Emil St. Chifu

Self-organizing Maps in Web Mining and Semantic Web

The nature inspired approaches represent a new trend in computer science in general and in the Semantic Web, due to their scalability and robustness. Neural networks represent one category of nature inspired solutions. The self-organizing map (SOM) is a very popular unsupervised neural network model (Kohonen, et al., 2000). It is a data mining and visualization method for complex high dimensional data sets. In the first part of the chapter, we present how the SOM model can be applied in Web mining, by giving sets of documents as input data space for SOM. The result of applying SOM on a set of documents is a map of documents, which is organized in a meaningful manner so that documents with similar content appear at nearby locations on the twodimensional map display. From the information retrieval point of view, our implemented SOM-based system creates document maps that are readily organized for browsing. A document map also clusters the data, resulting in an approximate model of the data distribution in the high dimensional document space. Some experimental results are included, where a couple of meaningful clusters have been discovered by our system in a subset of the “20 newsgroups” data set (Lang, K., 1995). The clustering capability of our system allows users to find out quickly what is new in a Web site of interest by comparing the clusters obtained from the site at different moments in time. In the rest of the chapter, we focus on how a more complex SOM based unsupervised neural network model is used for enriching a domain ontology. Building complete and reliable domain ontologies is the basis for the success of the Semantic Web. The ontology enrichment process consists in the addition of new concepts which will be attached as hyponyms for the existent nodes of the ontology (Pekar and Staab, 2002). The names of the new concepts are terms represented linguistically by common noun phrases. The enrichment process can also add new instances to existent concepts of the ontology. In this case, the process is also known in the literature as ontology population or named entity classification, where the named entities are represented linguistically by proper names of people, organizations, locations etc. (Cimiano and Volker, 2005). In both cases, the process is algorithmically the same, the only difference being the grammatical category of the linguistic entities to be classified: common noun phrases representing terms for new concepts to be added or proper noun phrases representing named entities, i.e. new instances for the existent 22

Automatic Web Service Composition Using OWL-S and Fluent Calculus

This paper presents a novel approach for semantic Web service composition based on the formalism of fluent calculus. We show how the planning capabilities of the fluent calculus can be used to automatically generate an abstract composition model. For describing Web service capabilities we have used an OWL-S ontology. Based on the OWL-S ontology semantics, we encode the Web service description in the fluent calculus formalism and provide a planning strategy for service composition. For testing our composition method, we have implemented an experimental framework that automatically composes and executes Web services.

Matching Semantic Web Services Using Learning Accuracy

The automatic discovery of suitable Web services for a given task is one of the key elements in implementing the Semantic Web vision. This paper presents a new matching algorithm for Semantic Web service discovery. Our matching algorithm allows for ranking the discovered Web services according to their relevance to the service request. The learning accuracy is proposed as a suitable metric for determining the semantic similarity between a service request and the service advertisements. The semantic similarity is computed by considering the semantic information encoded in a domain ontology, including both the concept hierarchy and the properties of the concepts. Evaluating the semantic similarity between a service request and a service advertisement is based on the concepts, their semantic relations, their common and distinguishing properties, and the semantic relations between their properties.

Unsupervised aspect level sentiment analysis using Ant Clustering and Self-organizing Maps

We present an approach for aspect based opinion mining, which uses an unsupervised neural network as the opinion classifier. To identify the aspects, we use the Ant Clustering Algorithm. It is able to group similar sentences into clusters and then to extract from each cluster one different aspect of the opinion target object. The neural model used for sentiment analysis is an extension of the Growing Hierarchical Self-organizing Maps. In our aspect based sentiment analysis method, we assume that different sentences in a product review refer to the different aspects of the reviewed product. We use the Growing Hierarchical Self-organizing Maps in order to classify the review sentences. This way we determine whether the various aspects of the target entity (e.g. a product) are opinionated with positive or negative sentiment in the review sentences. We classify the sentences against a domain specific tree-like ontological taxonomy of aspects and (positive/ negative) opinions associated with the aspects. As a consequence, we really classify the sentiment polarity about the different aspects of the target object, as expressed in the sentences. Moreover, being based on a classification against an ontology of aspects, our approach is semantic oriented, where the aspects themselves are also defined semantically. The approach proposed has been tested on a collection of product reviews, more exactly reviews about photo cameras.

Unsupervised Aspect Level Sentiment Analysis Using Self-Organizing Maps

This paper presents an unsupervised method for aspect level sentiment analysis that uses the Growing Hierarchical Self-organizing Maps. Different sentences in a product review refer to different aspects of the reviewed product. We use the Growing Hierarchical Self-organizing Maps in order to classify the review sentences. This way we determine whether the various aspects of the target entity (e.g. a product) are opinionated with positive or negative sentiment in the review sentences. By classifying the sentences against a domain specific tree-like ontological taxonomy of aspects and sentiments associated with the aspects (positive/ negative sentiments), we really classify the opinion polarity as expressed in sentences about the different aspects of the target object. The approach proposed has been tested on a collection of product reviews, more exactly reviews about photo cameras.

A neural model for semantically enhancing Web APIs

The paper describes an unsupervised neural model for classifying the methods of Web APIs into a large number of classes specified by a domain ontology. As a result of the classification, each method of a Web service is associated to one ontology concept, the name of the concept being further used to semantically annotate the method. The ontology concepts define some functionalities to be offered by different API methods. The names of these concepts are linguistically denoted by verbs or verb phrases that define the action performed by a method. The framework is based on a model of hierarchical self-organizing maps. The methods of the web APIs are encoded in a bag-of-words style, by counting the words that occur in their javadoc documentation. We experimented this automatic semantic annotation model with a data set consisting of APIs of RDF storage tools. The ontology and the APIs to be classified in our experiments are collected from this dataset.

A system for detecting professional skills from resumes written in natural language

In this paper, we present a new method for detecting professional skills (as noun phrases) from resumes written in natural language. The proposed method uses an ontology of skills, the Wikipedia encyclopedia, and a set of standard multi word part-of-speech patterns in order to detect the professional skills. First, the method checks to see if there are, in the text of the resumes, skills that are concepts in our ontology. The method also tries to identify possible new skills, which are not present in our ontology. This is done with the help of some specific, lexicalized, multi-word expression patterns (i.e. specific contexts) that could surround new, unknown skills. The specific expression patterns (specific contexts) are induced by training from a corpus of resumes. This induction of the possible specific contexts for new skills is based on a set of standard, generic part-of-speech patterns (found by hand) that usually contain the skills already present in the ontology. Hence our skill extraction method is based on a bootstrapping approach. The newly detected skills are validated by a human expert and then inserted automatically into the skill ontology. Populating the ontology with the new skills is performed with the help of the Wikipedia encyclopedia. The method proposed has been tested on a set of resumes written by users as well as on a corpus collected by automatically extracting resumes from specific Web sites.

Optimizing the generation of personalized healthy menus for elderly people using a crab breeding inspired method

In this paper we present a method for generating healthy diets for elderly people. The method proposed is based on the Crab Mating Optimization Algorithm, which is inspired from the breeding behavior of crabs in nature. In our case the generated diet is composed of several meals per day and it can be created for a number of 7 days. In generating a healthy diet we have considered the elders food preferences as well as the dietary restrictions associated with the diseases the elder suffers from. The method proposed has been integrated into an experimental prototype and evaluated on a set of profiles describing older people suffering from different diseases like diabetes and heart problems.

Hybrid Immune Based Method for Generating Healthy Meals for Older Adults

This paper presents a Hybrid Clonal Selection based method for generating healthy meals as starting from a given user request, a diet recommendation, and a set of food offers. The method proposed is based on a hybrid model, which consists of one core component and two hybridization components. The core component uses the CLONAG algorithm. One of the hybridization components is based on flower pollination, whereas the other utilizes tabu search and reinforcement learning. The flower pollination component is used for modifying the generated clones, while the tabu search and reinforcement learning component aims to improve the search capabilities of the core component by means of long-term and short-term memory structures. We integrated our method into an experimental prototype and we evaluated it on different older adult profiles.

Bird Mating Optimization method for one-to-n skill matching

This paper presents a Bird Mating Optimization method for one-to-n skill matching. The method proposed finds the optimal combination of skills from two or more CVs that best satisfies a job description. In our approach the CV sets as well as the job description are described semantically by using a skilltaxonomy. To evaluate the quality of a solution (i.e. a set of CVs that satisfies the job description considered) we have defined a fitness function that evaluates the degree of semantic matching of the combination of skills part of the considered solution to the set of skills of the job description. The method proposed has been tested on a set of 1000 CVs in the domain of computer science, the set being developed in house.