Tianyang Lv

Combining Multiple Clustering Methods Based on Core Group

As an unsupervised technique, clustering analysis has been widely applied in various fields. However, it is usually difficult to select an appropriate clustering method for an application, while no clustering method is suitable for all situations. This paper proposes a novel method to combine multiple clustering methods. First, the paper combines different agglomerative hierarchical methods in one clustering process to obtain core groups. Core group refers to the data that are always clustered together no matter what clustering method is applied. Then, it adopts other kind of clustering methods to refine the core groups and index database. In addition to conduct a series of experiments on the datasets from UCI, the paper applies the proposed method in a new research field, 3D model retrieval, to analyze and index the 3D model database.

Analysis on key nodes behavior for complex software network

It is important to understand software structural complexity and execution behavior in controlling the software development and maintenance process. Difference from previous work which based on structure network built on method association of software, we explore the topological characteristics of software execution behavior based on complex network and model the software execution network based on method invocation sequences. Taking typical open-source software under Linux for example, we build execution network based on the method call relationships, and then explore typical topology measurements of the key node and its adjacent network in software execution network. The result shows that the key nodes with high degree or high betweenness plays an important role in execution process of software system and the execution network can be divided into several levels, which has an important significance for maintenance and quality assurance for software.

Selective Feature Combination and Automatic Shape Categorization of 3D Models

It is the key problems in 3D model retrieval to obtain good feature and classify models efficiently. Although many feature extraction methods have been proposed, none is adapted to all models. Moreover, it still relies on manual work to classify models. To solve these problems, firstly, the paper proposes a series of selective combination methods which automatically decide each feature’s appropriate weight. The experiments conduct on PSB show that the combined feature performs much better than the best single feature. Secondly, the paper proposes the iterative clustering process to obtain the shape-based 3D models classification based on the combined feature. Experiment shows that the method can classify 91% models of Princeton Shape Benchmark.

Semantic 3D Model Retrieval Based on Semantic Tree and Shape Feature

3D model retrieval emerges as an important part of multimedia information retrieval. Current researches in 3D model retrieval concentrate on the shape-based way. However, its performance isn’t satisfying because of the semantic gap. The paper explores the semantic-based 3D model retrieval method based on semantic tree and the hybrid method based on content and semantic. First, the semantic tree is adopted to illustrate models’ semantic relationship and the accurate semantic of each tree node is described with several keywords. The semantic tree of all 1814 models of the Princeton Shape Benchmark is constructed. Second, the paper realizes the semantic retrieval based on the semantic tree which facilitates user’s feedback. Third, the hybrid retrieval process integrating semantic and content is proposed. The content similarity is computed by adopting the combined shape feature. The experiments conduct on Princeton Shape Benchmark shows the effectiveness of our method.

An auto-stopped hierarchical clustering algorithm for analyzing 3d model database

In the research of shape-based 3D model retrieval, the analysis and classification of 3D model database is an important topic for improving the retrieval performance. However, it encounters difficulties due to lack of valuable prior knowledge and the semantic gaps exist in 3D model retrieval. The paper proposes a new auto-stopped hierarchical clustering algorithm overcome these problems, which combines outlier detection with clustering. The Princeton Shape Benchmark along with 2 data sets from UCI is employed to evaluate the performance of the algorithm. And the new algorithm outperforms other auto-stopped algorithms and obtains better classification of 3D model database.

Clustering analysis of regional economy relationship based on the population flow social network

The economic development of a region is closely related with its population flow from/to other regions. And the regions, which have close relationship of the population flow, usually have a high economic relationship. Therefore, the analysis of the regional population flow can find the potential economic circles. Different from previous methods which are only rely on the analysis of experts to find economic circles, the paper proposes a method based on the weighted complex network clustering algorithm. The method analyzes the economic relationship based on the data of regional population flow. First, we construct the population flow social network according to the 1% national population sample survey in 2005; Second, we analyze the population flow social network using three clustering algorithms that are designed to analyze the weighted complex network; Third, we detect the regional economic circle and compare the result with prior knowledge.

Combined Grey Forecast Model with Particle Swarm Optimizer

Prediction and monitoring of the telephone traffic concerned by the telecom operators to predict accurately and reliability of the results directly affect the effectiveness and long-term development of enterprises. Based on the time series type of telephone traffic data for its small term, and characteristics of the trend increasing, for this reason this paper use grey model GM(1,1). The telephone traffic itself has less volatility and a strong seasonal characteristics, this paper presents a new prediction model, combined grey forecast model. The model combined the longitudinal and horizontal data of the projected trend of the data to achieve short-term forecasting and seasonal prediction of binding targets, and thus improve the prediction accuracy. At the same time by the introduction of particle swarm optimizer(PSO) to improve the grey forecast model to compare with the Season Exponent and Grey Forecast method. This paper picked a province of China from 2004 to 2009 the monthly telephone traffic data, through different angles using grey model GM(1,1) to predict and compare the predicted results, the experiment can see the improved grey prediction model have more accurately predict the future telephone traffic.

Studies on Shape Feature Combination and Efficient Categorization of 3D Models

In the field of 3D model retrieval, the combination of different kinds of shape feature is a promising way to improve retrieval performance. And the efficient categorization of 3D models is critical for organizing models. The paper proposes a combination method, which automatically decides the fixed weight of different shape features. Based on the combined shape feature, the paper applies the cluster analysis technique to efficiently categorize 3D models according to their shape. The standard 3D model database, Princeton Shape Benchmark, is adopted in experiment and our method shows good performance not only in improving retrieval performance but also in categorization.

A dynamic clustering algorithm based on artificial immune system for analyzing 3D models

In the field of content-based 3D model retrieval, classifying and organizing 3D model database is an important fundamental research, which is critical for improving the retrieval performance. Clustering is one of the most effective methods to classify 3D models. However, there has been little work on it. This paper proposes a dynamic clustering algorithm based on artificial immune system for classifying 3D models, which not only can classify existing models, but can deal with new incremental models. Experimental results show that our algorithm can obtain better classification of 3D models.

A Method of Legal Text Formalization

Knowledge representation and reasoning is a very critical issue on how to understand and storage knowledge. According to the clear logic structure and strict language structure in legal texts, this paper presents a method of legal text formalization using natural language understanding, based on which we modeled legal texts using CTL. After finishing these steps automatically, formal text could be a foundation of future knowledge discovering and formal reasoning. At last an experiment revealed that compared to human experts, this method is more effective and efficient on the aspect of formalization and reasoning.