Currently, keyword based Web searches are the preferred way to seek for resources of interest over the Web. Each resource, usually identified by its URL, can be accessed by one or more keywords describing its content. The most widespread methods to explore links between Web resources and
1.
Nowadays the Web represents a growing collection of an
enormous amount of contents where the need for better
ways to find and organize the available data is becoming
a fundamental issue, in order to deal with information
overload. Keyword based Web searches are actually the
preferred mean to seek for contents related to a specific topic.
Search engines and collaborative tagging systems make
possible the search for information thanks to the association of
descriptive keywords to Web resources. All of them show
problems of inconsistency and consequent reduction of
recall and precision of searches, due to polysemy, synonymy
and in general all the different lexical forms that can be used
to refer to a particular meaning. A possible way to face or
at least reduce these problems is represented by the
introduction of semantics to characterize the contents of Web
resources: each resource is described by one or more concepts
instead of simple and often ambiguous keywords. To support
these task the availability of a global semantic resource of
reference is fundamental. On the basis of our past experience
with the semantic tagging of Web resources and the SemKey
Project, we are developing Tagpedia, a general-domain
”encyclopedia” of tags, semantically structured for generating
semantic descriptions of contents over the Web, created by
mining Wikipedia. In this paper, starting from an
analysis of the weak points of non-semantic keyword based Web
searches, we introduce our idea of semantic characterization
of Web resources describing the structure and organization
of Tagpedia. We introduce our first realization of Tagpedia,
suggesting all the possible improvements that can be carried
out in order to exploit its full potential.
common sense of that keyword, it is often very difficult to
explore the search results space so as to find Web resources of
interest among those retrieved. For example, let us suppose
that we want to find all the resources dealing with ’ajax’
intended as the Greek hero: choosing ’ajax’ as search text
string, there are no links related to mythology among the
first 30 search results of Google. If we better specify the
search string in order to solve the problem, we partition the
space of relevant search results depending on the particular
word added to ’ajax’ to disambiguate its meaning. For
instance, depending on the addition of the word ’hero’ or the
word ’mythology’ to ’ajax’ in the search string, considering
the first 10 search results shown by Google, only two of them
are present in both cases. Besides polysemy, also synonymy
affects precision and recall of keyword based Web searches.
In fact, when a specific meaning can be accessed through
two or more keywords, the set of search results is
different depending on the particular keyword chosen. Moreover,
the different level of precision and the many possible users
points of view that can be considered describing a particular
resource, often cause a considerable loss of quality of Web
searches. For a deeper analysis of all the factors that affect
efficiency and effectiveness of keyword based Web search
systems see [
In order to face the different drawbacks of the systems just
analyzed, many distinct methods have been applied. The
aggregation of search results from different search
engines and their post elaboration is experimenting a
growing diffusion. Systems like Vivisimo [
Also considering tagging systems, we can find many
proposals to better organize search results to improve their
quality and the effectiveness of the search. FolkRank [
The rest of this paper is organized as follows. In Section 2 we describe our idea of semantic characterization of Web resources, underlining the need for a general-domain semantic resource of reference in order to support this task, taking into account also our past experience with the semantic tagging and SemKey. In Section 3 we introduce Tagpedia, the semantic resource of reference we have created by mining Wikipedia, explaining its organization and structure (Subsection 3.1). In Section 4 we describe how Tagpedia can be utilized, describing the Tagpedia Web API and showing all the possible improvements to Tagpedia to exploit its full potential. Conclusions are described in Section 5. 2.
We can solve, or at least substantially reduce, Web resources organization and classification problems by adding a further level of completeness in their characterization: the semantics. Instead of relying on post processing of search results, we can directly semantically describe resources thanks to their association with one or more properly chosen concepts. In this way we extend the characterization of resources introducing the semantic level: each resource (R) is described by one or more concepts (C) and in turn each concept can be accessed through one or more keywords (K) (see Figure 2). When we search for some informaton of interest, we can better specify our informative needs and we can easily and effectively access relevant results thanks to the support and the exploitation of the collection of concepts used to describe Web resources, referred to as semantic resource in what follows.
This way of improving Web contents organization represents an attempt to realize the semantic description of information that stands at the basis of the Semantic Web vision.
At present there are many proposal of semantic
classification methods for Web contents. FolksAnnotation [
During the first half of 2007, we have tried to realize the
possibility to semantically describe Web resources
developing SemKey [
We have implemented a working prototype of SemKey; by analyzing the usage patterns and the semantic classification support provided by our system, we have identified two key factors that need to be improved in order to really make possible semantic characterization of Web resources, as described in the previous part of this Section.
Both Wikipedia and WordNet, even if they show
important features to support the semantic description of Web
resources, are weakened by relevant lacks. WordNet presents
a rich set of parts of speech and a strongly structured set
of relations between them, but it lacks many data useful to
support proper names disambiguation and it is not
collaboratively edited. Wikipedia is an encyclopedia so its
content is composed mainly by a very rich set of names along
with their extended descriptions. Thus Wikipedia has strong
proper names coverage and it has been proposed as a named
entity disambiguation resource in [
Moreover, a great limit to the usability of SemKey and to an easy definition of new semantic metadata is represented by the different steps users must carry out to compose a semantic assertion. This often discourages them from creating semantic metadata. Some sort of automation is necessary in order to speed up the tag disambiguation process or to execute it through automated procedures. 3.
Starting from the need for a global semantic resource
exploitable as a reference to describe Web contents and
therefore comprehensive and updated, we have proposed a
possible solution to this demand, designing and building
Tagpedia. It is a semantic organization and classification of
tags, intended as words or in general brief textual
expressions, that people may use to describe Web
Resources. Tagpedia is based on the model of
termconcept networks [
We have chosen Wikipedia as the starting point because
it represents the most rich and constantly updated
encyclopedic reference over the Web with a huge
set of semantic contents included, even if not
explicitly exposed and easily accessible. During the last
few years many studies have been carried out finding new
ways to extract useful semantic data exploiting the great
amount of information contained in Wikipedia. Information
organizational patterns like infoboxes, internal and external
links, redirect and disambiguation pages have been analyzed
in order to extract valuable data. The DBPedia Project
[
The main aim of Tagpedia is the semantic characterization of data over the Web. In particular it must allow to describe a Web resource through the association with one or more univocally referenced concepts. Thus, the main constitutional unit of TagPedia is the concept. Each concept must be unequivocally identified but also easily accessed. The main way to point out a concept is through the words that refer to it. Such words will also be called tags in the following. As a consequence, each concept is identified by the set of all the words or, more generally, all the alfanumeric expressions of any kind that can be adopted by a community of users to refer to it, thus constituting a set of synonymous tags or syntag set. Syntag sets are the molecules which form Tagpedia.
The creation of an initial rich collection of syntag sets is the first necessary step that must be carried out to build our semantic resource. Wikipedia shows many features exploitable to create such a collection of syntag sets. In particular, in Wikipedia an article usually defines a specific concept. As a consequence in order to bootstrap Tagpedia, we create syntag sets from the articles of Wikipedia. In Figure 3 we show three examples of syntag sets made up by tags collected mining Wikipedia.
To be more precise, Wikipedia pages can be substantially divided into three groups: • article pages: each describes a particular concept, identified by the title of the same page; • redirect pages: each links an alternate literal expression, that constitutes the title of the redirect page, to the corresponding concept, usually identified by the title of an article page; • disambiguation pages: each lists all the possible concepts, usually identified through the titles of article or redirect pages, that can be referred by the literal expression constituting the title of the disambiguation page.
The redirect and the disambiguation page mechanisms are two important Wikipedia organizational solutions that can be exploited to build and enrich syntag sets.
Once identified a concept referring to a particular article page, we create an initial version of a syntag set, pointed out by a unique identifier, including only the tag corresponding to the title of the page (in Figure 3, considering the syntag set 1, the tag ’jaguar’ is the title of an article page). Then we collect all the words and expressions that may be used to refer to that concept.
As previuosly mentioned, in Wikipedia the redirect mechanism is used to link alternate literal expressions to the original encyclopedic article that describes a specific entity. It is usually used to manage synonyms, abbreviations, acronyms, misspellings, other spellings, different punctuations, particular capitalization rules and so on. In TagPedia we mine Wikipedia content and extract all the redirect information analyzing redirect pages; for each of them we enrich the syntag set related to the referred concept by adding the title of the page as a new tag (in Figure 3, considering the syntag set 1, the tag ’leo onca’ is extracted from a redirect page).
Moreover, Wikipedia usually manages polysemy through the disambiguation pages. As said, each disambiguation page represents a collection of links to all the different article pages that identify the distinct meanings pointed out by the page title (textual string). For example, the word ’ajax ’ is highly polysemous and has 49 different meanings in Wikipedia: its disambiguation page contains links to 49 distinct article pages; each one identifies a particular concept. We analyze Wikipedia disambiguation pages as a futher source of information to enrich the syntag sets of Tagpedia through the addition of new words that refer to a defined meaning. In particular, for every disambiguation page, we point out each syntag set related to the concepts referenced inside its Wikipedia text and we add the title of the same disambiguation page as a new tag exploitable to access to the selected syntag sets (in Figure 3, considering the syntag set 1, the tag ’panther’ is extracted from a disambiguation page).
Summarizing, let us define Ci a concept derived from a specific Wikipedia article page Pi. To populate with tags the syntag set for Ci we extract: • the title of Pi; • the title of every redirect page to Pi; • the title of every disambiguation page containing a link to Pi.
Starting from a dump of the English version of Wikipedia, we have developed a set of C++ routines, that automatically analyze the text of Wikipedia articles. By mining structural elements of Wikipedia syntax as well as by considering texts punctuation and by exploiting pattern matching techniques mainly based on regular expressions and string analysis, our routines gather all the concepts as well as all the possible tags used to refer to each single meaning, thus defining a huge collection of syntag sets. The meaning of each concept, identified by a syntag set, is also better specified by pointing to the corresponding article in Wikipedia.
All these data are collected in a relational database properly designed and optimized for a fast access. It is constituted by two basic collections: the concept table and the tag table. The first one gathers all the concepts of Tagpedia assigning to each of them a unique identifier, the Concept ID and a brief definition, extracted from the English version of Wikipedia. For every concept we also collect the URL of the corresponding Wikipedia article. On the other side, the tag table contains links between each concept, referenced through its identifier, and all the tags used to access to it.
By mining September 2007 dump of the English version of Wikipedia, we have obtained more than 1,9 millions of syntag sets and more than 4 millions of tags used to point out the intended concepts, each one referencing a specific Wikipedia article (see Figure 4).
Considering Figure 5, we can visualize the weight of the different sources of the 4.230.740 tags of Tagpedia. The number of tags extracted from article pages (P ) is equal to the number of syntag sets, that is 1.927.378. Among the 2.303.362 remaining tags, 481.250 have been generated by mining disambiguation pages and 1.822.112 by analyzing redirect ones. This group of syntag sets constitutes the basis of Tagpedia providing a way to unequivocally access and refer to concepts when users must semantically describe or search for Web resources.
In order to support the generation of semantic descriptions of Web resources or to semantically search for Web contents, the information contained in Tagpedia should be easily accessed, querying the whole collection of syntag sets. For this purpose we have developed the Tagpedia Web API that is a simple set of procedures that may be invoked via Web to exploit the semantic support offered by Tagpedia. These procedures carry out few fundamental tasks and may be composed to realize more complex functions; their execution can be easily requested by other external Web applications so as to integrate semantic features.
The main tasks that Tagpedia Web API supports are: • the definition of all the possible meanings for a given tag, i.e. all the syntag sets that contain the tag; • the collection of all the tags belonging to a specific syntag set, i.e. all the words or expressions exploitable to access that particular meaning; • the retrieval of the short textual description of a specific syntag set.
Exploiting Tagpedia Web API, we have integrated this semantic resource into SemKey, our semantic collaborative tagging system, substituting WordNet and Wikipedia so as to support the disambiguation of the meaning of tags. Once chosen one or more tags, the user specifies the right meaning for each of them, choosing a particular syntag set among those including the intended tag. An early prototypal Webbased interface useful to explore and interact with Tapgedia is accessible at the URL www.tagpedia.org.
In order to evaluate the coverage of Tagpedia and also to obtain suggestions to improve this semantic resource, we have tried to manually point out the right meaning of the tags associated to the 100 most popular Web resources over del.icio.us, tagged by more than 25000 users. Relying upon Tagpedia Web API, we have developed a Web based procedure that, starting from the URL of a Web resource retrieves all the related tags in del.icio.us. All the possible meanings of each tag are retrieved from Tagpedia along with their short descriptions and the user manually verify if the right concept is present. In this way, collecting all the results of our user based tests, we have obtained a first evaluation of the disambiguation effectiveness of our semantic resource. The results are shown in Table 1.
Tagpedia provides a valid support to the process of disambiguation for 84% of the total number of tags considered.
Anyway we have identified several different ways to improve its contents and, as a consequence, its semantic coverage and its usefulness. In the following part of this section we will describe these proposals for future works.
Despite its good disambiguation coverage, there are different particular tags like ’sem web’, ’inplaceedit’, ’web dev’ and similar ones that are not managed by Tagpedia, because they are non conventional words, often created by a user to describe a particular concept and then accepted and exploited by many others. One possible solution to this problem is the introduction of collaborative Web editing techniques for Tagpedia contents. Giving users the possibility to create new syntag sets or to merge or extend existing ones through new tags is fundamental for such a kind of resource. Indeed the effectiveness of Tagpedia in the description of Web resources is proportional to the possibility to adapt and enrich this semantic resource in respect to the variability of user descriptive needs. In this context, the introduction of the possibility to collaboratively collect and manage data, following a Wiki-like paradigm, represents a key factor of current Web and is a crucial issue considering Tagpedia.
Another aspect of Tagpedia that can be substantially improved is the enrichment of its semantic contents with the addition of semantic relations between syntag sets; they are useful to better identify concepts or to easily search for them. Each syntag set, representing a meaning, may be connected to other ones through relationships like specialization, generalization, relatedTo and similar ones. Possible ways to mine relevan relations between syntag sets are the analysis of the internal links between Wikipedia article pages as well as the exploitation of the hierarchy of Wikipedia categories. For instance, relying upon relations, when we specify the concept to search for or when we must choose a specific concept to semantically characterize a resource, the system can show the most general or the most specific ones to simplify this task. Similarly, during a semantic search, starting from a specific syntag set, if we can browse all the related ones, we can better specify our search needs and thus easily retrieve the desired information.
A third way to improve and enrich Tagpedia is the
definition of semi-automated procedures to extend its
data, exploiting other resources and importing their
contents into Tagpedia. Other relevant free Web thesauri
or dictionaries or other language tools can be valid sources
of information. For instance the Dictionary of Automotive
Terms [
Another aspect that must be further addressed in Tagpedia, is the support for multilinguism. In Tapgedia, each syntag set is language independent. The tags constituting that particular syntag set are specific to the particular language. Managing the possibility to collect different tags belonging to different languages into a syntag set, we can deal with different languages and once identified one or more particular concepts we can make language indipendent semantic searches. We think that this possibility should be better explored and defined, trying to determine specific semantic search patterns.
As already mentioned in the concluding part of Section 2, the definition and tuning of automated or semiautomated procedures to create semantic descriptions is a further important issue to be faced. Users should be allowed to semantically describe Web resources in an easy way; they must be supported in the task of turning simple keywords into concepts or browsing the collection of syntag sets constituting Tagpedia without complicating their usual interaction patterns or compromising the usability of the systems they interact with. Moreover, automated methodologies to derive semantic descriptions of Web resources from simple keyword based ones can also be tuned, so as to create an initial solid collection of semantic metadata and bootstrap this new way to characterize resources over the Web. 5.
In this paper we have presented Tagpedia, a collection of tags semantically structured, built ad hoc to describe Web contents.
Starting from a brief analysis of the weak points of keyword based methodologies for information organization and searching and considering also the current approaches to face these issues, we have introduced the possibility to semantically describe Web resources through concepts. To make it possible, we have developed an initial version of Tagpedia a general domain semantic resource of reference, created by mining Wikipedia. After a description of its structure and organization and an overview of the Tagpedia Web API, useful to easily access and exploit the information collected in Tagpedia, we have focused our attention on the possible improvements to this semantic resource. Collaborative wiki authoring, syntag set relations enrichment, automated procedures for content extraction from external sources, support for multilinguism and automated generation of semantic descriptions of Web resources are some of the many improvements considered that can be carried out, underlining its broad enhancement possibilities.
On the base of all these considerations, we believe that Tagpedia, despite its initial stage of development, represents an important attempt to support the introduction of semantics over the Web, trying to put in practice the principles of the Semantic Web on a global scale and to better structure and manage the huge amount of data constituting the actual Web.