International Journal of Scientific & Engineering Research, Volume 4, Issue 8, August-2013 607

ISSN 2229-5518

Semantic information retrieval from heterogeneous environments

Fisnik Dalipi1, Ilia Ninka2

1Department of IT, Faculty of Math-Natural Sciences, Tetovo State University e-mail: fisnik.dalipi@unite.edu.mk

2Department of IT, Faculty of Natural Sciences, University of Tirana e-mail: ilia.ninka@fshn.edu.al

Abstract — While most enterprise data is unstructured and file based, the need for access to structured data is increasing. In order to reduce the cost for finding information and achieve relevant results there is a need to build a very complex query which indeed is a serious challenge. Data volumes are growing at 60% annually and up to 80% of this data in any organization can be unstructured. In this paper we focus on describing the evolution of some modern ontology-based information retrieval systems. Further, we will provide a brief overview of the key advances in the field of semantic information retrieval from heterogeneous information sources, and a description of where the state-of-the-art is at in the field. Finally, we present and propose a novel use of semantic retrieval model based on the vector space model for the exploitation of KB (Knowledge Base) to enhance and sup- port searching over robust and heterogeneous environments.

Index Terms— ontology, information retrieval, vector space model, semantic web, metadata, semantic index, knowledge base.

——————————  ——————————

1 INTRODUCTION

he phrase “information retrieval - IR” dates back to the
1950s [1], but the concept was firstly used at the library cata- logues. Initial opinions on the subject emerged from librari- anship and information science. Originally, this opinion had phil- osophical nature, dealing with how information should be classi- fied and organized. Various schools held various positions and an ongoing debate was evident between them on philosophical and anecdotal rather than empirical grounds [2]. However, with the increasing volume of publication, and specifically of scientific literature, after the Second World War, practical concerns of how
to effectively access this literature became urgent [3,4].
There are two foundational projects that mark the creation of the
information retrieval field. The Cranfield testing was the first project, and consisted of several experiments starting at 1957 and lasted until 1966. The second project was SMART, starting in the early 1960s and running in various forms until the end of the twentieth century. The emphasis of SMART has always been on purely automatic text retrieval - starting from an arbitrary piece of natural language text from the user and matching against au- tomatically indexed documents [5].
One of the most influential methods was described by H.P. Luhn in 1957, in which (put simply) he proposed using words as index- ing units for documents and measuring word overlap as a criteri- on for retrieval [4]. The algorithms developed in IR were the first ones to be employed for searching the World Wide Web from
1996 to 1998.
In recent times, ontologies are widely used in IR systems. Never- theless, its main use has to do with query expansion, which con- sists in searching for the terms in the ontology more similar to the query terms, to use them together as a part of the query.
In this work, we present and propose a novel use of semantic retrieval model based on the vector space model to enhance and support searching over robust and heterogeneous environments.
IR represents a core component of the information systems. An information system must ensure that all the users who are meant to be served has the information needed to accomplish tasks, solve problems, and make decisions, no matter where that infor- mation is available. An information system must (1) actively find out what are the user’s requirements or needs, (2) find and access documents, which results in a collection, and (3) match or affili- ate documents with those requirements or needs. Realizing what type of information the user really needs to solve a problem is essential for successful retrieval.

The final goal of an IR system can be described as the representa- tion, storage, organization of, and access to information items [6]. A global, abstract view of these elements is displayed in Figure 1.

Fig. 1 Schematic view of the Information Retrieval process

2 IR concepts and models

The user formulates a query to the search engine. The format of the query can be search string, image, sound, etc. The search en-

International Journal of Scientific & Engineering Research, Volume 4, Issue 8, August-2013 608

ISSN 2229-5518

gine then retrieves documents directly from the data corpus, but only those that are important to the query. External elements in- clude the user interface, various query processing operations and resources for indexing (thesauri or controlled vocabularies). Next step is to rank them by decreasing probability or degree of rele- vance, and returns them to the user. This process can be iterated. According to [7], we can distinguish four processes in an IR sys- tem: indexing, query processing, searching and ranking. The ranking step aims to predict which how relevant the items are comparatively to each other, thus returning them by decreasing order of estimated relevance. Thus, in a way, ranking algorithms can be considered the core of IR systems, as they are a key to determine the performance of the system.
Depending on the type of query, different mechanisms can be used to refine it. The most common ones are based on additional user input. In this spectrum, relevance feedback approaches are generally the most efficient ones. However, they reduce the usa- bility of the systems, and therefore other external resources, such as taxonomies and thesauri, are often used instead (or comple- mentarily) to automatically classify, disambiguate or expand que- ry terms.

2.1 IR models and evaluation

The motivation for entering requests into an information retrieval system is an information need [8] and the success of a given re- trieval system depends on the system’s capacity to provide the user with the information needed with a reasonable time and with a straightforward interface for posing requests and collecting the results [9].
According to the definition in [6] an IR model is a quadruple
[D,Q,F, sim], where:
• D is a set of (logical representations of) documents.
• Q is a set of (logical representations of) queries.
• F is a framework for modeling documents, queries, and their relationships.
•sim: Q × D → U is a ranking function that defines an association between queries and documents, where U is a totally ordered set (commonly [0, 1], or Ρ, or a subset thereof). This ranking and the total order in U define an order in the set of documents, for a fixed query.
There are standard measures to evaluate the performance of IR
systems [10].

Precision: The ratio of documents retrieved by the system that are actually relevant to the query divided by the total number of documents retrieved.

For instance, if the system retrieved 6 documents for a query, where 3 of them were actually relevant, the precision perfor- mance for the system in that query is 0.5 or 50%. Polysemy may produce low precision rates, because irrelevant documents might be retrieved.

Recall: There may be many documents in the database that the user considers relevant, but only some of them will be retrieved by the system. The recall performance of a query is the number of relevant documents retrieved by the system divided by the total

number of relevant documents in the database.

Response time: The elapsed time between the submission of a query and the presentation of the documents retrieved by the sys- tem. Precision could be easily maximized by retrieving a single document that is certainly relevant, and recall by retrieving all documents in the database. Thus, a measure that combines both of them is preferred, for example, the F-measure:

where F-measure is the harmonic mean of precision and recall. The advantage of using F-measure is that maximizing it means maximizing a combination of recall and precision.
Three of the most important classic text IR models are: Boolean model, Vector space model and Probabilistic model.

2.2 Vector Space model

In the vector space model text is represented by a vector of terms [11]. The definition of a term is not inherent in the model, but terms are typically words and phrases. If words are chosen as terms, then every word in the vocabulary becomes an independ- ent dimension in a very high dimensional vector space. Any text can then be represented by a vector in this high dimensional space. If a term belongs to a text, it gets a non-zero value in the text-vector along the dimension corresponding to the term. Since any text contains a limited set of terms (the vocabulary can be millions of terms), most text vectors are very sparse. Most vector based systems operate in the positive quadrant of the vector space, i.e., no term is assigned a negative value [12].
Following the presented previous notation, we define the vector space model as below:

• D: documents are represented by a vector of words

or index terms occurring in the document. Each term in the document – or, for that matter, each pair (ti, dj ) – has a positive, non-binary associated weight wi,j .

• Q: queries are represented as a vector of words or

index terms occurring in the query. Each term in the query– or, for that matter, each pair (ti, q) – has a positive, non-binary associated weight wi ,q .

• F is an algebraic model over vectors in a t-

dimensional space.
• sim estimates the degree of similarity of a doc- ument dj to a query q as the correlation between the vectors dj and q. This correlation can be quantified, for instance, by the cosine of the angle between the two vectors:

sim (q,dj )=cos(q,dj )=(q•dj )/(|q|•|d j |).

International Journal of Scientific & Engineering Research, Volume 4, Issue 8, August-2013 609

ISSN 2229-5518

From the formula, sim(q, dj) varies from 0 to 1, i.e. 1.0 for identical vectors and 0.0 for orthogonal vectors. Thus, instead of attempting to predict whether a document is relevant or not, the vector model ranks the documents according to their degree of similarity to the query.

3 SEMANTIC INFORMATION RETRIEVAL

Recently, ontologies have been used in Information Retrieval to improve recall and precision [13]. Its principal use is related to query expansion, which consists in looking for the terms in the ontology more related to the query terms, to use them as a part of the query. Much ontology has been designed for the purposes of managing and extracting semantic knowledge from online litera- ture and databases.
IR systems that use semantic technologies for enhancing different parts of IR are called semantic search systems. Searching for the online ontologies, fact extraction from the ontologies, question answering, filtering and ranking retrieved information are usually put under the wing of semantic search. The introduction of ontol- ogies to move beyond the capabilities of current search technolo- gies has been an often portrayed scenario in the area of semantic- based technologies since the late nineties [14].

Based on literature review, below we provide the categories of semantic search engines. [15,16].

Fig 2. Classification of semantic search systems

In spite of the fact that that these concepts have shown several enhancements compared to classic keyword searching methodol- ogy, it is not clear among researchers that these techniques could be suitable to deal with large scale information sources.

4 SEMANTIC RETRIEVAL FROM HETEROGENEOUS ENVIRONMENTS

Semantic retrieval from distributed and heterogeneous environ-

ments is quite new concept and current ontology based retrieval technologies are very hypothetical, without having any well de- fined framework on applying ontology based search to the web as whole, which is consisted by unlimited number of domains. Some attempts have been made by [17], but they lack to address the potential use of ontology search beyond the organizational data corpus, as their models have difficulties to deal with the het- erogeneity of Web and are limited to a predefined set of ontolo- gies.

Fig 3. Semantic information retrieval framework

This model does not require users to know special purpose query language; rather, the system expects queries to be expressed in natural language. Another relevant aspect is that the set of un- structured (web) information is not needed to be adapted into conventional fragments of ontological knowledge. In order to answer the queries, the system uses available semantic data and other information from standard web pages. When dealing with such a large amount of semantic information, we need a semantic gateway which will pre-process, gather, store and access the online distributed semantic web information. One of the most popular semantic way gateways currently available in the state of the art are: Watson [18], and Swoogle [19].
Once the user poses the query, that query can further be pro- cessed by any ontology based system which ensures access to the online ontologies and that translates generic natural language queries into SPARQL. Such systems of choice could be Aqua- Log, proposed by [20,21], Querix [22], or QASYO [23].
After returning the fragments of relevant ontological knowledge as an answer, the system will perform a second step which in- cludes retrieving and ranking by their probability the documents which contain the needed information. The ranking process can
apply the concepts of vector space model ranking algorithm.
The proposed architecture in Figure 3 reflects the concept of het- erogeneity assuming large amount of semantic metadata online without having a pre-defined range of domains. We assume that the external element is not only a single knowledge base but in-

International Journal of Scientific & Engineering Research, Volume 4, Issue 8, August-2013 610

ISSN 2229-5518

volves online semantic web information.

5 CONCLUSIONS

In this paper, we focused on describing the evolution of modern IR systems with respect to their concepts and models and their application using ontologies. We define the IR evaluation and the vector space model. Further, we provided a brief overview of the key advances in the field of semantic information retrieval by describing where the state-of-the-art is at in the field.
Finally, we presented and proposed some ideas towards a novel use of semantic retrieval model which is based on the vector space model, aiming at enhancing and supporting the searching process over robust and heterogeneous environments with unlim- ited number of domains.

REFERENCES

[14] Luke, S., Spector, L., & Rager, D. (1996). Ontology-Based Knowledge Discovery on the World-Wide Web. Internet-Based Information Systems: Papers from the AAAI Workshop. AAAI, (pp. 96-102). Menlo Park, California.

[15] Mangold, C. (2007) “A survey and classification of semantic search approaches”, Int. J. Metadata, Semantics and Ontology, Vol. 2, No.

1, pp.23–34

[16] Esmaili, K.S. and Abolhassani, H. (2006) “A Categorization Scheme for Semantic Web Search Engines”, In Proc. of IEEE Conf. on Computer Systems and Applications, pp 171-178.

[17] Maedche, A., Staab, S., Stojanovic, N., Studer, R., & Sure, Y. (2003).

SEmantic portAL: The SEAL Approach. Spinning the Semantic

Web. MIT Press , 317-359

[18] D'Aquin, M., Gridinoc, L., Sabou, M., Angeletou, S., & Motta, E. (2007). Characterizing

Knowledge on the Semantic Web with Watson. 5th International

EON Workshop at International Semantic Web Conference

(ISWC'07). Busan, Korea.

[19] Ding, L., Finin, T., Joshi, A., Pan, R., & Cost, S. (2004). Swoogle: A Search and Metadata Engine for the Semantic Web. 13th Confer- ence on Information and Knowledge Management (CIKM 2004), (pp. 625-659). Washington, DC, USA.

[20] V. Lopez, M. Pasin, and Enrico Motta, “AquaLog: An Ontology- Portable Question Answering System for the Semantic Web,” Lec- ture Notes in Computer Science, Vol. 3532, Springer, Berlin, pp.

546-562, 2005.

[21] V. Lopez, and E. Motta, “Ontology-Driven Question Answering in AquaLog,” Lecture Notes in Computer Science, Vol. 3136. Spring- er-Verlag, Berlin, pp. 89–102, 2004.

[22] E. Kaufmann, A. Bernstein, and R. Zumstein., “Querix: A natu- ral language interface to query ontologies based on clarification dialogs,” In proceeding 5th International Semantic Web Confer- ence (ISWC 2006), pp 980–981, 2006.

[23] Abdullah M. Moussa and Rehab F. Abdel-Kader. QASYO: A Ques- tion Answering System for YAGO Ontology. International Journal

of Database Theory and Application Vol. 4, No. 2, June, 2011