Abstract: FAQ system can make user find answer to the problem that puzzles them. But now the research on Chinese FAQ system is still on the theoretical stage. This paper presents an approach to semantic inference for FAQ mining. To enhance the efficiency, a small pool of the candidate question-answering pairs retrieved from the system for the follow-up work according to the concept of the agriculture domain extracted from user input .Input queries or questions are converted into four parts, the question word segment (QWS), the verb segment (VS), the concept of agricultural areas segment (CS), the auxiliary segment (AS). A semantic matching method is presented to estimate the similarity between the semantic segments of the query and the questions in the pool of the candidate. A thesaurus constructed from the HowNet, a Chinese knowledge base, is adopted for word similarity measure in the matcher. The questions are classified into eleven intension categories using predefined question stemming keywords. For FAQ mining, given a query, the question part and answer part in an FAQ question-answer pair is matched with the input query, respectively. Finally, the probabilities estimated from these two parts are integrated and used to choose the most likely answer for the input query. These approaches are experimented on an agriculture FAQ system. Experimental results indicate that the proposed approach outperformed the FAQ-Finder system in agriculture FAQ retrieval.
Abstract: Due to the constant increase in the volume of information available to applications in fields varying from medical diagnosis to web search engines, accurate support of similarity becomes an important task. This is also the case of spam filtering techniques where the similarities between the known and incoming messages are the fundaments of making the spam/not spam decision. We present a novel approach to filtering based solely on layout, whose goal is not only to correctly identify spam, but also warn about major emerging threats. We propose a mathematical formulation of the email message layout and based on it we elaborate an algorithm to separate different types of emails and find the new, numerically relevant spam types.
Abstract: The management of the health-care wastes is one of
the most important problems in Istanbul, a city with more than 12
million inhabitants, as it is in most of the developing countries.
Negligence in appropriate treatment and final disposal of the healthcare
wastes can lead to adverse impacts to public health and to the
environment. This paper employs a fuzzy multi-criteria group
decision making approach, which is based on the principles of fusion
of fuzzy information, 2-tuple linguistic representation model, and
technique for order preference by similarity to ideal solution
(TOPSIS), to evaluate health-care waste (HCW) treatment
alternatives for Istanbul. The evaluation criteria are determined
employing nominal group technique (NGT), which is a method of
systematically developing a consensus of group opinion. The
employed method is apt to manage information assessed using multigranularity
linguistic information in a decision making problem with
multiple information sources. The decision making framework
employs ordered weighted averaging (OWA) operator that
encompasses several operators as the aggregation operator since it
can implement different aggregation rules by changing the order
weights. The aggregation process is based on the unification of
information by means of fuzzy sets on a basic linguistic term set
(BLTS). Then, the unified information is transformed into linguistic
2-tuples in a way to rectify the problem of loss information of other
fuzzy linguistic approaches.
Abstract: The most important property of the Gene Ontology is
the terms. These control vocabularies are defined to provide
consistent descriptions of gene products that are shareable and
computationally accessible by humans, software agent, or other
machine-readable meta-data. Each term is associated with
information such as definition, synonyms, database references, amino
acid sequences, and relationships to other terms. This information has
made the Gene Ontology broadly applied in microarray and
proteomic analysis. However, the process of searching the terms is
still carried out using traditional approach which is based on keyword
matching. The weaknesses of this approach are: ignoring semantic
relationships between terms, and highly depending on a specialist to
find similar terms. Therefore, this study combines semantic similarity
measure and genetic algorithm to perform a better retrieval process
for searching semantically similar terms. The semantic similarity
measure is used to compute similitude strength between two terms.
Then, the genetic algorithm is employed to perform batch retrievals
and to handle the situation of the large search space of the Gene
Ontology graph. The computational results are presented to show the
effectiveness of the proposed algorithm.
Abstract: The novelty proposed in this study is twofold and consists in the developing of a new color similarity metric based on the human visual system and a new color indexing based on a textual approach. The new color similarity metric proposed is based on the color perception of the human visual system. Consequently the results returned by the indexing system can fulfill as much as possibile the user expectations. We developed a web application to collect the users judgments about the similarities between colors, whose results are used to estimate the metric proposed in this study. In order to index the image's colors, we used a text indexing engine to facilitate the integration of visual features in a database of text documents. The textual signature is build by weighting the image's colors in according to their occurrence in the image. The use of a textual indexing engine, provide us a simple, fast and robust solution to index images. A typical usage of the system proposed in this study, is the development of applications whose data type is both visual and textual. In order to evaluate the proposed method we chose a price comparison engine as a case of study, collecting a series of commercial offers containing the textual description and the image representing a specific commercial offer.
Abstract: Most of researches for conventional simulations were
studied focusing on flocks with a single species. While there exist the
flocking behaviors with a single species in nature, the flocking
behaviors are frequently observed with multi-species. This paper
studies on the flocking simulation for heterogeneous agents. In order
to simulate the flocks for heterogeneous agents, the conventional
method uses the identifier of flock, while the proposed method defines
the feature vector of agent and uses the similarity between agents by
comparing with those feature vectors. Based on the similarity, the
paper proposed the attractive force and repulsive force and then
executed the simulation by applying two forces. The results of
simulation showed that flock formation with heterogeneous agents is
very natural in both cases. In addition, it showed that unlike the
existing method, the proposed method can not only control the density
of the flocks, but also be possible for two different groups of agents to
flock close to each other if they have a high similarity.
Abstract: Chinese Idioms are a type of traditional Chinese idiomatic
expressions with specific meanings and stereotypes structure
which are widely used in classical Chinese and are still common in
vernacular written and spoken Chinese today. Currently, Chinese
Idioms are retrieved in glossary with key character or key word in
morphology or pronunciation index that can not meet the need of
searching semantically. OCIRS is proposed to search the desired
idiom in the case of users only knowing its meaning without any key
character or key word. The user-s request in a sentence or phrase will
be grammatically analyzed in advance by word segmentation, key
word extraction and semantic similarity computation, thus can be
mapped to the idiom domain ontology which is constructed to provide
ample semantic relations and to facilitate description logics-based
reasoning for idiom retrieval. The experimental evaluation shows that
OCIRS realizes the function of searching idioms via semantics, obtaining
preliminary achievement as requested by the users.
Abstract: Due to the tremendous amount of information provided
by the World Wide Web (WWW) developing methods for mining
the structure of web-based documents is of considerable interest. In
this paper we present a similarity measure for graphs representing
web-based hypertext structures. Our similarity measure is mainly
based on a novel representation of a graph as linear integer strings,
whose components represent structural properties of the graph. The
similarity of two graphs is then defined as the optimal alignment of
the underlying property strings. In this paper we apply the well known
technique of sequence alignments for solving a novel and challenging
problem: Measuring the structural similarity of generalized trees.
In other words: We first transform our graphs considered as high
dimensional objects in linear structures. Then we derive similarity
values from the alignments of the property strings in order to
measure the structural similarity of generalized trees. Hence, we
transform a graph similarity problem to a string similarity problem for
developing a efficient graph similarity measure. We demonstrate that
our similarity measure captures important structural information by
applying it to two different test sets consisting of graphs representing
web-based document structures.
Abstract: In practice, we often come across situations where it is
necessary to make decisions based on incomplete or uncertain data.
In control systems it may be due to the unknown exact mathematical
model, or its excessive complexity (e.g. nonlinearity) when it is
necessary to simplify it, respectively, to solve it using a rule base. In
the case of databases, searching data we compare a similarity
measure with of the requirements of the selection with stored data,
where both the select query and the data itself may contain vague
terms, for example in the form of linguistic qualifiers. In this paper,
we focus on the processing of uncertain data in databases and
demonstrate it on the example multi-criteria decision making in the
selection of variants, specified by higher number of technical
parameters.
Abstract: In this study, a fuzzy similarity approach for Arabic
web pages classification is presented. The approach uses a fuzzy
term-category relation by manipulating membership degree for the
training data and the degree value for a test web page. Six measures
are used and compared in this study. These measures include:
Einstein, Algebraic, Hamacher, MinMax, Special case fuzzy and
Bounded Difference approaches. These measures are applied and
compared using 50 different Arabic web pages. Einstein measure was
gave best performance among the other measures. An analysis of
these measures and concluding remarks are drawn in this study.
Abstract: A dissimilarity measure between the empiric
characteristic functions of the subsamples associated to the different
classes in a multivariate data set is proposed. This measure can be
efficiently computed, and it depends on all the cases of each class. It
may be used to find groups of similar classes, which could be joined
for further analysis, or it could be employed to perform an
agglomerative hierarchical cluster analysis of the set of classes. The
final tree can serve to build a family of binary classification models,
offering an alternative approach to the multi-class SVM problem. We
have tested this dendrogram based SVM approach with the oneagainst-
one SVM approach over four publicly available data sets,
three of them being microarray data. Both performances have been
found equivalent, but the first solution requires a smaller number of
binary SVM models.
Abstract: This work proposes an approach to address automatic
text summarization. This approach is a trainable summarizer, which
takes into account several features, including sentence position,
positive keyword, negative keyword, sentence centrality, sentence
resemblance to the title, sentence inclusion of name entity, sentence
inclusion of numerical data, sentence relative length, Bushy path of
the sentence and aggregated similarity for each sentence to generate
summaries. First we investigate the effect of each sentence feature on
the summarization task. Then we use all features score function to
train genetic algorithm (GA) and mathematical regression (MR)
models to obtain a suitable combination of feature weights. The
proposed approach performance is measured at several compression
rates on a data corpus composed of 100 English religious articles.
The results of the proposed approach are promising.
Abstract: The method of gait identification based on the nearest neighbor classification technique with motion similarity assessment by the dynamic time warping is proposed. The model based kinematic motion data, represented by the joints rotations coded by Euler angles and unit quaternions is used. The different pose distance functions in Euler angles and quaternion spaces are considered. To evaluate individual features of the subsequent joints movements during gait cycle, joint selection is carried out. To examine proposed approach database containing 353 gaits of 25 humans collected in motion capture laboratory is used. The obtained results are promising. The classifications, which takes into consideration all joints has accuracy over 91%. Only analysis of movements of hip joints allows to correctly identify gaits with almost 80% precision.
Abstract: People detection from images has a variety of applications such as video surveillance and driver assistance system, but is still a challenging task and more difficult in crowded environments such as shopping malls in which occlusion of lower parts of human body often occurs. Lack of the full-body information requires more effective features than common features such as HOG. In this paper, new features are introduced that exploits global self-symmetry (GSS) characteristic in head-shoulder patterns. The features encode the similarity or difference of color histograms and oriented gradient histograms between two vertically symmetric blocks. The domain-specific features are rapid to compute from the integral images in Viola-Jones cascade-of-rejecters framework. The proposed features are evaluated with our own head-shoulder dataset that, in part, consists of a well-known INRIA pedestrian dataset. Experimental results show that the GSS features are effective in reduction of false alarmsmarginally and the gradient GSS features are preferred more often than the color GSS ones in the feature selection.
Abstract: Feature selection plays an important role in applications with high dimensional data. The assessment of the stability of feature selection/ranking algorithms becomes an important issue when the dataset is small and the aim is to gain insight into the underlying process by analyzing the most relevant features. In this work, we propose a graphical approach that enables to analyze the similarity between feature ranking techniques as well as their individual stability. Moreover, it works with whatever stability metric (Canberra distance, Spearman's rank correlation coefficient, Kuncheva's stability index,...). We illustrate this visualization technique evaluating the stability of several feature selection techniques on a spectral binary dataset. Experimental results with a neural-based classifier show that stability and ranking quality may not be linked together and both issues have to be studied jointly in order to offer answers to the domain experts.
Abstract: Text similarity measurement is a fundamental issue in
many textual applications such as document clustering, classification,
summarization and question answering. However, prevailing approaches
based on Vector Space Model (VSM) more or less suffer
from the limitation of Bag of Words (BOW), which ignores the semantic
relationship among words. Enriching document representation
with background knowledge from Wikipedia is proven to be an effective
way to solve this problem, but most existing methods still
cannot avoid similar flaws of BOW in a new vector space. In this
paper, we propose a novel text similarity measurement which goes
beyond VSM and can find semantic affinity between documents.
Specifically, it is a unified graph model that exploits Wikipedia as
background knowledge and synthesizes both document representation
and similarity computation. The experimental results on two different
datasets show that our approach significantly improves VSM-based
methods in both text clustering and classification.
Abstract: Genome profiling (GP), a genotype based technology, which exploits random PCR and temperature gradient gel electrophoresis, has been successful in identification/classification of organisms. In this technology, spiddos (Species identification dots) and PaSS (Pattern similarity score) were employed for measuring the closeness (or distance) between genomes. Based on the closeness (PaSS), we can buildup phylogenetic trees of the organisms. We noticed that the topology of the tree is rather robust against the experimental fluctuation conveyed by spiddos. This fact was confirmed quantitatively in this study by computer-simulation, providing the limit of the reliability of this highly powerful methodology. As a result, we could demonstrate the effectiveness of the GP approach for identification/classification of organisms.
Abstract: Clustering categorical data is more complicated than
the numerical clustering because of its special properties. Scalability
and memory constraint is the challenging problem in clustering large
data set. This paper presents an incremental algorithm to cluster the
categorical data. Frequencies of attribute values contribute much in
clustering similar categorical objects. In this paper we propose new
similarity measures based on the frequencies of attribute values and
its cardinalities. The proposed measures and the algorithm are
experimented with the data sets from UCI data repository. Results
prove that the proposed method generates better clusters than the
existing one.
Abstract: An important structuring mechanism for knowledge bases is building clusters based on the content of their knowledge objects. The objects are clustered based on the principle of maximizing the intraclass similarity and minimizing the interclass similarity. Clustering can also facilitate taxonomy formation, that is, the organization of observations into a hierarchy of classes that group similar events together. Hierarchical representation allows us to easily manage the complexity of knowledge, to view the knowledge at different levels of details, and to focus our attention on the interesting aspects only. One of such efficient and easy to understand systems is Hierarchical Production rule (HPRs) system. A HPR, a standard production rule augmented with generality and specificity information, is of the following form Decision If < condition> Generality Specificity . HPRs systems are capable of handling taxonomical structures inherent in the knowledge about the real world. In this paper, a set of related HPRs is called a cluster and is represented by a HPR-tree. This paper discusses an algorithm based on cumulative learning scenario for dynamic structuring of clusters. The proposed scheme incrementally incorporates new knowledge into the set of clusters from the previous episodes and also maintains summary of clusters as Synopsis to be used in the future episodes. Examples are given to demonstrate the behaviour of the proposed scheme. The suggested incremental structuring of clusters would be useful in mining data streams.
Abstract: Iris-based biometric authentication is gaining importance
in recent times. Iris biometric processing however, is a complex
process and computationally very expensive. In the overall processing
of iris biometric in an iris-based biometric authentication system,
feature processing is an important task. In feature processing, we extract
iris features, which are ultimately used in matching. Since there
is a large number of iris features and computational time increases
as the number of features increases, it is therefore a challenge to
develop an iris processing system with as few as possible number of
features and at the same time without compromising the correctness.
In this paper, we address this issue and present an approach to feature
extraction and feature matching process. We apply Daubechies D4
wavelet with 4 levels to extract features from iris images. These
features are encoded with 2 bits by quantizing into 4 quantization
levels. With our proposed approach it is possible to represent an
iris template with only 304 bits, whereas existing approaches require
as many as 1024 bits. In addition, we assign different weights to
different iris region to compare two iris templates which significantly
increases the accuracy. Further, we match the iris template based on
a weighted similarity measure. Experimental results on several iris
databases substantiate the efficacy of our approach.