Semantic Web
- 1. Methods inspired by nature and Semantic Web Rata Gheorghita Mugurel MOC2, Ivanciu Adriana MLC2 gheorghita.rata@infoiasi.ro, adriana.ivanciu@infoiasi.ro Semantic web is an extension of the current web, intended to provide an improved cooperation between humans and machines1. Genetic algorithms Genetic algorithms and search engines In the book Enhancing the power of the Internet, in the chapter Intelligent Information Search, the authors2 say that there were many approaches that were studied regarding the way of how this domain can be improved. There are two major problems, according to the authors: classical information models and information retrieval model itself. The most techniques were focused to the first problem. For the second one, probabilistic methods were the most popular in the past. Even if artificial intelligence and fuzzy theory had a great contribution, the evolving of genetic algorithms and neural networks gathered the attention. Although manual knowledge acquisition 1 Berners-Lee, T. Hendler, J. Lassila, O. The semantic web. Scientific American, 28-37 (2001). 2 Enhancing the power of the Internet By Masoud Nikravesh, Ben Azvine, Ronald Yager, Lotfi A. Zadeh
- 2. 2 Rata Gheorghita Mugurel MOC2, Ivanciu Adriana MLC2 process was the base for the search systems, data mining was an important technique for obtain knowledge in an automatic process. The power of genetic algorithm was proved when were used in the process of extracting keywords and establish its weights. The same authors say that genetic algorithms and genetic fuzzy system have great results regarding Search engines. In the same domain (Search engines), neural network-based methods are lesser extent. According to Hsinchun (1998), which is quoted in this paper, genetic algorithms are used to search in a dynamic manner on a keyword dictionary and return a list of related Web pages. The search process is described as following: The population is formed from chromosomes that have a fixed length Chromosomes represents user preferences A fitness value is associated with each chromosomes Genes contain the user keyword and a number that represents the frequency of the keyword occurrence in a web document (witch is a candidate for the solution) After the user evaluates the documents returned, the fitness value is adjusted, considering the score computed by the system. Going further, metagenetic algorithms are used to optimize the start population. One of these combines two genetic algorithms. The first is used to generate the start population with values from keywords index and the second creates a population with logic operators corresponding to each member from the first algorithm. The first 2
- 3. Methods inspired by nature and Semantic Web 3 algorithm can be easily replaced with a random selection for a faster search. SWARMS SWARMS3 (semantic web added rich mining systems) is a platform for knowledge management. It store the information in ontologies, can extract the network structure from the ontology and search (mining) the semantic data. This system is applied in many domains mainly in online news industry and social networking. To simple queries the SPARQL works great. But the more the queries became big and complicated, SPARQL will not satisfy the requirements anymore. In this case the developers appeal to methods inspired by nature. Another reason is that the metadata in Semantic Web is not always well structure, and a classic algorithm is hard to be adapted. The search in Semantic Web context is based on semantic similarity and it measure the similarity between objects from ontology. The semantic similarity is computed from hierarchy similarities, property similarities, label similarities and access similarities (Zongmin Ma, Huaiqing Wang, 2009). These can be computed with some probabilistic algorithms. The same authors propose a Semantic similarity based on cached models. The search algorithm should respect two rules: return an approximate optimal solution and the time spent on its searching 3 The Semantic Web for Knowledge and Data Management: Technologies and Practices By Zongmin Ma, Huaiqing Wang 3
- 4. 4 Rata Gheorghita Mugurel MOC2, Ivanciu Adriana MLC2 must be finite. The best algorithms class that fit these specifications is the one inspired from nature and genetics. The authors used a genetic algorithm for training the model and create the initial cache. The base elements of the genetic algorithm are: the population have 50 chromosomes; the mutation probability is 0.2; the algorithm will stop when the fitness is 0.9 or the generations number reach 100. Below is a chart that represents the two search ways and its time performance per number of requests: Performance of Ontology Cache Cache disabled Cache enabled 2500 T i m 2000 e 1500 C o n 1000 s u m 500 i n g 0 0 1000 2000 3000 4000 Request Count Performance of Time Consuming 4
- 5. Methods inspired by nature and Semantic Web 5 Details can be found in the document from the point 3 of the Bibliography. Human Similarity theories for the semantic web In the paper Human Similarity theories for the semantic web, the author4 shares his opinion about how human mind representation can be useful for making the web documents more ‘friendly’ for the computers. He thinks that the way of how human mind represents the data, in order to be easy to find similarities can be manipulated, studied and used for ontology building and other web semantic activities, generally speaking. Giving the fact that the users of the computers are human after all, he thinks that semantic web has a lot in common with humans and both humans and computers have to deal with a big quantity of information. One of the domains that can help Semantic Web is Psychology, in his opinion. In order to solve problems, humans are using inductive and deductive reasoning, they have to follow causal chains, to solve problems and to make decisions. In RDF, the data structure language for Semantic Web, the concepts witch are considered fundamental are resources, properties and statements. The first category is represented by objects. The objects can be anything like humans, books or activities. This resources have properties like names, chapters and physical locations. The statement is the link between the property and the resource. The author thinks that 4 Jose Quesada, Max Planck Institute, Human development 5
- 6. 6 Rata Gheorghita Mugurel MOC2, Ivanciu Adriana MLC2 psychologists and Semantic Web have the same interest in a certain way, represented by the fact that both tries to model the world using the formalism. Although there are big differences between the two domains, the author believes that there is a level of convergence between them. Conclusion In nature we can find an impressive number of algorithms that can be used to solve different problems from different domains including Semantic Web. Nature will always surprise and will offer patterns, algorithms, processes that will inspire solving technologies problems with a good result. 6
- 7. Methods inspired by nature and Semantic Web 7 Bibliography 1. Semantic web service composition based on ant colony optimization method Ghafarian, T.; Kahani, M. Networked Digital Technologies, 2009. NDT apos;09. First International Conference on Volume , Issue , 28-31 July 2009 2. Enhancing the Power of the Internet Series: Studies in Fuzziness and Soft Computing , Vol. 139 Nikravesh, M.; Azvine, B.; Yager, R.; Zadeh, L.A. (Eds.) 2004 3. The Semantic Web for Knowledge and Data Management: Technologies and Practices By Zongmin Ma, Huaiqing Wang, IGI Global, 2009 4. Human Similarity theories for the semantic web, Jose Quesada, Max Planck Institute, Human development presented in Nature inspired for the Semantic Web (NatuReS) October 27, 2008 7