In this paper we present a survey of web information extraction systems and semantic annotation platforms. The survey is concentrated on the problem of employment of these tools in the process of web semantization. We compare the approaches with our own solutions and propose some future directions in the development of the web semantization idea.
There exist many extraction tools that can process web pages and produce structured machine understandable data (or information) that corresponds with the content of a web page. This process is often called Web Information Extraction (WIE). In this paper we present a survey of web information extraction systems and we connect these systems with the problem of web semantization.
The paper is structured as follows. First we sketch the basic ideas of semantic web and web semantization. In the next two sections methods of web information extraction will presented. Then description of our solutions (work in progress) will continue. And finally just before the conclusion we will discuss the connection of WIE systems with the problem of web semantization.
The idea of the Semantic Web [4] (World Wide Web dedicated not only to human but also to machine – software agents) is very well known today. Let us just shortly demonstrate its use with respect to the idea of Web Semantization (see in next section).
The Fig. 1 shows a human user using the (Semantic) Web in three possible manners: a keyword query, a semantic query and by using a software agent. The difference between the first two manners (keyword and semantic query) can be illustrated with the question: “Give me a list of the names of E.U. heads of state.” This example from interesting article [16] by Ian Horrocks shows the big difference between use of a semantic query language instead of keywords. In the semantic case you should be given exactly the list of names you were requesting without having to pore through results of (probably more then one) keyword queries. Of course the user have to know the syntax of the semantic query language or have a special GUI1 at hand.
The last and the most important possibility (in the semantic or semantized setting) is to use some (personalized) software agent that is specialized to tasks of some kind like planning a business trip or finding the most optimal choice from all the relevant job offers, flats for rent, cars for sale, etc.
Both the semantic querying and software agents engagement is actually impossible to realize without any kind of adaptation of the web of today in the semantic direction. 1.2
The idea of Web Semantization [9] consist in gradual enrichment of the current web content as an automated process of third party annotation for mak1 Such handy GUI can be found for example in the KIM project [20]. General Applicable
Domain Specific
Structure of
Document e.g. HTML tables Specific
Form Specific
Text
ing at least a part of today’s web more suitable for The distinguishing between general applicable machine processing and hence enabling it intelligent methods and the others that have meaningful applitools for searching and recommending things on the cation only in some specific setting (specific domain, web (see [3]). specific form of input) is very important for Web Se
The most strait forward idea is to fill a seman- mantization because when we try to produce annotic repository with some information that is automat- tations in large scale, we have to control which web ically extracted from the web and make it available to resource is suitable for which processing method (see software agents so they could access to the web of to- in Sect. 5). day in semantic manner (e.g. through semantic search engine).
The idea of a semantic repository and a public ser- 2.1 General applicable vice providing semantic annotations was experimentally realized in the very recognized work of IBM Almaden Research Center: the SemTag [13]. This work demonstrated that an automated semantic annotation can be applied in a large scale. In their experiment they annotated about 264 million web pages and generated about 434 millions of semantic tags. They also provided the annotations as a Semantic Label Bureau – a HTTP server providing annotations for web documents of 3rd parties.
The most significant (and probably the only one) generally applicable IE task is so called Instance Resolution Task. The task can be described as follows: Given a general ontology, find all the instances from the ontology that are present in the processed resource. This task is usually realized in two steps: (1) Named Entity Recognition (see in Sect. 3.1), (2) Disambiguation of ontology instances that can be connected with the found named entities. Success of the method can be strongly improved with coreference resolution (see in Sect. 3.1).
Let us mention several good representatives of this approach: the SemTag application [13], the KIM project [20] and the PANKOW annotation method [7] based on smart formulation of Google API queries.
The task of a web information extraction system is to transform the web pages into program-friendly structures such as a relational database. There exists a rich variety of Web Information Extraction systems. The results generated by distinct tools usually can not be 2.2 Domain specific directly compared since the addressed extraction tasks are different. The extraction tasks can be distinguished Domain and from specific IE approaches are the typaccording several dimensions: the task domain, the au- ical cases. More specific information is more precise, tomation degree, the techniques used, etc. These di- more complex and so more useful and interesting. But mensions are analyzed in detail in the recent publica- the extraction method has to be trained to each new tions [6] and [18]. Here we will concentrate on a lit- domain separately. This usually means indispensable tle bit more specific division of WIE according to the effort. needs of the Web Semantization (see in Sect. 5). The A good example of domain specific information exdivision is demonstrated on the Fig. 2 and should traction system is SOBA [5]. This complex system is not be considered as disjoint division of the methods capable to integrate different IE approaches and exbut rather as emphasization of different aspects of the tract information from heterogeneous data resources, methods. For example many extraction methods are including plain text, tables and image captions but domain and form specific at the same time. the whole system is concentrated on the single domain of football. Next similarly complex system is ArtEquAKT [1], which is entirely concentrated on the domain of art. 2.3
Beyond general applicable extraction methods there exist many methods that exploit specific form of the input resource. The linguistic approaches usually process text consisting of natural language sentences. The structure-oriented approaches can be strictly oriented on tables [19] or exploit repetitions of structural patterns on the web page [21] (such algorithm can be only applicable to pages that contain more than one data record), and there are also approaches that use the structure of whole site (e.g. site of single web shop with summary pages with products connected with links to pages with details about single product) [17]. 3
In this section we will discuss the information extraction from textual resources. 3.1
There are classical tasks of text preprocessing and linguistic analysis like Text Extraction – e.g from HTML, PDF or DOC, Tokenization – detection of words, spaces, punctuations, etc., Segmentation – sentence and paragraph detection, POS Tagging – part of speech assignment, often including lemmatization and morphological analysis, Syntactic Analysis (often called linguistic parsing ) – assignment of the grammatical structure to given sentence with respect to given linguistic formalism (e.g. formal grammar),
resolving what a pronoun, or a noun phrase refers to. These references often cross boundaries of a single sentence.
Besides these classical general applicable tasks, there are further well defined tasks, which are more closely related to the information extraction. These tasks are domain dependent. These tasks were widely developed in the MUC-6 conference 1995 [15] and considered as semantic evaluation in the first place. These information extraction tasks are: WIE systems for Web Semantization
and classifies named entities such as persons, locations, date or time expression, or measuring units. More complex patterns may also be recognized as structured entities such as addresses.
plates describing entities with extracted roles (or attributes) about one single entity. This task is often performed stepwise sentence by sentence, which results in a huge set of partially filled templates.
late describes information about one single entity, this tasks identifies semantic relations between entities.
Template Unification: Merges multiple elementary templates that are filled with information about identical entities.
of Template Element Construction and Template Relation Construction into templates describing pre-specified event scenarios (pre-specified“queries on the extracted data”).
3.2
Contrary to the WIE methods based on the web page structure, where we (the authors) do not know about any well established benchmark for these methods2, the situation in the domain of text based IE is fairly different. There are several conferences and events concentrated on the support of automatic machine processing and understanding of human language in text form. Different research topics as text (or information) retrieval3, text summarization4 are involved.
On the filed of information extraction, we have to mention the long tradition of the Message Understanding Conference5 [15] starting in 1987. In 1999 the event of Automatic Content Extraction (ACE) Evaluation6 started, which is becoming a track in the Text Analysis Conference (TAC)7 this year (in 2009). 2 It is probably at least partially caused by the vital development of the presentation techniques on the web that is still well in progress. 3 e.g. Text REtrieval Conference (TREC)
http://trec.nist.gov/ 4 e.g. Document Understanding Conferences
http://duc.nist.gov/ 5 Briefly summarized in http://en.wikipedia.org/ wiki/Message Understanding Conference. 6 http://www.itl.nist.gov/iad/mig/tests/ace/ 7 http://www.nist.gov/tac
4 4.1
and stores the data in an ontology. We have made initial experiments in the domain of reports of traffic accidents. The results showed that this method can e.g. aid summarization of the number of injured people.
To avoid the need of manual design of extraction rules we focused on the data extraction phase and made some promising experiments [8] with the machine learning procedure of Inductive Logic Programming for automated learning of the extraction rules.
This solution is directed to extraction of information which is closely connected with the meaning of text or meaning of a sentence.
Our first approach for the web information extraction is to use the structural similarity in web pages containing large number of table cells and for each cell a link to detailed pages. This is often presented in web shops and on pages that presents more than one object (product offer). Each object is presented in a similar 5 The Web Semantization setting way and this fact can be exploited.
As web pages of web shops are intended for hu- In this section we will discuss possibilities and obstrucman usage creators have to make their comprehension tions connected with the employment of web informaeasier. Acquaintance with several years of web shops tion extraction systems in the process of web semanhas converged to a more or less similar design fashion. tization.
There are often cumulative pages with many products One aspect of the realization of the web semanin a form of a table with cells containing a brief de- tization idea is the problem of integration of all the scription and a link to a page with details about each components and technologies starting with web crawlparticular product. ing, going through numerous complex analyses (docu
Our main idea is to use a DOM tree representation ment preprocessing, document classification, different of the summary web page and by breadth first search extraction procedures), output data integration and encounter similar subtrees. The similarity of these sub- indexing, and finally implementation of query and pretrees is used to determine the data region – a place sentation interface. This elaborate task is neither easy where all the objects are stored. It is represented as nor simple but today it is solved in all the extensive a node in the DOM tree, underneath it there are the projects and systems mentioned above. similar sub-trees, which are called data records. The novelty that web semantization brings into ac
We8 have developed and implemented this idea [14] count is the cross domain aspect. If we do not want to on the top of Mozilla Firefox API and experimentally stay with just general ontologies and general applicatested on table pages from several domains (cars, note- ble extraction methods then we need a methodology books, hotels). Similarity between subtrees was Lev- how to deal with different domains. The system has to enshtein editing distance (for a subtree considered as support extension to a new domain in generic way. So a linear string), learning thresholds for decision were we need a methodology and software to support this trained. action. This can for example mean: to add a new ontology for the new domain, to select and train proper extractors and classifiers for the suitable input pages. 4.2
Our second approach [11, 12, 10] for the web information extraction is based on deep linguistic analysis. We have developed a rule-based method for extraction of information from text-based web resources in Czech and now we are working on its adaptation to English. The extraction rules correspond to tree queries on linguistic (syntactic) trees made form particular sentences. We have experimented with several linguistic tools for Czech, namely Tools for machine annotation – PDT 2.0 and the Czech WordNet.
Our present system captures text of web-pages, annotates it linguistically by PDT tools, extracts data An interesting point is the question: Whose effort will be used in the process of supporting new domain in the web semantization process? How skilled such user has to be? There are two possibilities (demonstrated on the Fig 3). The easier one is that we have to employ very experienced expert who will decide about the new domain and who will also realize the support needed for the new domain. In the Fig 3 this situation is labeled as Provider Initiated and Provider Trained because the expert works on the side of the system 8 Thanks go mainly to Duˇsan Maruˇsˇc´ak and Peter Vojt´aˇs. that provides the semantics.