (including the synonyms of these concepts) are presented to the manager. This facility is not available in traditional search engines. On the other hand, it allows for the manager specializing and generalizing queries in an intuitive way, according to the ontology hierarchy. For instance, searching for the concept “Laundry”, both sub-concepts “Laundry room” and “Laundry service” are included in the search as well as their synonyms (“pressing”, for instance) and corresponding terms in different languages (“Blanchisserie” and “Lavanderia”). 3
3.1
The development of Hontology has been carried out following seven main steps: 1. Identify existing ontologies on related domains: The first step was to search for ontologies on the hotel domain and related domains, such as tourism or traveling, in order to use them as a staring point for organizing the knowledge in our ontology. 2. Select the main concepts and properties: Based on the available related ontologies, we have filtered out the main concepts we have judged as interesting for considering in our ontology. These related ontologies are often built to be used on specific applications, using very specific concepts as well as they are monolingual. We compare Hontology with related ontologies in Section 4. 3. Organize concepts and properties hierarchically into categories: Hontology is neither a merge of existing ontologies nor an union of them. In the prior step, we have selected certain concepts and properties, which were manually re-structured into a hierarchy. 4. Translate the ontology: One of the main steps in the development of Hontology was to add, for each concept and property, different labels in different languages, including synonymous. This task was carried out by bilingual experts. 5. Expand concepts and properties based on comments: The process of expanding the ontology considers three main approaches, namely terms correlation, rules (lexical patterns) and synonyms, as detailed in Section 5. 6. Translate the new concepts and properties: The same as phase 4. For instance, the new concept “pillow” must have associated labels in French and Portuguese languages. 7. Export the ontology in several formats: Applications using Hontology explore different levels of formality. Some of them need to perform reasoning, while other just work with a flat list of concepts and properties. In order to satisfy these needs, we make Hontology available in OWL, RDF and XML. 3.2
Hontology contains seven top concepts, which represent the corresponding sub-domains: Category: contains all the types of categories into which a hotel can be classified, e.g., tourist, comfort and luxury.
Facility: includes the utility options offered by each hotel, e.g., beauty salon, kids club and pool bar.
Hospitality: contains the existing kinds of hotels, e.g., hostel, pension and motel.
Hotel: details the kind of hotels, e.g., bunker, cave and capsule. Leisure: lists the leisure options, e.g., gym, jacuzzi, and sauna. Points of interest: includes the main points, usually near to the hotel, which are most often mentioned in comments about the hotels, e.g., stadium, museum and monument.
Room: splits into Hostel Room and Hotel Room, which have different kinds and nomenclature for rooms.
In its current version, Hontology has 97 concepts, 9 object properties and 25 data properties. To the best of our knowledge, Hontology is the first multilingual ontology for the hotel domain. We have compared Hontology with related ontologies, as commented in the next section. 4
Few ontologies for the hotel domain and related domains have been proposed,
which cover different aspects in this domain. Furthermore, these ontologies are
monolingual, what makes their use limited in the context of Multilingual Social
Semantic Web. In this section, we compare Hontology with other relevant
ontologies which have some relation with the hotel domain: Mondeca4, HarmoNET5
and Travel Itinerary6. Hontology is freely available at mchaves.wikidot.com/
hontology. Ontologies describing concepts of hotels are often found as
subontologies of the tourism domain [
Mondeca ontology is the largest one in number of concepts. However, it is neither public nor freely available, which restrict its usage. HarmoNET and Travel Itinerary are public and freely available. The first one describes accommodation and events concepts and the latter is used for representing traveling concepts. One limitation of these ontologies, as well as in Hontology, is that they do not contain instances associated to the schema. It is another aspect that restrict their reuse. In this sense, we are currently working on how populate Hontology with instances.
Hontology was built based on the concepts and properties of these existing ontologies. However, Hontology is not a merge of them. For instance, concepts like “Transport” and “Multimedia Item” from HarmoNET and “Flight” and “Meal” from Travel Itinerary are out of the scope of Hontology.
5 harmonet.org 6 daml.org/ontologies/178
This comparison evidences the lack of the multilingual ontologies in the hotel domain. Hontology is public and freely available for the research community and then can be used as a baseline for constructing new ontologies. This is an important point for promoting the development of new multilingual ontologies. 5
We are working on the improvement of Hontology, in the main directions commented below.
First, the module ontology augmenter, commented in Section 2, aims at enriching Hontology with relevant information from comments. This task exploits the following heuristics: term correlation, rules (lexical patterns) and synonyms. Term correlation considers potential terms mentioned in the comments, which are present in Hontology. For instance, in a comment containing the sentence “Rooms are comfortable, but pillows are very hard” the terms “pillow” (in the ontology) and “room” (not in the ontology) should be probably related through a property linking them in Hontology. Once the ontology is enriched with the term “pillow”, a comment containing, for instance, only the sentence “Pillows are very hard” can be found under the concept “room”.
Rules (or lexical patterns) consider that comments usually contain a set of common adjectives, e.g., “good”, “cheap” and “soft”. This approach uses lexical patterns and extract relevant terms which are preceding or succeeding the adjective, e.g., “Air-conditioned is loud”, “Small bathroom”.
Furthermore, synonyms are important elements that must be considered in the improvement of Hontology. They have already being considered in the process of adding labels to the concepts. However, this task can be extended with the help of dictionaries and lexical resources within an automatic process.
Second, we plan to work on multilingual ontology matching [
Third, we are working on extending Hontology for including labels of
concepts in other languages such as Spanish and Italian. Two experts are currently
working on this task. Moreover, from the existing concepts and properties in
English, Portuguese and French, we intend to apply some techniques of ontology
localization. We are considering to create a linguist information repository, such
as in [
Finally, Hontology can be used as a multilingual resource to cross-language information retrieval. Cross-Language Evaluation Forum (CLEF)7 has challenged multilingual systems to search in documents written in several languages. Queries and questions on hotel domain can be supported by Hontology, since the main concepts and properties are present in it. For instance, a query containing “hotels with jacuzzi” can be automatically translated to Portuguese and French with the support of Hontology. 6
Multilinguality in ontologies has been exploited on different perspectives. First,
tools for supporting automated inclusion of multilingual labels in ontologies have
been proposed. Espinoza et al. describe a tool for automatically localizing
ontologies, i.e., adapting an ontology to a concrete language and cultural community
[
Another approach involves to interface ontologies and lexical resources [
Although these efforts, no concrete ontologies have been made publicly available, as commented in Section 4. Our aim is to share with the community a multilingual ontology that can be extended by using the tools and approaches described above. Moreover, Hontology can be used as a baseline for evaluating this kind of tools.
This paper has presented Hontology, a multilingual ontology for the hotel domain. Hontology gives support for annotating comments from Social Web sites in the context of a framework for Customer Knowledge Management.
In its preliminary version, Hontology has been manually created and supports three languages. Our main contribution is to make available for the community, a multilingual ontology that can be used as a baseline for many usages and applications in the context of the Multilingual Semantic Web, promoting its realization.
As future work, we plan to extend Hontology, in the following main directions:
enrich Hontology by using potential terms from comments themselves; exploit
Hontology in Multilingual Ontology Matching; include labels in other languages;
explore issues related to ontology localization and internationalization. In
addition, we plan to apply some machine-learning methods for the sentiment analysis
on comments. The main idea is to classify a comment as “positive”, “negative” or
“neutral”, for instance, what can hep hotel managers in their analysis. We plan
to exploit method, such as SO-polarity (Subjective Objective) and PN-polarity
(Positive-Negative), in order to determine the strong of comment PN-polarity,
i.e., weakly positive, mildly positive, or strongly positive [