Many Facial expression recognition methods have achieved great success, but they only considered front facial images or facial images close to the front. Besides, unlike in-the-laboratory datasets, the facial images in the real world (or in the wild) are without lighting and pose control, so that it is a big challenge to recognize these facial expressions. In this paper, the PIDViT (i.e., Pose-Invariant Distilled Vision Transformer) using the teacher-student model for the probability distributions of facial expressions of frontal and multi-pose faces was proposed and solved the pose variance and occlusion issues on expression recognition. First, the multi-pose face dataset FairFace-3D was built from the original FairFace and then used to train pose-invariance on the PIDViT. The PIDViT was trained in two stages; stage 1 is to train the PIDViT to achieve the consistency of facial expressions between frontal faces and multi-pose faces, and stage 2 is to use the student model pre-trained in stage 1 and train facial expressions further on target datasets. Finally, comprehensive experiments were conducted on three in the wild facial expression datasets, and the results validates the generalization of the PIDViT and its superiority over most state-of-the-art models.

The combination of computer vision and natural language processing is still a very challenging issue. In contrast to previous models focusing on generating only a single sentence for a video, we think that describing a longer video is an important application. In this paper, we propose a video scenario description system that considers video genres to generate multiple sentences. First, the semantics and genres of videos are analyzed. Next, video descriptions are also analyzed. Then, relevant semantic features are selected and translated into the corresponding video descriptions through deep learning. In the experiments, we compare the generated video descriptions based on four evaluation metrics. The results reveal our method is comparable with the state-of-the-art methods.

For image matching, the scale invariant feature transform (SIFT) algorithm is a commonly used one. They are invariant to image rotation, scale zooming, and partially invariant to change in illumination and 3D camera viewpoint. Affine SIFT (ASIFT) is an extension of SIFT, which solves the problem when images are captured at different angles. However, ASIFT has higher computational complexity than SIFT, due to a huge amount of features in the images. Therefore, in this study, a Hadoop‐based image retrieval system is proposed to solve the ASIFT shortcomings of high computation by the MapReduce technology. The system uses a combination of the Bag‐of‐Words method and support vector machine. Finally, the experimental results verify that the proposed method is more effective than the other state‐of‐the‐art methods for a variety of datasets.

In this paper, a movie title recognition system is proposed to extract the movie title from a movie poster and transform it into words. First, the title region in a poster image is located using gradient-conversion method, and then all character images in the region are segmented using the watershed approach. Next, two kinds of features about the texture and shape defined in the MPEG-7 specification are extracted to train an SVM classifier used in recognizing those character images into words. Finally, we employ PyEnchant on each word for spelling check or even spelling revision. The experimental results show that the correct rate of title recognition would reach 90%, if the title region in a poster image can be located and segmented successfully.

In this paper, to help users find the domain information from tweets they are interested in and further support the work of domain explorers and social content retrieval, a classification method on the social contents in Twitter is proposed, based on news categories. The classification method uses traditional "bag of words" features and explicit features extracted from tweets to facilitate the classification. Since data sparseness is always a serious problem when considering "bag of words" features in the text classification, we further employ dimensionality reduction methods on "bag of words" features and observe their performances. The experimental results show that 1) our proposed method can achieve good performances in the tweet classification and 2) using dimensionality reduction methods can achieve higher accuracy than not using them.

In this paper, we propose a protein secondary structure prediction method based on the support vector machine (SVM) with position-specific scoring matrix (PSSM) profiles and four physicochemical features, including conformation parameters, net charges, hydrophobic, and side chain mass. First, the SVM with the optimal window size and the optimal parameters of the kernel function is found. Then, we train the SVM using the PSSM profiles generated from PSI-BLAST and the physicochemical features extracted from the CB513 data set. Finally, we use the filter to refine the predicted results from the trained SVM. For all the performance measures of our method, Q3reaches 79.52, SOV94 reaches 86.10, and SOV99 reaches 74.60; all the measures are higher than those of the SVMpsi method and the SVMfreq method. This validates that considering these physicochemical features would exhibit better performances.

Online shopping is becoming increasingly important as more and more manufacturers sell products on the Internet, and many users are using the Internet to express and share their opinions. However, it is impossible for consumers to read all product reviews. Therefore, it is necessary to design effective systems to summarize the pros and cons of product characteristics, so that consumers can quickly find their favorable products. In this paper, we present a product ranking system using opinion mining techniques. Users can specify product features to get back the ranking results of all matched products. In this system, we consider three issues while calculating product scores: 1) product reviews, 2) product popularity, and 3) product release month. Finally, the experimental results show that the system is practical and the ranking results are interesting.

The main characteristic of collaborative filtering is to provide personalized recommendations to a customer based on the customer profile, without considering content information about domain items. In this paper, we investigated to use a relevance feedback mechanism in the collaborative recommendation system. First, we used the Self-organizing Map (SOM) method to avoid suffering from the scalability and sparsity problem in the collaborative filtering. In addition, we adopted the Statistical Attribute Distance (SAD) method which uses the similarity in statistics of customers' ratings to calculate customer correlations, instead of using the statistics of customers that rate for similar items. Then, the multi-tier granule mining algorithm was used to find association rules. Finally, with the relevance feedback mechanism and the association rules, the recommendations could be refined to provide customers more relevance information.

Studies on frequent pattern mining have often considered the existence of an item in a transaction but ignored its profit and purchase quantity. As a result, utility mining has been introduced to resolve this issue. However, the resulting patterns simply show the correlation among high utility items without quantity information of such items. In this study, a novel method is proposed to discover high utility fuzzy itemsets from quantitative databases which considers both profits and quantities of items. The quantities are fuzzified into linguistic regions, thus the quantitative concept is consistent with human cognition and easy to interpret. In addition, we provide theoretical and empirical analysis of the proposed method. The simulation results demonstrated that our method can efficiently and effectively mine high utility fuzzy itemsets from quantitative databases.

With the widespread use of the Extensible Markup Language (XML), more and more applications store and query XML documents in XML database systems. So how to efficiently process a query and find specified patterns from XML documents is a crucial issue. In the paper, we proposed an efficient query processing algorithm for XML documents. We build T-Bitmap for each element in an XML document to filter out useless nodes. Moreover, we also build tag-index and value-index to improve the search efficiency while processing the ancestor-descendant axis and value nodes. Finally, through the experiments, we compared our method with others, and found that our method has better performances than others.

In the Internet, proxy servers play the key roles between users and Web sites, which could reduce the response time of user requests and save network bandwidth. Basically, an efficient buffer manager should be built in a proxy server to cache frequently accessed documents in the buffer, thereby achieving better response time. In the paper, we developed an access sequence miner to mine popular surfing 2-sequences with their conditional probabilities from the proxy log, and stored them in the rule table. Then, according to buffer contents and the rule table, a prediction-based buffer manager also developed here makes appropriate actions such as document caching, document prefetching, and even cache/prefetch buffer size adjusting to achieve better buffer utilization. Through the simulation, we found that our approach has better performance than the others, in the quantitative measures such as hit ratios and byte hit ratios of accessed documents.

A disk scheduler called MS-EDF is proposed not only to meet the deadlines of all requests, but also to offer the optimal seek time. Through the experiments, the results show our disk scheduling algorithm outperforms the others for multimedia systems in all cases.

Sequential patterns discovery had emerged as an important problem in data mining. In this paper, we propose an effective GST algorithm for mining sequential patterns in a large transaction database. Different from the apriori-like algorithms, the GST algorithm can out of order find large k-sequences (k >= 3);i.e., we can find large k-sequences not directly through large (k-1)-sequences. This leads to that our algorithm has much better performance than the Apriori-like algorithms. Besides, we also propose the method to find new sequential patterns by scanning only new transactions since the database was increased. Through several comprehensive experiments, the GST algorithm gains a significant performance improvement over the Apriori-like algorithms. Also we found as long as the ratio of the items purchased in new transactions is always much better than scanning the entire database.

The goal of the paper is to mine generalized association rules using pruning techniques. Given a large transaction database and a hierarchical taxonomy tree of the items, we try to find the association rules between the items at different levels in the taxonomy tree under the assumption that original frequent itemsets and association rules have already been generated beforehand In the proposed algorithm GMAR, we use join methods and pruning techniques to generate new generalized association rules. Through several comprehensive experiments, we find that the GMAR algorithm is much better than BASIC and Cumulate algorithms.

The average lock range (ALR) is proposed as an evaluation factor for measuring the strengths and weaknesses of locking-based concurrency control methods, for both structural and nonstructural locking. The methodology provides a simple and general way to analyze the performance of any locking method, and requires no queueing model. Based on the concept of the ALR, two popular locking protocols, the 2PL protocol and the tree protocol, are analyzed and a simulation is done to validate the correctness of the ALR model.>

The 'straw man' analysis (see J.N. Gray et al., 1981) is based on the assumption that the accessible unit and locking unit are both a record occurrence (or a tuple); consequently the results cannot be applied when a locking unit is a large granule such as an area, a file, or an index (or a data) block. In this paper, the probabilities of a transaction's waits and deadlocks are derived when the accessible unit is a data object and the locking unit is a granule. The results can be used to explain a transaction's wait and deadlock situations for: (1) various sizes of a granule; and (2) different distributions of data objects accessed by a transaction.>