Showing papers in "Transactions of The Japanese Society for Artificial Intelligence in 2011"
TL;DR: A novel formalization of CSG is proposed, which assumes that the value of a characteristic function is given by an optimal solution of a distributed constraint optimization problem (DCOP) among the agents of a coalition, and an approximation algorithm is developed that can find a CS with quality guarantees.
Abstract: Forming effective coalitions is a major research challenge in AI and multi-agent systems. Coalition Structure Generation (CSG) involves partitioning a set of agents into coalitions so that social surplus (the sum of the rewards of all coalitions) is maximized. A partition is called a coalition structure (CS). In traditional works, the value of a coalition is given by a black box function called a characteristic function. In this paper, we propose a novel formalization of CSG, i.e., we assume that the value of a characteristic function is given by an optimal solution of a distributed constraint optimization problem (DCOP) among the agents of a coalition. A DCOP is a popular approach for modeling cooperative agents, since it is quite general and can formalize various application problems in MAS. At first glance, this approach sounds like a very bad idea considering the computational costs, since we need to solve an NP-hard problem just to obtain the value of a single coalition. To optimally solve a CSG, we might need to solve O(2n) DCOP problem instances, where n is the number of agents. However, quite surprisingly, we show that an approximation algorithm, whose computational cost is about the same as solving just one DCOP, can find a CS whose social surplus is at least max(2/n, 1/(w*+1)) of the optimal CS, where w* is the tree width of a constraint graph. Furthermore, we can generalize this approximation algorithm with a parameter k, i.e., the generalized algorithm can find a CS whose social surplus is at least max(2k/n, k/(w*+1)) of the optimal CS by exploring more search space. These results illustrate that the locality of interactions among agents, which is explicitly modeled in the DCOP formalization, is quite useful in developing efficient CSG algorithms with quality guarantees.
66 citations
TL;DR: This paper presents a new way of formalizing the Coalition Structure Generation problem (CSG), so that it can be applied constraint optimization techniques to it, and describes the complexity of the CSG under recently developed compact representation schemes for characteristic functions.
Abstract: This paper presents a new way of formalizing the Coalition Structure Generation problem (CSG), so that we can apply constraint optimization techniques to it. Forming effective coalitions is a major research challenge in AI and multi-agent systems. CSG involves partitioning a set of agents into coalitions so that social surplus is maximized. Traditionally, the input of the CSG problem is a black-box function called a characteristic function, which takes a coalition as an input and returns the value of the coalition. As a result, applying constraint optimization techniques to this problem has been infeasible. However, characteristic functions that appear in practice often can be represented concisely by a set of rules, rather than a single black-box function. Then, we can solve the CSG problem more efficiently by applying constraint optimization techniques to the compact representation directly. We present new formalizations of the CSG problem by utilizing recently developed compact representation schemes for characteristic functions. We first characterize the complexity of the CSG under these representation schemes. In this context, the complexity is driven more by the number of rules rather than by the number of agents. Furthermore, as an initial step towards developing efficient constraint optimization algorithms for solving the CSG problem, we develop mixed integer programming formulations and show that an off-the-shelf optimization package can perform reasonably well, i.e., it can solve instances with a few hundred agents, while the state-of-the-art algorithm (which does not make use of compact representations) can solve instances with up to 27 agents.
34 citations
TL;DR: It is shown that the performance of the proposed model is superior to that of the k-means and PLSI in terms of category mining, and is applicable for marketing support, service modeling, and decision making in various business fields, including retail services.
Abstract: This paper describes a computational customer behavior modeling by Bayesian network with an appropriate category. Categories are generated by a heterogeneous data fusion using an ID-POS data and customer's questionnaire responses with respect to their lifestyle. We propose a latent class model that is an extension of PLSI model. In the proposed model, customers and items are classified probabilistically into some latent lifestyle categories and latent item category. We show that the performance of the proposed model is superior to that of the k-means and PLSI in terms of category mining. We produce a Bayesian network model including the customer and item categories, situations and conditions of purchases. Based on that network structure, we can systematically identify useful knowledge for use in sustainable services. In the retail service, knowledge management with point of sales data mining is integral to maintaining and improving productivity. This method provides useful knowledge based on the ID-POS data for efficient customer relationship management and can be applicable for other service industries. This method is applicable for marketing support, service modeling, and decision making in various business fields, including retail services.
16 citations
TL;DR: The results indicated that there is moderately significant correlation between HDS-R score and synthesis of several selected prosodic features, which suggests that prosody-based speech sound analysis has possibility to screen the elderly with cognitive impairment.
Abstract: This paper presents a new trial approach to early detection of cognitive impairment in the elderly with the use of speech sound analysis and multivariate statistical technique. In this paper, we focus on the prosodic features from speech sound. Japanese 115 subjects (32 males and 83 females between ages of 38 and 99) participated in this study. We collected speech sound in a few segments of dialogue of HDS-R examination. The segments corresponds to speech sound that is answering for questions on time orientation and number backward count. Firstly, 130 prosodic features have been extracted from each of the speech sounds. These prosodic features consist of spectral and pitch features (53), formant features (56), intensity features (19), and speech rate and response time (2). Secondly, these features are refined by principal component analysis and/or feature selection. Lastly, we have calculated speech prosody-based cognitive impairment rating (SPCIR) by multiple linear regression analysis. The results indicated that there is moderately significant correlation between HDS-R score and synthesis of several selected prosodic features. Consequently, adjusted coefficient of determination R2=0.50 suggests that prosody-based speech sound analysis has possibility to screen the elderly with cognitive impairment.
11 citations
TL;DR: The experimental results show that the predictive performance of the proposed kernel is competitive with that of the existing efficient tree kernel proposed by Vishwanathan et al., and is also empirically faster than the existing kernel.
Abstract: Kernel method is one of the promising approaches to learning with tree-structured data, and various efficient tree kernels have been proposed to capture informative structures in trees. In this paper, we propose a new tree kernel function based on ``subpath sets'' to capture vertical structures in tree-structured data, since tree-structures are often used to code hierarchical information in data. We also propose a simple and efficient algorithm for computing the kernel by extending the Multikey quicksort algorithm used for sorting strings. The time complexity of the algorithm is O((|T_1|+|T_2|)log(|T_1|+|T_2|)) time on average, and the space complexity is O({|T_1|+|T_2|)}, where |T_1| and |T_2| are the numbers of nodes in two trees T_1 and T_2. We apply the proposed kernel to two supervised classification tasks, XML classification in web mining and glycan classification in bioinformatics. The experimental results show that the predictive performance of the proposed kernel is competitive with that of the existing efficient tree kernel proposed by Vishwanathan et al., and is also empirically faster than the existing kernel.
9 citations
TL;DR: The key ideas of this new protocol are to 1) compute global information with a spanning tree, 2) update step length simultaneously with a synchronization protocol, and 3) estimate lower bounds during the search.
Abstract: Distributed Lagrangian Relaxation Protocol (DisLRP) has been proposed to solve a distributed combinatorial maximization problem called the Generalized Mutual Assignment Problem (GMAP). In DisLRP, when updating Lagrange multipliers (prices) of goods, the agents basically control their step length, which determines the degree of update, by a static rule. A merit of this updating rule is that since it is static, it is easy to implement even without a central control. Furthermore, if we choose this static rule appropriately, we have observed empirically that DisLRP converges to a state providing a good upper bound. However, it must be difficult to devise such a good static rule for updating step length since it naturally depends on problem instances to be solved. On the other hand, in a centralized context, the Lagrangian relaxation approach has conventionally computed step length by exploiting the least upper bound obtained during the search and a lower bound obtained through preprocessing. In this paper, we achieve this approach in a distributed environment where no central control exists and name the resultant protocol Adaptive DisLRP (ADisLRP). The key ideas of this new protocol are to 1) compute global information with a spanning tree, 2) update step length simultaneously with a synchronization protocol, and 3) estimate lower bounds during the search. We also show the robustness of ADisLRP through experiments where we compared ADisLRP with the previous protocols on the critically hard benchmark instances.
7 citations
TL;DR: A novel approach that uses conditional random fields and self-supervised learning to automatically extract all of the basic attributes and the transition between activities derived from sentences in Japanese web pages is proposed.
Abstract: In our definition, human activity can be expressed by five basic attributes: actor, action, object, time and location. The goal of this paper is describe a method to automatically extract all of the basic attributes and the transition between activities derived from sentences in Japanese web pages. However, previous work had some limitations, such as high setup costs, inability to extract all attributes, limitation on the types of sentences that can be handled, and insufficient consideration interdependency among attributes. To resolve these problems, this paper proposes a novel approach that uses conditional random fields and self-supervised learning. Given a small corpus sample as input, it automatically makes its own training data and a feature model. Based on the feature model, it automatically extracts all of the attributes and the transition between the activities in each sentence retrieved from the Web corpus. This approach treats activity extraction as a sequence labeling problem, and has advantages such as domain-independence, scalability, and does not require any human input. Since it is unnecessary to fix the number of elements in a tuple, this approach can extract all of the basic attributes and the transition between activities by making only a single pass. Additionally, by converting to simpler sentences, the approach can deal with complex sentences retrieved from the Web. In an experiment, this approach achieves high precision (activity: 88.9%, attributes: over 90%, transition: 87.5%).
5 citations
TL;DR: This paper proposes an unsupervised detection algorithm based on an entropy-like measure called document complexity, which reflects how many similar documents exist in the input collection of documents, and substitutes an estimated occurrence probability of each document for its complexity.
Abstract: In this paper, we study content-based spam detection for spams that are generated by copying a seed document with some random perturbations. We propose an unsupervised detection algorithm based on an entropy-like measure called document complexity, which reflects how many similar documents exist in the input collection of documents. As the document complexity, however, is an ideal measure like Kolmogorov complexity, we substitute an estimated occurrence probability of each document for its complexity. We also present an efficient algorithm that estimates the probabilities of all documents in the collection in linear time to its total length. Experimental results showed that our algorithm especially works well for word salad spams, which are believed to be difficult to detect automatically.
4 citations
TL;DR: It is argued that task-oriented spoken dialogue systems or communication robots do not need to quickly respond verbally as long as they quickly respond non-verbally by showing their internal states by using an artificial subtle expression, and both the slow reply speed and the blinking light expression can reduce speech collisions, and improve a user's impression.
Abstract: We argue that task-oriented spoken dialogue systems or communication robots do not need to quickly respond verbally as long as they quickly respond non-verbally by showing their internal states by using an artificial subtle expression. This paper describes an experiment whose results support this point. In this experiment, 48 participants engaged in reservation tasks with a spoken dialogue system coupled with an interface robot using a blinking light expression. The blinking light expression is designed as an artificial subtle expression to intuitively notify a user about a robot's internal states (such as processing) for the sake of reducing speech collisions as consequences of turn-taking failures due to end-of-turn misdetection. Speech collisions harm smooth speech communication and degrade system usability. Two experimental factors were setup: the blinking light factor (with or without a blinking light) and the reply speed factor (moderate or slow reply speed), resulting in four experimental conditions. The results suggest that both the slow reply speed and the blinking light expression can reduce speech collisions, and improve a user's impression. Meanwhile, contrary to expectation, no degradation of evaluation due to the slow reply speed was found.
4 citations
TL;DR: Results from the experiments implied that online negotiation involving haptic interaction can increase the sense of presence and is also helpful for expressing one's emotions, which play major role in online negotiations.
Abstract: In this study, we bring the haptic technology into the online negotiation system to improve the method of conveying nonverbal information. In this system, subjects can convey the nonverbal information by changing the ball's size as well as the force feedback. We conducted two experiments and compared them to verify the effect that the haptic interaction brings about. Results from the experiments implied that online negotiation involving haptic interaction can increase the sense of presence and is also helpful for expressing one's emotions, which play major role in online negotiations.
4 citations
TL;DR: The way to treat multi-agent systems uniformly using the fixed-point operator of the extended BDI logic \ omatoes is shown.
Abstract: In multi-agent environments, to model cooperations among autonomous agents, many notions such as mutual beliefs and joint intentions, recognition of possibilities to achieve a goal with cooperation, and team formations, should be formally represented. In the traditional BDI logics, it is hard to treat them uniformly. We show the way to treat them uniformly using the fixed-point operator of the extended BDI logic \ omatoes. We also give some examples to apply it to the proof of some behaviors of multi-agent systems.
TL;DR: G-Monaka substantially increases the performance of semantic category acquisition compared to conventional methods, including distributional similarity, bootstrapping-based Espresso, and its graph-based extension g-Espresso, in terms of F-value of the NE category task from unsegmented Japanese newspaper articles.
Abstract: Extraction of named entitiy classes and their relationships from large corpora often involves morphological analysis of target sentences and tends to suffer from out-of-vocabulary words. In this paper we propose a semantic category extraction algorithm called Monaka and its graph-based extention g-Monaka, both of which use character n-gram based patterns as context to directly extract semantically related instances from unsegmented Japanese text. These algorithms also use ``bidirectional adjacent constraints,'' which states that reliable instances should be placed in between reliable left and right context patterns, in order to improve proper segmentation. Monaka algorithms uses iterative induction of instaces and pattens similarly to the bootstrapping algorithm Espresso. The g-Monaka algorithm further formalizes the adjacency relation of character n-grams as a directed graph and applies von Neumann kernel and Laplacian kernel so that the negative effect of semantic draft, i.e., a phenomenon of semantically unrelated general instances being extracted, is reduced. The experiments show that g-Monaka substantially increases the performance of semantic category acquisition compared to conventional methods, including distributional similarity, bootstrapping-based Espresso, and its graph-based extension g-Espresso, in terms of F-value of the NE category task from unsegmented Japanese newspaper articles.
TL;DR: A new traffic light control system based on multi-agent model is proposed, where offset value, one of the main traffic light parameters, is controlled by using only local information, and green-wave formation is formed through the coordination of each intersection agent.
Abstract: Traffic jam is one of critical issues in urban life. By which, many social problems, for example, time loss, economical loss, and environmental pollution are caused. There are two typical methods for solving traffic jam, improvement of car navigation system and control of traffic lights. We focus on control of traffic lights. Existing traffic light control system is basically centralized control type and lacks robustness and scalability. If the central computer becomes breakdown, all traffic lights received the damage of it. In this paper, we propose a new traffic light control system based on multi-agent model. The offset value, one of the main traffic light parameters, is controlled by using only local information, and green-wave formation is formed through the coordination of each intersection agent.
TL;DR: A new text-mining method that could forecast in higher accuracy about both the level and direction of long-term market trends and showed high returns with annual rate averages as a result of the implementation test.
Abstract: In this study, we propose a new text-mining method for long-term market analysis. Using our method, we performe out-of-sample tests using monthly price data of financial markets; Japanese government bond market, Japanese stock market, and the yen-dollar market. First we extract feature vectors from monthly reports of Bank of Japan. Then, trends of each market are estimated by regression analysis using the feature vectors. As a result of comparison with support vector regression, the proposal method could forecast in higher accuracy about both the level and direction of long-term market trends. Moreover, our method showed high returns with annual rate averages as a result of the implementation test.
TL;DR: This paper develops an algorithm for solving an adversarial/cooperative agent for good/nogood problem by generalizing the asynchronous backtracking algorithm used for solving a DisCSP and develops a method that improves this basic algorithm.
Abstract: In this paper, we extend the traditional formalization of distributed constraint satisfaction problems (DisCSP) to a quantified DisCSP. A quantified DisCSP includes several universally quantified variables, while all of the variables in a traditional DisCSP are existentially quantified. A universally quantified variable represents a choice of nature or an adversary. A quantified DisCSP formalizes a situation where a team of agents is trying to make a robust plan against nature or an adversary. In this paper, we present the formalization of such a quantified DisCSP and develop an algorithm for solving it by generalizing the asynchronous backtracking algorithm used for solving a DisCSP. In this algorithm, agents communicate a value assignment called a good in addition to the nogood used in asynchronous backtracking. Interestingly, the procedures executed by an adversarial/cooperative agent for good/nogood are completely symmetrical. Furthermore, we develop a method that improves this basic algorithm. Experimental evaluation results illustrate that we observe an easy-hard-easy transition by changing the tightness of the constraints, while very loose problem instances are relatively hard. The modification of the basic algorithm is also effective and reduces the number of cycles by approximately 25% for the hardest problem instances.
TL;DR: It is shown that the utility and the success of the team formation is deeply affected by depth of the tree structure and number of tasks, and an effective method of dynamic reorganization using reinforcement learning is proposed.
Abstract: We propose an effective method of dynamic reorganization using reinforcement learning for the team formation in multi-agent systems (MAS). A task in MAS usually consists of a number of subtasks that require their own resources, and it has to be processed in the appropriate team whose agents have the sufficient resources. The resources required for tasks are often unknown \ extit{a priori} and it is also unknown whether their organization is appropriate to form teams for the given tasks or not. Therefore, their organization should be adopted according to the environment where agents are deployed. In this paper, we investigated how the structures of network and the number of tasks affect team formations of the agents. We will show that the utility and the success of the team formation is deeply affected by depth of the tree structure and number of tasks.
TL;DR: It is confirmed that the web-based production system for education works sufficiently in a standard computer facility and students learned important features of human cognitive processing by meta-monitoring their own thinking processes.
Abstract: In learning cognitive science, students must learn how to handle an actual production system that runs on a computer. We developed a web-based production system for education that can be used from anywhere such as class rooms, offices, and homes. The system as a web-based application has many advantages as a learning support system. It furnishes students with learning support information for if-clause matching to facilitate learning. We confirmed that our system works sufficiently in a standard computer facility and students learned important features of human cognitive processing by meta-monitoring their own thinking processes.
TL;DR: This research model membership services as club goods and analyze the reward programs and suggests that the subjects' behavior might be based on the value of how many times they should use services in order to recover payment of the entry fee.
Abstract: In service industries, service providers offer various reward programs to consumers with the aim to build customer loyalty and increase sales In this research, we model membership services as club goods and analyze the reward programs Club goods are defined as excludable and nonrival public goods We conduct equilibrium analysis, experiments with human subjects and multi-agent simulation In theoretical equilibrium, all consumer players become a member But the results of the experiments indicate that provider players do not properly set the entry fee and the service price for members and consumer players sometimes make irrational decisions of membership entry Moreover, we elucidate subjects' behavior mechanism of the membership entry by simulations Results of simulations suggest that the subjects' behavior might be based on the value of how many times they should use services in order to recover payment of the entry fee
TL;DR: Through various types of test simulations using the developed simulator, it is demonstrated that the simulator with the cognitive error model is a powerful tool to quantitatively evaluate traffic accidents and to discover such a dangerous situation that accidents frequently occur.
Abstract: Road traffic is a key portion of infrastructure to support mobility and transportation of human beings and goods. At the same time, it includes various kinds of risks. One of the most critical ones is a traffic accident. To evaluate traffic accidents quantitatively, we have newly developed a cognitive error model and implemented it into a multi-agent based traffic simulator. In the traffic simulator, each component creating traffic phenomena is modeled as an agent, and interaction among numerous agents simulates nonlinear behaviors of urban traffics. An actual traffic accident often occurs when a car driver overlooks something to watch, such as other cars, pedestrians, traffic signals, or obstacles. In the cognitive error model we developed, a driver agent has its own field of view and a gazing point, and cannot recognize objects off the gazing point. Through various types of test simulations using the developed simulator, we demonstrate that the simulator with the cognitive error model is a powerful tool to quantitatively evaluate traffic accidents and to discover such a dangerous situation that accidents frequently occur.
TL;DR: It is found that the market in which short-selling was allowed was more stable than the market with short- selling regulation, and a bubble emerged in the regulated market.
Abstract: Since the subprime mortgage crisis in the United Sates, stock markets around the world have crashed, revealing their instability To stem the decline in stock prices, short-selling regulations have been implemented in many markets However, their effectiveness remains unclear In this paper, we discuss the effectiveness of short-selling regulation using artificial markets An artificial market that is an agent-based model of financial markets is useful to observe the market mechanism That is, it is effective for analyzing causal relationship between the behaviors of market participants and the transition of market price We constructed an artificial market that allows short-selling and an artificial market with short-selling regulation and have observed the stock prices in both of these markets We have demonstrated that our artificial market had some properties of actual markets We found that the market in which short-selling was allowed was more stable than the market with short-selling regulation, and a bubble emerged in the regulated market We evaluated the values of assets of agents who used three trading strategies, specifically, these agents were fundamentalists, chartists, and noise traders The fundamentalists had the best performance among the three types of agents
TL;DR: A method for supporting an exploratory analysis of spatiotemporal trend information by focusing on comparative analysis using visualization cube, of which the effectiveness is evaluated through experiments with test participants.
Abstract: This paper proposes a method for supporting an exploratory analysis of spatiotemporal trend information by focusing on comparative analysis. Recent growth of the Web has brought us various kinds of information, among which trend information is one of crucial information for our decision-making. In order to develop an exploratory analysis support system for spatiotemporal trend information, visualization cube has been proposed. It is an abstract data structure with 4 axes, on which 5 operations for generating views are defined. The prototype system has been developed based on the concept of visualization cube, of which usability has been evaluated through the experiments conducted at an elementary school. However, it was also pointed out that the data space that can be explored with a single interaction by the existing system is not so large. In order to solve this problem, the proposed method employs dual views by considering two visualization cubes. By operating two visualization cubes at the same time, six kinds of comparative analyses in a broad sense can be supported. The exploratory analysis support system is developed based on the proposed concept, of which the effectiveness is evaluated through experiments with test participants. Although no synchronization mechanism between views is implemented in the prototype system, effective synchronization functions are discussed based on the experimental results.
TL;DR: The evolution of OntoGear is described, and the functionality of physical process integration model (hereafter ppim) that allows engineers to describe whole processes of any artifacts in the form of function decomposition trees in their product lifecycle is realized.
Abstract: In this paper, we describe the evolution of OntoGear, which has ever been discussed in the previous research, and newly developed software tools. OntoGear is an engineering knowledge management software platform based on ontology engineering, and its previous system has provided the most basic functionalities based on systematization framework of functional knowledge; i.e. describing a function decomposition tree, and building way knowledge base to share and reuse organized and generalized functional knowledge. Compared to the previous one, our new system realized the functionality of physical process integration model (hereafter ppim) that allows engineers to describe whole processes of any artifacts in the form of function decomposition trees in their product lifecycle. New OntoGear system is appended two client tools and a server--a modeling tool for ppim, a viewer for ppim and OntoGear server. Furthermore, we introduce an application using the system for design support of application system of SOFC (Solid Oxide Fuel Cell), which is a kind of fuel cell and expected for its quick realization. The application contributes to clarify whole functional structures and the relationships among them through SOFC system's lifecycle. Since one of the tools of OntoGear software environment has presently released as a software product, which is named OntoloGear SE (Standard Edition), by MetaMoJi Corporation, we briefly report the productization status.
TL;DR: A fast machine learning platform which reduces the processing overheads at iterative procedures of machine learning, and shows the performance of Variational Bayes clustering and linear SVM implemented on this platform.
Abstract: We propose a computing platform for parallel machine learning. Learning from large-scale data has become common, so that parallelization techniques are increasingly applied to machine learning algorithms in order to reduce calculation time. Problems of parallelization are implementation costs and calculation overheads. Firstly, we formulate MapReduce programming model specialized in parallel machine learning. It represents learning algorithms as iterations of following two phases: applying data to machine learning models and updating model parameters. This model is able to describe various kinds of machine learning algorithms, such as k-means clustering, EM algorithm, and linear SVM, with comparable implementation cost to the original MapReduce. Secondly, we propose a fast machine learning platform which reduces the processing overheads at iterative procedures of machine learning. Machine learning algorithms iteratively read the same training data in the data application phase. Our platform keeps the training data in local memories of each worker during iterative procedures, which leads to acceleration of data access. We evaluate performance of our platform on three experiments. Our platform executes k-means clustering 2.85 to 118 times faster than the MapReduce approach, and shows 9.51 times speedup with 40 processing cores against 8 cores. We also show the performance of Variational Bayes clustering and linear SVM implemented on our platform.
TL;DR: It is demonstrated that introducing off-diagonal elements to singular value matrix in pLSI is equal to permitting joint probability between different hidden variables, and that this extension showed tolerance for over-learning and over-fitting problems.
Abstract: probabilistic Latent Semantic Indexing (pLSI) is a fundamental method for the analysis of text and related resources which is based on a simple statistical model. This method has high extendibility and scalability due to its simplicity. pLSI is also known as matrix factorization method such as Singular Value Decomposition(SVD) or Non-negative Matrix Factorization. Using pLSI, three matrices which include one diagonal matrix as SVD are achieved. The diagonal elements of this diagonal matrix represent singular values in SVD. However it is not entirely clear what the diagonal matrix of pLSI represents. Then it is also unclear whether the diagonalization constraint is necessary in pLSI. This question is the starting point of this paper. To make an answer for this question, we demonstrated that introducing off-diagonal elements to singular value matrix in pLSI is equal to permitting joint probability between different hidden variables. Although permitting joint probability in pLSI does not lose scalability and simplicity, our experiments demonstrated that this extension showed tolerance for over-learning and over-fitting problems.
TL;DR: A method to model the sense of value of each medical staff as his/her understanding about medical service workflow, and to obtain the practical knowledge using the models is proposed.
Abstract: It is ideal to provide medical services as patient-oriented. The medical staff members share the final goals to recover patients. Toward the goals, each staff has practical knowledge to achieve patient-oriented medical services. But each medical staff has his/her own sense of value that comes from his/her expertness. Therefore the practical knowledge sometimes conflicts. The aim of this research is to develop an intelligent system to support externalizing practical knowledge, and sharing it among medical staff members. In this paper, the author propose a method to model the sense of value of each medical staff as his/her understanding about medical service workflow, and to obtain the practical knowledge using the models. The method was experimented by an implementation of knowledge-sharing system base on the method and by its trial use in Miyazaki University Hospital.
TL;DR: This framework provides query evaluation rules based on the proposed lexicographic semantics, which guarantees that each query using such a rich vocabulary is correctly evaluated over the underlying Semantic Web.
Abstract: This paper describes a new framework for querying the Semantic Web using a rich vocabulary. This framework consists of two mechanisms; one for building a rich vocabulary based on lexicographic semantics, and the other for evaluating queries using such a vocabulary. A vocabulary built by the former mechanism has the following two features: (a) its richness because of its expandability and (b) the lexicographic-semantic definition of its words. Query expressions using such a rich vocabulary satisfy the following two properties: (c) no need to use nested query structures, and (d) no need to use variables. In our framework, a new word, i.e., a derived word, can be defined as a character string label given to an expression that combines already defined words with operators. This expression, or phrase, works as a lexicographic definition of this derived word. Each vocabulary consists of basic words and derived words. A lexicon of a vocabulary denotes a set of lexicographic definitions of all of its derived words. Once someone defines a lexicon of a large vocabulary with all of its basic words being mapped to an ontology of the Semantic Web, users can query this Semantic Web using this vocabulary. The same lexicon can be reused for the Semantic Web that has a different ontology if all of its basic words are newly mapped to its ontology. Use of a rich vocabulary in querying a Semantic Web simplifies the query sentence structure and removes the necessity of using variables from each query, which makes it much easier for users to query the Semantic Web. This framework provides query evaluation rules based on the proposed lexicographic semantics, which guarantees that each query using such a rich vocabulary is correctly evaluated over the underlying Semantic Web.
TL;DR: Two algorithms named FAGV-gSpan and FAGE- gSpan are developed for finding frequent patterns from a single graph with numerical attributes, by effectively combining techniques of graph mining and quantitative itemset mining.
Abstract: In this paper, we discuss pattern mining problems from a single graph whose vertices or edges contain a set of numerical attributes. Several networks can be naturally represented in this kind of complex graph. A typical example is a social network whose vertex corresponds to a person with some numerical attributes such as age, salary and so on. Another example is a communication network whose edge represents a communication between devices. We can associate the numbers of communications per certain time period to each edge. For these kinds of complex graphs, it is meaningful to consider not only graph strucuture but also the internal informations. Although it can be expected that these kinds of data will increase rapidly, most of current graph mining algorithms do not handle these complex graphs directly. Motivated by the above background, we developed algorithms named FAGV-gSpan and FAGE-gSpan for finding frequent patterns from a single graph with numerical attributes, by effectively combining techniques of graph mining and quantitative itemset mining. The effectiveness of the proposed algorithms was confirmed by experiments using real world datasets.
TL;DR: This work proposes a method for verifying a relational search result that exploits the symmetric properties in proportional analogies to improve the reliability of the verification process and shows that the proposed method improves the accuracy of a relationalsearch engine by a wide margin.
Abstract: Relational similarity can be defined as the similarity between two semantic relations R and R' that exist respectively in two word pairs (A,B) and (C,D). Relational search, a novel search paradigm that is based on the relational similarity between word pairs, attempts to find a word D for the slot ? in the query {(A,B), (C,?)} such that the relational similarity between the two word pairs (A, B) and (C, D) is a maximum. However, one problem frequently encountered by a Web-based relational search engine is that the inherent noise in Web text leads to incorrect measurement of relational similarity. To overcome this problem, we propose a method for verifying a relational search result that exploits the symmetric properties in proportional analogies. To verify a candidate result D for a query {(A, B), (C, ?)}, we replace the original question mark by D to create a new query {(A,B),(?,D)} and verify that we can retrieve C as a candidate for the new query. The score of C in the new query can be seen as a complementary score of D because it reflects the reliability of D in the original query. Moreover, transformations of words in proportional analogies lead to relational symmetries that can be utilized to accurately measure the relational similarity between two semantic relations. For example, if the two word pairs (A,B) and (C, D) show a high degree of relational similarity then the two word pairs (B,A) and (D,C) also have a high degree of relational similarity. We apply this idea in relational search by using symmetric queries such as {(B, A), (D, ?)} to create six queries for verifying a candidate answer D to improve the reliability of the verification process. Our experimental results on the Scholastic Aptitude Test (SAT) analogy benchmark show that the proposed method improves the accuracy of a relational search engine by a wide margin.
TL;DR: A new similarity is proposed that is the summation of similarities based on the logistic regression that performs good accuracy for social network, citation network, dictionary network, biological network and transfer network and accuracy of proposed methods reaches higher than HRG.
Abstract: Recently, network analysis has been intensively investigated in several fields of science. Link prediction is a problem of predicting the existence of a link between two entities based on observed links, and it is one of the popular link mining tasks. Although many link prediction methods have been proposed, they have their merits and demerits. In this paper, we present two topics as follows: 1) In order to obtain the strategies of selecting the best link prediction methods, we perform experiments of six link prediction methods (Common Neighbors (CN) , Jaccard's Coefficient (JC) , Adamic/Adar (AA) , Shortest Path (SP) , Preferential Attachment (PA) and Hierarchical Random Graph (HRG) ) for 39 real networks. 2) We propose a new similarity that is the summation of similarities based on the logistic regression. We used 10-fold cross validation and bagging for model selection of proposed method. We estimate the accuracy and computation time of HRG, proposed method (bagging) and proposed method (10-fold cross validation) for 28 data sets. As a result of 1) , CN, JC and AA achieve good performance for the networks that has higher clustering coefficient than 0.4. SP achieves good performance for the network that has higher average shortest path length than 3. PA underperforms the random predictor for the network has lower variance of degrees than 0.5. HRG performs consistently well. As a result of 2) , accuracy of proposed methods (both of bagging and 10-fold cross validation) are reached higher than the accuracy of HRG for 17 data sets and finishes the calculation faster than HRG. Proposed methods perform good accuracy for social network, citation network, dictionary network, biological network and transfer network (journey). Proposed methods underperform for trade network, circuit network, and food web network. Sometimes, proposed method (bagging) reaches higher accuracy than the accuracy of proposed method (10-fold cross validation). Proposed method (10-fold cross validation) finishes the calculation faster than proposed method (bagging). In conclusion, proposed methods finish the calculation faster than HRG and accuracy of proposed methods reaches higher than HRG.