smartmuseum Bayesian Network-user-centered approach for computational storytelling (Sparacino, 2003: 2008) Neighbourhood based methods-for example, Self-Organizing Map-a user activity detection (Laerhoven, et al., 2000)and location detection(Schmidt, et al. But usually all kind of methods are combined to get the best results. Like a project constructing an appropriate user profile uses three different machine learning algorithms the model building process: Bayesian network, clustering and rule-based models(Sugiyama, etal.,2004) Data Mining (Hand, et al., 2001) have defined data mining as follows: Data mining is the analysis of (often large)observational data sets to find unsuspected relationships and to summarize the data in novel ways that are both understandable and useful to the data owner. The relationships and summaries derived through a data mining exercise are often referred to as modeIs or patterns In data mining, association rule learners are used to discover elements that co-occur frequently within a data set consisting of multiple independent selections of elements(such as purchasing transactions), and to discover rules, such as implication or correlation, which relate co-occurring elements Agrawal, Imielinski and Swami(1993) have introduced the problem of mining association les between sets of items in a large database of customer transactions. Each transaction consists of items purchased by a customer in a visit. We are interested in finding those rules that have: Minimum transactional support s- the union of items in the consequent and antecedent of the rule is present in a minimum of s of transactions in the database Minimum confidence c - at least c of transactions in the database that satisfy the antecedent of the rule also satisfies the consequent of the rule Commonly the confidence threshold value is chosen. But the fixed confidence threshold has little basis in statistics, since some sets may exceed it simply by random coincidence(thereby defeating the goal of finding meaningful correlations), and some meaningful associations may occur in the data without reaching the threshold(Zaki, 1999). However, in practice it does eliminate vast numbers of insignificant sets An example of data mining, often called the market basket analysis, relates to its use in retail sales(Agrawal, Imielinski, Swami, 1993). If a clothing store records the purchases of customers, a data mining system could identify those customers who favour silk shirts over apprise H Heritage Malta Grant Agreement Number: FP7-216923 国如 Platform
Grant Agreement Number: FP7-216923 Acronym: SMARTMUSEUM Project title: Cultural Heritage Knowledge Exchange Platform Bayesian Network– user-centered approach for computational storytelling (Sparacino, 2003; 2008); Neighbourhood based methods–for example, Self-Organizing Map–a user activity detection (Laerhoven, et al., 2000) and location detection (Schmidt, et al.). But usually all kind of methods are combined to get the best results. Like a project constructing an appropriate user profile uses three different machine learning algorithms in the model building process: Bayesian network, clustering and rule-based models (Sugiyama, et al., 2004). Data Mining (Hand, et al., 2001) have defined data mining as follows: Data mining is the analysis of (often large) observational data sets to find unsuspected relationships and to summarize the data in novel ways that are both understandable and useful to the data owner. The relationships and summaries derived through a data mining exercise are often referred to as models or patterns. In data mining, association rule learners are used to discover elements that co-occur frequently within a data set consisting of multiple independent selections of elements (such as purchasing transactions), and to discover rules, such as implication or correlation, which relate co-occurring elements. Agrawal, Imielinski and Swami (1993) have introduced the problem of mining association rules between sets of items in a large database of customer transactions. Each transaction consists of items purchased by a customer in a visit. We are interested in finding those rules that have: Minimum transactional support s — the union of items in the consequent and antecedent of the rule is present in a minimum of s % of transactions in the database. Minimum confidence c — at least c % of transactions in the database that satisfy the antecedent of the rule also satisfies the consequent of the rule. Commonly the confidence threshold value is chosen. But the fixed confidence threshold has little basis in statistics, since some sets may exceed it simply by random coincidence (thereby defeating the goal of finding meaningful correlations), and some meaningful associations may occur in the data without reaching the threshold (Zaki, 1999). However, in practice it does eliminate vast numbers of insignificant sets. An example of data mining, often called the market basket analysis, relates to its use in retail sales (Agrawal, Imielinski, & Swami, 1993). If a clothing store records the purchases of customers, a data mining system could identify those customers who favour silk shirts over
smartmuseum cotton ones. Although some explanations of relationships may be difficult, taking advantage of it is easie The ability to track user ng behaviour down to individual mouse clicks has brought the vendor and end customer lose than ever before(Mobasher, Cooley, Srivastava, 2000). It is now possible for vendors to personalize their product messages for individual customers on a massive scale, a phenomenon referred to as " mass customization. The usage mining tasks can involve the discovery of association rules Local Mode! Aggregation Fant. Mode Data Miming To Transfomer o cinerator Mode Extractor DMm如)( Date Mimm Exactor EDs cor Algorith Agori Data Source Dam Soute Dit Source DIn Source Figure 1-a data warehouse architecture and distributed data mining framework (Park, Kargupta, 2003) An overview of mining systems for distributed applications like data mining primitives in ad hoc wireless networks of mobile devices like PDAs, cell phones and wearable computers (Park, et al., 2003). Zaki(1999) surveys the state of the art in parallel and distributed ociation-rule-mining algorithms and concludes that such systems are still in their infancy and a lot of exciting work remains to be done in system design, implementation, and deployment In the Smartmuseum project the association rules can be used to identify user preferences and to give him some advices which objects in the museum might be interesting. The generation of association rules is very fast and it can be applied to a large database. But it is noted that the selection of reasonable rules is rather difficult Conceptual Clustering Clustering is the classification of objects into groups by using sufficient number of variables The classified objects can be connected to predefined categories that characterize the object in the best way. As(Rosch, 1978) pointed out the purpose of the categorization system of organism is to provide maximum information about the environment with the least cognitive effort. Categorization must use features that maximize similarity between the samples in the cluster and at the same time separate different clusters as well as possible apprise H Heritage Malta Grant Agreement Number: FP7-216923 国如 Platform
Grant Agreement Number: FP7-216923 Acronym: SMARTMUSEUM Project title: Cultural Heritage Knowledge Exchange Platform cotton ones. Although some explanations of relationships may be difficult, taking advantage of it is easier. The ability to track user browsing behaviour down to individual mouse clicks has brought the vendor and end customer closer than ever before (Mobasher, Cooley, & Srivastava, 2000). It is now possible for vendors to personalize their product messages for individual customers on a massive scale, a phenomenon referred to as ―mass customization.‖ The usage mining tasks can involve the discovery of association rules. Figure 1 - A data warehouse architecture and distributed data mining framework (Park, Kargupta, 2003) An overview of mining systems for distributed applications like data mining primitives in ad hoc wireless networks of mobile devices like PDAs, cell phones and wearable computers (Park, et al., 2003). Zaki (1999) surveys the state of the art in parallel and distributed association-rule-mining algorithms and concludes that such systems are still in their infancy, and a lot of exciting work remains to be done in system design, implementation, and deployment. In the Smartmuseum project the association rules can be used to identify user preferences and to give him some advices which objects in the museum might be interesting. The generation of association rules is very fast and it can be applied to a large database. But it is noted that the selection of reasonable rules is rather difficult. Conceptual Clustering Clustering is the classification of objects into groups by using sufficient number of variables. The classified objects can be connected to predefined categories that characterize the object in the best way. As (Rosch, 1978) pointed out the purpose of the categorization system of an organism is to provide maximum information about the environment with the least cognitive effort. Categorization must use features that maximize similarity between the samples in the cluster and at the same time separate different clusters as well as possible
smartmuseum There are a variety of theories specifying the best way to categorize objects. They range from the classical theory(Laurence, et al., 1999) or the prototype theory(Rosch, 1978)to a dynamic approach as the situated simulation theory(Barsalou, 2003). The classical theor holds that the category membership is defined as a set of necessary and sufficient features. If an object does not exhibit the necessary and sufficient features, then it does not belong to the category. Whether an object belongs to a category in the prototype theory is decided by usin its similarity with a prototype(rosch, 1978) To measure similarity with the prototype, a set of features is needed. Some of the features have higher weight than others and the membership of the sample in a category is determined by measuring similarity between a sample and the category representation. The situated simulation theory states that conceptual representation and the category membership is highly contextualized and dynamical and is dependant on the previous experience and the current situation( Barsalou, 2003) Conceptual clustering is a machine learning task for unsupervised classification of objects (Michalski, 1980). A most known example of the conceptual clustering approach is a conceptual clustering system COBWEB(Fisher, 1987). COBWEB is an incremental system for hierarchical conceptual clustering. The system carries out a hill-climbing search through a space of hierarchical classification schemes using operators that enable bidirectional travel through this space(Fisher, 1987). The COBWEB algorithm is an incremental method and connected to the observation that most of human learning can be viewed as a gradual process of concept accusation and human ability for incorporating knowledge from new experiences Into existent concept structures There are two ways of clustering(Kim Chan, 2008) First, agglomerative(bottom-up) hierarchical clustering algorithms initially put every object in its own cluster and then repeatedly merge similar clusters together, resulting in a tree shape structure that contains clustering information on many different levels Second, divisive(top-down) hierarchical clustering algorithms are similar to agglomerative ones, except that initially all objects start in one cluster which is repeatedly split. The conceptual clustering belongs to the latter one There is a number of different approaches to the conceptual clustering ( Godoy, et al., 2006) have presented a document clustering algorithm, named WebDCC (Web Document Conceptual Clustering)that carries out incremental, unsupervised concept learning over Web documents in order to acquire user profiles.( Perkowitz, et al. have built adaptive Web sites: sites that automatically improve their organization and presentation by learning from visitor access patterns apprise H Heritage Malta Grant Agreement Number: FP7-216923 国如 Platform
Grant Agreement Number: FP7-216923 Acronym: SMARTMUSEUM Project title: Cultural Heritage Knowledge Exchange Platform There are a variety of theories specifying the best way to categorize objects. They range from the classical theory (Laurence, et al., 1999) or the prototype theory (Rosch, 1978) to a dynamic approach as the situated simulation theory (Barsalou, 2003). The classical theory holds that the category membership is defined as a set of necessary and sufficient features. If an object does not exhibit the necessary and sufficient features, then it does not belong to the category. Whether an object belongs to a category in the prototype theory is decided by using its similarity with a prototype (Rosch, 1978). To measure similarity with the prototype, a set of features is needed. Some of the features have higher weight than others and the membership of the sample in a category is determined by measuring similarity between a sample and the category representation. The situated simulation theory states that conceptual representation and the category membership is highly contextualized and dynamical and is dependant on the previous experience and the current situation (Barsalou, 2003). Conceptual clustering is a machine learning task for unsupervised classification of objects (Michalski, 1980). A most known example of the conceptual clustering approach is a conceptual clustering system COBWEB (Fisher, 1987). COBWEB is an incremental system for hierarchical conceptual clustering. The system carries out a hill-climbing search through a space of hierarchical classification schemes using operators that enable bidirectional travel through this space (Fisher, 1987). The COBWEB algorithm is an incremental method and is connected to the observation that most of human learning can be viewed as a gradual process of concept accusation and human ability for incorporating knowledge from new experiences into existent concept structures. There are two ways of clustering (Kim Chan, 2008). First, agglomerative (bottom-up) hierarchical clustering algorithms initially put every object in its own cluster and then repeatedly merge similar clusters together, resulting in a tree shape structure that contains clustering information on many different levels. Second, divisive (top-down) hierarchical clustering algorithms are similar to agglomerative ones, except that initially all objects start in one cluster which is repeatedly split. The conceptual clustering belongs to the latter one. There is a number of different approaches to the conceptual clustering. (Godoy, et al., 2006) have presented a document clustering algorithm, named WebDCC (Web DocumentConceptual Clustering) that carries out incremental, unsupervised concept learning over Web documents in order to acquire user profiles. (Perkowitz, et al.) have built adaptive Web sites: sites that automatically improve their organization and presentation by learning from visitor access patterns
smar museum (Kim, et al., 2008)are extracting a continuum of general (long-term) to specific(short-term) interests of a user. The proposed approach is to learn a user interest hierarchy (UIH) from a set of web pages visited by a user. They have developed a divisive hierarchical clustering (HC) algorithm to group words(topics)into a hierarchy where more general interests are represented by a larger set of words Their divisive algorithm does not necessarily generate binary splits and uses a minimum cluster size as one of the stopping criteria. However, instead of using category utility to determine if child clusters are generated, we use a graph-based method and a different similarity function. (Kim, et al., 2008) (Singh, et al. ) are attempting to model the user's interests for ludic news reading behaviour i.e., general reading of the news with basic themes of interest that may change slowly over time. It is adaptive approach in the sense that once the initial profile or interest hierarchy built, the leaf categories of the hierarchy are updated after each session with the explicit feedback of the user. This adaptive phase continues the learning and can also model the drift"" in user's interests over time Another approach for building visual ontologies is to hierarchically cluster a visual training data set. Such clustering methods have been used earlier for ontology discovery in textual and numerical data (Clerkin, et al., 2001) The main advantages of the method of conceptual clustering are that it uses unsupervised learning method and it is incremental and can adapt new items into a conceptual structure. But there is not known how such methods behav the case of large data sets and whether they are applicable if the number of variables is rather large end their mutual relations are complicated Naive Bayes a graphical approach to classification is called the naive Bayes model, in which conditional independence assumptions are used to simplify the model structure ( Bishop 2006). The structure of the naive bayes classifier is a Directed Acyclic Graph(dag) that represents the conditional probabilities with arrows between variables, and independence of variables if the arrow is missing(Korpipaa, et al., 2003) As the method is a very simple estimation of likelihood it is widely used in application of be interesting to a user a naive Bayesian classifier le. for identifying which Web sites would model for news story classification, named News Dude (billsus, et al., 1999). the naive Bayesian networks are applied to classify nine contexts of a mobile device user in their normal daily activities(Korinpaa et al., 2003) Naive Bayes framework is feasible for context recognition. In real world conditions, the recognition accuracy using leave-one-out cross validation was 87% of true positives and 95% apprise H Heritage Malta Grant Agreement Number: FP7-216923 国如 Platform
Grant Agreement Number: FP7-216923 Acronym: SMARTMUSEUM Project title: Cultural Heritage Knowledge Exchange Platform (Kim, et al., 2008) are extracting a continuum of general (long-term) to specific (short-term) interests of a user. The proposed approach is to learn a user interest hierarchy (UIH) from a set of web pages visited by a user. They have developed a divisive hierarchical clustering (DHC) algorithm to group words (topics) into a hierarchy where more general interests are represented by a larger set of words. Their divisive algorithm does not necessarily generate binary splits and uses a minimum cluster size as one of the stopping criteria. However, instead of using category utility to determine if child clusters are generated, we use a graph-based method and a different similarity function. (Kim, et al., 2008) (Singh, et al.) are attempting to model the user‘s interests for ludic news reading behaviour, i.e., general reading of the news with basic themes of interest that may change slowly over time. It is adaptive approach in the sense that once the initial profile or interest hierarchy is built, the leaf categories of the hierarchy are updated after each session with the explicit feedback of the user. This adaptive phase continues the learning and can also model the ―drift‖ in user‘s interests over time. Another approach for building visual ontologies is to hierarchically cluster a visual training data set. Such clustering methods have been used earlier for ontology discovery in textual and numerical data (Clerkin, et al., 2001) The main advantages of the method of conceptual clustering are that it uses unsupervised learning method and it is incremental and can adapt new items into a conceptual structure. But there is not known how such methods behave in the case of large data sets and whether they are applicable if the number of variables is rather large end their mutual relations are complicated. Naive Bayes A graphical approach to classification is called the naive Bayes model, in which conditional independence assumptions are used to simplify the model structure (Bishop, 2006). The structure of the naive Bayes classifier is a Directed Acyclic Graph (DAG) that represents the conditional probabilities with arrows between variables, and independence of variables if the arrow is missing (Korpipää, et al., 2003). As the method is a very simple estimation of likelihood it is widely used in application of personalization and context awareness. For example, for identifying which Web sites would be interesting to a user a naïve Bayesian classifier is used (Pazzani, et al., 1997) and a model for news story classification, named News Dude (Billsus, et al., 1999). the naive Bayesian networks are applied to classify nine contexts of a mobile device user in their normal daily activities (Korinpää et al., 2003). Naive Bayes framework is feasible for context recognition. In real world conditions, the recognition accuracy using leave-one-out cross validation was 87% of true positives and 95%