Skip to Main content Skip to Navigation
Theses

Extending semantic nets using concept-proximity  

Abstract : The past few years has witnessed tremendous upsurge in information availability in the electronic form, attributed to the ever mounting use of the World Wide Web (WWW). For many people, the World Wide Web has become an essential means of providing and searching for information leading to large amount of data accumulation. Searching web in its present form is however an infuriating experience for the fact that the data available is both superfluous and diverse in form. Web users end up finding huge number of answers to their simple queries, consequentially investing more time in analyzing the output results due to its immenseness. Yet many results here turn out to be irrelevant and one can find some of the more interesting links left out from the result set. Chapter1 Introduces our motivation behind the research: One of the principal explanations for the unsatisfactory condition in information retrieval is the reason that majority of the existing data resources in its present form are designed for human comprehension. When using these data with machines, it becomes highly infeasible to obtain good results without human interventions at regular levels. So, one of the major challenges faced by the users as providers and consumers of web era is to imagine intelligent tools and theories in knowledge representation and processing for making the present data, machine understandable. Chapter 2 evaluates and studies the existing methods and their short falls: Several researches has been carried out in enable machines to understand data and some of the most interesting solutions proposed are the semantic web based ontology to incorporate data understanding by machines. The objective here is to intelligently represent data, enabling machines to better understand and enhance capture of existing information. Here the main emphasis is given to the thought for constructing meaning related concept networks for knowledge representation. Eventually the idea is to direct machines in providing output results of high quality with minimum or no human intervention. In recent years the development of ontology is fast gaining attention from various research groups across the globe. There are several definitions of ontology purely contingent on the application or task it is intended for. Chapter 3 presents the platform ToxNuc-E and positioning of our research around this platform: Given the practical and theoretical importance of ontology development, it is not surprising to find a large number of enthusiastic and committed research groups in this field. Extended Semantic Network is one such innovative approach proposed by us for knowledge representation and ontology like network construction, which looks for sets of associations between nodes semantically and proximally. Our objective here is to achieve semi-supervised knowledge representation technique with good accuracy and minimum human intervention, using the heuristically developed information processing and integration methods. The main goal of our research is to find an approach for automatic knowledge representation that can eventually be used in classification and search algorithms in the platform ToxNuc-E. Chapter 4 elaborates on the concept of Proximal Network modeling, generated by mathematical models: As stated earlier the basic idea of Extended Semantic Network is to identify an efficient knowledge representation and ontology construction method to overcome the existing constraints in information retrieval and classification problems. To realize this we put our ideas into practice via a two phase approach. The first phase consists in processing large amount of textual information using mathematical models to make our proposal of automatic ontology construction scalable. This phase of our proposal is carried out by realising a network of words mathematically computed using different statistical and clustering algorithms. Thus creating a proximal network computationally developed, depending essentially on word proximity in documents. The proximal network is basically representing the recall part of our approach. Chapter 5 investigates the semantic network modelling and introduces a design model proposed by us to enable efficient cost effective design: Semantic Network is basically a labelled, directed graph permitting the use of generic rules, inheritance, and object-oriented programming. It is often used as a form of knowledge representation where concepts represented by nodes are connected to one another using the relational links represented by arcs. Semantic network is constructed with the help of expert knowledge and understanding of a domain. Hence it is mainly a human constructed network with very good precision. Chapter 6 in effect details the extended semantic network: The second phase of our research mainly consists in examining carefully and efficiently the various possibilities of integrating information obtained from our mathematical model with that of the manually developed mind model. This phase is ensured by a heuristically developed method of network extension using the outputs from the mathematical approach. This is achieved by considering the manually developed semantic mind model as the entry point of our concept network. Here, the primary idea is to develop a innovative approach obtained by combining the features of man and machine theory of concepts, whose results can be of enormous use in the latest knowledge representation, classification, retrieval, pattern matching and ontology development research fields. In this research work we illustrate the methods used by us for information processing and integration aimed at visualising a novel method for knowledge representation and ontology construction. Chapter 7 illustrates some of the experiments carried out using our extended semantic network and opens directions for future perspectives: The question on knowledge representation, management, sharing and retrieval are both fascinating and complex, essentially with the co-emergence between man and machine. This research presents a novel collaborative working method, specifically in the context of knowledge representation and retrieval. The proposal is to attempt at making ontology construction faster and easier. The advantages of our methodology with respect to the previous work, is our innovative approach of integrating machine calculations with human reasoning abilities. The resulting network so obtained is later used in several tools ex: document classifier to illustrate our research approach. We use the precise, non estimated results provided by human expertise in case of semantic network and then merge it with the machine calculated knowledge from proximal results. The fact that we try to combine results from two different aspects forms one of the most interesting features of our current research. We view our result as structured by mind and calculated by machines. One of the main future perspectives of this research is finding the right balance for combining the concept networks of semantic network with the word network obtained from the proximal network. Our future work would be to identify this accurate combination between the two vast methods and setting up a benchmark to measure our prototype efficiency.
Document type :
Theses
Domain :
Complete list of metadatas

Cited literature [97 references]  Display  Hide  Download

https://pastel.archives-ouvertes.fr/pastel-00005840
Contributor : Ecole Mines Paristech <>
Submitted on : Friday, February 26, 2010 - 8:00:00 AM
Last modification on : Monday, October 19, 2020 - 10:55:27 AM
Long-term archiving on: : Wednesday, July 28, 2010 - 8:23:18 PM

Identifiers

  • HAL Id : pastel-00005840, version 1

Citation

Reena Shetty. Extending semantic nets using concept-proximity  . domain_other. École Nationale Supérieure des Mines de Paris, 2008. English. ⟨pastel-00005840⟩

Share

Metrics

Record views

478

Files downloads

460